Это видео недоступно.
Сожалеем об этом.
This Tiny GPU Packs Some SERIOUS Power!!! - RTX 4000 Ada Generation SFF Review
HTML-код
- Опубликовано: 26 дек 2023
- Links to items below may be affiliate links for which I may be compensated
Check out the RTX 4000 Ada Generation SFF here: amzn.to/3NKozjb
Erying 1th Gen i9-11800H (ES 0000) 8-Core Motherboard Combo: s.click.aliexpress.com/e/_Dki...
Patriot Viper Blackout 2x8GB 3600MHz: amzn.to/479hAqU
nVidia RTX A2000 (Refurb) for only $299!: amzn.to/3RCb32c
Small form factor GPUs are a thing of the past. These days, GPUs are pushing 4 slots tall, and like Pontiac in the 90s, it seems wider is better. The RTX 4000 Ada Generation SFF is a professional card that challenges every concept of what makes a graphics card worth buying, and crams it all into a half height, half length package that draws only 70W of power. This thing is absolutely insane, and delivers unmatched power/L in your system.
Shout out to @ychto for making this video possible.
Grab yourself a Pint Glass at craftcomputing.store
What am I drinking???
Las Pinas Barrel Aged Ale from Pfriem Brewing (Portland, OR)
My review of the nVidia Tesla P4: • What is the nVidia Tes...
My review of the nVidia A2 / A16: • Cloud Gaming on Ampere...
Follow me on Mastodon @Craftcomputing@hostux.social
Support me on Patreon and get access to my exclusive Discord server. Chat with myself and the other hosts on Talking Heads all week long.
/ craftcomputing
A price to performance comparison of the RTX 4000 and the new 4060 low profile would be interesting.
there is a 4060 low profile ?
@@NunoAlmeida-sh9ru GIGABYTE GeForce RTX 4060 OC Low Profile 8G Graphics Card, 3X WINDFORCE Fans, 8GB 128-bit GDDR6, GV-N4060OC-8GL Video Card
@@NunoAlmeida-sh9ru yeah, Gigabyte RTX 4060 OC Low Profile 8G. The RTX 4000 Ada Generation has more VRAM and uses the RTX 4070 gpu chip with lower clocks, lower memory bandwidth/bus width, and less power.
@darkace256 In this video we have the RTX 3070 performance comparison, where it scores -15% on Graphics. However on the 3dmark Fire Strike database it shows other computers with much higher scores, so your overclocking headroom may vary it seems.
@@NunoAlmeida-sh9ru yup, I bought one for my uSFF rig.
@@NunoAlmeida-sh9ru Yeah, but it does need an 8 pin power connector, instead of getting all its power from the PCIe slot.
Thank you for reviewing these! I love seeing a gpu get outclassed by a lower power draw, more vram, and better performance - to me that is what progress is all about
Great performance. In about 3 years I'll be able to afford and buy one.
I hope that SFF cards will be more frequent in the future not just for workstation cards buty also consumer. For cunsumer right now we only have the RTX 4060 from Gigabyte and no one else and that needs 4 pin power.
Would be fun to have that sort of options, but there is no market for higher end SFF consumer GPU's really, at least not enough to support competition in it. Consumers have no reason generally to fork out a premium to go for SFF in particular not if it is actively cooled which makes it louder. Aside from SFF enthusiasts but that is a very niche market split even smaller if you take out the HTPC guys who'll go for passively cooled lower spec cards since they don't need it to do 3D stuff really. I don't think the LP 4060 is a money maker even as the only card in it's bracket.
I wanted one of these since it came out. Thanks for reviewing it !!!
Could you include some numbers on image generation with Stable Diffusion, running an LLM, Blender, along with a graph showing the relationship between power consumption and results for this GPU, please? Excellent video, as always.
Hi Jeff! This particular gpu isn't my cup of tea, but it is always good to know what it out there 😀 . Personally,I keep hoping we get desktop motherboards with a mobile gpu on board. That feels like it would be killer since it could be done great for small form factor, or for plex/storage build.
Been waiting for this ever since we drag raced the ex-boxes, those numbers are damn impressive!
Around 3060 to 3060 Ti performance for 70W is actually interesting, depending on what folks want them for. I assume they come with full height brackets as well? They'll be interesting to see what they fetch on the used market in a few years, and what the 5000-series SFF will bring in 2025.
36GB GDDR7 and the performance of a 3070 ti 😅
@@justinpatterson529136? That would be weird. 32 seems more likely.
@@justinpatterson5291 there’s actually someone on Reddit that has made a power mod for the 4000 sff that makes it a slightly overclock 3070ti. They’re also working on a copper cooler mod for this and I’ve talked to him about and I plan to get these two mods installed onto my card once it’s finalized. So far I’m loving the sff Ada but I’d really like the 3070ti performance since that’s actually the card side graded from
There will be no 5000 series, Nvidia is scrapping that branding.
@@HeyNavi I'm all for SFX PCs. Love the tiny form factor and twice or more now, nvidia has proved you dont need a massive card to get good performance.
Best GPU review of the past year! Really interesting for future SFF-builds.
Love this video. Amazing content as always Jeff!
[Angrily shakes fist] I settled for the 4060 LP personally
Edit: i kinda wish you had used a 4060 for comparison instead, but, 🤷
lol better than me, im stuck between deciding if i use a new case, but forces me to use either a a2000 6gb or a rtx 3050. since they are the only cards i own that fit my sub 215mm gpu requirement.
I had been curious about this GPU since it's release. Currently have a precision 3660 with an A2000 12GB in it. Have looked into the RTX 4000 and the RTX 4000 Ada versions. But for games, not much if anything I could find. Until now! It really shows just how good the performance per watt is on ada!
Love this video, thank you for covering these cards!
Hey man, absolutely amazing breakdown comparison video I have seen! I am looking to build a workstation and haven't paid tech any attention for 2 years😂 So here we go! Thank you for the content, and you just got another sub!
I love it when you switch tones completely when talking about the beer! Clearly... you LOVE beer! I do too! :)
We installed one of these in a build we just did and when we opened the package out first reaction was "there's no way..." Thing is a beast.
I would love to see this card in a dedicated gaming rig! I was so excited seeing this vid because I have played with slim factors for years and seriously want to go small with my next build. Bigger is not always better so yes...go for it on the next transition of this awesome card!!! And you're a beer guy?!?!? Subbed.😎👍
Ooo I want to see this in your cloud gaming build. This would be awesome in a 1U, split between two people it may give good performance with enough VRAM. I’d love to see two 1080p RTX games running on VMs at the same time!
The 20GB of VRAM would make this a perfect GPU for handling VR-modded Resident Evil or Cyberpunk, and being 4070/4080 class guarantees it'll handle it without too much interpolation.
This is nowhere near the performance of a 4080. It might have the cores, but it uses way less power. The 3dmark section shows the 4000 Ada sff losing to a 3070. Not great. Instead, I'd buy a 3090 or 4090 if you need the ram and the performance, just don't expect them to be efficient.
4070 and 4080 aren’t in the same class. There’s even a card, 4070TI, between them. It would have to be in one class or the other.
It's 3060ti performance so more like 4050 class
Excellent! Well done. Seeing how we don't have that many alternatives in this low profile and low power category (75W), I'll be upgrading from the A2000 to the A4000 when the 4k gets cheaper in the future.
I personally would like to see more gaming cards in such a form factor, unfortunately it may come at a steep premium unfortunately. Anyway great vid and it's always interesting to see how much you can get out of different architectures with such a low wattage and it honestly surprises me with this one.
I got an old 1050ti in this form factor, its still running strong, used to over heat though, had to mod it with an extra fan and heatsync. works good like that, funny thing it is.
RTX 3050 low profile is out now
I really like my Instinct Mi25 I purchased and flashed to a WX9100. Works great. Its onboard fan header runs the .8 amp fan no problem. Thank you for all your info you put out on that card. I may get one of these now. :D
Looking forward to getting one of these in about 2031.
Good to know SFF cards are still made. The last half-height single slot card I was aware of was the gtx 750 ti.
I think they're great. I'd love to build an SFF gaming machine around one of these. It's just... so cute! And yes, I'd absolutely support nVidia releasing SFF gaming cards.
great, now I actually need to pay attention to background stuff to see if I can figure out what you will review in the future.
would be interested in performance with cracking hashes, and ai acceleration. how would it fair in a zima edge? would be interesting if there was a way to couple it to a flashstor ssd nas for vms, docker containers, streaming media, etc
Thank you for the review.
Loved your video, and I’m subscribing. A great analysis, and I’m happy you got the card for testing.
Next year, I’m looking to build a low-power (
Great stuff as always Geoff (Jeff?). Re: the 4000, just amazing power in such a small package! But the one thing I was totally blown away by was the A2000's RT performance in Spiderman Remastered. Sure the 0.1% lows are pretty poor but I would not have thought it would be averaging 78fps
When did you get so many subs? almost 350k? Wow. Nice work man. You had a few thousand and a day job when i started watching you.
A great little card. I've been thinking about getting the single-slot full-height version of this for a server where I don't want to lose two expansion slots to one card. The price premium is spicy for both versions of this card but it certainly packs a punch for its size.
Plex server?
@@sagejpc1175 I was thinking local LLM's using Serge in a docker, image generators and also Plex. So a lot of machine learning (thus the 20GB of VRAM and good single slot performance) and some Plex transcoding for good measure. If I only needed a Plex card I'd definitely look at something way lower down the scale like a RTX 4060 (just for the newest transcoder).
you also got the RTX A4000 that is a single slot Card and it has The core of a RTX 3070 with 16GB GDDR6 ECC, and a Fan, ive been wanting one of those for my ai server, but their like 500-600$ Dx
Use 6 RTX 4000, 2 in my workstation and 2 in each server that are built as render farm, but I don't play games
The A-series card is also possible to NVlink for dubble the Vram and use two cards as one
Been wanting to test one of these. Really would love to see Nvidia release a low power gaming GPU with Ada architecture.
MSI Cyborg with 4060 45watt
I'd love to see how it performs as split to vGPU. Lots of VRAM gives hopes for some nice perfrormance. Next episode of gaming server?
@Jeff, you should take a look at the Intel ARC A30, A40, A50 and A60 cards with the A50 being the same size and form factor the RTX A4000 SFF. Obviously, they won't be as powerful BUT are cheaper options and the same form factor
I really would have loved some professional benchmarks like token rate in text generation, and it’s performance in stable diffusion
I really like my A2000 12G version for Stable diffusion. It's just nice to see more sff cards, since nearly all my cases are Media PCs/Servers with 2U
Also would be nice, to see Stable Diffusion Automatic1111 speedtest comparisons, because it uses other Strengths of the GPU, Especially vram Speeds.
I would love to see some more high performance low power cards. It might be out of the question for now, but one day we may be able to have a handheld with its own dedicated GPU.
As is, this could be a viable option for someone with enough budget who likes to do high tech camping due to the lower power draw. I personally would stick to current handhelds, but I could absolutely see this in a rig that someone takes on road trips with a camper.
I have the A2000 6GB and using it in an HP office PC with Intel 8700 CPU, it was my temporary replacement PC while i was rebuilding my gaming rig. It's solid enough for gaming, I even played PC VR games on it, but the more demanding VR games sometimes crash. I will keep it in that office PC or get a better one with better CPU and reuse it, it is a great GPU to have. It was good to see the comparison with the A4000 SFF card and i have been waiting for it for months already. Sadly the A4000 SFF is too pricey and just not worth it for us mere mortals.
Very interesting video. It’s awesome to see a jump as big as the RTX a2000 to RTX a4000 sff ada. That’s the difference between a gtx 1660 6gb and a 20gb RTX 3070
I have the 4000 sff ada in formd t1 v2.0 but now that I have such a small gpu, I really want to downsize even more and go for a velka 3.
There’s also a guy on Reddit who is currently working on a copper cooler and power mod for the 4000 sff Ada and it makes it draw more power and making it a slightly overclocked 3070ti.
I plan to get these installed once they’re finalized because my previous card was a 3070 ti fe and I definitely miss having those extra frames in single player games, especially since I’m on 1440p 240hz oled. For competitive games this thing is awesome and dead silent.
I currently got an rtx a4000 i have heard it does have heating issues any ideas on that and if it does is there anything that can be cone to compensate on a full air cooled system
I think the 4000 would be great for small but mighty video editing systems like Avid or Da Vinci
Very impressive performance, I wish I had the professional workload to justify a new gpu. With electricity costs and space heater temps in mind, 75 watts tdp would be a big upgrade. Sure 4k gaming is fun but in the summer 😢
I'd love to see what that thing can do on an ML compute analysis and rendering workload, since that is my use case, granted i have a VERY tiny multi-purpose DB machine that is primarily a compute and storage machine. It's an EPYC 3451 and only is a PCIE 3 system, but it does have 8 20 TB Exos drives, 64 gigs of DDR4 So-Dimm and currently has a 2070 ASUS ITX graphics card in it, primarily offloading long render time stuff, and ML Compute based DB analysis. and YES i still refer to it as ML, since i don't train for LLM, or graphics analysis, my ML is all Neural net Cuda compute.
Only speaking from the CAD software that I have deployed; Solidworks, NX, and AutoCAD for my users at work. These CAD packages run FAR better on workstation graphics that are speced below your average 3060 and they kind of run like crap on your normal off the shelf "gaming" cards. The reason is the workstation graphics drivers. There is some magic in the drivers that is not available on consumer cards that allows them to run better on these CAD packages. Even on just a single 3D, relatively non complicated, model (not a large assembly) the performance difference is visibly noticeable. Exactly the same performance difference that can be likened to seeing a lot frame drops in a game. If all you need is CAD performance for these CAD packages, it just needs to be workstation class and I would never recommend that price tag for this use case. I'd just spend the dollars on something 3-4 generations old in the mid tier of workstation class.
I have two of the A2000s and they're slick! (its comparable to the 3050) but seeing this, I might have to get one for a future SFF build
This video popped up on algorithm and I got my beer prepared! Let's go!
2:28 isnt there something in Skylake onwards taht lets you output that cards output through the CPU hardware to the mobo video output? im pretty sure i saw TechYesCity do this with a Polaris 21 mining card.
I would really love to see a side by side PCB comparison between this card and A2000! They look almost identical, I would be curious to know if they used the same layout.
How were the gpu and memory temps on the rtx 4000 sff?
Cool stuff and quite stacked
Nice to see some testing on this. According to Tom's Hardware it should be as fast as the 4060 LP, and it would be nice to see how far off they are.
My bet is this is faster than the 4060 LP. I'm curious to get my hands on one and find out.
I'm grateful for any card with credible performance that does not need a power station to run. Amazing what Nvidia did with just 70W.
Nice review of what looks like an interesting move from Nvidia future is bright for single slot lower power consumption cards it’s seems
That is an insane GPU. Super cool
I know that gaming benchmarks get the views, but I would have been interested in seeing how it compares in AI acceleration/inference tasks compared to the A2000 (both due to increased cores and memory). I mean, that's what these were designed for, as you say.
I made a similar post. The only reason I watched this review was for gauging AI performance to determine what card to buy for a server I'm building for local LLMs and Diffusion. I hadn't searched the entire internet, but it seems like AI benchmarks are relegated to the hardcore review blogs like Tom's Hardware and Anandtech, but I'm hoping these metrics will start getting more mainstream attention.
Well, whatever, it's annoying, but maybe it's fine if this stuff (like a lot of developments in AI) are murky and in the dark to the general public for now for "job security" reasons anyway. I suppose it's qualitatively harder to performance benchmark for as well, considering that there's no retail software available that generalizes all of the tasks into a discrete number that I'm aware of (for games, it's mainly just "how many triangles can it draw and how fast"), and the amount of VRAM and bus bandwidth in a card can be just as or more important for performance implications than raw compute power (as opposed to gaming where it's mainly just a bottleneck-preventer). @@reviewaccount469
How do they compare to a 1070 mini itx? I think that's kinda the last medium/high end card that was made in itx form factor and it's what I'm using in my compact travel pc
Can it or any SFF GPU do video transcoding in a HP Microserver Gen8 without needing to disable the onboard video or iLO?
This thing is sick! It would have been cool to see how it performs in a home media server serving a Plex media server etc.
I have a 4 litre uSFF gaming PC that has been home to a GT1650 LP, an RTX A2000 and, most recently, the Gigabyte RTX 4060 LP, all of them run fine on a 250W GaN PSU. I would love the RTX 4000 SFF, but it cost four time the price of the 4060 LP and, if my A2000 was anything to go by, the 4060 LP is much much quieter. The one thing I didn't like about the A2000 was the minimum 30% fan speed ( yeah, I modded mine to run the fan off a motherboard header, via an adapter cable).
However, even undervolted to ~ 90W, the 4060LP still draws more power than the A2000 & A4000 SFF. I'd love to see a third party A4000 SFF with the same cooler as Gigabyte uses on the 4060 LP. That'd be the current pinnacle of the SFF GPU world. We can dream.
I want to know what the best 2 slot gpu out there. I had a 3080 ti fe. Anything faster than that?
Brilliant information as always guv . Have you tried the intel range of GPUs. I think they are cheap? Might be interesting.
Is SR-IOV enable? If it is how many vGPU are enable and what the licence costs?
Can they use the output on the motherboards?
Picked up an a2000 for $250 to play around with, and use in a home server I want to build. I’m keeping my eye on the new Ada generation for a potential future upgrade
I'm finding myself wondering about transcode capability.... Throwing 1 or 2 into a HL15 storage server and slaving out to VMs to handle Handbrake or a Stream Transcode VM....
I would love to see you doing a gaming benchmark of the card like how well does it do in 4k with frame generation turned on
I wanted to see some 4k benchmarks as well
There also is a Tesla L4 and it uses AD104 with 7680 cuda cores and also 24GB of memory with 192bit bus
How about using iGPU as output for P4 or T4 cards? Would it be possible or is that blocked in Nvidia drivers?
Could something like this power a golf simulator and software like the FSX Pro? The requirements for FSX say no Quadro cards, are these newer cards different in that regard?
i would like to know how many 4k streams the rtx 4000 20gb can do on plex?
Isn't there a 4050 low profile PCIe slot power GPU coming out soon? I have been trying to find the new article on it
I was more exited with the beer than the card haha.
That little thing is almost as fast as my 2080 was and it was a big thick boy
Interesting. You cover all sides of the spectrum. I'm a lot closer to the Budget End of that Spectrum. 🧐 Big Thumbs Up.
Finally, a high performance, low power PC that makes sense for gaming and pro work!! for people like me living in a country like the UK where energy is expensive this is an incredibly attractive prospect!
Insightful review. I am guessing the single engineering gamers can justify the price. I’ll keep rocking my A2000 that I got on Amazon as NOS for $250 before the prices got silly.
I would love to see some Plex transcoding 4K>1080p or TDARR transcoding numbers FPS/number of transcodes possible.
That's amazing efficiency.
That price is a steal considering the 1600+ euro we have to pay over here 😂
Since the RTX 4000 Ada Generation SFF is around the same price as the RTX 4080, I would love to see it against the RTX 4080 in media creation and 3D modeling. I don't think I would ever buy the card for gaming, but if it can keep up with a RTX 4080 in media creation and 3D modeling I would buy it over a RTX 4080.
It won't keep up, the rtx 5000 ada is the 4080 equivalent. This one mirrors the 4070.
@@tejiriamrasa3258 Thanks for the info, and I noticed they have jacked the price up too. So, I would be better off getting a RTX 4080 for content creation. The way things are looking it looks like I should wait until Jan for the release of the RTX 4080 super and see where prices and availability will be?
Is the rxt 4000 a good card for converting blu ray with ai to UHD 4k with dvdfab and Uifab video enlarger?Thanks
You know windows allows you to use the integrated video output for dedicated GPU. I've done that before with a Quadro p600. You just have to select iGFX as primary boot device and tell windows to process graphics with dedicated GPU. Pretty nice.
I'm looking forward to when these are 2 or 3 generations old and going for ~1/3 of the price so I can slap 4+ of them in a homelab for gobs of VRAM to load LLMs. Bit too pricey right now but it's hard to not be impressed with such power at 70w.
I have an older supermicro server chassis with a threadripper pro in it. the Mobo takes the 2 available 12v EPS lines and the rest...molex (bleh). these would be great for passthrough gaming VMs! Super niche use case, but not much else is out there for decent gaming gpus that run solely off of pcie slot power.
How does it measure up to the Low pro 4060TI
Looking to stream, encode and decode H265 and AV1 codecs, its PERFECT
What would be an interesting test would be a comparison of the RTX 4000 Ada SFF (70W) vs the RTX 4000 Ada (130W), because they both have the same chip and memory, but the wattage is different.
that tesla low profile its so cute just makes want one :3
for the a4000, would appreciate if you could do a 5 gaming virtual machine and run at the same time, and see whether the FPS is acceptable?
So excited to watch this …
As a person with a P4, T400, and P620 all in use... Yes.
The efficiency is crazy
i like how you keep picking up a different card even though you're talking about the same card
Just a heads up for Optiplex 9020/7060 sff owners:
With these cards and others like the Low Profile 4060, the x16 PCI slot is on the "wrong" side for this card.
Also you need an 8pin/6min ATX PCIe power connector and power supply capable of powering these.
I think currently the "best" card you can get for those Optiplexs is a Low Profile RX 6400
Thanks for review, it's an excellent card for price to performance for what it is and what and who it is meant for.
Given that for me, it's comparing Enterprise to consumer use cases, which shows when comparing it to the 3070.
Perhaps should've shown the gaming test results including the 3070 which would have emphasized the difference.
Then again the 3070 would not have shown good performance results in enterprise applications in comparison to the RTX 4000.
Im a hige fan of graphics cards that dont require external power. I keep an MSI GTX 1650 Ventus around that has come in extremely handy for quick troubleshooting and occasionally just playing around with for fun. I really really really wish we would get at least a modern 75w pci slot powered card with each new generation with the accompanying performance per watt improvements. Preferably a low profile version for the people dropping a graphics card into an office PC to reduce e-waste. Plus a normal height version for better cooling and quieter opperation. We know there is a market for them based off the used price for said cards but somehow companies seem to be blind to this.
Side note, the ASRock low profile version of the Intel ARC a380 goes on sale for $99 occasionally for a brand new in the box card. According to Techpowerup it is roughly around the same performance as a GTX 1650. A comparable low profile 1650 seems to be selling for about $175 used on ebay currently which is absurd.
hold on a minute, is that a sour beer? i love that brand! if you get the chance try their apricot sour, it's kind of wild in flavours.
This is the coolest thing Nvidia has made in a while (IMO) and I couldn't help but order one. It's going in my Docker host. NVR, Jellyfin video transcoding, LLM powered HA smart assistant, and Steam headless. Cannot wait to play with this thing.
I'm super curious about how you achieve Steam headless on Docker, and if it can at all work with Sunshine? If you have a blog post or a public repo with compose files!
@@PuzzledMinds I haven’t dug into it yet but there are premade containers for it. I assume it’s just enough to run and stream games. I honestly don’t know what goes into making that possible, but I’ve seen full Linux distros in Docker containers so it makes sense that this is a thing you can do with it. Almost seems like it could be exploited for pseudo-vgpu capabilities on non-vgpu capable cards.