RX Vega Owners Thread, Tests, Mods, BIOS & Tweaks !

Got my bench scores in using a stock Vega 64 bios and all base default driver settings.

Fire Strike
https://www.3dmark.com/fs/18277311

Fire Strike Extreme
https://www.3dmark.com/fs/18277380

Fire Strike Ultra
https://www.3dmark.com/fs/18277443

Time Spy
https://www.3dmark.com/spy/6199544

Time Spy Extreme
https://www.3dmark.com/spy/6187400

Time Spy Stress Test (twice)
https://www.3dmark.com/tsst/386548
https://www.3dmark.com/tsst/386574

Edit.
For comparison, I would suggest only looking at the Fire Strike Ultra and Time Spy GPU rscores because 3DMark is incorrectly reporting Time Spy CPU scores for the 2600K.

 

With reference cards. I don't think flashing the Vega 64 BIOS is worth it.
Undervolting does WAY better.

Your FireStrike Extreme vs mine.
https://www.3dmark.com/compare/fs/18277380/fs/17864731#

Your TimeSpy vs mine.
https://www.3dmark.com/compare/spy/6181916/spy/5886998#

 

I'm not sure if you are looking at GPU scores, but they are about the same.

This is also just the first step.

 

Yes, I saw.
I'm almost 6% higher with only undervolting.
Less clock in the core and same memory clock.
That's my point. Flashing is kinda pointless.

 

Okay, I''ll list it out.

1. Your CPU is influencing the scores, so very little difference in games.
2. I'm at default settings.
3. You have 100 mV lower to the memory.
4. I do not need to worry about any software related issues, or need to use any overclock tools.

Now, as for efficiency, I'm not done yet.

p.s.
If you post your Fire Strike Ultra score, you'll see less CPU interference.

 

@BlackZero that 2600k is holding back your Vega card, not by much though.

That said Vega56 is so much faster than Polaris, damn.

Vega56 Firestrike 22.6k, Timespy 6.9k
580 Firestrike 16.9k, Timespy 4.8k

 

Thanks, @Undying

Yes, although I game at 1440p, scores are being held back at lower resolutions. Also, I have a water cooled system, so just trying to keep it as silent as possible.

5K in Time Spy is pretty good, and not that far at all.

Fire Strike isn't far either.

p.s.
This was from a few years ago

Spoiler

https://www.3dmark.com/fs/1091750

 

My undervolt is 980v for the memory, 1030v for P7

Anyway, I'm not trying to disregard your PC or card.
I just don't see the point of flashing the BIOS if you can achieve the same with just tweaking voltages/clocks.

When you are done tweaking, if you don't mind, post the settings with OverdriveNTool so I can test the same.

 

The Vega 56 and Vega 64 only achieve similar final overclocks after you flash, because the Samsung HBM voltage is bios locked. In addition, the Vega 64 bios allows full board power, which is not available with a Vega 56. Finally, the Vega 56 will use less power and generate less heat due to the slightly smaller die package, so will usually boost higher than a Vega 64, and this will result in very similar final performance.

So, unless you flash the bios, none of that is possible. However, for undervolting, that won't matter all that much, and it will depend on your individual card.

 

The main reason we suggest people with a Vega 56 WITH SAMSUNG HBM2 (I write it again, WITH SAMSUNG HBM2) to flash Vega 64’s BIOS is to overclock the HBM2 to around 1100MHz, providing a boost of bandwidth and FPS, since Vega chips LOVE memory bandwidth (Radeon VII is an example of that). Remember, Vega 56’s BIOS has lower memory voltage than Vega 64s, capping their maximum stable frequency to around 900-930MHz if your card is a lucky one. If the card has a dual BIOS it isn’t even a problem, because you can switch to the backup BIOS in case of failed flashing.

 

@OnnA

Why isn't it recommended to go over 100 Mv for the memory setting in Wattman?

 

IMO good for HBM2 is @
1150 -> max 975mV
1120 -> 931mV
1200 -> 1025mV

It's temp. related, also IF floor rises tW a little (+/- 20tW if set >1.050mV)
Im not using now more than 1125-1135MHz at 931/943mV (Good is to set it at max 950mV with ~1150MHz but not every game can handle it)

 

Is it me or latest AMD drivers for Vega64 19.2.1 are crap.
So up and down clock in TimeSpy etc.

 

Not bad, but overall 19.1.1 WHQL is better IMO.
IMO ATI is implementing new Render paths (new approach) in 19.2
and we need to wait for them to do it right (1-2months) and we can easily add +10% performance to our Vegas (first for Vega 1 and 2nd Gen. then Polaris)

DAL (Display Abstraction Layer):
-> https://www.x.org/wiki/Events/XDC2016/Program/amd_dal.pdf

So TimeSpy for my setup at ~8800pts (was 8.5k)

 

For me TimeSpy score was 7600 vs 8400. Guite a big difference!

 

@OnnA I'm having trouble trying to get a good UV/OC on my card. I've tried to apply the same settings that are throughout this forum, without much luck

I picked up a Sapphire Nitro+ Vega 64 and I can't get HBM2 stable above 1050MHz. If I don't set HMB2 to at least 951mV then it won´t go higher than 800MHz.

Do you think flashing the latest bios would help?

Here's my current setup which is only semi-stable. Seems to crash every now and then:

 

First try to more safe approach.
1680MHz/1050 and test it.

First find some stable GPU OC then go for HBM2 OC
Small step
Nitro+ is very good Vega (one of the best)

Note:
You can go for 1050MHz HBM2 with Mem.Timing Profile 1 (it's faster, and yields same results as higher Auto/P2)
Try it -> look at Giantmonkey Videos he has 'only' 1040HBM2 OC

 

OnnA I have one question. I have the Nitro+ vega 64 like lyno, and the new fan control in Adrenalin 2019 just doesnt work properly and zero rpm stops working. If I install just the driver from 19.2.1 and get the old control panel from 18.11 will I get the old fan control with just max-min temperature+rpm sliders? Or is it driver related? thanks

 

Driver bug IMO
Remember to set 1st stage as 0% (no RPM) and it should work NP.
Test & observe -> If works OK, if not, set Min as 16% (i have it like this on my LC)

 

@lyno

The most obvious reason for crashes when under volting would be the voltage setting for each power state and driver latency.

So, to find a stable UV, the first step would be to set each of the power states between p7 and p5 at the same voltage. By doing this, you eliminate driver latency, and can then work around to find a good compromise.

I would just leave it like that, because then you're not open to driver changes, and the power saving will be very little either way.


Edit.
Changed p4 to p5.