You want AMD HIP or Nvidia Cuda and as much ram as possible. The local AI space changes quickly so a benchmark made today could be outdated in a few weeks. @@hermesvoeglein535
I snagged an arc A380 low profile for $110 a few months ago for my Plex server. Thing is a transcoding beast and only pcie power required. The few bucks more for the few extra gigs of GDDR6 ram really is worth it when the family is streaming multiple videos.
The vram isn't really used in media applications. That's why an integrated GPU with zero vram still works fine for multiple streams and transcodes. It's also why the p400/600 2GB is a great deal.
My first checks for getting the A310 transcoding to work on Ubuntu: 1) kernel version, 2) necessary firmware bundles. And if you have jellyfin installed it has its own version of ffmpeg. But I dunno, I'm running it on Debian testing and it just works after I installed the necessary bits.
AV1 is honestly a game-changer in my opinion. I got an A380 for $110 and the performance in Jellyfin is excellent and the AV1 streams it produces look genuinely amazing. At one point I was comparing different stream resolutions in Jellyfin and it seemed like the resolution wasn't changing, then I realized that it WAS changing but was having trouble seeing it because I was looking for compression artifacts rather than resolution. I am so used to associating low resolution with compression artifacts (blockiness) that seeing 720p or 480p without compression artifacts genuinely broke my brain for several minutes. So basically, if you're looking for a GPU for your media server, I'd really recommend prioritizing one that has AV1 enc/dec. Intel has made AV1 very accessible.
@@Amwfilms following up this. So the A380 didn’t have the fan issue at all and I’d never even heard of it. Well, I just got a Sparkle A310 ECO and it definitely does have the fan issue. The fan is very audible even at idle and it’s constantly doing this ramp up/ramp down thing which was very annoying. Haven’t tried updating the firmware yet like others have recommended, but hopefully that fixes it.
Really useful video! Many of the issues you were having with the Intel A310 is probably due to the lack of Resizable BAR on your motherboard/CPU. Transcoding still works great (I currently have a $110 A380 in my Jellyfin server) but any 3D rendering will suffer or fail.
As a German I do love the efficiency angle of this video. Our power is expensive so I am currently using a N100 mini PC that takes care of all my transcoding. The idle power and full system power of it is unbeaten.
I loved seeing Arc being finally included into these benchmarks but I would have really loved to see the A380 perform against the other cards instead of the A310, especially considering its price point is much closer to the other cards. I think some of the problems you had on Arc could've been solved through some tinkering with the linux kernel but that's not really straight forward ( I'm saying this cuz I've heard of people getting the A380 down to 1W reported power consumption ). Also I would've loved some AV1 decoding benchmarks! Considering that's one of Arc's biggest strengths.
I mean media encoding and decoding performance should be the same since all arc cards have the same two media engines but the extra vram could've helped in some cases potentially. I do wish more time was spent getting the arc card working because I feel that those are the main reason people would be watching this video. Not too relevant but it would be interesting redoing these tests with a 12th gen intel CPU or higher to see how the arc card would perform with deeplink/hyperencode.
@@GaiaGoddessOfTheEarth sadly hyper encode/decode are only supported on windows an last time I checked ( months ago, should probably check again ) support for them on Linux was considered non priority ( aka you'll be lucky if you ever get this ) edit: just checked, last year Intel said Linux support currently not planned as they'd rather focus getting drivers finished and stable first ( which is reasonable considering they've only recently almost finished a complete driver rewrite ) but last month an Intel employee asked for the use case of hyper encode on Linux so not all hope is lost
Yes the A310 is indeed "made for" home plex and jellyfin servers if you ask me. But i also found out, there is also a very low powered version of the A310. That has a base clock of 1ghz. Think it was also the sparkle.
I think you're confusing it. the A380 is the one thats at 2Ghz, the A310 is the one thats base 1Ghz. I think Asrock sells them, tho they use the exact same design
I believe the mentioned i5 and i7 has the same integrated iGPUs, so should really matter which one you tested, however would nice to see a comparison between iGPUs at this point, as it seems like a very suitable choice just for jellfin
Arc310 hands down for less than $100. Now the only reason why I don't run this is because TrueNAS scale doesn't support it yet. Eagerly waiting for 6.2... Until then my Tesla P4 has been doing great for $50 GPU.
@@ramtek55 I probably should have mention I bought this last year. So prices have probably slowly ticked up lol. Sadly low watt PCIE powered GPU's are a hot commodity now. Everything has to be power hungry now.
Literally have spent hours looking at different models and guides regarding low profile cards that don’t need a power supply connector. This is a great resource! Thanks!
I recently purchased (for $123) a new Nvidia T400 for use as my daily driver (an upgrade from Intel 14500's UHD 770). I'm not a gamer, so integrated graphics is generally powerful enough for my typical usage (fine for triple 2K monitors playing TH-cam videos), but performance with a few dozen Chrome tabs open is noticeably improved with the T400. The T400, which employs the TU117 (same chip as the GTX 1650 and the T1000), benchmarks more than 50% faster than the UHD 770. The card's 4GB of dedicated GDDR6 video memory makes a difference. I did not want to heat up my office space, so I sought a low-power card, and the T400's 30W max power draw ticked the box. I had built my PC nine months ago with the intention of using only the integrated graphics, and finding a suitable motherboard with triple display outputs really constrained my options. Desirable motherboards generally offered at most two video outputs. Had I known that I'd be using a T400, I would have selected a more sophisticated motherboard with Thunderbolt, not caring about integrated video ports.
I hope you don't mean... visual quality. If you're impressed by QS's visual quality vs NVENC, then I'd strongly suggest availing yourself of the services of several optometrists. (Yay, unstable and buggy drivers right into the Linux kernel!)
@@tim3172About quality, for H264 and H265, Nvidia is clearly better than AMD and Intel, but on the AV1, Nvidia and Intel are pretty much equal, and AMD is not that far behind. So if we're switching more and more toward AV1, Intel iGPUs are actually gonna be good for transcoding
If you're wondering why there's no AMD cards like the rx 6400 or 6500 it's because they don't have dedicated video encoding hardware at all. Well done, AMD!
@physxme NVIDIA doesn't in the 4000 series either. I think the 3050 is the newest desktop card from AMD or NVIDIA which is. I think the 3050 and 6400 are the only bus powered cards as well?
@@nighthawkvc25a With unraid and 8 spinning disks in idle, about 65Watts (disks spun downs) in full use with 9 dockers and transcoding about 115Watts. Also using a script to clock the p400 down when in idle
I think that one thing that you forgot to mention for each card are the CUDA cores/Shader cores/Tensor cores as this is what most people with a homelab would ultimately be looking at for things like transcoding or local AI. Overall, a good video as there aren't too many videos out that will go over all of this in a non-gaming capacity. I use a P400 for Plex transcoding and I just setup Jellyfin and run a P400 with that one as well. I also have a pair of M40 12GB Teslas that I run for Local AI for Ollama and InvokeAI.
Great video! I would love to see these tested with a local LLM of some sort. I'm interested to see how many use cases a person can get out of one of these.
I got a Tesla P4 for $75 and the cool part about it is that it has 2 encoder chips on the card. It also runs off the PCIE slot so no power cables. The downside is I had to 3d print a shroud to use a blower fan into it to lower temps since no built on fan. Works great though.
One caveat on the A310 (and other Intel Arc GPUs) though: Make sure your PC supports Resizable BAR (ReBAR) or Smart Access Memory (SAM), or else its performance will tank pretty hard even on decoding/encoding. On another note, if you really want to go AMD, the Radeon Pro WX4100 does have decent hardware decoding/encoding capabilities, just be aware it's a Polaris GPU so its not gonna be as feature rich as the newer cards. You do get a more hassle free driver experience in Linux though.
Indeed for a multituden of reasons Many of those of us inthe industry have refused to support team greed for many many yrs nwo due to their pathetic bs tactics It's many many yrs now since they tried blacklisting me for example since I refused to play their bs games and forgive them editorial control and dictating how we would do tests and only their approved tests I and many other told them to shove it of course The sad part is they have only consistiently gotten worse over the yrs
Fun fact, at 15:50 "samples/minute per watt" is actually measuring energy per sample, or rather its reciprocal. Divide by 60 for samples per (watt second), which is the same as samples per joule. 1 divided by that gives joules per sample. Doesn't tell you anything different, of course. I just like the way these units work!
The intel arc-line of products seem like a crazy good deal. I guess I should quickly get one before Intel starts feeling entrenched and monopolistic and starts pulling Intel CPU division / nVidea stuff
they pulled their gpu line already, no more intel gpus will be released. the designs that are out are available from board partners,but intel is not designing another dedicated gpu.
@@dylantrom4333 I’d bet he’s getting that from MLID, who has it out for Intel in general, and Arc GPUs in particular. I mean just compare how he talks about RDNA 4 and Intel 20a: Amd has scaled back on RDNA 4 so they can put more effort into RDNA5, which MLID says is going to follow much quicker after RDNA4 (like 6 months instead of 2 years): MLID can’t stop talking about how smart this plan is. Intel announces that development of 18a is going so well that they’re halting all development on 20a to focus all of their efforts on 18a, which will be out some 3-6 months after when 20a was planned on launching: MLID: “the sky is falling over at Intel folks, they’re basically in chapter 7 bankruptcy, we’re going to go from a Intel v AMD duopoly to a Qualcomm v AMD duopoly. “ I mean, FFS, I’m pretty sure Dave2D’s video showed Lunar Lake CPUs beating current gen desktop CPUs in single core performance, and MLID: “Intel has given up on Single threaded performance folks, Intel CPUs only exist to host Nvidia GPUs from this point on”
the P600 (90 USD in 2020) has been excellent for my Plex server. I always transcode on the fly for external streams, and it's handled everything without missing a beat! The CPU transcoding (Haswell Xeon) before was crazy hot, loud and power consuming. It's what I'd buy again if I needed to replace it, especially at today's prices.
Something to remember about putting a gaming class card in your server is that you are leaving almost all of that performance on the table. The encode/decode parts are separate from the general processing parts, so you're paying for performance you may never use, on top of the far greater power draw.
If you're holding on to the A310, I'd love a Battlemage comparison after the comparable version is released for a few months, personally want to see if idle power has been reigned in and are less issues
Nvidia: Famous inventors and scientists that changed our way of life. AMD: Exotic locations and star clusters Intel: Neckbeard DnD loser nonsense. Shame.
Always excellent. FWIW, another option: You can often pick up a LP Tesla P4 for very little - superglue a 40mm blower to the end and you're good. To get one to run as a GFX card is not hard. My home server has one in passthrough to a VM running a stripped-out Win 11 + Sunshine > Moonlight for network emulation / games. Also as a transcoder in my Jellyfin server - these cards come with lots of RAM, making sharing between VM's practicable.
Was that the 3050 that didn't fit in the HP EliteDesk? Did the A2000 fit? I'm also surprised that there was so much issues with A310. I know Intel has been putting in a lot of work to get it supported with more things so I figured it would have been able to do all the tests that you did for everything else. I'm planning on using Plex and as long as the A310 works that's really all I need in the short term.
The A2000 and 3050 6GB (at least the one I had from MSI) were the exact same size essentially. And it wasn't the EliteDesk, it was a lenovo P330. I don't currently have an EliteDesk to test fit it in sadly, otherwise I would let you know!
Thank you for putting this together! I ran into problems when my P400 died (well, I THOUGHT it died anyway) and I was trying to decide between a P620, A310 and an RTX 3050. After a bunch of reading I got a P620 based on it's acceptable level of performance, low price and low idle power draw when compared to the A310. Only THEN did I realize that a TrueNAS update had altered the way the NVIDIA driver was configured in JellyFin. I had been using CPU for transcoding for a month because I never function tested after the update...
now im very interested in seeing how the igpu in the 8700k compares to that of a newer gen intel desktop cpu (maybe the new core ultra 200 series that is rumored to come out soon) and the aliexpress intel laptop cpus that are frankensteined onto desktop boards since laptop cpus tend to be more power efficient and have better igpus than desktop cpus.
Did you not find a suitable Intel A380 option? I've been told the 380s are easier to find than the 310s, and that they're even faster and better....and They also do AV1 encoding.
I bought an A380 for my old Lenovo office PC (i7-3770), but haven't even tested it in that system yet. I can confirm that it's a bit weird. In my test bench, it would blow through some benchmarks and then just choke on others. Not sure why. I guess if you support Intel, you'll have to put up with a few snags for a while. I've been using the A770 since it's release and love it.
The one thing about these videos that everyone makes a mistake on is it's more about single slot than it is about low profile. Just getting a cheap case with full height PCI slots isn't a problem, but trying to fit a GPU in with a bunch of HBA's and NICs is a lot more difficult.
Hey. Thank you for the great videos. I Just had a little nitpick with the object detection value proposition ( 19:10 ) chart. The dollar/ms metric is not a useful parameter since we want to minimize both of these values under ideal circumstances. Therefore, they should either both be in the nominator or the denominator. The dollar*ms would have been a useful parameter and the card with the lowest metric would provide the best value proposition. Also the samples/minute per watt ( 15:52 ) is basically equivalent to samples/joules which is a very useful metric (you have to divide the values by 60 to be precise) .
Fantastic work, and very interesting. Thank you for this. Something which would help a lot would be a simple text on each graph saying "higher is better" or "lower is better" to make it easier to follow along. When watching while doing chores, it's very easy to get confused.
Any chance we can get a Tesla P4 added to the mix? It has 8GB of VRAM, a 256 bit bus, 2 NVENC chips and 1 NVDEC; and can be had for under $150, including a cooling solution, all day long.
Nvidia is still king for Pass Through in Linux IMO. I make due with a P400 and its only job is to transcode files for a few family members who are remote and when I want to watch something on the go. Can even handle several transcode at once.
Thanks for the confirmation that i chose the 3050 for what i want correctly. Still thinking of getting a A310 for my Jellyfin though. Now it is looking a bit better also.
Arc A310 best for pure max encoding/Transcode performance. P400/P600/P620 for "value" transcode For a blend of Raster AND Encode with a little DLSS in there... maybe the 3050? I would say if you have a headless multipurpose workstation and remote 1080p gaming... that 3050 is actually a good price. The A310 can actually game too, but keep expectations VERY low. And if you can find an A2000 near the 3050 price... get that instead if you were considering the 3050.
thank you for talking about lower powered controllers. I have never really been a true gamer. My needs are small. Just doing internet, home work ( office type of stuff ), doing sw mainteance of computer, watching videos, and lately - a little transcoding. If I play games, the type are 20+ years in sophistication. My most advanced card is an NVIDIA Geoforce 730 and the integrated graphics on an Intel I7-8700. the 730's cost me about $40 on ebay - and I think that was newish - old cards - never used. Very happy with both levels of video.
I went p620 to replace some older “bigger compute” GPUs that were drawing 34 watts at idle. P620 is drawing only 7.8 watts for me in a TrueNAS Scale server. With my electricity of 37¢/kWh I will save $84 per year.
Thank you heaps for putting in the time and effort for this video! This helps quite a bit for when I'm able to move away from a card I picked up for cheap as it was used for mining.
Do any of these cards support package c states below c3? I can’t find any cards that do. You would gain massive system idle consumption gains if such a card exists.
As a feedback to your graphs, while they are looking nice, I lack the information of y axis. E.g: 8:19 where the y axis represented frames per second, I wasn't sure what I was looking at as I missed the FPS part that you said during the video. Having that shown somewhere on the graphs would be helpful.
A benchmark for llama 3 or some other llm would be handy, its a little bit of a hat trick right now but home assistant lets you use a locally hosted llm as the conversation agent
I use intel integrated gpu for ffmpeg by using the jellyfin version of ffmpeg. It usually works given that jellyfin also works. I *think* you'll need to re-compile ffmpeg to add a few flags for those intel cards to work.
How well supported are the A3x0 GPUs in linux/bsd at this point? Because last i looked into it, there was little to none. Also, AMD just works on Linux/BSD, so it seems incredibly shortsighted to limit yourself to nVidia and Intel only.
This is exactly what I’m looking for. Many channels tend to overlook sff cards and don’t really talks about idle power consumption at all. I already owned half of the cards on your list and you successfully convinced me to get a A310. 😂😂
In my Handbrake testing, I noticed intel quicksync using an A380 renders about as fast as my nvidia cards but quality per bitrate was much better with quicksync over nvenc. Like 25%+ better. So I've stuck with the A380 for that. (I'm encoding everything to .264 for compatibility) Until basically everything is HVEC compatible, I'm not going to re-encode. But storage keeps getting cheaper, so I'm caring less and less about file size. 🤷♂ I haven't had any issues with the intel card using handbrake GUI. I also have the 3050 6GB for one of my SFF computers, it's not bad, even for light gaming.
I bought the a310 for transcoding av1 and it is by far the fastest. However, must use newer linux kernels since the i915 driver needs to be on the newer side.
For the blender tests I think score per watt is more relevant. It makes no sense to have a column of watts and one of score without a correlation. You could show a horizontal graph with score and power mentioned at the end (no representation on the graph) and then a graph with score per watt.
The performance of the integrated graphics makes me wonder how this comparison would have done with an even slightly more recent cpu (or an amd chip). Also, maybe im misremembering, but I think I've heard that some systems are able to cut power off to certain components; with such a system-- or even with just a system that let's you use two gpus-- I imagine the best option for efficiency might be to use the igpu for transcoding videos and only turn to the discrete gpu for the rest.
The iGPU staying at least in shouting distance of the discrete cards for many cases is kind of even more impressive to me. And then there are the 720p cases where it wins over all but the A310! Sure, video decoding is probably one application is was specifically designed for but still...
One thing I am missing in your video, is mentioning the different codec support between nvenc and nvdec generations. Also there is quite some difference between those generations, in terms of picture quality. You just compared the framerates. I know this would make the video much more complicated to create, but it should be mentioned. In this context I wouldnt recommend a P620 anymore. (T1000 also not really)
Great video, much appreciated! I am curious about the Quadro P1000, seems to be another single slot, low power option that is a bit newer than the P620. Also I have seen tons of complaints about the fan in the Sparkle A310 Eco card; did you experience any fan noise issues with the card?
I would have liked to see a benchmark of the 4060 low profile since it is not that expensive and has more VRAM and can do a ton of the new Nvidia things like ai super resolution and such
I would have loved you to have included the Tesla P4. They were designed for encoding servers and are low profile. I got a new old stock on eBay a few months back for about 100€ delivered with a blower fan. It’s basically a down locked GTX 1070 with no output/6-pin power/etc. I do use it for gaming as well, but mostly for ML and encoding and it is fast. Perhaps in the middle of the pack, but tiny, sips power and literally 100€. I think I have 6 watts idle and even Balder’s Gate 3 on Ultra it’s around 40 W. Says it will max at 70, but never saw that.
Very nice video with a lot of useful info, thank you. Please consider in future adding even failed items onto graphs(with some fake value and remark that it failed the test). This way you increase graphs consistency and columns order.
Thanks for the work on this. I am not sure if you had your power management set for the Intel card correctly. My plex server is on Windows OS with an Arc A380 GPU. Idle system with monitor turned off, pulls around 3.5w. This is with PCIe power management set to on in the OS.
I got an Nvidia T1000 GPU for my home lab PC, and it's been working great, especially for Jellyfin transcoding. The best part is I got it for only 100 USD, which is a steal considering how well it performs. It handles the media tasks smoothly, and its power efficiency is perfect for my setup.
I would like to see tests regarding multiple simultanious transcoding (2 and 4) tests since your already using jellyfin and more AI tests - also with more streams and AI tasks. Anyway - great video!
Fantastic video! Not quite Project Farms level of comparison, but very good. I've got a few SFF PCs and would like to put a GPU in one, but it'd have to be half-height and single-slot. This is a video I'll have to watch again if I get the chance to purchase something.
More like this! I home lab primarily to learn. As such performance per watt is the metric I pay most attention to. I genuinely appreciate that you included power draw at idle as well. While I may be trans coding video a few days a week, I am certainly not doing so 24x7. Is the A310 the most power efficient option? Clearly it depends on how much time it is active versus idle.
I’ve been looking for “upgrades” for my old 1060 6gb in my once gaming pc now home server for plex transcoding and hardware encoding. Thanks for this video!
I have seen someone get the Quatro P400 working on a Synology NAS, and I assume the P1000 would also work, but not sure the Intel A310 would. It sucks that there are no options for AV1 transcoding on Synology NAS, and they don't even have any CPUs with QuickSync either.
Awesome video, I just got a A380 that I'm hoping to use. I wonder if Intel wanting ReBAR support from the CPU/Mobo was the source of some of your Arc issues.
Great video. I don't plan to do much of transcoding or even video editing. But I do plan to build an NVR with several reolink cameras. So it would be great to see a video comparing the inference times for object / person detection and comparing the performance of these GPUs against the Coral TPUs shown in a previous video. Keep up the great work. 👍
Thoughts on the A380 vs A310, seems like some more performance was available within a similar price class? Also, have you looked into cartwheel-ffmpeg by intel?
the i7 8700 was released in 2017. How would the integrated graphics of a more modern cpu perform, including the low power/budget n95/n100/n200/n97 options found in most mini pcs? To be devils advocate, it looked like the performance of the igpu was adequate if the system was only transcoding for one stream, perhaps multistream performance would increase the differences. In my home set up, there are only likely 1-2 streams at any one time, so anything on top of that is just wasted power (W). Thanks for a great video
From what ive heard theres some sff(small form factor) PCs that have the x16 pcie slot up against the power supply which means you can only 1 slot cards in that slot. You can mod them if you know how with the right type of power supply. Thanks for the information. Im looking at setting up a home server for family files which wont be very big and as a media server. Its a i5 4690 which will have a rx 480 to start with. I used to use it as a plex server. Now im thinking maybe getting a different GPU thats designed for the job. Hmm i need to research the topic, im on a tight budget so maybe cheap quadro.
I guess this is very correct for a PC Server. However I'd say an Intel-based Synology streams 4K and several 1080p transcodes smoothly 24/7 through plex and the iGPU. How can we compare the two? If it works fine, do we need huge numbers or are we "future proofing" right now for 8K that we don't have?
For anyone looking for a low profile card for the Minisforum MS-01, the Intel ARC 310 ECO from Sparkle, featured in the video, fits and works well. Some/many/most/all (not sure) of the low profile Nvidia Quadro cards fit but do not work in the MS-01. The machine does not turn on with the card installed or otherwise does not function properly. My environment: Two MS-01 machines, Core i9, each with 96GB RAM, three M.2 SSDs, running ESXi 8.0 Update 3. Each has the Arc 310 ECO passed through to a Windows 11 VM running Plex. I was unable to get the Arc 310 ECO running under Windows Server. No working drivers.
This is EXACTLY something I've been wondering about - especially if any of these would help with transcoding - or local AI
Well hopefully it was helpful for you!
Would be really nice to have some metrics for local AI usecases :D
@@hermesvoeglein535 There you need as much gddr as possible
You want AMD HIP or Nvidia Cuda and as much ram as possible.
The local AI space changes quickly so a benchmark made today could be outdated in a few weeks.
@@hermesvoeglein535
Me to I was doing some research yesterday 😂
The Raid Shadow legend joke was amazing, made my day.
No way you used the 3D printed PCIe display stand I put on Printables for the GPUs, I'm so surprised to see it in the wild.
Thank you
I snagged an arc A380 low profile for $110 a few months ago for my Plex server. Thing is a transcoding beast and only pcie power required. The few bucks more for the few extra gigs of GDDR6 ram really is worth it when the family is streaming multiple videos.
Just got the same thing! Excited to try this out!
Do you know how it handles emulation? I've been wondering for a ps3 emulator, and then maybe lowend steam games at 1080p
That's the one I'm aiming for
The vram isn't really used in media applications. That's why an integrated GPU with zero vram still works fine for multiple streams and transcodes. It's also why the p400/600 2GB is a great deal.
Are these intel graphics cards any good for image processing in Photoshop?
My first checks for getting the A310 transcoding to work on Ubuntu: 1) kernel version, 2) necessary firmware bundles. And if you have jellyfin installed it has its own version of ffmpeg. But I dunno, I'm running it on Debian testing and it just works after I installed the necessary bits.
Does your intel arc have the fan revving issue and if so did you find a fix it?
AV1 is honestly a game-changer in my opinion. I got an A380 for $110 and the performance in Jellyfin is excellent and the AV1 streams it produces look genuinely amazing. At one point I was comparing different stream resolutions in Jellyfin and it seemed like the resolution wasn't changing, then I realized that it WAS changing but was having trouble seeing it because I was looking for compression artifacts rather than resolution. I am so used to associating low resolution with compression artifacts (blockiness) that seeing 720p or 480p without compression artifacts genuinely broke my brain for several minutes.
So basically, if you're looking for a GPU for your media server, I'd really recommend prioritizing one that has AV1 enc/dec. Intel has made AV1 very accessible.
Does your intel arc have the fan revving issue and if so did you find a fix it?
@@Amwfilms tbh I haven’t so much as heard the fan at all. I’m also not stressing it very hard, just jellyfin at the moment, so maybe that’s why.
@@Amwfilms update the firmware on your card, it is likely running an older firmware.
I love AV1 but I'm not sure if decode is all the way there yet...
@@Amwfilms following up this. So the A380 didn’t have the fan issue at all and I’d never even heard of it. Well, I just got a Sparkle A310 ECO and it definitely does have the fan issue. The fan is very audible even at idle and it’s constantly doing this ramp up/ramp down thing which was very annoying. Haven’t tried updating the firmware yet like others have recommended, but hopefully that fixes it.
Bro were you in my search history this morning?
Same thing happened to me hahahaha
Nah but TH-cam was
Nah I definitely said the same thing. Was quite literally looking up good server GPU's last night before bed.
Exactly the same ahahahah
@@ironcrafter54 You took the words out of my mouth.
Really useful video! Many of the issues you were having with the Intel A310 is probably due to the lack of Resizable BAR on your motherboard/CPU. Transcoding still works great (I currently have a $110 A380 in my Jellyfin server) but any 3D rendering will suffer or fail.
As a German I do love the efficiency angle of this video. Our power is expensive so I am currently using a N100 mini PC that takes care of all my transcoding. The idle power and full system power of it is unbeaten.
I loved seeing Arc being finally included into these benchmarks but I would have really loved to see the A380 perform against the other cards instead of the A310, especially considering its price point is much closer to the other cards. I think some of the problems you had on Arc could've been solved through some tinkering with the linux kernel but that's not really straight forward ( I'm saying this cuz I've heard of people getting the A380 down to 1W reported power consumption ). Also I would've loved some AV1 decoding benchmarks! Considering that's one of Arc's biggest strengths.
I mean media encoding and decoding performance should be the same since all arc cards have the same two media engines but the extra vram could've helped in some cases potentially. I do wish more time was spent getting the arc card working because I feel that those are the main reason people would be watching this video. Not too relevant but it would be interesting redoing these tests with a 12th gen intel CPU or higher to see how the arc card would perform with deeplink/hyperencode.
@@GaiaGoddessOfTheEarth sadly hyper encode/decode are only supported on windows an last time I checked ( months ago, should probably check again ) support for them on Linux was considered non priority ( aka you'll be lucky if you ever get this )
edit: just checked, last year Intel said Linux support currently not planned as they'd rather focus getting drivers finished and stable first ( which is reasonable considering they've only recently almost finished a complete driver rewrite ) but last month an Intel employee asked for the use case of hyper encode on Linux so not all hope is lost
That's a FAT card lol
@@zenginellc not really, ASRock has a really nice pair of low profile a380 models
@@fedferhow the heck does an A380 fit Into a pc?
0:52 When I heard Raid Shadow, I almost skipped ahead
ya, raid shadow is trash.
You know a game is bad if you've never seen porn of it
Aahaha
@@Reckless150681what about overwatch?
It was definitely a Good laugh. Almost as good at raid owls bottom mount pdu solution, the power bottom.
Yes the A310 is indeed "made for" home plex and jellyfin servers if you ask me. But i also found out, there is also a very low powered version of the A310. That has a base clock of 1ghz. Think it was also the sparkle.
A310 ECO
@@tim3172 i think its thats one. But in this video, does he use the Eco version or the "nomral 2ghz" version?
The fan on his says eco
@@wojtek-33 then it performs damn good!
I think you're confusing it. the A380 is the one thats at 2Ghz, the A310 is the one thats base 1Ghz. I think Asrock sells them, tho they use the exact same design
I believe the mentioned i5 and i7 has the same integrated iGPUs, so should really matter which one you tested, however would nice to see a comparison between iGPUs at this point, as it seems like a very suitable choice just for jellfin
The extra L3 cache might have a very minimal impact?????
Arc310 hands down for less than $100. Now the only reason why I don't run this is because TrueNAS scale doesn't support it yet. Eagerly waiting for 6.2... Until then my Tesla P4 has been doing great for $50 GPU.
$50? Hard to find that thing for less than $100 on eBay unfortunately.
@@ramtek55 I probably should have mention I bought this last year. So prices have probably slowly ticked up lol. Sadly low watt PCIE powered GPU's are a hot commodity now. Everything has to be power hungry now.
Truenas Scale definitely supports the Arc cards in Dragonfish stable train.
I was sad to see the arc wouldn't work in frigate, unfortunately i guess its the 3050 for my Truenas
Wait.. TNS doesnt support the A310?
Literally have spent hours looking at different models and guides regarding low profile cards that don’t need a power supply connector. This is a great resource! Thanks!
I recently purchased (for $123) a new Nvidia T400 for use as my daily driver (an upgrade from Intel 14500's UHD 770). I'm not a gamer, so integrated graphics is generally powerful enough for my typical usage (fine for triple 2K monitors playing TH-cam videos), but performance with a few dozen Chrome tabs open is noticeably improved with the T400. The T400, which employs the TU117 (same chip as the GTX 1650 and the T1000), benchmarks more than 50% faster than the UHD 770. The card's 4GB of dedicated GDDR6 video memory makes a difference.
I did not want to heat up my office space, so I sought a low-power card, and the T400's 30W max power draw ticked the box. I had built my PC nine months ago with the intention of using only the integrated graphics, and finding a suitable motherboard with triple display outputs really constrained my options. Desirable motherboards generally offered at most two video outputs. Had I known that I'd be using a T400, I would have selected a more sophisticated motherboard with Thunderbolt, not caring about integrated video ports.
But is the T400 good? Even though it's 4 GB, it's still good for games, right?
@@ggkkskmom would be comparable to GTX 1650, so not a AAA gamer by today's standards.
People REALLY underestimate the speed and quality of intel's quicksync. Plus they commit drivers right to the linux kernel.
I hope you don't mean... visual quality. If you're impressed by QS's visual quality vs NVENC, then I'd strongly suggest availing yourself of the services of several optometrists.
(Yay, unstable and buggy drivers right into the Linux kernel!)
@@tim3172About quality, for H264 and H265, Nvidia is clearly better than AMD and Intel, but on the AV1, Nvidia and Intel are pretty much equal, and AMD is not that far behind. So if we're switching more and more toward AV1, Intel iGPUs are actually gonna be good for transcoding
@@tim3172Cpu encoding is even better
For the price and power draw, the iGPU really came off well in this video
Perfect timing. Setting up my first server as we speak!
For frigate object tracking, I wish you had compared the GPUs versus a few different Corals (m.2, usb, etc).
If you're wondering why there's no AMD cards like the rx 6400 or 6500 it's because they don't have dedicated video encoding hardware at all. Well done, AMD!
They have the Radeon Pro W7500 with AV1 encode and staff.
@@lharsay yea but that's no longer an SFF (or low profile) card. And they don't have an SFF card in the entire 7000 series.
@physxme NVIDIA doesn't in the 4000 series either. I think the 3050 is the newest desktop card from AMD or NVIDIA which is. I think the 3050 and 6400 are the only bus powered cards as well?
RX 550 has both
@@pkt1213 Nvidia has the RTX 2000 ada which is technically an RTX 40 series card and it's bus powered. It's just expensive as hell.
P400 here, runs great in my unraid/plex machine. Multiple transcodes is not a problem.
What are you getting for the power draw of the P400 at idle and load?
@@nighthawkvc25a curious with the power draw as well
@@nighthawkvc25a With unraid and 8 spinning disks in idle, about 65Watts (disks spun downs) in full use with 9 dockers and transcoding about 115Watts. Also using a script to clock the p400 down when in idle
don't forget about the tesla P4 love that little card
Agreed! I love the low profile, single slot p4. Great card at 100usd with 8 gigs of vram
I think that one thing that you forgot to mention for each card are the CUDA cores/Shader cores/Tensor cores as this is what most people with a homelab would ultimately be looking at for things like transcoding or local AI. Overall, a good video as there aren't too many videos out that will go over all of this in a non-gaming capacity. I use a P400 for Plex transcoding and I just setup Jellyfin and run a P400 with that one as well. I also have a pair of M40 12GB Teslas that I run for Local AI for Ollama and InvokeAI.
Great video! I would love to see these tested with a local LLM of some sort. I'm interested to see how many use cases a person can get out of one of these.
I could be wrong, but don't most LLMs need at least 12gb of VRAM? So none of them would work with it.
@@Afsafs123At 8GB of VRAM you can actually start using quite some models, but yeah 6 and less won't be able to run much
I got a Tesla P4 for $75 and the cool part about it is that it has 2 encoder chips on the card. It also runs off the PCIE slot so no power cables. The downside is I had to 3d print a shroud to use a blower fan into it to lower temps since no built on fan. Works great though.
One caveat on the A310 (and other Intel Arc GPUs) though: Make sure your PC supports Resizable BAR (ReBAR) or Smart Access Memory (SAM), or else its performance will tank pretty hard even on decoding/encoding.
On another note, if you really want to go AMD, the Radeon Pro WX4100 does have decent hardware decoding/encoding capabilities, just be aware it's a Polaris GPU so its not gonna be as feature rich as the newer cards. You do get a more hassle free driver experience in Linux though.
Hey, sorry just starting along this journey. I have a dell server, I was wondering how I check if it supports ReBAR or SAM?
Anything but NVIDIA. Glad that option exists in this segment now
Indeed for a multituden of reasons
Many of those of us inthe industry have refused to support team greed for many many yrs nwo due to their pathetic bs tactics
It's many many yrs now since they tried blacklisting me for example since I refused to play their bs games and forgive them editorial control and dictating how we would do tests and only their approved tests I and many other told them to shove it of course
The sad part is they have only consistiently gotten worse over the yrs
Fun fact, at 15:50 "samples/minute per watt" is actually measuring energy per sample, or rather its reciprocal. Divide by 60 for samples per (watt second), which is the same as samples per joule. 1 divided by that gives joules per sample.
Doesn't tell you anything different, of course. I just like the way these units work!
The intel arc-line of products seem like a crazy good deal. I guess I should quickly get one before Intel starts feeling entrenched and monopolistic and starts pulling Intel CPU division / nVidea stuff
they pulled their gpu line already, no more intel gpus will be released. the designs that are out are available from board partners,but intel is not designing another dedicated gpu.
@jmwintenn searching Google for Intel Battlemage yields nothing of the sort.
@@jmwintenn source?
@@dylantrom4333 I’d bet he’s getting that from MLID, who has it out for Intel in general, and Arc GPUs in particular. I mean just compare how he talks about RDNA 4 and Intel 20a:
Amd has scaled back on RDNA 4 so they can put more effort into RDNA5, which MLID says is going to follow much quicker after RDNA4 (like 6 months instead of 2 years): MLID can’t stop talking about how smart this plan is.
Intel announces that development of 18a is going so well that they’re halting all development on 20a to focus all of their efforts on 18a, which will be out some 3-6 months after when 20a was planned on launching: MLID: “the sky is falling over at Intel folks, they’re basically in chapter 7 bankruptcy, we’re going to go from a Intel v AMD duopoly to a Qualcomm v AMD duopoly. “
I mean, FFS, I’m pretty sure Dave2D’s video showed Lunar Lake CPUs beating current gen desktop CPUs in single core performance, and MLID: “Intel has given up on Single threaded performance folks, Intel CPUs only exist to host Nvidia GPUs from this point on”
the P600 (90 USD in 2020) has been excellent for my Plex server. I always transcode on the fly for external streams, and it's handled everything without missing a beat! The CPU transcoding (Haswell Xeon) before was crazy hot, loud and power consuming. It's what I'd buy again if I needed to replace it, especially at today's prices.
Something to remember about putting a gaming class card in your server is that you are leaving almost all of that performance on the table. The encode/decode parts are separate from the general processing parts, so you're paying for performance you may never use, on top of the far greater power draw.
If you're holding on to the A310, I'd love a Battlemage comparison after the comparable version is released for a few months, personally want to see if idle power has been reigned in and are less issues
Nvidia: Famous inventors and scientists that changed our way of life.
AMD: Exotic locations and star clusters
Intel: Neckbeard DnD loser nonsense.
Shame.
Always excellent. FWIW, another option:
You can often pick up a LP Tesla P4 for very little - superglue a 40mm blower to the end and you're good. To get one to run as a GFX card is not hard. My home server has one in passthrough to a VM running a stripped-out Win 11 + Sunshine > Moonlight for network emulation / games. Also as a transcoder in my Jellyfin server - these cards come with lots of RAM, making sharing between VM's practicable.
Was that the 3050 that didn't fit in the HP EliteDesk? Did the A2000 fit? I'm also surprised that there was so much issues with A310. I know Intel has been putting in a lot of work to get it supported with more things so I figured it would have been able to do all the tests that you did for everything else. I'm planning on using Plex and as long as the A310 works that's really all I need in the short term.
The A2000 and 3050 6GB (at least the one I had from MSI) were the exact same size essentially. And it wasn't the EliteDesk, it was a lenovo P330. I don't currently have an EliteDesk to test fit it in sadly, otherwise I would let you know!
@@HardwareHaven ah ok the hard drive cage opening made me think that scene was from an EliteDesk.
Another commenter said the issues were because his motherboard doesn't have a resizable BAR, or at least didn't have it enabled in the BIOS.
Thankyou for this vid I had been longing for this!
Thank you for putting this together! I ran into problems when my P400 died (well, I THOUGHT it died anyway) and I was trying to decide between a P620, A310 and an RTX 3050. After a bunch of reading I got a P620 based on it's acceptable level of performance, low price and low idle power draw when compared to the A310.
Only THEN did I realize that a TrueNAS update had altered the way the NVIDIA driver was configured in JellyFin. I had been using CPU for transcoding for a month because I never function tested after the update...
now im very interested in seeing how the igpu in the 8700k compares to that of a newer gen intel desktop cpu (maybe the new core ultra 200 series that is rumored to come out soon) and the aliexpress intel laptop cpus that are frankensteined onto desktop boards since laptop cpus tend to be more power efficient and have better igpus than desktop cpus.
Did you not find a suitable Intel A380 option? I've been told the 380s are easier to find than the 310s, and that they're even faster and better....and They also do AV1 encoding.
0:52 you had me in the first half
I bought an A380 for my old Lenovo office PC (i7-3770), but haven't even tested it in that system yet. I can confirm that it's a bit weird. In my test bench, it would blow through some benchmarks and then just choke on others. Not sure why. I guess if you support Intel, you'll have to put up with a few snags for a while. I've been using the A770 since it's release and love it.
The one thing about these videos that everyone makes a mistake on is it's more about single slot than it is about low profile. Just getting a cheap case with full height PCI slots isn't a problem, but trying to fit a GPU in with a bunch of HBA's and NICs is a lot more difficult.
As a follow up I suggest stressing each card to see how many streams from Plex or jelly they can run simultaneously
I've been trying to figure this out for over a year now, thank you so much man.
Hey. Thank you for the great videos. I Just had a little nitpick with the object detection value proposition ( 19:10 ) chart. The dollar/ms metric is not a useful parameter since we want to minimize both of these values under ideal circumstances. Therefore, they should either both be in the nominator or the denominator. The dollar*ms would have been a useful parameter and the card with the lowest metric would provide the best value proposition.
Also the samples/minute per watt ( 15:52 ) is basically equivalent to samples/joules which is a very useful metric (you have to divide the values by 60 to be precise) .
This is the exactly video I needed to see today to help me with my build choices.
Thanks!
loving the new logo
Fantastic work, and very interesting. Thank you for this. Something which would help a lot would be a simple text on each graph saying "higher is better" or "lower is better" to make it easier to follow along. When watching while doing chores, it's very easy to get confused.
I would be interested in seeing more of the Intel Arc GPUs being benchmarked. As well, see how they perform on some AI workloads.
Any chance we can get a Tesla P4 added to the mix? It has 8GB of VRAM, a 256 bit bus, 2 NVENC chips and 1 NVDEC; and can be had for under $150, including a cooling solution, all day long.
Nvidia is still king for Pass Through in Linux IMO. I make due with a P400 and its only job is to transcode files for a few family members who are remote and when I want to watch something on the go. Can even handle several transcode at once.
Thanks for the confirmation that i chose the 3050 for what i want correctly. Still thinking of getting a A310 for my Jellyfin though. Now it is looking a bit better also.
I was literally wanting this video last week. Great timing!
Arc A310 best for pure max encoding/Transcode performance.
P400/P600/P620 for "value" transcode
For a blend of Raster AND Encode with a little DLSS in there... maybe the 3050?
I would say if you have a headless multipurpose workstation and remote 1080p gaming... that 3050 is actually a good price.
The A310 can actually game too, but keep expectations VERY low.
And if you can find an A2000 near the 3050 price... get that instead if you were considering the 3050.
thank you for talking about lower powered controllers. I have never really been a true gamer. My needs are small. Just doing internet, home work ( office type of stuff ), doing sw mainteance of computer, watching videos, and lately - a little transcoding. If I play games, the type are 20+ years in sophistication. My most advanced card is an NVIDIA Geoforce 730 and the integrated graphics on an Intel I7-8700. the 730's cost me about $40 on ebay - and I think that was newish - old cards - never used. Very happy with both levels of video.
I went p620 to replace some older “bigger compute” GPUs that were drawing 34 watts at idle. P620 is drawing only 7.8 watts for me in a TrueNAS Scale server. With my electricity of 37¢/kWh I will save $84 per year.
Thank you heaps for putting in the time and effort for this video!
This helps quite a bit for when I'm able to move away from a card I picked up for cheap as it was used for mining.
For me the A310 wins, when considering, it can also transcode AV1 very fast.
Do any of these cards support package c states below c3? I can’t find any cards that do. You would gain massive system idle consumption gains if such a card exists.
As a feedback to your graphs, while they are looking nice, I lack the information of y axis. E.g: 8:19 where the y axis represented frames per second, I wasn't sure what I was looking at as I missed the FPS part that you said during the video. Having that shown somewhere on the graphs would be helpful.
did you try the jellyfin version of ffmpeg for the a310 for the encoding tests?
A benchmark for llama 3 or some other llm would be handy, its a little bit of a hat trick right now but home assistant lets you use a locally hosted llm as the conversation agent
I use intel integrated gpu for ffmpeg by using the jellyfin version of ffmpeg. It usually works given that jellyfin also works. I *think* you'll need to re-compile ffmpeg to add a few flags for those intel cards to work.
How well supported are the A3x0 GPUs in linux/bsd at this point? Because last i looked into it, there was little to none.
Also, AMD just works on Linux/BSD, so it seems incredibly shortsighted to limit yourself to nVidia and Intel only.
This is exactly what I’m looking for. Many channels tend to overlook sff cards and don’t really talks about idle power consumption at all. I already owned half of the cards on your list and you successfully convinced me to get a A310. 😂😂
The HDMI cable being plugged in was a good catch. Appreciated
In my Handbrake testing, I noticed intel quicksync using an A380 renders about as fast as my nvidia cards but quality per bitrate was much better with quicksync over nvenc. Like 25%+ better. So I've stuck with the A380 for that. (I'm encoding everything to .264 for compatibility) Until basically everything is HVEC compatible, I'm not going to re-encode. But storage keeps getting cheaper, so I'm caring less and less about file size. 🤷♂
I haven't had any issues with the intel card using handbrake GUI. I also have the 3050 6GB for one of my SFF computers, it's not bad, even for light gaming.
I bought the a310 for transcoding av1 and it is by far the fastest. However, must use newer linux kernels since the i915 driver needs to be on the newer side.
For the blender tests I think score per watt is more relevant. It makes no sense to have a column of watts and one of score without a correlation. You could show a horizontal graph with score and power mentioned at the end (no representation on the graph) and then a graph with score per watt.
The performance of the integrated graphics makes me wonder how this comparison would have done with an even slightly more recent cpu (or an amd chip).
Also, maybe im misremembering, but I think I've heard that some systems are able to cut power off to certain components; with such a system-- or even with just a system that let's you use two gpus-- I imagine the best option for efficiency might be to use the igpu for transcoding videos and only turn to the discrete gpu for the rest.
Cool to see Intel carving out a legitimate niche for themselves. Great video, low profile stuff doesn't get enough love.
Running an A380 for Plex is currently unstable w/ Plex 1.4.x, as it's "technically" unsupported for now, despite working well on older releases.
The iGPU staying at least in shouting distance of the discrete cards for many cases is kind of even more impressive to me. And then there are the 720p cases where it wins over all but the A310! Sure, video decoding is probably one application is was specifically designed for but still...
One thing I am missing in your video, is mentioning the different codec support between nvenc and nvdec generations.
Also there is quite some difference between those generations, in terms of picture quality. You just compared the framerates. I know this would make the video much more complicated to create, but it should be mentioned. In this context I wouldnt recommend a P620 anymore. (T1000 also not really)
Great video, much appreciated! I am curious about the Quadro P1000, seems to be another single slot, low power option that is a bit newer than the P620. Also I have seen tons of complaints about the fan in the Sparkle A310 Eco card; did you experience any fan noise issues with the card?
I would have liked to see a benchmark of the 4060 low profile since it is not that expensive and has more VRAM and can do a ton of the new Nvidia things like ai super resolution and such
I would have loved you to have included the Tesla P4. They were designed for encoding servers and are low profile. I got a new old stock on eBay a few months back for about 100€ delivered with a blower fan. It’s basically a down locked GTX 1070 with no output/6-pin power/etc.
I do use it for gaming as well, but mostly for ML and encoding and it is fast. Perhaps in the middle of the pack, but tiny, sips power and literally 100€. I think I have 6 watts idle and even Balder’s Gate 3 on Ultra it’s around 40 W. Says it will max at 70, but never saw that.
Love to see what the amd 8700g cpu with igpu looked like when put next to the chip & gpu combos used here in both price & performance.
Got a Quadro K2200 and it does everything I need it to do in my Z440.
I have a friend who's been looking at options for a plex server. I'll be sending him this video. Thanks for being awesome and making this so simple!
Very nice video with a lot of useful info, thank you. Please consider in future adding even failed items onto graphs(with some fake value and remark that it failed the test). This way you increase graphs consistency and columns order.
Thanks for the work on this. I am not sure if you had your power management set for the Intel card correctly. My plex server is on Windows OS with an Arc A380 GPU. Idle system with monitor turned off, pulls around 3.5w. This is with PCIe power management set to on in the OS.
I got an Nvidia T1000 GPU for my home lab PC, and it's been working great, especially for Jellyfin transcoding. The best part is I got it for only 100 USD, which is a steal considering how well it performs. It handles the media tasks smoothly, and its power efficiency is perfect for my setup.
I would like to see tests regarding multiple simultanious transcoding (2 and 4) tests since your already using jellyfin and more AI tests - also with more streams and AI tasks. Anyway - great video!
I like the vocal disclaimer since i listen to some of your videos like a podcast for the info dump
What I really missed is the Tesla P4 and also some AMD offerings like their APUs to see if its even an option.
Fantastic video! Not quite Project Farms level of comparison, but very good. I've got a few SFF PCs and would like to put a GPU in one, but it'd have to be half-height and single-slot.
This is a video I'll have to watch again if I get the chance to purchase something.
More like this! I home lab primarily to learn. As such performance per watt is the metric I pay most attention to. I genuinely appreciate that you included power draw at idle as well. While I may be trans coding video a few days a week, I am certainly not doing so 24x7. Is the A310 the most power efficient option? Clearly it depends on how much time it is active versus idle.
I’ve been looking for “upgrades” for my old 1060 6gb in my once gaming pc now home server for plex transcoding and hardware encoding. Thanks for this video!
Yep. This answered a question I had, which is "Do I need a GPU for that?" and that answer is: Not really, nope.
I have seen someone get the Quatro P400 working on a Synology NAS, and I assume the P1000 would also work, but not sure the Intel A310 would. It sucks that there are no options for AV1 transcoding on Synology NAS, and they don't even have any CPUs with QuickSync either.
Awesome video, I just got a A380 that I'm hoping to use. I wonder if Intel wanting ReBAR support from the CPU/Mobo was the source of some of your Arc issues.
Great video. I don't plan to do much of transcoding or even video editing. But I do plan to build an NVR with several reolink cameras. So it would be great to see a video comparing the inference times for object / person detection and comparing the performance of these GPUs against the Coral TPUs shown in a previous video. Keep up the great work. 👍
Thoughts on the A380 vs A310, seems like some more performance was available within a similar price class? Also, have you looked into cartwheel-ffmpeg by intel?
the i7 8700 was released in 2017. How would the integrated graphics of a more modern cpu perform, including the low power/budget n95/n100/n200/n97 options found in most mini pcs? To be devils advocate, it looked like the performance of the igpu was adequate if the system was only transcoding for one stream, perhaps multistream performance would increase the differences. In my home set up, there are only likely 1-2 streams at any one time, so anything on top of that is just wasted power (W). Thanks for a great video
Would have loved to see the 4060 make an appearance, even though it has the 8p supplemental power
From what ive heard theres some sff(small form factor) PCs that have the x16 pcie slot up against the power supply which means you can only 1 slot cards in that slot. You can mod them if you know how with the right type of power supply.
Thanks for the information. Im looking at setting up a home server for family files which wont be very big and as a media server. Its a i5 4690 which will have a rx 480 to start with. I used to use it as a plex server. Now im thinking maybe getting a different GPU thats designed for the job. Hmm i need to research the topic, im on a tight budget so maybe cheap quadro.
I guess this is very correct for a PC Server. However I'd say an Intel-based Synology streams 4K and several 1080p transcodes smoothly 24/7 through plex and the iGPU. How can we compare the two? If it works fine, do we need huge numbers or are we "future proofing" right now for 8K that we don't have?
I have 4 Quadro P600s and 2 P400s deployed in Tdarr nodes. Great performance and for the price/power consumption pretty hard to beat for my use-case.
For anyone looking for a low profile card for the Minisforum MS-01, the Intel ARC 310 ECO from Sparkle, featured in the video, fits and works well. Some/many/most/all (not sure) of the low profile Nvidia Quadro cards fit but do not work in the MS-01. The machine does not turn on with the card installed or otherwise does not function properly. My environment: Two MS-01 machines, Core i9, each with 96GB RAM, three M.2 SSDs, running ESXi 8.0 Update 3. Each has the Arc 310 ECO passed through to a Windows 11 VM running Plex. I was unable to get the Arc 310 ECO running under Windows Server. No working drivers.