Rendering with 4 RTX3090 in Blender
ฝัง
- เผยแพร่เมื่อ 10 ก.ค. 2021
- #Nvidia #H3 Platform #GPU
How many GPUs do you need for your rendering?
Testing out 4 RTX3090 with Blender to see if this will increase your productivity.
H3 Platform:
www.h3platform.com/product-de...
AMD Threadripper Pro:
www.amd.com/en/processors/ryz...
Discord:
/ discord
Facebook:
/ mediamanstudioreview - ภาพยนตร์และแอนิเมชัน
is so strange seeing someone live without both of his kidneys, is truly remarkable.
hi Jose, thanks for watching
🤣🤣🤣🤣🤣
He's actually full Cyborg in a human skin. Every thing under the skin is sold
😂
Lol good one
5:28 nice jetplane
Woah! This test was on CUDA, I can only imagine how fast it would be with Optix + Cycles X... Probaly 5sec or less on classroom using 4x 3090
I wish I still had the hardware to test it out on CyclesX
@@MediamanStudioServices please can you test it with Optix selected on Preferences > Devices instead of CUDA?
@@henrique-3d I do not have the 4 GPUs anymore, but I will test it with two RTX3090 and post the results.
@@MediamanStudioServices Thank you very much! :)
@@henrique-3d the Classroom did 24 sec with Optix and two RTX3090
3:33 so realistic it adds the engine noise
lol 😂
My guy has a whole jumbo jet in his computer lmao.
We need a updated version of this video with 4 RTX 4090. You might need a power grid handy
My pc takes only 35-40sec. ............ To Boot-up
🤣🤣 same here, with the sli bridge for the O's to stand up.
Me 7min bloatware and Malware lmao
Blender 3.0 is gonna be sick with 4 of those
yes, I will be nice when we get the full release of Blender 3.1
man. 4 3090s in cyclesX will be faster than eevee
Thanks for producing some of the best GPU related videos out there
So *this* is why there was a 3090 shortage!
you hear those cards spin up like a danm jet engine? lol
ya my one complaint but this is not designed to sit beside you at you desk.
Wow , that thing sound like a super vacum cleaner!! Very impresive demostraition of Gpu Power
My laptop takes about 30 seconds to render the default cube..
Dude, your channel never disappoints! I'm still learning 3D and animation using Blender, and like most people, haven't been able to upgrade my GPU (980Ti). I have been trying to get the 3080 since it came out. I have two questions, one personal, and one video related:
1. As a beginner, is the 10GB VRAM enough in Blender? The junk blender scene is too big for the 3080
2. For the single work station, did you have all four cables connected to the workstation's host card?(Does that matter?)
Happy to hear that!
1) yes 10gb is enough for most GPU rendering, you can address Ram limitations when your skill reach the level you are creating scene that require more ram.
2) yes i did test 4 cables and also 2 cables to see if there was a performance reduction. It we about the same results for the basic benchmark testing i did. But in theory, there should be a reduction for complex scenes when using just two cables.
Cool video Mike... Nice intro! It seems the big jump is from 1 to 2 gpus which has a major decrease in rendering. Thank u for the showcasing the falcon products!
Cool, thanks. it was an interesting project to see if the external case was any good.
Four fighter Jets flying at high speed ..... thats awkwardly loud ....
yes, it was loud, but the performance was great. not for sitting right beside you on your desk.
thanks for watching
@@MediamanStudioServices loved the Video ... thanks alot for the Great Content ... very informative.
@@Dr.Hypersonic Thanks for watching
Nice intro 👌 i like your videos 📹 you are kinda totally different in topic rather then always for gaming. Also i like you use real scene render. Hi 👋 your new subscriber
The 3090 may be getting out of fashion but damn it's a freaking powerful card.
way more bank for the buck when compared to 4090 tho. still a must buy
@@bigbrawler7078 I just got it for $750 more worth it compare to RTX 4070 Ti for same price yet less Cuda and RT cores than RTX 3090
@@alpzepta bruh rtx 4060 (non ti) will have 4 gb less ram than 3060 and less cuda cores too. And they will most probably say it's 3x better than 3060. They rely on dlss for everything now while forgetting that 3d artists like me don't want the same card all over again and dlss is dogshit to us. It's absolutely useless.
@@bigbrawler7078 I haven’t check out DLSS (my last laptop have RTX 2060 and now I’m switching to desktop) but it’s probably useless in making CGI Animation and 4K Video editing
Now try 4 4090's!
me watching this while struggling to get enough money for a 3060
Real deal, just earned a sub.
Thanks
This is damn funny at the start...well done Mike
Thanks
Great Job!
Feels like an aircraft getting ready to take off
PS4
I would love to know the power it required to run these benchmarks with these 4 3090s 😂
Excellent video, bloody shame I only discovered this channel today.....
happy you enjoyed the video. Please pass to all your friends. And keep watching
Sure is a purdy video... 4 3090s looks like a wonderful thing.!
one trick you can do on animations is render each freame in a different computer.
eg: one computer render frames 1 to 10 while the other render 11 to 20, you can even put the final result on a network and inform blender to create the file as soon as it start rendering but not replace existing files, with that setup you can put both machines to render from frame 1, but the second machine will realize another machine is already rendering the first frame so it will start from the second, and as soon as one machine finishes it will see what is the next non existing frame and render it.
this setupt is especially usefull because you dont know how much time it will take to render each frame, but it dont matter because as soon as one frame is finished the other will start automatically, unlike in an scenario where you put one machine to render say, from 1 to 10 and the second from 11 to 20, this other scenario would be very bad if the first 10 frames are faster to render but the 11 to 20 are slow (or the opposite)
I advise you to manually subtitle your videos. As much as TH-cam tries to do this automatically, it's not perfect. And with your caption set up, you can use your caption as a reference point for TH-cam to translate into every language in the world. Your videos are very interesting and a lot of people should have access to these videos
Bro when did this guy get four kidneys
He stole it on the highway😭
now waiting for RTX 4000 series to see how long it will take to render
It’ll cook the case before it ever finishes
@@CVerse overpriced space heater
not how long it'll take, but how short it'll take
Nice sound lol am i the only one who likes that sound?
Yes, you can hear that work is being done
that sounds intimidating
My man living my dream 😢
Your channel is a gem keep it up! Have you tested or do you have intentions to test thunderbolt gpu enclosures?
Hi Ozztheforster, i would love to give it a try, but currently I do not have any access so such equipment.
Thanks for watching
You're a mad man. 4 RTX 3090
I wish I can keep them. But they are on loan.
thanks for watching
@@MediamanStudioServices 👍🏿
Hey can you tell me what 3090 models are you using?
Eres el mejor ! ♥
can you test a scene that goes out of core and compare render times on external and internal cards? Would like to see if the bandwidth becomes an issue on external cards when it has to keep transferring new data into the GPU ram during rendering.
Sorry, you know you are asking for hours of testing time. and Sorry, I do not have the GPU anymore for this type of testing.
I am just wondering, why would you want to do out of band memory testing. This increases the time to render greatly. Most GPU rendering should fit within the Memory limitations. If not then go back and optimize you scene. Rendering out of band is crazy. get bigger Vram GPUs is the answer.
I just did a test of the class room Scene with release version of Blender 3.0, a single RTX 2080 TI and GPU rendering. It took 29.46 seconds. So the 3.0 cycles rendering speed improvements are worth looking into. It would be interesting to see what Blender 3.0 does with 4 RTX 3090 cards.
I am looking into Blender 3.0 and the new cycles render. stay tuned for a video
People see classroom render in seconds and all they can talk about is fan noise lol
I'll put my head directly in a jet engine for that speed.
ya hard to understand some people, but its great to see GPUs are getting so much better than the Gens in the past
thanks for watching
Would you happen to know if NVlinking these cards successfully increases the VRAM ?
Hi, I appreciate your video. Will you gonna test a 1 minutes of 3D animation video for rendering in same way?
What's the software you're using for monitoring of PCIE bandwidth and other stuff? Is it open source?
BTW, excellent video, liked and subbed!
Hi. Thanks for watching. It's HWinfo and free
It's a pity that Nvidia killed the 2-slot 3090s. For the single reason that people then would not buy their overpriced Quadros for workstations/servers. The hilariously oversized 4-slot coolers only exist to prevent people using these GPUs in servers.
I totally agree. Thanks for watching
A pity is one word, slimy is another
very cool
Excellent speed increase... though sounds like you're at the dentist. These will go in my padded closet. :)
i imagine at the speed of the 3080 and faster the cpu, ram, and storage speeds start to become a real factor on cycles. hence why i always suggest doing a single instance of blender per gpu. it's what i do when i'm given access to a multi gpu setup to render some more intesive projects and from the few tests i've done i have seen a little uplift with slower gpus but if using cycles on a 3080 or faster gpu it gives me a WAY faster render using this method.
you should install Deadline for the free 10 nodes
How much does the H3 Platform cost? I can't find a price anywhere. Also, how many 3 slot 3090's can you fit in there?
This external gpu is new to me. I wonder if my meager powerdirector will even have any boost on such device?
Hey mike have you ever thought of trying this unit on some complex mapping software like Bentley context capture? I’d love to see how this would preform with it. Also i can provide you the files to try out that has taken my computer a week to process. I would love to see what this would do and how much time it would cut down.
Sounds like a great experiment
but I do not have the equipment anymore.
Well spoken.
I have 3090 and 64 GB, and a scene with 1 million faces makes the scene heavy and difficult to deal with. Can I know the reason or advice?
Can you combine a bunch of older GPUs together to render? Must they all be the same model?
Hey Mediaman... Would you consider doing a texture bake with dual 2080ti's vs dual 3090s if possible
with that falcon 4205 is like having your own home rendering farm and you only just need 2x or 4x RTX 3090 to make a few movies with now ain't that nice?
Mike am hoping these are studio setups because this is crazy as hell to have these render nodes as a freelancer 🤣🤣
ya this is enterprise solution and not cheap.
Nice
thanks for watching
Next, the Falcon 4210 with 8 RTX A6000.
$50,000. Better be instant rendering.
great video
thanks
Please share on your FB to help grow the channel
he got the 4090 early
Great, could you do a video on how you connected all 4 GPUs to the workstation?
sure when I can get a ASUS WRX80 MOBO again and the 4 GPUs I will make a video
thanks for watching
Does the h3 module connect to the workstation physically or does it function as a network-only rendering server?
have a look at this video.
th-cam.com/video/HSzBzKJBSlY/w-d-xo.html
Thanks for watching and please share on your FB to help grow the channel.
you need to increase the bucket size to actually see any performance lift
I know
Hi, I got a question. Would an nvidia quadro p4000 2gb render in blender 3.00 ?
Am looking to swap with my msi armour rx570 4gb with a nvidia quadro 4000 or should I keep my rx570
For rendering you might want a graphic card with the most CUDA cores and most ROPs. It doesn't matter if it's quadro. quadro 4000 is ancient and you should throw it away.
My single RTX3060 renders the BMW in 13 seconds using optix
You're probably using a much newer build of Blender. Version 3.1 for instance vastly improves render speeds over earlier builds. My 3090 (Optix) takes roughly 7.2 seconds for the BMW with the stable 3.1.
true, b2.9,rtx3060 12sec
@@shoohratyuldashev I'm going to get architectural renders, I was thinking of buying a 3080, do you have any suggestions about it?
@@mimarfatihcan sure go for it, i would try having 2 3060 instead. good luck with purchase.
Can you do some AI training tasks? Using the same external setup?
nice
thanks
..wow.. amazing render speed..... that the correct way for muti powerful cards to use, not for coins.
don't you wanna change those noise fans on 120-140mm ones?
Hi what do Pixar or any big company uses, both Cpu and Gpu?
it all depends on the workload required, but most big studios use a mixture of CPU and GPU rendering
can you replace the H3 with something else?
If I had this, I would be unstoppable
its good to dream
@@MediamanStudioServices it’s also good to wake up
@@LuSteyx if we could dream while awake, that would be great too
what is that blue-ish 3090? looks sweet, imagine one of those on an XTIA XProto chassis
leadtek and they don't make them anymore. Sad.....
Badass! (A lesser man would say: "But they are not ti's")
But they are not ti's
good
thanks
Where the heck did they get the blower style 3090‘s? Are these nvidia‘s? Would be great for my deep learning box but I only find the fan style type on the used market. What I heared is nvifia hold back these blower type cards and offer them only for professional companies, right?
try this link
videocardz.com/newz/galax-re-releases-geforce-rtx-3090-rtx-3080-graphics-cards-with-blower-type-coolers
Nvidia killed them, because they were replacing their overpriced Quadro cards. They are now only available in ugly 3-slot coolers that don't physically fit.
@@IvanNedostal He's specifically talking about air cooled cards fam
The tile sizes were also really small. You can get much faster rendering with optimal tile sizes.
Subscribed
Thanks
Hello Mediaman, would it be possible to see the 4 RTX3090 running Blender in Linux?
I would if I still had the equipment. Sorry
Crysis: "Finally, a worth opponent, our battle will be legendary!"
lol its a plane on your room
How can one buy the Falcon 4210?!
Would appreciate the input.
I would contact H3 directly
actual jet engine
what manufacturer are those clean looking 3090s?
Leadtek and gigabyte. But both are not in the market anymore
1. Can I do the same with a DGX A100 (DGX-OS/Ubuntu only) ? 2. is it using cuda cores or tensor cores or RTX cores ? 3. can I do the same with vray 5/cinema4D in windows 10 ?
yes you can use the DGX A100 but at a cost of about 199k USD. The render I did was on Win10 with redshift, but the H3 Chassis does not care what OS or software you are using, It is just an extension of your host computer with 4 GPU to use.
Can you try this with the new Blender 3.0 Cycles X? That sh!t would be almost realtime
Exactly!
Looking to buying one, this fixes my solution to a lot of gpu problems since I’m a digital artist, how much do they go for, I tried getting in contact with a supplier but no awnser
If you can, WAIT! Intel is going to realese their new gpu this coming month, nvidia in october their new 4k, and now they are still over priced due to crypto mining demand. You could find a 3090 for half a price if you wait until then. I am in the same situation than you... Waiting to get a new gpu in these dark times...
These units are not cheap. The chassis is about 5k USD and then you need to put GPUs into it. I would look at the Asus sage WRX 80 motherboard and getting pcie riser cables to put in GeForce GPUs. Or you could also fit 4 of the rtxa5000 two slot, fan style GPUs plugged directly into the motherboard. Now this set up would require a very big PSU, and lots of fans to push air thru the GPUs. But the full system would be cheaper than the H3 chassis but the GPUs would be a extra cost.
How much is the H3 Falcon 4205?
And where can I get one?
just look on amazon or something lol
Could you do this similar test with AMD 6000 or 7000 series? and if possible RTX 4090?
They don't fit due to their hilariously oversized ugly coolers. You need 2-slot cards for such servers.
@@ProjectPhysX I'm sure that if you're willing to fork over this kind of money for even one of these cards, you are also willing to figure out your own cooling solution even if it means using something aftermarket.
God i wish that was me
Can you redo these tests with CyclesX?
I mean.. it would be real time haha
Didn't think about this. A use for the mining cards with no video output. Basically you get a small rendering farm in one box. Neat idea.
Oh how times have changed. The classroom scene would have taken at least an hour with the 3 Amiga mini rendering farm I had in the late 90's.
90's... Any gpu I can think of from that time would take days. Correct me if I'm wrong.
@@Zeke__02 Notice, I mentioned a 3 Amiga rendering farm (68040 with math coprocessors, Intels would have taken much longer). Keep in mind a big render for video at that time was 720x480, not 1920 by 1080 or 4K which is even bigger. Standard video at the time would be 16% of the number of pixels rendered for blu-ray. Having 3 computers working on that one scene, which the Amiga software like Lightwave and Real 3D could split, would have taken about an hour. So no, that scene would not have taken days (on that platform) as it's not as complicated as it appears. However there were scenes I did that took several hours or more.
And I am speaking from direct experience, not conjecture, as I modeled scenes equivalent to the class room scene.
@@swdw973 I think I have a much better understanding now 🤝
are you using NVLink for 4 gpu ?
hi Anirudh, no I am not. Linux is the only OS that supports 4 GPU using NvLink
i have a 850w power supply, will it be enough for rendering with a 3090? i have one but i cant get definitive answers if that is enough power for it
Yes, its enough. 850 is what nvidia is asking for. Im assuming you dont have other huge power draws in your system like a second gpu.
yes also make sure you have enough wattage for everything in your pc not just gpu what I do is search the watts needed for everything then go up by 100 (for future upgrades), just don't go overboard you don't need 1k watts for an rtx3060 for obvious reasons
WHAT, I couldn't hear you over the jet in the room. Unattainable toys.
hi Othoap, ya this type of "Toy" is more for the studio or enterprise artist.
@@MediamanStudioServices Yeah we gave up on in-house render farms and just buy cloud time. Why invest in equipment that is out of date even before you can buy it. "enterprise artist" good one
PM me at the mediaman FB page and lets talk about remote render farms
facebook.com/mediamanstudioreview
blender 4.02 15s with 3060ti :D 50' samples ; 128 sampels 24.6s :D
You should test more viewport performance too. But great video
Great idea. I have been look for a method to represent this performance in the viewport. I need to look at animation as well as environment creation and modeling. So I need a scene that will be a good sample for this project. You know of anyone that has such a scene?
@@MediamanStudioServices not specifically, but I believe there’s some good animated demo projects on the blender website, you can also showcase the viewport performance of each project you’re benchmarking in rendered view with optic Denoising so it gives us a better insight to how quick/responsive the cards are before hitting render. Ps: love ur videos and really appriciate the effort!
what conection do you use? x16 or I can use x4? pci-e
@Kartoffelpuffer can't use X16 on all 4 cards. Unless you go HEDT, or xeon/epyc.
I checked their website and looks like they're running 4 x8 lanes.
Can you do the same benchmarks with the 6900XT ?
if I had them I would