GPU VEGA NANO

[Majestic]

Got mine today from Newegg (via the eGPU enclosure combo).

I’m getting power cuts (PC turns off) when running 3D Mark Sky Diver benchmark. This is both with standard and power saver profiles in Wattman.

My setup is a S4M (Ryzen 5 2600) with HDPLEX 400W + Dell 330w brick. Suspect that undervolting will be necessary.

Going to hook it up to a wattage meter and do some more digging this weekend.

I could have sworn that I read a post previously of someone saying that the HDPLEX 400W was not enough to power even the older R9 Nano. I think the issue was the Nano would sometimes have power spikes where it would draw more than usual. (Sorry, I'm going from memory.) This is why I decided to get a standard sfx psu for my LZ7 case instead of the HDPLEX. Kind of glad now though because I'm pretty confident the Corsair SF 600W should be able to handle the new nano.

I think I might want to try undervolting it anyway though. Interested in what you find @E-Rod.

 

[quetzacoatlx]

everything i've read indicates roughly 30 watts higher on the full-size vega56 compared to the 1070 ti. there's no way a card as small as the nano is going to be as power hungry. it's limited by the size of the cooler.



i can't think of almost any scenario where it would be a better option, but i do see it being the best choice in a brickless s4m build with the new hdplex 400 ac-dc. i'm sure, after saying that, someone will make a brickless 1080 ti mini build though.

Vega cards sometimes run into trouble even with some pretty strong PSU. My Vega 64 regularly reboots when 4K gaming with the most powerful SFX-L PSU people can buy. Similar problems can even happen on systems with 1000+w PSU.

 

[Phuncz]

Vega cards sometimes run into trouble even with some pretty strong PSU. My Vega 64 regularly reboots when 4K gaming with the most powerful SFX-L PSU people can buy. Similar problems can even happen on systems with 1000+w PSU.

If your 700-1000W PSUs can't handle a 275W TDP card, something else is going on.
I've run a 275W TDP card for years on a 450W SFX PSU without issue.

 

[Majestic]

If your 700-1000W PSUs can't handle a 275W TDP card, something else is going on.
I've run a 275W TDP card for years on a 450W SFX PSU without issue.

I think there were little messages like this all over internet I saw of some people not being able to run the R9 Nano on the HDPLEX 400W and 330 Dell brick. I can't find the more legit one that made me decide not to do it. Maybe this is an issue only certain people see due to the games, system configurations and software loads they have.
https://kit.com/product/hdplex-400w-hifi-dc--311570
https://smallformfactor.net/forum/t...i-dc-atx-the-next-generation.3501/#post-66508
https://smallformfactor.net/forum/threads/hdplex-400w-dc-atx.2071/page-13#post-55199
Again, nothing really definitive. It was just messages like this I kept seeing that made me shy away from the HDPLEX for my nano when I was doing research for my build.

 

[Phuncz]

My point was towards your claim that a 275W TDP card (Vega 64) can have issues on >1000W PSUs. I don't even understand how you could have an issue with the Silverstone 800W PSU (you say "the most powerful SFX-L PSU") with your Vega 64, I powered two Radeon 290X cards (275W TDP each) with that PSU on my Ryzen 1800X setup.

So it either means the card is defective, another component is defective, overclocking is applied and not proven stable, something in the system is overheating or there's a software issue. The Polaris and Vega cards have short spikes in power drawn during heavy load, which most PSUs can handle. But generally with SFF builds the PSU isn't oversized as usual. But that shouldn't be a problem

It's indeed true that the R9 Nano and possibly the Vega Nano will not work reliably on a 330W power brick, there's little overhead for the rest of the system when just the GPU can almost draw that amount of power on its own.

 

[E-Rod]

To anyone who is running the Vega Nano, have you been able to undervolt the card while simultaneously boosting performance? Curious to know if an undervolted VNano would sit between the 1070ti and 1080 from nVidia.

Yes! I have been toying with Wattman and doing just this. So far I am at:

• GPU P6 1537 MHz @ 1050 mV
• GPU P7 1612 MHz @ 1100 mV
• HBM 900 MHz @ 950 mV
• Power Limit -50%
• Default Fan Curve (85 Max, 74 Target, Fan Auto)

I could have sworn that I read a post previously of someone saying that the HDPLEX 400W was not enough to power even the older R9 Nano. I think the issue was the Nano would sometimes have power spikes where it would draw more than usual. (Sorry, I'm going from memory.) This is why I decided to get a standard sfx psu for my LZ7 case instead of the HDPLEX. Kind of glad now though because I'm pretty confident the Corsair SF 600W should be able to handle the new nano.

I think I might want to try undervolting it anyway though. Interested in what you find @E-Rod.

I remember that R9 thread too and tried one which yielded similar results in my setup. At first the Vega Nano was showing to be very similar with the power spikes causing cuts in power. Also learned about the AMD Radeon Perf Overlay (CTRL+SHIFT+O), good stuff.

However, I have made some good progress this afternoon. With the above undervolt I was able to get the Civ VI Gaming and AI benchmarks to run to completion at Ultra Settings 1080p. Here are the results for the gaming benchmark:

• 83 FPS AVG
• ~82W GPU
• ~220W from the wall
• 65-70 C Temps (S4M horizontal, with feet)
• Quiet fan operation (<2200 rpm)
• No Coil Whine!

Definitely some good results and I'm liking the temps. Of course time will tell whether the undervolt is stable but I have a good working point to start from. At this point I can put it through the ringer more and see how it fares in 3D Mark.

UDPATE: Set the Frame Limiter in the Radeon Controls to 60 FPS, which is probably the key aspect to all of this. Seems like a no-brainer after the fact, but huge improvements in power consumption and temps.

Was able to get 3D Mark Sky Diver to complete without any issue this time around. I think it was the super high FPS jumps that would peg the power consumption.

Here is the overview of Sky Diver @ -25% Power Limit:

• Graphics Tests: 60fps
• Physics Tests: 60fps in 8/24/48 thread test, 47fps in 96 thread test
• Combined Tests: 60fps
• Overall Score: 13374 (with 60fps cap)
• 65 C Peak GPU Temp
• 59 C Peak CPU Temp
• 110-180W from the wall

Firestrike does solid 50fps+ (30 fps combined) on -50% PL but will power cut in between at -40% PL. Seems there's not much extra room here.

I'm pretty content with the current limitations, which seem to be power-related. Maybe this year we'll see some better bricks and DC-ATX PSUs that can handle spikes better. I'll be keeping an eye out and seeing if they let the Vega Nano stretch its legs a bit.

From what I can tell, the Vega Nano is definitely doable in the S4M but does require you pretty much run the card at half power. That said, the temps and fan are pretty low so it's far from bothersome.

 

[lhl]

Nice info in the German review site. Showing the power draw down to 321W for the system after the under volt gives some confidence.

Once the card’s sweet spot is found maybe a bios flash with those settings to make it permanent?

Just as an FYI, AFAIK flashed BIOS's are still locked out due to firmware signing and there aren't any workarounds. However, you can use OverdriveNTool to do your tweaks and run them on boot.

 

[lhl]

Yes! I have been toying with Wattman and doing just this. So far I am at:

• GPU P6 1537 MHz @ 1050 mV
• GPU P7 1612 MHz @ 1100 mV
• HBM 900 MHz @ 950 mV
• Power Limit -50%
• Default Fan Curve (85 Max, 74 Target, Fan Auto)

Just some tips since I've done a lot tuning my Vega since last year (grabbed a Vega 64 at launch for the $500 MSRP).

• You should use GPU-Z to measure your clocks, loads, and power performance. At -50% PL, I can guarantee that you aren't sustaining anywhere close to 1500MHz.
  
• Maybe a little different for brownouts, but in general for optimizing your Vega (you may want to move it onto a bench rig w/ a beefy power supply as this will make life a lot easier), set your PL to +50% (max) but then *lower* your core voltage. You're going to be primarily thermally limited I suspect
  
• Note that you must also lower the "memory voltage" (really, memory voltage controller) as that controls the core floor
• My general advice would just be to drop everything to 900-950mV and see where you get (you may have to drop P7 clock to something more reasonable like 1300-1400MHz if you get lockups)
• You'll get a lot of "free" performance by oc'ing HBM performance. V56 HBM is volted lower than for V54 so you might only be able to get to 950-1000MHz, but worth pushing/paying attention to the hotspot temps
• Here's a good summary of undervolting methodology and here's a more detailed writeup.

I ended up swapping my Vega 64 in my workstation for a 1080Ti shortly after, but w/ a Morpheus 2 heat sink replacement (not an SFF build obvs), but for my "low power" mode, I ended up being happy w/ 1360Mhz/940mV P7 (actual sustained speed 1338MHz) and 1100MHz/900mV HBM w/ 60C GPU temps and about 170W or so from the wall.

 

[Sean Crees]

If you look at the following article (translated from German) you can see how well this GPU can undervolt:

https://translate.google.de/translate?sl=de&tl=en&js=y&prev=_t&hl=de&ie=UTF-8&u=https://www.hardwareluxx.de/index.php/artikel/hardware/grafikkarten/44084-amd-radeon-rx-vega-56-und-vega-64-im-undervolting-test.html&edit-text=

Obviously there is an element of Silicon Lottery here as evidenced by the results on this video, but you can see that when undervolted, the card consistently outperforms the Vega 64 on stock settings and routinely trades punches with the GTX 1080. All of this with a reduction in power consumption in the neighbourhood of 75W.

I'm planning to put my Nano in the S4M with the largest stable undervolt I can.

I just want 1070 level performance at around 150w in the smallest form factor possible. This is obviously the smallest size yet, but i'd like to see someone do a more head to head comparison between this when undervolted and the 1070 and 1080, to show exactly what sort of performance/watt it can do. Also i'd like to see more on how much noise the card generates under that sort of condition.

Right now I think i'm going to wait for the next gen graphics cards in the fall and make my decision at that time.

 

[hyp36rmax]

VEGA GPU's tend to have power spikes and seems to not have the efficient power management that Nvidia GPU's utilize. For those of you considering using the VEGA 56 NANO, i highly recommend undervolting if possible. I believe this is also using a custom BIOS.

 

[Gutzy]

Also I am playing with the idea of a 3D printed shroud and two of these as a mod kit:

http://www.blacknoise.com/site/en/p...fans/nb-blacksilent-pro-series/80x80x15mm.php

If all goes well with the 3D printed shroud, will you sell them?
If not, are you willing to share the 3D diagram files via something like thingiverse.com?

 

[Kmpkt]

If we can pull it off then I'll be selling them.

 

[Boil]

Also I am playing with the idea of a 3D printed shroud and two of these as a mod kit:

http://www.blacknoise.com/site/en/p...fans/nb-blacksilent-pro-series/80x80x15mm.php

If all goes well with the 3D printed shroud, will you sell them?
If not, are you willing to share the 3D diagram files via something like thingiverse.com?

If we can pull it off then I'll be selling them.

The stock fan is an 80mm dealio, but the heat sink is cut out to 'countersink' the fan...

If you are doing a shroud for dual 80mm fans, will you be trimming the heat sink any...?


I feel this is the perfect Vega GPU (well, the ONLY...) for an all-AMD build in the CCD MI-6...

For that use-case (MI-6 chassis) I would strip the shroud (no replacement shroud at all), trim the heat sink 'opening' to allow for a NF-A12x15 & use a converter cable to hook the fan to the GPU fan header...

 

[ChinStrap]

No more "auto-notify" button. I hope they re-stock at some point.

 

[Majestic]

No more "auto-notify" button. I hope they re-stock at some point.

Yeah, I thought that was odd. Kind of glad I took the plunge for the combo even though I don't know what to do with the eGFX enclosure. I was thinking that maybe PowerColor is looking for a new merchant to sell the Vega Nano through like Amazon. However, it is more likely that they have low stock of the part. I feel like manufacturers are still under estimating how much interest there is in the sff space.

 

[hyp36rmax]

If we can pull it off then I'll be selling them.

I would love to test this. Money is ready

 

[Kmpkt]

So apparently the stock fan profile on this card is shit and needs to be tuned. This reviewer saw a 7% increase in performance on Superposition simply by allowing the fan to max out based on the curve.

 

[EdZ]

I could have sworn that I read a post previously of someone saying that the HDPLEX 400W was not enough to power even the older R9 Nano. I think the issue was the Nano would sometimes have power spikes where it would draw more than usual. (Sorry, I'm going from memory.)

Short version: HDPLEX fine, Fury Nano (and likely Vega Nano, as Vega and Fury are extremely similar) have issues with laptop bricks.
Long version: here.

 

[TheHig]

Vega is a fickle beast. It’s still up to the user to dial the card in for maximum performance.

Keep it under 74C . HBM needs to be 900-950. Gpu needs to hold 1300+. All of those things together to get your 1070-1070ti numbers. Any less in one area and it drops off. At least in my experience with a reference one. I REALLY like this card but man that hassle.

Bring 7nm GPUs to the masses sooner than later please AMD.

 

[Majestic]

Short version: HDPLEX fine, Fury Nano (and likely Vega Nano, as Vega and Fury are extremely similar) have issues with laptop bricks.
Long version: here.

Thanks @EdZ . I think that was the post I was referring to.
So, everyone in this thread is talking about undervolting the vega nano, but I've never under volted a GPU. I will google search once my nano comes in, but can anyone share a guide or video tutorial they've found really helpful for undervolting AMD GPU's?

Edit: Oh @lhl posted some guides and stuff in his post above. I'll look at that. Thanks!