Is this STILL the Fastest SFF GPU??? - RTX 4060 Low-Profile Review
Vložit
- čas přidán 30. 04. 2024
- Grab yourself a Pint Glass at craftcomputing.store
In the world of Small Form Factor GPUs, there aren't a ton of options, especially if you're aiming for high performance. While cards like the RX 6400 and Intel A380 exist, they even struggle at 1080P gaming. Then along came the RTX 4000 ADA Generation SFF, and brought with it a level of performance and size never before seen together.
But it is also $1250.
But, now we have the RTX 4060 Low Profile from Gigabyte, and it runs only $319 at retail. Is this the SFF GPU of my dreams, or is it just another compromise in a long list of sub-par performers?
But first... What am I drinking???
Belgian-Style Blonde Ale (Pfriem Brewing, Portland, OR) 7.5%
Check out the Gigabyte RTX 4060 LP on Amazon: amzn.to/44qqalo
NVIDIA RTX 4000 ADA Gen SFF: amzn.to/44sgNBQ
Sparkle Intel Arc A380 Genie: amzn.to/3WuM6cW
XFX Speedster SWFT105 RX6400: amzn.to/4dkjA3Y
Gigabyte RTX 3050 LP 6GB: amzn.to/3wfCato
Today's LAN Rig Test Bench:
• Raptor Lake Mobile on ... + • Fixing the 13900H (ES)...
Parts from the Test Bench
10729 13900H ES motherboard on Aliexpress: www.aliexpress.us/item/325680...
Fractal Design Terra: amzn.to/43VAdi8
ASUS TUF Gaming RTX 4070 Ti Super: amzn.to/43Ve8QO
Patriot Viper VP4300 1TB: amzn.to/43YSATt
Western Digital SN770 2TB: amzn.to/43YSQBz
Cooler Master V850 SFX: amzn.to/3JiNhV8
be quiet! BK034 Cooler: amzn.to/3TZlIFg
OLOy 2x16GB DDR4 3600 MT/s: amzn.to/3Jd8uQx
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 - Věda a technologie
TL;DW - do you have infinite money? No? Buy the 4060 and never look back
Forget infinite money, it's barely faster at all.
The 4000 ada is much better for productivity. rtx 2000 ada is probably closer to 4060 raw compute power (though its not a gaming card either so 4060 will probably beat it in games). It just came out 2 months ago. Its still double its price, but it has 3 benefits: half the power draw, double vram for productivity tasks with ecc, and nerdware is coming out with a modded cooler for it and the 4000 ada to make them single slot. This can't be done with the 4060 because of the cooling for its power draw and its stacked 2 slot i/o, making the 2000 and 4000 ada the only performant options for certain sff 1 slot cases
@@mtrebtreboschi5722was VERY disappointed to see it be 2slot only. so close.. so close
I'd argue both are a bad deal.
i'd argue a 10 series is better value
The 4000 was almost perfect!
✅ Lots of memory
✅ Bus powered
✅ ECC memory
✅ Small form factor
❌vGPU support for Prodmox
Jeff - please tell me you have a fix for the last one. 😅
Nope, nvidia started using SR-IOV for vgpu from ampere so the unlock does not work for it....
Is Vgpu. = passtrough ?
Not quite. Vgpu is all about splitting the gpu into multiple virtual gpus (that can then be passed through). Passthrough otherwise works on basically everything these days (barring weird bugs)
@@xpatrikpvp So my A2000 12GB from work doesn't work with vGPU on proxmox as well? For god's sake nVidia what the hell.
As an added note vGPU has been typically available on quadro 5000 and 6000 or higher when there was a 8000. The thing is to inlock vGPU is also requires a software license from Nvidia.
"In the end, the conclusion should surprise nobody". True. But I still appreciate the time you took to prove it.
Precisely. Even if it's pretty obvious to us enthusiasts, somebody actually building a tiny workstation might not be so clued in, and now this information directly comparing them exists.
Still waiting on a Voodoo 5500 benchmark :P
I was about to say "I know a guy".... then I saw the user name.
Well played sir.
This might be Mandela effect , but i remember sli was a passthough externaly on the rear of the case, but weren't the cards also powered by external power adapters? (wall worts) ?
As a SFF only builder who appreciates low wattage draw and reduced heat generation, thank you very much for the comparison. As a non gamer, I would appreciate some photo processing speed comparisons such as RAW to JPEG time for 50 to 100 images.
Doesn't that usually use CPU though?
@@spakeschannel Depends on the software.
@@marksulloway5669 Well, I dunno about Capture One, but I'm pretty sure Lightroom uses CPU for exporting.
Just downclock a more reasonably priced gpu.
I got my 4060 LP back in September - I had to order it from the states via Amazon UK. It's pretty damn good for such a small card.
Curious - do you need a SFF GPU? Personally, I don't... I'd like to need one, in the sense that the machine I'll be picking up probably an A2000 for soon isn't SFF and the reason is most of all price, not only, but mostly. I have a Dell T5810 with a 14core, 96Gb ram currently, 5xGbe NIC's, a quad M.2 SSD card, and a Quadro P4000.
@@noth606 What an odd question - nobody *needs* a SFF gaming GPU. I chose to build a 4 litre low power gaming rig and the 4060 LP is clearly the best choice for that use case.
Been waiting for this since the announcement 😁 Keep it up man! Loving the content 👌
You do GPU reviews correctly. The way they should be and i really appreciate that.
What's your definition of "correct"? I believe there is a discussion video on this channel with Steve from Hardware Unboxed on that topic.
This is the video I've been waiting for, Thank you
Thank you very much for this comparison!
I love your reviews and videos in general!
Do you think you can through some Premiere, Davinci Resolve and Blender benchmarks in there for the future? This should stretch the Videomemory if done right.
I see you post, i make my tea and start to watch with you. Cant drink, but I still enjoy!
I made tea a couple videos back :-)
Great video Jeff!
Great video Jeff. Any chance you'd be willing to do a comparison of the 2000, 4000, 4060 LP, and maybe any other LP/SFF cards you can get your hands on, in regards to their performance vs cost for things like video transcoding, CAD, 3D rendering? I'm working on building a small server on a budget for hosting Plex/Jellyfin for a handful of people (up to maybe 5 or 6 simultaneous streams, depending on cost) and I'm not sure yet which way I should go in regards to transcoding. Most of my content is stored in 4k and is streamed to a range of devices from tablets to TVs.
Shuttle PC lol that brings back memories. Those piped heatsinks were cool though.
they are still around or not?
@@scudsturm1 I guess so look over on shuttlecomputers lol.
Great review, informative, fun and relevant to a tech nerd like me.
While on its face this comparison may seem crazy, it’s the exact comparison we all would have done in our heads. Thanks for making the magic happen, as always!
Thank you for this, i hate how few video reviews/testings these A-Cards get.
Great comparison! I'd be very interested to see how they compare in workstation type loads.
Great video. is it possible to benchmark the fastest SFF gpu for AI inference tasks on LLM and SDXL. Let’s put the RTX4060lp vs RTX A2000-12GB vs RTX 4000 on the track. Really interested to see which gpu provides the $/perf value.
One question for you. Did you use the gaming setting on the RTX 4000 ada. NVidia builds profiles in the RTX XXXX Ada that allow it to run in gaming mode. I like how you mentioned that the RTX 4000 Ada is not a gaming card. Also the "Quadro" offers ISV certifications that are important for productivity and engineering applications. I think you should explain that a lot of the cost difference is in the ECC memory used in the "Quadro". Good overall review for gaming.
Great video thanks a lot for the comparison. Which drivers did you use for the RTX 4000 ADA? By default it will install the professional drivers but it also works with "game ready" drivers for gaming GPUs.
I used the same driver version for both cards, not the 'Quadro' drivers.
Thanks mate, this is interesting. 4060 LP owner, but have always wondered how it would stack up vs the 4000 Ada. Thought the latter would be well ahead, tbh!
LOL, Quadro!! I love it! And, by the way, totally agree!
Got me a 4060LP back in January. Lots of power in a small package. I can finally game decent on my HTPC.
I bought one of these in September for a super skinny build. I think i paired it with a 7600x. Ran super hot until i flipped the CPU fan and used it as an exhaust. Very nice card for the size!
Would like to see how the rtx 2000 Ada Generation 16gb stacks up in the SFF lineup
slightly worse than the 4060 LP in gaming performance for double the price, not worth it
I'm certainly intrigued by the performance of other small GPUs, especially stuff that might only fit within a *single* expansion slot, though I doubt many others woould
You should review the new RTX 2000 ada alongside these. It just released in february and and you'd be the first to do a legitimate review of it as far as I can tell, but its half the price of the 4000 ada while still boasting a massive 16gb of vram.
it's about double the price of the 4060 LP though and offers slightly worse gaming performance than the 4060. That 16GB of VRAM is the only thing that separates it from the 4060
Hi there Jeff!
It would be also interesting to see the difference between these two cards in a PCIe 3.0 system, as I think it is the one most people use as daily driver (if I am not mistaken). Another interesting thing to see is the gap between these cards and the RTX A2000 (both the 6 and 12 GB VRAM versions) for people that play games and do some kind of video editing.
Would be awesome to see you make a small form factor build around the 4060!
I'd love to see you an ollama llm benchmark along side the games.
Very thankful for you doing these tests, we’re you able to undervolt the ada for higher clocks on average?
Edit: $350 for the 4060 is crazy in comparison to the 4000 ada sff. i think no matter the difference in performance, $900 price difference could build me another pc or two
Good Job, only missed the TEMP comparison
14:14 what about those temps? I thought noise was missing though
@@Incommensurabilities indeed also missing noise readings, to me it's obvious the blower fan makes more sound....
Both cards are virtually silent. You'd be hard pressed to hear either of them over your power supply fan.
I love the electric cost comparison
I like SFF. Great you did this comparison 😊👍
So when are you power modding the Quadro? Should be the same as the a2000 that not from concentrate did a video on, alongside a supplementary power connector.
My supermicro server case doesn't have extra pcie power cables. basically two eps for the motherboard and a bunch of molex for the backplane. The 4000 SFF allows some decent passthrough gaming VM action.
That cooler is ADORABLE!
Thank you for this comparison! The other advantage of the 4000 ADA SFF is you don't need the PCIE cable so this will be easier to install into tiny SFF cases.
Could you compare these two cards in a few Blender rendering tests?
Awesome video! Thanks for the numbers
maurice Moss got a laugh :)
Marsellus wallace was the boss
and Jules Winefield was the "badmother fu...."
With NV Clean Install, could you force install the Game ready driver and still use the Rtx 4000?
IIRC A2000 used to overclock nicely without the need to rise the voltage, have you tried it on this one?
I have been genuinely interested in this graphics card as I have several 2u servers and plan to add one using the Silverstone 2u chasis for a cloud gaming machine and test out dedulication as I reinstall games fairly often and have friends over fairly often.
So voltage mod on the quadro when?
I have an RTX A5000 in my personal rig and when I get the time I want to see what kind of monster it could become if allowed to draw anywhere close to it's consumer brethren.
4:00 Correction. AD104 is the 4070, but not the 4080 unless you mean 4080 mobile. The desktop 4080 uses AD103.
You are correct.
will you test the new RTX 2000 Ada SFF?
Can you test the Plex transcoding performance of the SFF, with its DUAL encoders? In x265 and AV1 please 🙏🙏
Quite the unexpected and nice review tho! Cheers :)))
Something you used to be able to do with old Quadro cards was disabling ECC for the memory on the card. Theoretically this was supposedly improving the memory performance. Don't know if the RTX 4000 use ECC for the memory or if it can be disabled. But if it does use ECC and it can be disabled then it might be interesting to see if it really changes performance.
I guess you didnt like the beer.... the glass was too full during the video!!!
Nooooo.... no no no no no.
no.
Love this topic as I'm in the market for a SFF GPU to pair with a cheap old Dell office PC secondhand for purposes of building an Emulation box that can handle up to PS3 and Xbox 360. This 4060 LP could be just what I need.
Would love to know if there is a difference in transcoding between the cards, specifically on Plex
What about a video editing score? Capcut or davinci resolve. Both are free.
I have a soft spot for a decently performing graphics card that doesn't require external power. Really wish we would get more updated options as such. I keep an old GTX 1650 laying around that I use for testing purposes on new system builds and occasionally playing with a low power system for fun.
Comes down to if you need a blower style cooler or typical gpu style cooler.
The ADA card is a godsend for AI. You would be a silly goose to buy it for anything else. I would love to see AI benchmarks in your reviews since my personal homelab depends on AI services more and more. I bet I am not the only viewer who feels this way.
I would like to have a comparison in terms of 3D Archviz and Stable Diffusion of these two cards, especially in terms of 3dsmax UI speed and V-Ray performance and generation / scaling to 4k using Stable Diffusion🤔
20GB of VRAM in the 4000 and basically 2x as many cuda cores. It’ll be interesting to see a direct comparison but I’m gonna guess that the 4000 trounces the 4060 in AI applications for those reasons alone.
@CraftComputing
Jeff, while this suggestion may seem like a waste of time, a run-down comparison of the 4060SFF and the 3050SFF would show interesting results. Also remember, that some people, myself included, consider 720p/768p perfectly acceptable and valid gaming resolutions. Perhaps show us a few benchmarks for that res bracket. EDIT: forgot to mention, good show on this one! This video was interesting and your points about cost vs performance vs power usage are spot on!
those resolutions are moot, pretty much every game can be set at 1080p high/ultra with both of these cards with more then acceptable frame rates.
@@Rustyer266 And for the few that can't (Alan Wake 2?), DLSS is still an option and is usually better than 720p.
@@Rustyer266
Not true. Games older than 2019? Sure, you might get away with a few settings on max at 1080p, but 720p/768p is a given for those systems that can't really get high end performance 1080p. I'm a custom system builder with 35+ years experience, I know my craft..
Hey Jeff, have you done any more testing on the MS-01? I'm getting frequent hangs for no obvious reason, even got it once when I had all E cores disabled. Updated Microcode. Just now I'm running some experimental BIOS that MinisForum sent me. Looks promising, but still. People are saying it's a bad batch or defective unit. Memtest showed no errors, the only other thing I'm doing that not every one is doing is runing containers which does assign cores a bit strangely and SR-IOV on the 10GB nics.
I'm interested in this. I just got mine and I'm still in the process of setting mine up. If you get resolution on this, please let me know!
Any chance you can compare the cards for PLEX encoding?
Tasty video 👍
If 4060 draws little or no power from PCIe slot (dunno?), then maybe that's a tiny win rel. to Quadro, saving a smidge on wear and tear of mobo's power components. Down the road, similar replacement mobo probably harder (and costlier) to source than a PSU.
Kindest regards, neighbours and friends.
Is the 4060 LFP usable in Proxmox and can be shared as vGPU for VMs?
1 Liter PC enjoyer here, the point here is that there's simply no auxiliary power input to be found in chassis so the GPU must be powered by the PICE slot alone. I have a ThinkStation P3 Tiny from Lenovo and it can take LP single-slot PCIE cards. I am currently pairing it with an A2000 with a modified one-slot cooler.
The man making the single slot cooler for A2000 is also doing so for Tesla P4/T4, A4000 Ada SFF, and 4060LP(by removing 1 HDMI port). If you could limit the power budget of the 4060LP to 75W and compare the performance, I think the result would be interesting.
I compared the prices of A2000 and A4000 Ada SFF when I built this system, and I just picked A2000 as A4000 Ada SFF costs twice as much.
I'm kind of wondering how the Ada 4000 does for Plex transcoding. I obviously know price to performance makes no sense, but I just want to know if it's better than quick sync on an Intel 13700k.
Just built a small formfactor 13400F system that would be killer with this class of card. From an efficiency stand point AND from a "tinkerer" / Homelab enthusiast standpoint, I am VERY curious as to how that ada card would fair in GPU-p / vGPU scenario.
Just out of curious, which GPU ran cooler through most of the testing?
Would you test the 5700xt vs rx-6600 vs Rtx-4060? I would love to see how this plays out!
Could either allow passthrough into multiple vgpu in kvm-qemu?
Im a PC enthusiast but am new to the AI workloads, they would benefit from the VRAM increase of the quadro, wouldnt they, or would they have a similar showing? From what ive seen, AI workloads want more RAM
I would love to know how many concurrent camera feeds each card can handle in blue iris.
Are there any tools to measure ML performance? I’d loved to see how they perform when the RAM is saturated.
When are you going to try out the intel flex 170 Gpu with proxmox? Unofficial support for sr-iov.
As soon as I can get my hands on one. I've been told multiple times 'they're in the mail'...
I like your vdi vids. but i watch a lot of all your stuff. good work. enterprise for all
Since I have a powermodded rtx A2000, I'm wondering if you can get another 30w or so out of the 4000 SFF....
shunt mods allowed more power on the A2000 for mining and had big performance upgrade.
Curious if the power limit increase would give the quadro the easy win.
Not saying you should do it, but find it interesting.
there is a intel sff form factor for its arc series, not saying it will be close but be interesting to see its numbers, especially with av1 encoding and decoding.
Yeah, I mentioned the A380. It struggles to keep up with even the RX 6400 in gaming. Definitely a decent option if you're looking for video encode/decode though.
16:47 Idk, but my *impression* from watching the video was that the 4000 had a trend of miserable 1% / 0.1% lows in the games where it lost out in terms of overall experience in a more pronounced way than the 4060 did when it lost out?
Either way, if you're getting an SFF card for gaming, the 4060 appears to be the obvious choice given the price difference and overall similar performance.
I needed one that was bus powered (refurbished office/PoS unit that doesn't have PCIe aux power), so I was forced to go with a 3050 6GB. It's fine for what it is. Plenty of power for an old 6th gen intel CPU.
I would have gone with an Arc A380 but I wanted CUDA capability for folding when I'm not using it.
can you test rtx ada 2000? just saw them, must be new.
I wonder if these would fit with slim coolers and SFF brackets in a Minisforum MS01. That would be such a killer combo especially for the 20GB card since it has no power connectors
There is a company that makes a mod for the A2000 and A4000 that turns them into single slot cards for about $89 so they will fit Minisforums MS01.
is it like the games only take advantage of 2 or 4 cores instead of the full 6 or 8?
The 4060 faired really well compaired to the A4000. I read an article a while back that Asus was gonna release a 4060LP card but haven't heard anything since .
It's out already, has slightly higher base clock than Gigabyte at 2490MHz and says it will OC to 2520, planning to pick one up in a month or so.
Planning to run 3-4 of the Gigabyte 4060 SFF in the Asus ESC4000 G3 as a local LAN gaming setup. Been looking at buying these over the past month but now worried about availability after this lol. Otherwise i'll wait for turbo variants of the 3080 to finally come down in price to reality with the rest of the 3080s.
They also power mode the a4000 to 110 watts. And single slot coolers will be coming out.
if you run AI or CAD or image elaboration, opencv and similar you need a lot of RAM and you should go with A4000 but as you said if you wanna game the 4060 is for you :)
Can you compare LLM performance?
I've been saying since the launch of the 4060 and 4060 ti that eventually nVidia would have to make a low profile version if they ever wanted to sell through their stock of these chips at MSRP or slightly higher pricing (since they don't seem to want to drop prices)... I'm only wondering what took so long for an OEM to take advantage of the efficiency of these chips to help distinguish themselves in the market? Now all we need is for N3rdware to come out with a single slot cooler designed for this card!
Boy, I'd love to see a 4060 ti 16GB version of this card, but I doubt nVidia will allow that to happen...
Waiting for RTX 2000 vs 4060 LP
Is there any way to overclock the 4000?
Shunt mod maybe? :D
There are some people who have shunt modded it to 130W....
if you ever get a wild ale desire you should check out de Garde Brewing out of Tillamook,OR. All they do is open air fermentation.
Oh, I'm well aware of DeGarde :-) I make it to Tillamook at least one a year.
@@CraftComputing Of course you do. I'm still going to be trying to name drop those eccentric and great PNW breweries.
Got one of these around xmass, probably the only variant of 4060 in existence that is not a meme 😂
4:16 Another thing that differentiates the 4000 Ada from the 4060 is the former has 2 hardware encoders and decoders instead of just 1. That even beats the 4090, which has 2 encoders but only 1 decoder. Though I think a more fair comparison would be the 4060 against the 2000 Ada instead, since it has closer specs and 16 GB of VRAM instead of 20.
(Just got to the end, rough glass this time, ‘thank you for your service’)
I am confused. My case is TU 150 and is a SFF case. I have a 3080 Ti reference card installed in it. Isn't 3080 Ti faster than low profile 4060?
For gaming, your conclusion would be accurate.
For AI, the 8 GB of VRAM on the 4060 might be able to load some basic models, but you can either do text-based AI or AI generated pictures, but not both, simultaneously (whereas with the RTX 4000 Ada SFF, you'd be able to run both basic models, simultaneously).
Was the same driver used for both cards? If not that could explain some of the results
'Game Ready' drivers were used for both cards.
@@CraftComputing I wasn't aware they offered game ready drivers for professional cards. When I install them it only gives me the option to use production or new feature not game ready
What about running some benchmarks that are compute oriented and would also benefit from more memory? The tradeoff of core count vs frequency works quite nicely at these GPUs for gaming, and you can see that games are trying really hard to fit in 8GB to please the average kids.
someone does a shunt mod service for the a4000 ada. apparently it runs faster than a 3070ti and pulls just over 100w under load.
Hey can you report on 4k gaming with DLSS? I bet it would be somewhat ok with 60fps
HEEEY! I was turning my gaming pc into gaming pc + home server and i able to passthrough my 6800xt gpu but limitation is i only can passthrough 1 vm at a time. Looks Amd's MxGPU still sucks.. what options do we have? I can accept %10-20 performance loss. But not more than that. ThXX
BTW u better drink that beer, looks it evaporates!