Just the video I wanted. Thank you very much sir! You've helped a lot in clearing my doubts. I went with 5900x setup instead of Threadripper Workstation purely bcoz of price factor. Also because as an Independent artist I could afford the performance loss as compared to spending a thousand bucks extra for threadripper workstation which I couldn't afford at this point anyway. But I surely get the point that at some point later I will need to shift to a workstation. I also get that apart from performance, I also don't have much hardware flexibility and upgrade path with a desktop but for now 5900x setup is just the optimum sweetspot for me.
Hi vaibhav tailang, I understand you situation as I myself have a 5900x for the same reasons. The system is great for most work. I edit all my videos on the 5900x and I really like the performance i get for the money. Good luck Thanks for watching
@@phasepanther4423 I ment that it is the only video about using dual gpus on consumer motherboards, which usually have 16x pcie lanes, in this case you can use 2gpus with x8 lanes for each of them. There is no such video about this topic. I didn't say anything about not supporting sli of crossfire
Hey Mike! Using PCIe Gen4 x8 is the same bandwidth as PCIe Gen3 x16. A 3090 at fullspeed will be 1% (aprox) bottlenecked by PCIe 3 x16, so a 3070 shouldnt be bottlenecked at all with PCIe 4 x8, However using a older system with only PCIe Gen3 (Intel 10th Gen or Ryzen 2000) and dual graphics card may result in more of a performance difference. Also I believe that during Rendering if the scene fits within the GPU VRAM the PCIe interface is only used to feed the GPU during loading, so PCIe speed could really hurt is when the scenes don't fit in the VRAM and the GPU need the PCIe interface needs to access system RAM as a cache. So maybe in huge scenes the performance difference may actually be huge aswell.
Something to keep in mind especially in RS case is how out of core works with slower PCIE bandwidth. PCIE speed isn't too much of an issue when everything is loaded at once in a single go and fits in VRAM. But as Out of core is pretty robust in RS you can work on much much larger scenes but the PCIE speed can make a large difference in the total bandwidth of data moving back and forth from System RAM / Scratch Disk (worst but not bad on M.2) vs a single load to the VRAM.
Hi Saul, I am not an expert on Out of Bang Vram usage, but if this video is any indication of how the GPU performance when they run out of memory, this I say more Vean is always better. At the end of this video I render a very large scene with RS and the RTX A6000 crushed the RTX3090 due to the limited Vram. th-cam.com/video/M2I69uK9WVU/w-d-xo.html Thanks for watching
I've 5950x and was thinking of switching to a workstation with next gen stuff, this vid though was exactly what I was looking for, getting a second GPU and I'll be staying on my current system for a long while.
You can actually go as low as 4x with pcie 4.0 without losing performance. I have rtx 3070 on pcie 3.0 8x and it works flawlessly with no performance loss.
Hi Mike! Thanks for the video, always has lots of useful information. Just one thing, at 4:03 you mention VRay benchmark scene but that is a Redshift benchmark scene. Is that correct? Cheers.
yups, my bad. but can not change it now. TH-cam will not allow any edits after you submit the video. I would have to delete the video and repost it. Thanks for pointing that out
Thanks for this, I recently have upgraded my cpu to a 5900x and I have been thinking of just plugging a second 3070 in to improve my render times and give a second one for rendering while I stream (do to some complicated set up for post editing I have to dual record my streams, one for the split audio and a second just to get raw game footage at a higher bit rate) as well as just maybe squeeze out two more years before needing to change to a new cpu socket. gaming is secondary to me and I only run at 1080 anyways. This was super helpful for that decision , thank you.
Thanks for this I recently have upgraded my cpu to a 5900x and I have been thinking of just plugging a second 3070 in to improve my render times and give a second guy for rendering while I stream (do to some complicated set up for post editing I have to dual record my streams one for the split audio and a second just to get raw game footage at a higher bit rate) gaming is secondary to me and I only run at 1080 anyways. This was super helpful for that decision , thank you.
I think most of the differences you are running into is caused by CPU differences between the two builds, since Light Cache and some Global Illumination are still using CPU even during GPU renderings.
Thank you for doing this. This helped me understand what the impact is, if I go down the desktop dual gpu scenario. I don't need 2 at the moment(And have never done that on any of my builds) but I would like the option in my next build.
Two GPU does work in a desktop if you can fit the cards into the system and MOBO. and for 3D rendering, the 8x PCIe speeds does not affect the performance that much. It only about 2-3% in final time to render due to the transfer of the scene to the GPU. But desktop is a viable solution for dual GPU set ups. thanks for watching
Very useful information! Thanks for the awesome benchmarks you do! It would be interesting to see how much simulation (Houdini Pyro or Vellum) speed is impacted using x16 vs x8 PCIE
Well even though this is 2 years old, still very helpful. My situation is using a NVlink bridge on two 3090TI's for a client that wants the VRAM of a A6000 without the price. Was able to find a nvlink bridge, but it difficult finding a motherboard with the 81mm spacing I need between the two 5.0 x 16 slots
@@hareram4233I ended up getting the TRX 50 Aero-D by Gigabyte, has a 3 slot spacing and found a 3 slot NvLink online. Still in the works, but looks like Linux can support NvLink even if windows does not....at least I hope so
GPU prices are way down now, and theused market is starting to flood. I really want a 3090 mainly fo the Vram, but the prices on used GPUs is hard to ignore. I already have a 2070 and used they are starting to pop up for $250-ish bucks. It may not increase my Vram and let me worry less about scene size, but if it can shave some significant time off my renders in Blender 3.xx I'd call it a win. Would be curious to see what you think.
Well I think the 3090 would shave off quite a bit of time vs. the 2070. if you can get the 3080 or 3090 at a good price. then you should get one. thanks for watching
Thanks for the sharing. I had the same experience. I used Supermicro 2024US-TRT with two A100-40GB cards for GPU-accelerated computing (simulation). I made a mistake by putting one of the card into a X8 slot. I can see from nvidia-smi that both cards can run perfectly to 99%, until oneday I noticed that one GPU in X16 and the other in X8. I had the question whether if affects anything. So I load task individually to the card in X16, and to the one in X8, separately. I didn't see any difference at all! Then I was wondering if it changes with different load, by which I meant the different usage of memories. To confim that, I test different load from occupying 10GB memories until 40 GB in full. I didn't see any difference. Later I did the same tests on two 3090 cards (turbo fan) on a 10980XE workstation (of Dell 5820X). Sorry I can't test A100 in workstation because of heat issue. I put one of the RTX3090 card only in a X16 slot and later a X8 slot. This time I could see about 20% of difference in terms of simulation speed. Note that on 5820X, only PCIE 3.0 is provided. The I was wondering if the difference is caused by the difference between Geforece and Tesla. But I can't put them into the AS-2024-US chasis because the RTX 3090 cards are too long (msi-aero). So I roughly had a conclusion that, PCIE3.0X16 or PCIE4.0X8 may be good enough for my applications, but it may vary, subject different user cases. Thanks alot for your sharing.
There shouldn't be any issue... I have an rtx 2080 and a 3080 ti both running 8x. I was actually shocked that it worked perfectly immediately, I literally didn't have to do anything, just plugged the card in, turned on my computer and it just worked. Shocking.
@@KazCanning I have two 3070ti back to back connected to my mobo. Do you think they're gonna overheat once I put them underload? If so, what should I do to cool them down better? I have a 680x corsair case. Thank you!
@@thienhofficial hard to say. I know my 3080tis get pretty hot when working on renders. I also have the founders edition cards so they have the pass through heat fins, basically blowing hot air on the top card.
Hi Eric, not that I am a Mac hater, but there is not much support for GPU rendering on a Apple system, so I do not have any desire creating a Mac video on 3D rendering performance. Thanks for watching
This was just the video I'd hoped you'd make! Super interesting, thank you Mike. It seems that an 570X board would do the trick with relatively minimal sacrifice. There's a pretty significant price difference in Australia between the 3060 Ti (~ $1200) and the 3070 (~ $1700), so for a dual GPU rig for Blender, two 3060 Tis would be cheaper and faster than a single 3080 ($2300), 3080 Ti ($2500) and even a 3090 ($3300). Yes, GPUs are expensive down under. I was thinking of pairing them with an MSI Tomahawk x570, a Ryzen 5950x and potentially an 850w PSU?
hi Guido, I was looking at the asrock taichi x570 when I built my system, but it was not available when I purchased my equipment. The Tomahawk is also a nice board. Good luck on the GPU purchase. Thanks for watching.
Hi Guido, Just a word of advice. If you have triple slot cards (air cooled) then it is recommended to have at least one empty slot between them. This way, the top card won't be choked of air. I personally recommend the MSI x570 Godlike. Another thing is for dual 3060ti a 1000 Watt gold or higher PSU is recommended from a reliable psu maker like Seasonic or EVGA (EVGA gpu's are bad but their PSU's are excellent). Generally, using dual GPU's in rasterization based engines like EEVEE or Unreal engine (by using multiple instances to speed up rendering / multi instance rendering) will use more power than a pathtraced engine like cycles or octane would use. Thus tripping the OCP on your PSU. Speaking from experience. Hope this helps.
@@nawafalrawachy394 thanks, that's really good to know! That motherboard costs $1200 here 😵 more expensive than TRX40 boards, could even get an WRX80 for that. So maybe I'll stick to a single GPU after all.
@@GuidoGautsch It could still work with a regular layout board if the cards are between 2.7-2.8 slots wide. Though the top card will struggle a bit to get air which will increase its temps. To combat this, I reduce the power limit on the top card to 95% and give it a custom fan curve to keep it from going above 80c.
That's the info! Thank you so much! I want to build this for rendering and modelling in Blender, PSU and case with upgradeability in mind. 5950x is out of budget so I hope the 5900x will be able to handle mobo and future 2 GPUs. My list Ryzen 5900x, X570 Taichi, 32Gb Vengeance LPX, RTX 3060, plus a 2nd GPU in the Future, Case Thermaltake Level 20 HT, Deepcool 360EX, 2NVMe WD 750SN, Seasonic Prime PX1300. Thank you for any input.
Hi dulichauch, if you reduce the PSU to 1000watts you could get some extra ram. I would suggest 64GB. Even with two RTX3080 at 350 watts that still leaves room for the 105watt CPU. All other parts are a good choice. The Taichi is a good choice for two GPUs.
@@MediamanStudioServices Oh, thats surprising! At beQuiet's PSU Calculator, adding the watercooler, 2 GPUS, 3 additional fans and 2xM.2 and one SSD I received 1000W usage, so I got the big PSU, still think of keeping it. For RAM: would it be totally absurd to order 2 separate LPX Kits of 16x2GB for 170€ or is it strongly recommmended to go with one kit only 2x32gb for 280€?
@@dulichauch it was just a suggestion to save money and get the much needed RAM required for creative workloads. More PSU is always better if you can get it. and I would get the 2x32gb and leave room for upgrade later.
I think you don't have a big hit in performance because on the x570 you have two x8 pcie 4.0 that is equal to 2 x16 pcie 3.0 the big hit is gonna be when you use two gpu that are pcie 3.0 or use a board that doesn't support pcie 4.0
hi Eriuz , I just assumed that most people looking to get two GPU would be using a current MOBO. But you are correct, the older MOBO's are using PCIe V3 and a lower bandwidth. thanks for watching
Thank you so much! It's hard for many of us investing our savings in hardware trying to get the best performance at a lowest cost. Have you ever considered power consumption? because having to many machines working 24/7 affects the cost of the electricity bill. So, is it cheaper having many GPUs in one PC vs Many PCs with single GPUs?.
Sir, you are an ABSOLUTE LEGEND! I appreciate this clarification. I am building my first-ever custom-loop PC and was very curious about the two important points you've made. Moving from Apple after 20 years is a little daunting but willing to try out the new and more powerful machines. My new PC will have the following specs: • Intel 14900k (water-cooled by the EK-Direct Die, metal thermal paste) • Z790 Master X Motherboard • x2 Navidia 4090 Cards (water-cooled, metal thermal paste) • 4 TB, M2, SSDs • 92GB Ram at 8600
Interesting information, so the bandwidth of the pci4.0 is at this moment hardly used if a 16* slot and a 8* slot can stay withing 10% of eachother. And now we already have PCIe5.0, while PCIe4.0 is hardly fully used, maybe storage attached to the PCIe bridge, makes better use of the available bandwidth.
Guys could someone help? I have a Gigabyte Z690 Aorus Elite DDR4 CPU, and have a rtx 3080 in there. Is it worth it if I put another one in there (I already have another 3080, and also a 1070 ti)? I use the PC for gaming / video editing.
It sounds like a lot of that performance reduction comes from switching time loss. I was going to build a gaming system and just add a second GPU later, but I ended up cowering out and just throwing a pair of 2080tis into a T7820. $1k and I got a rig that happily does both Blender with RT and AI production work.
can u make a new test but with pci 5.0? Im planning to build a new rig with 7950x and x670 mainboard. I read the manual that if i use 2 pcie 5.0 slot, It will share each x8, but im not sure if there s any reduction considering it s faster
I'm using an RTX 3070 with a Ryzen 2700X on B450 motherboard. It's running at x8 because I have a second GPU. I've measured about 2-3% performance drop on x8. I also have it overclocked +200MHz on the core and +1300MHz on the memory with power limit set to 100% and I gain 14% performance boost on my Blender renders.
Thank you for this video. It helped explain alot. I have a dual gpu setup with two 4070 supers. My current board has 1 pcie x 16 5.0 and 1 pcie x 16 4.0. I am in the process of getting a new motherboard with 2 pcie x16 5.0 slots. Would this improve performance?
No. 4070 SUPER runs @PCIe 4.0, your motherboard will downgrade all lanes to accommodate your GPU, so PCIe 5.0 would be disabled/ downgraded to PCIe 4.0.
hi Dewayne Dailey, for video editing, Unless you use Resolve Studio, the paid version. most video editing software does not use the GPU for much acceleration. And rendering in the viewport is a single GPU task. Just final 3D rendering can use the multi GPU set ups
Hey Mike, I'm looking to build a PC for workflow mainly in Premiere Pro and Filmora, is it more wise to go for an RTX 3090 or 2 RTX 3070's which would benefit me the most? I don't care for gaming this is strictly for my workflow. Also, what boards and which ram would you recommend? Obviously faster ram makes a difference? What about hard drive? I would prefer ones with at least 3K write speeds. And which water cooled system should I use? I've also narrowed it down to either the Ryzen 5900X or 5950X which I have access to. I will await your reply and thx a million for your hard work.
Premiere Pro isn't able to take advantage of GPUs efficiently and is more CPU-limited us my understanding, so even a single 3070 would be fine. Invest in the 5950x out of those two options. If you were to drop PrePro for Resolve, however, two 3070s would beat out the single 3090 by ~30%
@@ChrisDallasDualped yeah, I hear you. I prefer PP over Resolve as well for editing - just wish it'd be able to take advantage of GPU grunt as effectively as Resolve does.
Hi Mike, Could you please say what motherboard do you use for the Desktop machine in this video, and what cooling system? As I see you use blower versions of GPUs in this rig, right? Also, what do you think about GIGABYTE X570S AORUS MASTER motherboard for using with two fan style GPU and Ryzen 5950x? Looks like it has enough room for airflow between two GPU and it has decent VRM for this processor? Tring to figure out best config for my situation. Thank you for your channel, there is a lot of useful and unique information here!
Hi Vlr. I am using the Gigabyte X570 pro wifi. www.gigabyte.com/Motherboard/X570-AORUS-PRO-WIFI-rev-10#kf for the cooling system. I did my own custom loop. check out the video i did here. th-cam.com/video/r_Rl30LhE2M/w-d-xo.html th-cam.com/video/dwfn5IQXnIE/w-d-xo.html The GIGABYTE X570S AORUS MASTER is a good choice. Thanks for watching
Thanks!, Great information and tests!, I have one doubt, Is it possible to render in V-ray with RT cores and 2 RTX GPUs but from different generation (E.g. RTX 2090 + RTX 3090) or different models (E.g. RTX 3090 + RTX 3080)? obviously without nvlink or sli. Well, thanks in advance!
@@TILK3D I haven't solve my specific doubt regarding 2 RTX GPUs but from different generation (E.g. RTX 2090 + RTX 3090) , But I have found this: 1.-You can render with several Nvidia GPU's using Vray and Vray will manage to use the available CUDA cores in the GPUs (This should work with CUDA cores of the same GPU generation even if they are different model cards (E.g. RTX 3090 + RTX 3080), but I still I don't know if different CUDA cores generations can work (E.g. RTX 2090 + RTX 3090) I have a RTX 3090 and a 980 Ti, I need to make some experiments. 2.-It is no necessary to use SLI or NVLink in order to use all the available CUDA cores of those same generation Cards, 3.-The NVLink will help sharing GPU's RAM but the improvements in rendering time is small (+/- 10%), 4.-I'm deducting (I still don't have evidence) that you can render with Vray and RT cores of several RTX GPUs of the same generatios just like the cuda cores (E.g. RTX 3090 + RTX 3080) also without an NVLink, Vray will manage to use the available RT cores. 5.-I don't know if you can render with VRAY using different RT cores generations GPU's (E.g. RTX 2090 + RTX 3090) Hope this is useful to you, let me check the sources where I found this information and I'll share it here. Please let me know if you have found something else. Best regards.
@@antoniosomera THANK YOU FOR YOUR INFORMATION I WILL BE LOOKING FOR MORE INFORMATION AND IF I SOLVE MY DOUBT THE SAME I WILL SHARE IT GREETINGS, SUCCESS
You can use any combination of supported GPUs. The CUDAs will stack, the VRAM will not. So if you use a 2070 with a 3090, and the scene exceeds the 2070's VRAM, performance will suffer. But aside from this, you can mix and match however you want, and with different generations. This should be the case with most CUDA based renderers.
great channel. for the video, I was expecting a benchmark comparison between one desktop gpu vs 2 gpu. is it worth to add a second gpu for rendering or is it worst or is it the same ? maybe you have another video that deal with this question, otherwise would be nice to do one. keep up the good work
I'm about to run 2x RTX 3090 FE in NVlink configuration for a total of 48GB Vram pool. il let you know how it goes, would be cool to see this on your channel before I rig it up for Blender and Iclone.
I currently have a x399 TR 2950x platform, all pcies are the same here, I used to run 4 graphics cards on them for rendering tests, I'm going to switch to Amd 9950x and I have a problem choosing a board, whether to the B650 Aorus v2 board, it's similar in Aorus, x670, x870, can I connect 2 graphics cards? I have concerns about the second Pcie connector, because the specification says that one is connected to the CPU and the rest to the chipset, proart boards have it described that they have two pcie connectors connected to the CPU
Most don't slip the lanes from the cpu What they have is a chip that splits it. If you look in some mother board manuals it says which have it. But the cpu it self cannot split any of its lanes. I only found this out recently
I run 2 GPU, on my gaming rig. Non SLI, etc. bring the 16x to 8x and I barely notice any performance loss as in FPS. 6950xt for gaming and etc 3050 for stream. But 5950x and Dark Hero VIII. Core count and good MOBO help a lot, I am sure. But I game at 2k ultra and stream 2k ultra
i have a question for you. did you have to make any bios changes or were you able to just plug in and play(stream) with no issues? i have noticed that when i plugged in my second GPU (1080) i noticed some lag on the monitors connected to it. i did a clean install of drivers for my 4090 and the 1080 and still noticed some stutters on the monitors connected to the 1080. my monitors arent the same hz because my gaming monitor is a 170hz when the other 4 are just 60hz. i can get the stutters to stop when i make my gaming monitor the same hz but whose trying to game at 60hz when you have a 4090.. lol. any tips?
@@itschops I did not, but I did not plug any monitors in to the stream gpu. The stream gpu is just for work. Give that a try, if not. You make have to fresh install windows. Because when I upgraded to a 4080s, to replace my 6950xt. It would only read the 3050 out of the box. Yet the 4080s would output displays lol. I uninstalled all drivers even and, and same thing. I uninstalled all drivers again and reseated GPUS and amame thing. Fresh install helped, but couldn’t fully test it as the 4080s is huge and my 6950xt was liquid cooled via AlphaCool. So I am just using the 4080s and had to do a fresh install of windows. Works flawless
Thanks for letting me know. the main reason for wanting to have the 1080 added to my PC was to have the extra display connections since i have 5 monitors but the stutters were happening too often so i removed it for now. I didnt try a fresh install of windows but are you running 2 4080s now? do you think its running smooth because they are the same series card? @@si1entshotzdtom
I have the same set up, except I don't have a 3050. Is the 3050 even PCIe gen 4? Because if it's not, I think the 6950xt will have to downgrade to gen 3 too. I'm still building my PC, was thinking it would be cool to get a GTX for the ray tracing (and DLSS/FRS combo). But I don't really know if it's nessecary. Why a second card for streaming?
Hi Mike, another great video from you. Do you think on the instances when you rendering heavy rendering, is it likely that there will be significant difference between desktop processor vs workstation processor, just like the test you did for RTX A6000 vs RTX 3090, where the difference was negatable until you tested with memory intense rendering?
This is extreamly valuable info! Thank you for your content! I have a following setup: Ryzen 5950X, Asus Prime Pro570X, 64Gigs of Ram (DDR4 Kingston 3200mhz), Asus Tuf Rtx3060 (12Gb), 750watt PSU. Do you know is it possible to add a second gpu (I have a leftover Asus Rtx3060ti), how it would effect my workflow and rendertimes in following programs: Blender, After Effects, Premiere Pro and also Unreal Engine 5? I was also thinking to get a bigger PSU (1000 watt). Thank you sir for your time and tips! This channel is pure gold!
hi George, adding the second GPU will limit the Vram in the 3960 (12gb) to the same amount of Vram in the 3060ti (8GB) as 3D render software sends two packages, one for each GPU and they need to be the same package size. As for After Effects, Premiere Pro, you will not get much benefit as both of these programs have very little multi GPU support. Same with UE 5. thanks for watching
Hi Lucas, this is a great idea for a video. Let me look into the specs required. But I am thinking it will mostly be the same as 3D and VFX computers. But I will do some research to confirm this. Thanks for watching.
Can you please tell the system requirements for a simple rtx 3060 graphics card, as i have one of the desktops which is quite old enough lets say 5th gen intel i5 with motherboard such old to, would this graphics card fit in it, or would i need to change my motherboard and processor as well to an 10-11th gen?
Hey Mike, thank you for your videos, they help me a lot concerning my future build. Indeed, as a 3D Artist, I am looking to upgrade my build that has lasted me for my studies since 2016. As I'll work with a workstation at work, I am looking to build a personal desktop PC for 3D rending (gpu based Octane Render/C4D & Blender/Cycles), video editing (adode suite) and gaming (most likely 1440p, 34' ultrawild). My current build would be : - 12th gen i7 (liquid cooled AIO) - 64gb RAM DDR4 (DDR5 looks a little too expensive) - 2TB m.2 nvme + 500gb ssd sata - RTX 3080 10gb (+ GTX 1060 6gb from current build) - 1200W PSU Which motherboard would you recommand that can handel 2 GPUs with at least 2x 16 PCIE 4.0 or 5.0 ? Doesn't look like a lot of choise yet for z690 chipset that doesn't break the bank... maybe I should wait a bit ? And do you think 1200w psu is enough for 2 gpus and the overall build ? Thanks a lot Mike, just subscribed, your content is priceless !
hi Colin, no desktop MOBO will support 2 GPUs in 16x. You would need to move up to a HEDT or Threadripper MOBO. Now you can still get good performance from a X570 running the GPUs at 8X. And I would recommend the Asrock X570 Taichi for running two GPU. This board have great VRMs to manage the power and the spacing to house the two GPU. Good luck with the build
Hi! Thanks for your comparison. This is exactly what I am looking for ! I have a question though. Do you think the CPU difference actually contributes to the time difference?
Hey mate! I am bit confused whether to buy gaming PC or workstation PC, for CAD works. So I searched for custom build pcs website and found Z690 chipset supports both workstation(Nvidia rtx a2000) & gaming(Nvidia rtx 3080ti) graphics. But my question is can I put both these graphics on chipset at same time & while working on CAD software it uses workstation graphics (for compatibility as per hardware certification) & while gaming it uses gaming graphics.
Looking to build a machine to help my realtime cycle rendering in the viewport. I've seen the 3090 nearly rendering certain scenes in realtime I want this level of performance but getting a 3090 now a days is hard. would 2 3080s out perform 1 3090? in rendering, video editing etc?
hello, good video, sorry, an sli bridge is needed to join the two graphics cards for 3d rendering or simply by connecting the two cards to the board, it can be used, I have a 3090 ti and a 3080 zotac
Does the gen of the PCIE slot factor in this as well? Will two gpus that are meant to work on PCIE 4.0X16 slots still get a performance hit on each of them even if they're on PCIE 5.0x8 lanes? So no matter what, the overall performance of the lane is the highest MUTUAL rating between the slot and the gpu that is on it? So a gpu that is meant to run on a 4.0x16 slot will be reduced to a 4.0x8 performance when mounted to a 5.0x8 slot?
if your graphic card is PCIE gen 4 , then putting it in a Gen 5 PCIE slot will not increase your performance cuz your card only know how to talk in "GEN 4" language, now what if you put two PCIE gen 4 graphic cards into 2 PCIE gen 5 slots with only 20 or 24 PCIE lanes, you will have two PCIE gen 4 x8 . But for rendering or AI that speed reduction wont impact your performance that much. And for gaming, you will never want to use 2 graphics cards cuz almost no game has coded to work with multi GPUs. for gaming maybe you can assign Physics to process on one card and everything else on your main graphic card , but as i said , don't do it for gaming too much heat and electricity for a little or no improvement (sometime the heat will even reduce your gaming FPS)
Great content as always! I don't know if this is worth making a video - but I would love to see a video about NVlinked GPUs - performances,installation,configuring the settings I was creating a scene in Blender the other day, I ran out of Vram on my 3090 lol...I might have used the wrong settings, didn't optimized the scenes better or used high res textures that I shouldn't be. My future plan is to build a workstation with multiple gpus (currently I have my eyes on the H3 platform, but I might need more vram + gpu power to render) I'm using 3970x with a 3090 Thanks!
Hi theWanderer521, I would live to do a Video in two RTX3090, trouble is that I do not have two 3090 anymore. But if I can get my hands on two gpus I will do a video on the set up of the NVlink. Thanks for watching
@@MediamanStudioServices Waiting to see this kind of video. Please make one. I have planned to build pc with i9 12900k with 2 x RTX 3090. l want to know what would be the performance with NVlink.
Thanks, was going insane not knowing why my 3080 was running on PCIE 3.0 x8 instead of x16. 2 M.2 SSDs and intel i9-10900k chip. Cant get it to PCIE 4.0 since the chip doesn’t support it
Awesome video, however i have a question. Do Sata drives use any pcie lanes, or are they using the dedicated 4x that is needed for the chipset? I have an asus b450-f gaming motherboard that i currently use with a gtx 1070. I have ordered an rtx 3080Ti and i want to use them simultaneously in blender, but i'm not sure whether i should upgrade my motherboard. The motherboard manual says that in a dual gpu configuration GPU1 would get 8x, GPU2 would get 4x, the NVME SSD would get 4x. I'm also using 4 Sata HDD-s for storage, but if they use any additional PCIE lanes other than the 4 required for the chipset i will loose a lot of performance on the GPUs. Does it even worth using a secondary 1070 with a 3080Ti? It's still cuda cores though... Thank you!
Sata uses your chipset and not pcie lanes. As for using the 1070. There is some advantages but with only 4 lanes. It may be better to just use the 3080ti at full speed and the 16 lanea
Hey I want to know !! is that possible i can se 1660 super and 3060 ti at the same pc !!! i want to use it for blender render like can i render it with one graphic card work with the other one ??
Blender will look for the card with the least memory so it doesn't crash. My understanding is if you have a GPU with 8GB, and one with and 12GB, it will treat both as 8GB GPUs when assigning rendering. The Blender forums will be able to give you a definite answer.
@@swdw973 It appears that when you have a GPU with more VRAM, and the less powerful GPU runs out of memory while your system has ample RAM, Blender can utilize the system's RAM to support the less powerful GPU. Consequently, the more powerful GPU can continue to operate at full capacity using your system's RAM. For instance, if you possess both a 3090 and a 3060 Ti GPU, along with a system equipped with 64 GB of RAM, Blender might use the 64 GB of system RAM to match the performance of the 3090. The efficiency of this process is enhanced by faster RAM. Additionally, optimal performance would benefit from a well-configured RAID system with satisfactory HDD speeds. This setup should allow Blender to function smoothly without constraining the performance of the more powerful GPU, correct?
@@viniarteeSystem RAM will never get anywhere near the speed of vRAM. My answer was based on info from a Blender developer on one of the Blender forums. I actually run 2 different GPUs, a 3060 and 4070, but I made sure they both are the 12GB versions, because of what the developer told me.
I don't know if you can answer this but I'm just rendering with a high-end gaming PC using stable diffusion and I just want to know if x 8 pcie Gen 4 would reduce the quality because I want to put more nvme drives into the machine and I don't want to do it if it's going to compromise my Ai rendering.
Hey.. thanx for the video. I have a question . Is there a difference in video quality when you export with hardware acceleration enable (nvidia, amd, Qsync) compared to just using cpu for vedio decoding and encoding..?
Hi Dasun Gunathilaka, there is always a difference when using a different encoder, be it hardware or software, but the difference should be very small and the quality should be about the same. thanks for watching
@Claus Bohm I bought an NVlink from Nvidia 3 Slot for A5000/6000 that is compatible with the 3090Ti but I there is no way to make it work.. I tried everything. fresh install windows 10, drivers update, bios update, anything.. Not sure what's the problem..
@@georgeluna6217 Maybe its dependant on the motherboard.(ones with sli support are sure to be compatible for this kind of setup) So check that out the motherboard specs from the company site. And that nvlink for A5000/A6000 should work with rtx 3090/ rtx 3090ti.
Hey. I have a problem connecting two 3090s through the Nvlink bridge. I use blender 3.0 stable to render my works b550m pro4 motherboard Processor x5900 I can not find information on the Internet on how to correctly connect the bridge and speed up the blender. Please help
your mother board needs to support Nvlink. But have you tried to run a render with the two separate GPU? Connecting with Nvlink is not going to necessary speed up your rendering.
@@MediamanStudioServices hey I saw that the 3080 or less only uses about 8x. But the 3090 uses more and would lose a lot of performance. Have you tested this out yet? Or do you know it's true?
@@djuansayjones6178 I am not sure of the PCIe bandwidth for each card, but I am making a good guess here, the RTX3080 does use more than the bandwidth of just 8x when at full performance, so yes untrue.
Hey Mike. Is there a way to stack 6 GPU RTX 3080/3090 blower on a single workstation motherboard ? I've been looking as well towards mining motherboard (that can fit 6-12 GPU) - but they reduce the pcie lanes speed x1 - Your video here showed a very little difference between x16 and x8 - So i'm curious if x1 is pushing it way too far for a render gpu farm? need your advice.
My question to you is why would you want to do this. Just buy a GPU server for this task. Gigabyte makes a great server for just this style of workflow. Anything over 3-4 GPU in one workstation requires a server chassis.
Just the video I wanted. Thank you very much sir! You've helped a lot in clearing my doubts. I went with 5900x setup instead of Threadripper Workstation purely bcoz of price factor. Also because as an Independent artist I could afford the performance loss as compared to spending a thousand bucks extra for threadripper workstation which I couldn't afford at this point anyway. But I surely get the point that at some point later I will need to shift to a workstation. I also get that apart from performance, I also don't have much hardware flexibility and upgrade path with a desktop but for now 5900x setup is just the optimum sweetspot for me.
Hi vaibhav tailang, I understand you situation as I myself have a 5900x for the same reasons. The system is great for most work. I edit all my videos on the 5900x and I really like the performance i get for the money.
Good luck
Thanks for watching
Great video. It is the ONLY video about this topic on TH-cam. Thank you guys.
But why though?
Just because sli nv link and crossfire aren't supported anymore there's no reason to ignore this at all.
@@phasepanther4423 I ment that it is the only video about using dual gpus on consumer motherboards, which usually have 16x pcie lanes, in this case you can use 2gpus with x8 lanes for each of them. There is no such video about this topic. I didn't say anything about not supporting sli of crossfire
@@phasepanther4423all workstation boards still support both. S.l.o and nv-link. Even trx50 & wrx90 stil havel support.
@@tojiyev_cg many am4 boards only had 16x/4x hard wired.
Amazing info you've given us and have subbed to your channel today, you have the best rendering info on the net that I've seen, thank you for this.
thanks for watching Chris.
THANK YOU FOR YOU DIRECT TO THE POINT VIDEOS!!!!!
Your welcome
You guys are lifesavers. Literally the only video on this topic on TH-cam. Thank you so much!
thanks for the kind comment
I greatly appreciate the way you produced this video. To the point. No click bait title. Straight forward. Thank you!
Thank you for the video. I'm here for comparing the two methods for Video Transcoding performance.
Hey Mike! Using PCIe Gen4 x8 is the same bandwidth as PCIe Gen3 x16. A 3090 at fullspeed will be 1% (aprox) bottlenecked by PCIe 3 x16, so a 3070 shouldnt be bottlenecked at all with PCIe 4 x8, However using a older system with only PCIe Gen3 (Intel 10th Gen or Ryzen 2000) and dual graphics card may result in more of a performance difference.
Also I believe that during Rendering if the scene fits within the GPU VRAM the PCIe interface is only used to feed the GPU during loading, so PCIe speed could really hurt is when the scenes don't fit in the VRAM and the GPU need the PCIe interface needs to access system RAM as a cache. So maybe in huge scenes the performance difference may actually be huge aswell.
Hi Guillermo Cosio, I totally agree, thanks for sharing this with the channel.
Something to keep in mind especially in RS case is how out of core works with slower PCIE bandwidth. PCIE speed isn't too much of an issue when everything is loaded at once in a single go and fits in VRAM. But as Out of core is pretty robust in RS you can work on much much larger scenes but the PCIE speed can make a large difference in the total bandwidth of data moving back and forth from System RAM / Scratch Disk (worst but not bad on M.2) vs a single load to the VRAM.
Hi Saul, I am not an expert on Out of Bang Vram usage, but if this video is any indication of how the GPU performance when they run out of memory, this I say more Vean is always better. At the end of this video I render a very large scene with RS and the RTX A6000 crushed the RTX3090 due to the limited Vram.
th-cam.com/video/M2I69uK9WVU/w-d-xo.html
Thanks for watching
Amazing content, thank U, Mike !!!
thanks Andrey. Please keep watching the channel.
@@MediamanStudioServices sure!!!
I've 5950x and was thinking of switching to a workstation with next gen stuff, this vid though was exactly what I was looking for, getting a second GPU and I'll be staying on my current system for a long while.
You can actually go as low as 4x with pcie 4.0 without losing performance. I have rtx 3070 on pcie 3.0 8x and it works flawlessly with no performance loss.
Sir you are really making youtube usefull
thanks Rocky Vicky
Hi Mike! Thanks for the video, always has lots of useful information. Just one thing, at 4:03 you mention VRay benchmark scene but that is a Redshift benchmark scene. Is that correct? Cheers.
yups, my bad. but can not change it now. TH-cam will not allow any edits after you submit the video. I would have to delete the video and repost it.
Thanks for pointing that out
@@MediamanStudioServices No problem. Good content.
Really informative!
thanks GET
Thanks for this, I recently have upgraded my cpu to a 5900x and I have been thinking of just plugging a second 3070 in to improve my render times and give a second one for rendering while I stream (do to some complicated set up for post editing I have to dual record my streams, one for the split audio and a second just to get raw game footage at a higher bit rate) as well as just maybe squeeze out two more years before needing to change to a new cpu socket. gaming is secondary to me and I only run at 1080 anyways. This was super helpful for that decision , thank you.
Thanks for investigating, but boy was the Laughing Man name drop the GITS nostalgia hit I needed today!
id love to see a video on how to setup the 2 gpus and how to run them together
Very helpful as always!
Thanks for this I recently have upgraded my cpu to a 5900x and I have been thinking of just plugging a second 3070 in to improve my render times and give a second guy for rendering while I stream (do to some complicated set up for post editing I have to dual record my streams one for the split audio and a second just to get raw game footage at a higher bit rate) gaming is secondary to me and I only run at 1080 anyways. This was super helpful for that decision , thank you.
Thank you for making this video its kinda really detailed Theorie that always comes in my mind
I think most of the differences you are running into is caused by CPU differences between the two builds, since Light Cache and some Global Illumination are still using CPU even during GPU renderings.
Very useful test. I guess with older gen GPUs like RTX 2070 or 1660, the performance drop will be negligible.
the performance difference should be the same across the generation of GPUs.
thanks for watching
Thank you for doing this. This helped me understand what the impact is, if I go down the desktop dual gpu scenario. I don't need 2 at the moment(And have never done that on any of my builds) but I would like the option in my next build.
Two GPU does work in a desktop if you can fit the cards into the system and MOBO. and for 3D rendering, the 8x PCIe speeds does not affect the performance that much. It only about 2-3% in final time to render due to the transfer of the scene to the GPU. But desktop is a viable solution for dual GPU set ups. thanks for watching
@@MediamanStudioServices Thank you Mike, that is great to know.
Really wanted to know how two different GPUs would work for production workloads. especially rendering in xpu.
thank you I needed this
your most welcome
Very useful information!
Thanks for the awesome benchmarks you do!
It would be interesting to see how much simulation (Houdini Pyro or Vellum) speed is impacted using x16 vs x8 PCIE
in what way, a storage drive using the PCIe or the GPU?
@@MediamanStudioServices
Well both.....
I second this
Well even though this is 2 years old, still very helpful. My situation is using a NVlink bridge on two 3090TI's for a client that wants the VRAM of a A6000 without the price. Was able to find a nvlink bridge, but it difficult finding a motherboard with the 81mm spacing I need between the two 5.0 x 16 slots
Did u find one? I also need suggestions for this motherboard
@@hareram4233I ended up getting the TRX 50 Aero-D by Gigabyte, has a 3 slot spacing and found a 3 slot NvLink online. Still in the works, but looks like Linux can support NvLink even if windows does not....at least I hope so
GPU prices are way down now, and theused market is starting to flood. I really want a 3090 mainly fo the Vram, but the prices on used GPUs is hard to ignore. I already have a 2070 and used they are starting to pop up for $250-ish bucks. It may not increase my Vram and let me worry less about scene size, but if it can shave some significant time off my renders in Blender 3.xx I'd call it a win. Would be curious to see what you think.
Well I think the 3090 would shave off quite a bit of time vs. the 2070. if you can get the 3080 or 3090 at a good price. then you should get one.
thanks for watching
Thanks for the sharing. I had the same experience. I used Supermicro 2024US-TRT with two A100-40GB cards for GPU-accelerated computing (simulation). I made a mistake by putting one of the card into a X8 slot. I can see from nvidia-smi that both cards can run perfectly to 99%, until oneday I noticed that one GPU in X16 and the other in X8. I had the question whether if affects anything. So I load task individually to the card in X16, and to the one in X8, separately. I didn't see any difference at all! Then I was wondering if it changes with different load, by which I meant the different usage of memories. To confim that, I test different load from occupying 10GB memories until 40 GB in full. I didn't see any difference. Later I did the same tests on two 3090 cards (turbo fan) on a 10980XE workstation (of Dell 5820X). Sorry I can't test A100 in workstation because of heat issue. I put one of the RTX3090 card only in a X16 slot and later a X8 slot. This time I could see about 20% of difference in terms of simulation speed. Note that on 5820X, only PCIE 3.0 is provided. The I was wondering if the difference is caused by the difference between Geforece and Tesla. But I can't put them into the AS-2024-US chasis because the RTX 3090 cards are too long (msi-aero). So I roughly had a conclusion that, PCIE3.0X16 or PCIE4.0X8 may be good enough for my applications, but it may vary, subject different user cases. Thanks alot for your sharing.
@3:26 dear the screenshot score is differnt from what ur audio said.. thanks for the great work.
I finally got a workstation
ya, i rushed this video and there is a few mistakes. But thanks for watching
what workstation did you get?
I was looking for this answer and you're the only one that could tell me.... Subscribing!
Glade I could help
Is there a problem if I use different cards? i have a 3070 ti but i want to buy a 3080ti, all this mounted on a z690 motherboard with an i7 12700f
Following!
There shouldn't be any issue... I have an rtx 2080 and a 3080 ti both running 8x. I was actually shocked that it worked perfectly immediately, I literally didn't have to do anything, just plugged the card in, turned on my computer and it just worked. Shocking.
@@KazCanning I have two 3070ti back to back connected to my mobo. Do you think they're gonna overheat once I put them underload? If so, what should I do to cool them down better? I have a 680x corsair case. Thank you!
@@thienhofficial hard to say. I know my 3080tis get pretty hot when working on renders. I also have the founders edition cards so they have the pass through heat fins, basically blowing hot air on the top card.
@@KazCanning which psu you have? I have 3080 and 3090 woth a 1300w and when i render my monitors go off
Would love to see a review of the new Mac Pro/Max vs RTX 3090/A6000. Great content btw.
Hi Eric, not that I am a Mac hater, but there is not much support for GPU rendering on a Apple system, so I do not have any desire creating a Mac video on 3D rendering performance.
Thanks for watching
This was just the video I'd hoped you'd make! Super interesting, thank you Mike. It seems that an 570X board would do the trick with relatively minimal sacrifice. There's a pretty significant price difference in Australia between the 3060 Ti (~ $1200) and the 3070 (~ $1700), so for a dual GPU rig for Blender, two 3060 Tis would be cheaper and faster than a single 3080 ($2300), 3080 Ti ($2500) and even a 3090 ($3300). Yes, GPUs are expensive down under. I was thinking of pairing them with an MSI Tomahawk x570, a Ryzen 5950x and potentially an 850w PSU?
hi Guido, I was looking at the asrock taichi x570 when I built my system, but it was not available when I purchased my equipment. The Tomahawk is also a nice board.
Good luck on the GPU purchase.
Thanks for watching.
Hi Guido,
Just a word of advice. If you have triple slot cards (air cooled) then it is recommended to have at least one empty slot between them. This way, the top card won't be choked of air. I personally recommend the MSI x570 Godlike. Another thing is for dual 3060ti a 1000 Watt gold or higher PSU is recommended from a reliable psu maker like Seasonic or EVGA (EVGA gpu's are bad but their PSU's are excellent). Generally, using dual GPU's in rasterization based engines like EEVEE or Unreal engine (by using multiple instances to speed up rendering / multi instance rendering) will use more power than a pathtraced engine like cycles or octane would use. Thus tripping the OCP on your PSU. Speaking from experience. Hope this helps.
@@nawafalrawachy394 thanks, that's really good to know! That motherboard costs $1200 here 😵 more expensive than TRX40 boards, could even get an WRX80 for that. So maybe I'll stick to a single GPU after all.
@@MediamanStudioServices thank you. The Taichi looks solid
@@GuidoGautsch It could still work with a regular layout board if the cards are between 2.7-2.8 slots wide. Though the top card will struggle a bit to get air which will increase its temps. To combat this, I reduce the power limit on the top card to 95% and give it a custom fan curve to keep it from going above 80c.
That's the info! Thank you so much! I want to build this for rendering and modelling in Blender, PSU and case with upgradeability in mind. 5950x is out of budget so I hope the 5900x will be able to handle mobo and future 2 GPUs. My list Ryzen 5900x, X570 Taichi, 32Gb Vengeance LPX, RTX 3060, plus a 2nd GPU in the Future, Case Thermaltake Level 20 HT, Deepcool 360EX, 2NVMe WD 750SN, Seasonic Prime PX1300. Thank you for any input.
Hi dulichauch, if you reduce the PSU to 1000watts you could get some extra ram. I would suggest 64GB. Even with two RTX3080 at 350 watts that still leaves room for the 105watt CPU.
All other parts are a good choice. The Taichi is a good choice for two GPUs.
@@MediamanStudioServices Oh, thats surprising! At beQuiet's PSU Calculator, adding the watercooler, 2 GPUS, 3 additional fans and 2xM.2 and one SSD I received 1000W usage, so I got the big PSU, still think of keeping it. For RAM: would it be totally absurd to order 2 separate LPX Kits of 16x2GB for 170€ or is it strongly recommmended to go with one kit only 2x32gb for 280€?
@@dulichauch it was just a suggestion to save money and get the much needed RAM required for creative workloads. More PSU is always better if you can get it.
and I would get the 2x32gb and leave room for upgrade later.
Excellent! Thank you very much for the information, it is very helpful
Glad it was helpful!
Huge thanks on this informative video!
your welcome
I think you don't have a big hit in performance because on the x570 you have two x8 pcie 4.0 that is equal to 2 x16 pcie 3.0 the big hit is gonna be when you use two gpu that are pcie 3.0 or use a board that doesn't support pcie 4.0
hi Eriuz , I just assumed that most people looking to get two GPU would be using a current MOBO.
But you are correct, the older MOBO's are using PCIe V3 and a lower bandwidth.
thanks for watching
Thank you so much! It's hard for many of us investing our savings in hardware trying to get the best performance at a lowest cost. Have you ever considered power consumption? because having to many machines working 24/7 affects the cost of the electricity bill. So, is it cheaper having many GPUs in one PC vs Many PCs with single GPUs?.
Sir, you are an ABSOLUTE LEGEND! I appreciate this clarification. I am building my first-ever custom-loop PC and was very curious about the two important points you've made. Moving from Apple after 20 years is a little daunting but willing to try out the new and more powerful machines.
My new PC will have the following specs:
• Intel 14900k (water-cooled by the EK-Direct Die, metal thermal paste)
• Z790 Master X Motherboard
• x2 Navidia 4090 Cards (water-cooled, metal thermal paste)
• 4 TB, M2, SSDs
• 92GB Ram at 8600
What's your use case?
Really informative. It would be great to note your full specs in your videos.
Interesting information, so the bandwidth of the pci4.0 is at this moment hardly used if a 16* slot and a 8* slot can stay withing 10% of eachother. And now we already have PCIe5.0, while PCIe4.0 is hardly fully used, maybe storage attached to the PCIe bridge, makes better use of the available bandwidth.
I wish you included card temps
Guys could someone help?
I have a Gigabyte Z690 Aorus Elite DDR4 CPU, and have a rtx 3080 in there.
Is it worth it if I put another one in there (I already have another 3080, and also a 1070 ti)? I use the PC for gaming / video editing.
For gaming, it's not going to help you as you can not sli/nvlink the GPUs. For video editing. Only resolve can use multiple gpus
It sounds like a lot of that performance reduction comes from switching time loss. I was going to build a gaming system and just add a second GPU later, but I ended up cowering out and just throwing a pair of 2080tis into a T7820. $1k and I got a rig that happily does both Blender with RT and AI production work.
awsome been eyeballin a threadripper pro/r9 dual system build :)
Hi Jeffe, I wish you luck finding a system in this messed up market.
All the best.
Thanks for watching
can u make a new test but with pci 5.0? Im planning to build a new rig with 7950x and x670 mainboard. I read the manual that if i use 2 pcie 5.0 slot, It will share each x8, but im not sure if there s any reduction considering it s faster
I would love to see this as well. Similar build I am considering.
I'm using an RTX 3070 with a Ryzen 2700X on B450 motherboard. It's running at x8 because I have a second GPU. I've measured about 2-3% performance drop on x8. I also have it overclocked +200MHz on the core and +1300MHz on the memory with power limit set to 100% and I gain 14% performance boost on my Blender renders.
Hi Narek, thanks for sharing your experiences with the channel.
Thank you for this video. It helped explain alot. I have a dual gpu setup with two 4070 supers. My current board has 1 pcie x 16 5.0 and 1 pcie x 16 4.0. I am in the process of getting a new motherboard with 2 pcie x16 5.0 slots. Would this improve performance?
It would increase, however it's still going to run at 8x
No. 4070 SUPER runs @PCIe 4.0, your motherboard will downgrade all lanes to accommodate your GPU, so PCIe 5.0 would be disabled/ downgraded to PCIe 4.0.
are you going to make a new dual gpu video with 4080 or 4070 ti?
Try fitting 2 nvidia 4090 fe cards in lol
THANKS! Im less confusing. Question does the MB auto balance work load? I’m planning to use 2 different Nvidia for AI gen text & images
Love your videos. new subscriber..
hi Dewayne Dailey, for video editing, Unless you use Resolve Studio, the paid version. most video editing software does not use the GPU for much acceleration. And rendering in the viewport is a single GPU task. Just final 3D rendering can use the multi GPU set ups
Hey Mike, I'm looking to build a PC for workflow mainly in Premiere Pro and Filmora, is it more wise to go for an RTX 3090 or 2 RTX 3070's which would benefit me the most? I don't care for gaming this is strictly for my workflow. Also, what boards and which ram would you recommend? Obviously faster ram makes a difference? What about hard drive? I would prefer ones with at least 3K write speeds. And which water cooled system should I use? I've also narrowed it down to either the Ryzen 5900X or 5950X which I have access to. I will await your reply and thx a million for your hard work.
Premiere Pro isn't able to take advantage of GPUs efficiently and is more CPU-limited us my understanding, so even a single 3070 would be fine. Invest in the 5950x out of those two options. If you were to drop PrePro for Resolve, however, two 3070s would beat out the single 3090 by ~30%
@@GuidoGautsch thx for the input, I hate resolve btw Adobe is so much easier to work with and so is Filmora btw. Yeah 5950X is what I am aiming for.
@@ChrisDallasDualped yeah, I hear you. I prefer PP over Resolve as well for editing - just wish it'd be able to take advantage of GPU grunt as effectively as Resolve does.
I agree with PP as this is my choice for editing the Mediaman videos. Just wish Adobe would get better GPU support.
What if i put my videocard on 2nd gpu slot of motherboard, will this affect the performance?
Hi Mike,
Could you please say what motherboard do you use for the Desktop machine in this video, and what cooling system? As I see you use blower versions of GPUs in this rig, right?
Also, what do you think about GIGABYTE X570S AORUS MASTER motherboard for using with two fan style GPU and Ryzen 5950x? Looks like it has enough room for airflow between two GPU and it has decent VRM for this processor? Tring to figure out best config for my situation.
Thank you for your channel, there is a lot of useful and unique information here!
Hi Vlr. I am using the Gigabyte X570 pro wifi.
www.gigabyte.com/Motherboard/X570-AORUS-PRO-WIFI-rev-10#kf
for the cooling system. I did my own custom loop. check out the video i did here.
th-cam.com/video/r_Rl30LhE2M/w-d-xo.html
th-cam.com/video/dwfn5IQXnIE/w-d-xo.html
The GIGABYTE X570S AORUS MASTER is a good choice.
Thanks for watching
If you use one of the GPUs for gaming, will having 2 GPUs affect its performance?
Thanks!, Great information and tests!, I have one doubt, Is it possible to render in V-ray with RT cores and 2 RTX GPUs but from different generation (E.g. RTX 2090 + RTX 3090) or different models (E.g. RTX 3090 + RTX 3080)? obviously without nvlink or sli. Well, thanks in advance!
@andrew I have the same question, maybe you already got an answer
@@TILK3D I haven't solve my specific doubt regarding 2 RTX GPUs but from different generation (E.g. RTX 2090 + RTX 3090) , But I have found this: 1.-You can render with several Nvidia GPU's using Vray and Vray will manage to use the available CUDA cores in the GPUs (This should work with CUDA cores of the same GPU generation even if they are different model cards (E.g. RTX 3090 + RTX 3080), but I still I don't know if different CUDA cores generations can work (E.g. RTX 2090 + RTX 3090) I have a RTX 3090 and a 980 Ti, I need to make some experiments. 2.-It is no necessary to use SLI or NVLink in order to use all the available CUDA cores of those same generation Cards, 3.-The NVLink will help sharing GPU's RAM but the improvements in rendering time is small (+/- 10%), 4.-I'm deducting (I still don't have evidence) that you can render with Vray and RT cores of several RTX GPUs of the same generatios just like the cuda cores (E.g. RTX 3090 + RTX 3080) also without an NVLink, Vray will manage to use the available RT cores. 5.-I don't know if you can render with VRAY using different RT cores generations GPU's (E.g. RTX 2090 + RTX 3090) Hope this is useful to you, let me check the sources where I found this information and I'll share it here. Please let me know if you have found something else. Best regards.
@@antoniosomera THANK YOU FOR YOUR INFORMATION I WILL BE LOOKING FOR MORE INFORMATION AND IF I SOLVE MY DOUBT THE SAME I WILL SHARE IT GREETINGS, SUCCESS
You can use any combination of supported GPUs. The CUDAs will stack, the VRAM will not. So if you use a 2070 with a 3090, and the scene exceeds the 2070's VRAM, performance will suffer. But aside from this, you can mix and match however you want, and with different generations.
This should be the case with most CUDA based renderers.
@@Outrider42 Thanks!
great video thanks!
great channel. for the video, I was expecting a benchmark comparison between one desktop gpu vs 2 gpu. is it worth to add a second gpu for rendering or is it worst or is it the same ? maybe you have another video that deal with this question, otherwise would be nice to do one. keep up the good work
Hi Comodore, I have a few dual GPU video's on my channel. Check them out.. Thanks for watching
I'm about to run 2x RTX 3090 FE in NVlink configuration for a total of 48GB Vram pool. il let you know how it goes, would be cool to see this on your channel before I rig it up for Blender and Iclone.
aaaaaand how is going? thanks
I currently have a x399 TR 2950x platform, all pcies are the same here, I used to run 4 graphics cards on them for rendering tests, I'm going to switch to Amd 9950x and I have a problem choosing a board, whether to the B650 Aorus v2 board, it's similar in Aorus, x670, x870, can I connect 2 graphics cards? I have concerns about the second Pcie connector, because the specification says that one is connected to the CPU and the rest to the chipset, proart boards have it described that they have two pcie connectors connected to the CPU
Most don't slip the lanes from the cpu
What they have is a chip that splits it. If you look in some mother board manuals it says which have it. But the cpu it self cannot split any of its lanes. I only found this out recently
Would Directx12 use both GPUs as one for gaming?
I run 2 GPU, on my gaming rig. Non SLI, etc. bring the 16x to 8x and I barely notice any performance loss as in FPS.
6950xt for gaming and etc 3050 for stream. But 5950x and Dark Hero VIII.
Core count and good MOBO help a lot, I am sure.
But I game at 2k ultra and stream 2k ultra
i have a question for you. did you have to make any bios changes or were you able to just plug in and play(stream) with no issues? i have noticed that when i plugged in my second GPU (1080) i noticed some lag on the monitors connected to it. i did a clean install of drivers for my 4090 and the 1080 and still noticed some stutters on the monitors connected to the 1080. my monitors arent the same hz because my gaming monitor is a 170hz when the other 4 are just 60hz. i can get the stutters to stop when i make my gaming monitor the same hz but whose trying to game at 60hz when you have a 4090.. lol. any tips?
@@itschops I did not, but I did not plug any monitors in to the stream gpu. The stream gpu is just for work. Give that a try, if not. You make have to fresh install windows.
Because when I upgraded to a 4080s, to replace my 6950xt. It would only read the 3050 out of the box. Yet the 4080s would output displays lol.
I uninstalled all drivers even and, and same thing. I uninstalled all drivers again and reseated GPUS and amame thing.
Fresh install helped, but couldn’t fully test it as the 4080s is huge and my 6950xt was liquid cooled via AlphaCool.
So I am just using the 4080s and had to do a fresh install of windows. Works flawless
Thanks for letting me know. the main reason for wanting to have the 1080 added to my PC was to have the extra display connections since i have 5 monitors but the stutters were happening too often so i removed it for now. I didnt try a fresh install of windows but are you running 2 4080s now? do you think its running smooth because they are the same series card? @@si1entshotzdtom
I have the same set up, except I don't have a 3050. Is the 3050 even PCIe gen 4? Because if it's not, I think the 6950xt will have to downgrade to gen 3 too. I'm still building my PC, was thinking it would be cool to get a GTX for the ray tracing (and DLSS/FRS combo). But I don't really know if it's nessecary. Why a second card for streaming?
I hope we get more lanes on AM5 boards
ya me too...
Thanks for watching
Not sure if it was propaganda, but I read that AMD desktop chips are designed to process a max of 12 pci lanes at once. 2x12*=4 what ever that means.
Hi Mike, another great video from you. Do you think on the instances when you rendering heavy rendering, is it likely that there will be significant difference between desktop processor vs workstation processor, just like the test you did for RTX A6000 vs RTX 3090, where the difference was negatable until you tested with memory intense rendering?
I dont think so as the CPU is close is specs. So it should not be much difference in the test results.
This is extreamly valuable info! Thank you for your content!
I have a following setup: Ryzen 5950X, Asus Prime Pro570X, 64Gigs of Ram (DDR4 Kingston 3200mhz), Asus Tuf Rtx3060 (12Gb), 750watt PSU.
Do you know is it possible to add a second gpu (I have a leftover Asus Rtx3060ti), how it would effect my workflow and rendertimes in following programs: Blender, After Effects, Premiere Pro and also Unreal Engine 5? I was also thinking to get a bigger PSU (1000 watt).
Thank you sir for your time and tips! This channel is pure gold!
hi George, adding the second GPU will limit the Vram in the 3960 (12gb) to the same amount of Vram in the 3060ti (8GB) as 3D render software sends two packages, one for each GPU and they need to be the same package size. As for After Effects, Premiere Pro, you will not get much benefit as both of these programs have very little multi GPU support. Same with UE 5.
thanks for watching
@@MediamanStudioServices Thank you. Yes, I desided to build second backup rig which will help me with Blender rendering.
@@MediamanStudioServices Is this the same for davinci resolve? Davinci being much more GPU reliant.
Hi Mike,
Thank you for taking the time to provide this valuable information.
If I may ask. What is the keyboard you use in your videos. Thank you.
It just the Lenovo cheap keyboard
Great videos. Can you make one talking about pcs and workstation to architectural work
Hi Lucas, this is a great idea for a video. Let me look into the specs required. But I am thinking it will mostly be the same as 3D and VFX computers. But I will do some research to confirm this.
Thanks for watching.
@@MediamanStudioServices thanks, it will be very helpful. I really miss content for small 3d artists. You have been helping me a lot!!!
@@TheMrJackzilla Glad to be of assistance. Please share on your FB page to help grow the channel.
Does it affect southbridge temperature? significantly I mean
Can you please tell the system requirements for a simple rtx 3060 graphics card, as i have one of the desktops which is quite old enough lets say 5th gen intel i5 with motherboard such old to, would this graphics card fit in it, or would i need to change my motherboard and processor as well to an 10-11th gen?
Idk if it helps but I use a 3060 ti with a intel I7 and a 6 year old motherboard
@@seanspliff is your processor i7 even 6 years old?
@@SubirajSinghFitness yup bought it in a prebuilt with the same mother board. My motherboard is a gigabyte x99 UD3
@@seanspliff thanks bro 🍻
Hey Mike, thank you for your videos, they help me a lot concerning my future build.
Indeed, as a 3D Artist, I am looking to upgrade my build that has lasted me for my studies since 2016.
As I'll work with a workstation at work, I am looking to build a personal desktop PC for 3D rending (gpu based Octane Render/C4D & Blender/Cycles), video editing (adode suite) and gaming (most likely 1440p, 34' ultrawild).
My current build would be :
- 12th gen i7 (liquid cooled AIO)
- 64gb RAM DDR4 (DDR5 looks a little too expensive)
- 2TB m.2 nvme + 500gb ssd sata
- RTX 3080 10gb (+ GTX 1060 6gb from current build)
- 1200W PSU
Which motherboard would you recommand that can handel 2 GPUs with at least 2x 16 PCIE 4.0 or 5.0 ?
Doesn't look like a lot of choise yet for z690 chipset that doesn't break the bank... maybe I should wait a bit ?
And do you think 1200w psu is enough for 2 gpus and the overall build ?
Thanks a lot Mike, just subscribed, your content is priceless !
hi Colin, no desktop MOBO will support 2 GPUs in 16x. You would need to move up to a HEDT or Threadripper MOBO. Now you can still get good performance from a X570 running the GPUs at 8X. And I would recommend the Asrock X570 Taichi for running two GPU. This board have great VRMs to manage the power and the spacing to house the two GPU.
Good luck with the build
Can you make a tutorial video on how to setup 2 different graphics card (rtx 4090 and 4080) in on pc
I will be making a new content really soon. Thanks for the idea for a video
Hi! Thanks for your comparison. This is exactly what I am looking for !
I have a question though. Do you think the CPU difference actually contributes to the time difference?
Hey mate! I am bit confused whether to buy gaming PC or workstation PC, for CAD works. So I searched for custom build pcs website and found Z690 chipset supports both workstation(Nvidia rtx a2000) & gaming(Nvidia rtx 3080ti) graphics. But my question is can I put both these graphics on chipset at same time & while working on CAD software it uses workstation graphics (for compatibility as per hardware certification) & while gaming it uses gaming graphics.
Why use two GPUs. Just get the 3090 or 4090 as they also work well with CAD application.
Looking to build a machine to help my realtime cycle rendering in the viewport. I've seen the 3090 nearly rendering certain scenes in realtime I want this level of performance but getting a 3090 now a days is hard. would 2 3080s out perform 1 3090? in rendering, video editing etc?
hello, good video, sorry, an sli bridge is needed to join the two graphics cards for 3d rendering or simply by connecting the two cards to the board, it can be used, I have a 3090 ti and a 3080 zotac
No SLI or Nvlink required to render on both GPUs
thank you very much for answering and clearing my doubt greetings and success
@@TILK3D so i cant use 2 rtx 2060?
it doesnt support sli
Does the gen of the PCIE slot factor in this as well?
Will two gpus that are meant to work on PCIE 4.0X16 slots still get a performance hit on each of them even if they're on PCIE 5.0x8 lanes?
So no matter what, the overall performance of the lane is the highest MUTUAL rating between the slot and the gpu that is on it?
So a gpu that is meant to run on a 4.0x16 slot will be reduced to a 4.0x8 performance when mounted to a 5.0x8 slot?
if your graphic card is PCIE gen 4 , then putting it in a Gen 5 PCIE slot will not increase your performance cuz your card only know how to talk in "GEN 4" language, now what if you put two PCIE gen 4 graphic cards into 2 PCIE gen 5 slots with only 20 or 24 PCIE lanes, you will have two PCIE gen 4 x8 .
But for rendering or AI that speed reduction wont impact your performance that much.
And for gaming, you will never want to use 2 graphics cards cuz almost no game has coded to work with multi GPUs. for gaming maybe you can assign Physics to process on one card and everything else on your main graphic card , but as i said , don't do it for gaming too much heat and electricity for a little or no improvement (sometime the heat will even reduce your gaming FPS)
Great content as always! I don't know if this is worth making a video - but I would love to see a video about NVlinked GPUs - performances,installation,configuring the settings
I was creating a scene in Blender the other day, I ran out of Vram on my 3090 lol...I might have used the wrong settings, didn't optimized the scenes better or used high res textures that I shouldn't be. My future plan is to build a workstation with multiple gpus (currently I have my eyes on the H3 platform, but I might need more vram + gpu power to render) I'm using 3970x with a 3090
Thanks!
Hi theWanderer521, I would live to do a Video in two RTX3090, trouble is that I do not have two 3090 anymore. But if I can get my hands on two gpus I will do a video on the set up of the NVlink.
Thanks for watching
@@MediamanStudioServices Waiting to see this kind of video. Please make one.
I have planned to build pc with i9 12900k with 2 x RTX 3090. l want to know what would be the performance with NVlink.
@@MediamanStudioServices On your experience, tell me is that a good idea or should I go with workstation.
Here's two questions, one if two x8 is only 5-9 percent slower why and second is there a way to fit complex scenes in limited memory cards?
Thanks, was going insane not knowing why my 3080 was running on PCIE 3.0 x8 instead of x16.
2 M.2 SSDs and intel i9-10900k chip. Cant get it to PCIE 4.0 since the chip doesn’t support it
glad I could help John.
Awesome video, however i have a question. Do Sata drives use any pcie lanes, or are they using the dedicated 4x that is needed for the chipset? I have an asus b450-f gaming motherboard that i currently use with a gtx 1070. I have ordered an rtx 3080Ti and i want to use them simultaneously in blender, but i'm not sure whether i should upgrade my motherboard. The motherboard manual says that in a dual gpu configuration GPU1 would get 8x, GPU2 would get 4x, the NVME SSD would get 4x. I'm also using 4 Sata HDD-s for storage, but if they use any additional PCIE lanes other than the 4 required for the chipset i will loose a lot of performance on the GPUs. Does it even worth using a secondary 1070 with a 3080Ti? It's still cuda cores though... Thank you!
Sata uses your chipset and not pcie lanes. As for using the 1070. There is some advantages but with only 4 lanes. It may be better to just use the 3080ti at full speed and the 16 lanea
would like to see when one GPU is 16x and the 2nd one is only 4x
Hey I want to know !! is that possible i can se 1660 super and 3060 ti at the same pc !!! i want to use it for blender render like can i render it with one graphic card work with the other one ??
you can ues to different models working together. but i never tried a 1660
Blender will look for the card with the least memory so it doesn't crash. My understanding is if you have a GPU with 8GB, and one with and 12GB, it will treat both as 8GB GPUs when assigning rendering. The Blender forums will be able to give you a definite answer.
@@swdw973 It appears that when you have a GPU with more VRAM, and the less powerful GPU runs out of memory while your system has ample RAM, Blender can utilize the system's RAM to support the less powerful GPU. Consequently, the more powerful GPU can continue to operate at full capacity using your system's RAM. For instance, if you possess both a 3090 and a 3060 Ti GPU, along with a system equipped with 64 GB of RAM, Blender might use the 64 GB of system RAM to match the performance of the 3090. The efficiency of this process is enhanced by faster RAM. Additionally, optimal performance would benefit from a well-configured RAID system with satisfactory HDD speeds. This setup should allow Blender to function smoothly without constraining the performance of the more powerful GPU, correct?
@@viniarteeSystem RAM will never get anywhere near the speed of vRAM. My answer was based on info from a Blender developer on one of the Blender forums.
I actually run 2 different GPUs, a 3060 and 4070, but I made sure they both are the 12GB versions, because of what the developer told me.
I don't know if you can answer this but I'm just rendering with a high-end gaming PC using stable diffusion and I just want to know if x 8 pcie Gen 4 would reduce the quality because I want to put more nvme drives into the machine and I don't want to do it if it's going to compromise my Ai rendering.
Hey.. thanx for the video.
I have a question .
Is there a difference in video quality when you export with hardware acceleration enable (nvidia, amd, Qsync) compared to just using cpu for vedio decoding and encoding..?
Hi Dasun Gunathilaka, there is always a difference when using a different encoder, be it hardware or software, but the difference should be very small and the quality should be about the same.
thanks for watching
I need a setup that has 48gb vram gpu for llm, I thought about 2 rtx 3090, could you recommend me an economical configuration? b550 x570
I am using 2 x 3090 Ti with a 5950X (so x8 PCIe) but wondering what are the benefits from NVlink... if any..
@Claus Bohm I bought an NVlink from Nvidia 3 Slot for A5000/6000 that is compatible with the 3090Ti but I there is no way to make it work.. I tried everything. fresh install windows 10, drivers update, bios update, anything.. Not sure what's the problem..
@@georgeluna6217 Maybe its dependant on the motherboard.(ones with sli support are sure to be compatible for this kind of setup) So check that out the motherboard specs from the company site. And that nvlink for A5000/A6000 should work with rtx 3090/ rtx 3090ti.
I am using a 5950x with 3 rtx 3090. The system is water cooled.
Do you get a huge reduction in performance by adding a third gpu? I mean does it scale back. I'm sure it still renders faster than 2 3090
thanks this is very informative! would it be possible for you to conduct a similar test, using tb3 egpu solutions daisy chained?
Hi ozztheforester, if I could get my hands on such equipment I will do a video. Here is a link to some reviews.
egpu.io/best-egpu-buyers-guide/
Thanks for the great content .. what is the power supply capacity for the desktop?
its a 850watt
Hey. I have a problem connecting two 3090s through the Nvlink bridge. I use blender 3.0 stable to render my works
b550m pro4 motherboard
Processor x5900
I can not find information on the Internet on how to correctly connect the bridge and speed up the blender. Please help
your mother board needs to support Nvlink. But have you tried to run a render with the two separate GPU? Connecting with Nvlink is not going to necessary speed up your rendering.
@@MediamanStudioServices hey I saw that the 3080 or less only uses about 8x. But the 3090 uses more and would lose a lot of performance. Have you tested this out yet? Or do you know it's true?
@@djuansayjones6178 I am not sure of the PCIe bandwidth for each card, but I am making a good guess here, the RTX3080 does use more than the bandwidth of just 8x when at full performance, so yes untrue.
@@MediamanStudioServices ok thanks a lot!!!
Great video! How do rendering softwares handle two GPUs in a non SLI/Crossfire/NVlink configuration vs a SLI/Crossfire/NVlink??
Thank You
when I get some more GPU I will test this out and do a video
thanks for watching
Hey Mike. Is there a way to stack 6 GPU RTX 3080/3090 blower on a single workstation motherboard ? I've been looking as well towards mining motherboard (that can fit 6-12 GPU) - but they reduce the pcie lanes speed x1 - Your video here showed a very little difference between x16 and x8 - So i'm curious if x1 is pushing it way too far for a render gpu farm? need your advice.
My question to you is why would you want to do this. Just buy a GPU server for this task. Gigabyte makes a great server for just this style of workflow. Anything over 3-4 GPU in one workstation requires a server chassis.
really informative
thanks Dom v
Ok, how about gpu and an accelerator, an instinct mi100 maybe? , x8 x8 🤔..
Does the mobo HAVE to be crossfire/sli ready to have two gpu cards?
No. Just support 16x or 8x pcie. Preferably directly from the CPU
Could you make a video on quadro cards, can they handle multiple 4K monitors
Thank You😊