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

I have C0 rev.
I saw 1760MHz at ~1.150mV
So i have some headroom, now mostly playing at 1680MHz 1.087v or 1.093v with -5% (GPU is @ 150->200tW)

 

Early but not first batch of the Vega 56 Pulse here, doing 1500 to 1550 Mhz at 1.0v which due to Hynix memory that's about the limit performance wise. (Which saw a higher use during the subsequent batches for this GPU though Samsung does pop up occasionally still.) Early testing was mostly around 1.2v to 1.25v and 1600 to 1700 Mhz but it saw limited gains in most titles, reversed it and noticed I could drop to around 1400Mhz before it really started decreasing performance and after some reading and learning about the bandwidth limitations and memory speed importance and how the GPU scales settled for 1500 Mhz up to 1550 for boost and a 1.0v voltage which has held since I dialed it in with some tweaking for the fan "curve" AMD Adrenaline 2019 introduced to maintain OK cooling though at a slightly higher noise level and higher fan speed as a trade-off until this stepping stuff is fixed.

Benchmarks differ a bit but from game testing 1300 to 1400 Mhz is where performance really starts dropping off, memory goes much quicker since it's already hitting some limits at the default 800 Mhz now raised to 900 Mhz which is holding so far.
Some games differ a bit, Far Cry 5 and New Dawn recently and it's rapid packed math is a good stress tester and lower-level API titles can help push the GPU and also dial in exactly where stability issues start occurring and then tuning either clock speeds or voltage and stabilizing everything even with the GPU under or near full load and heavy activity when the cores are actually being used efficiently.

 

Is there any available information/documentation regarding Vega silicon revisions out there? A few quick tries on Google didn't give anything.

According to GPU-Z My Sapphire Vega 56 Pulse has C3 silicon but I have no idea how reliable that is?

Anyway, my card seem to do really well with both undervolting and overclocking, though I haven't pushed hard on the latter as I try to stick to a reasonable power consumption and noise level.

During my many hours benching I seem to be able to reach 1630-1640 MHz at about 1000-1010 mV, I'm confident I could get the card stable at 1680-1700 MHz but even a bump to 1030 or 1050 mV would take me past my power target so I haven't really tried finding the required voltage. Regardless, it's safe to say that this card is definitely the best quality silicon out of any graphics cards I've owned.

And Vega is really fun to play with, which in this particular situation means tweaking and benching.

 

Got a Vega 56 8GB with 3 games (RE2, DMC5 and The Division 2) for €269. Can't wait to install it tomorrow. This is 1 part meant for a complete system overhaul somewhere in July (Ryzen R9 3850X). I love the smell of new hardware in the morning.

In case someone from the NL as well is interested:
https://www.megekko.nl/product/1962...yte-Radeon-RX-VEGA-56-GAMING-OC-8G-Videokaart

 

"I love the smell of new hardware in the morning." -> Me too

 

I really hope my card comes with Samsung memory. I've seen several gpuz screens of the Gigabyte Vega56 Gaming OC that it has Samsung memory. But even if it has Hynix, still an awesome deal.

 

All this talk got me in the mood to do a couple of quick Time Spy runs just to see where my 24/7 settings land me on clocks vs. voltage.

Note; This is a cold run and the power limit has been increased to +20 rather than the +5 I apply for my daily use. The clock and voltage settings are set in a custom WattMan profile and I use Time Spy and GPU-Z to bench/log.

Below is a couple of snippets from the GPU-Z log, the top one being a selection from GT1 and the bottom from GT2.

Notable is that neither hits the power limit (165W +20% should land at ~200W) though GT2 is, as expected, further away from it.

I would have to increase either voltages or the p7 clocks (currently 1742 MHz) for the card to boost higher under this particular workload, probably both. As I generally stick to a power limit of +5 that's not a limitation I have to deal with under normal conditions however.

Anyway... with these settings I seem able to consistently hit ~1650MHz at, or just under, 1000mV. Pretty darn good if I say so myself.

Other items of note;

• Still using Adrenalin 18.12.2 as I've seen few reasons to upgrade at this point.
• My card has Samsung memory but I've not flashed any 64 BIOS to push voltages, nor do I intend to.
• The HBM is benchmark stable at 950MHz but I get random lockups in, of all games, Thea: The Awakening when running 950MHz or above regardless of what memory timing level I use.
• I'm currently running the memory at 925MHz with Memory Timing level 2, as that's both stable in all tested workloads and seem to generate the best performance out of the different memory timing options.
• Graphics score of 7363, skipping the CPU test as it's irrelevant for GPU testing.

Code:

        Date        , GPU Clock [MHz] , Memory Clock [MHz] , SOC Clock [MHz] , GPU Temperature [°C] , GPU Temperature (Hot Spot) [°C] , Memory Temperature [°C] , GPU only Power Draw [W] , VDDC [V] ,
2019-03-05 14:14:02 ,        1650.0   ,            925.0   ,        1107.0   ,               36.0   ,                          49.0   ,                  36.0   ,                 186.0   , 0.9938   ,
2019-03-05 14:14:03 ,        1650.0   ,            925.0   ,        1107.0   ,               36.0   ,                          50.0   ,                  37.0   ,                 186.0   , 1.0000   ,
2019-03-05 14:14:04 ,        1650.0   ,            925.0   ,        1107.0   ,               37.0   ,                          50.0   ,                  38.0   ,                 183.0   , 1.0000   ,
2019-03-05 14:14:05 ,        1650.0   ,            925.0   ,        1107.0   ,               38.0   ,                          51.0   ,                  38.0   ,                 183.0   , 0.9875   ,
2019-03-05 14:14:06 ,        1650.0   ,            925.0   ,        1107.0   ,               38.0   ,                          52.0   ,                  39.0   ,                 182.0   , 0.9875   ,
2019-03-05 14:14:07 ,        1649.0   ,            925.0   ,        1107.0   ,               39.0   ,                          53.0   ,                  40.0   ,                 184.0   , 0.9875   ,
2019-03-05 14:14:08 ,        1649.0   ,            925.0   ,        1107.0   ,               39.0   ,                          53.0   ,                  41.0   ,                 184.0   , 0.9938   ,
2019-03-05 14:14:09 ,        1648.0   ,            925.0   ,        1107.0   ,               40.0   ,                          54.0   ,                  41.0   ,                 183.0   , 0.9938   ,
2019-03-05 14:14:10 ,        1646.0   ,            925.0   ,        1107.0   ,               40.0   ,                          56.0   ,                  42.0   ,                 191.0   , 0.9813   ,
2019-03-05 14:14:11 ,        1645.0   ,            925.0   ,        1107.0   ,               41.0   ,                          56.0   ,                  43.0   ,                 189.0   , 0.9938   ,
2019-03-05 14:14:12 ,        1646.0   ,            925.0   ,        1107.0   ,               42.0   ,                          57.0   ,                  43.0   ,                 187.0   , 0.9938   ,
2019-03-05 14:14:13 ,        1646.0   ,            925.0   ,        1107.0   ,               43.0   ,                          57.0   ,                  44.0   ,                 186.0   , 0.9875   ,
2019-03-05 14:14:14 ,        1646.0   ,            925.0   ,        1107.0   ,               44.0   ,                          58.0   ,                  44.0   ,                 186.0   , 0.9875   ,
2019-03-05 14:14:15 ,        1646.0   ,            925.0   ,        1107.0   ,               44.0   ,                          58.0   ,                  45.0   ,                 186.0   , 1.0000   ,

Code:

        Date        , GPU Clock [MHz] , Memory Clock [MHz] , SOC Clock [MHz] , GPU Temperature [°C] , GPU Temperature (Hot Spot) [°C] , Memory Temperature [°C] , GPU only Power Draw [W] , VDDC [V] ,
2019-03-05 14:15:57 ,        1647.0   ,            925.0   ,        1107.0   ,               56.0   ,                          68.0   ,                  57.0   ,                 176.0   , 0.9938   ,
2019-03-05 14:15:58 ,        1648.0   ,            925.0   ,        1107.0   ,               56.0   ,                          69.0   ,                  58.0   ,                 173.0   , 0.9875   ,
2019-03-05 14:15:59 ,        1649.0   ,            925.0   ,        1107.0   ,               55.0   ,                          68.0   ,                  57.0   ,                 173.0   , 0.9938   ,
2019-03-05 14:16:00 ,        1649.0   ,            925.0   ,        1107.0   ,               55.0   ,                          69.0   ,                  58.0   ,                 175.0   , 0.9938   ,
2019-03-05 14:16:01 ,        1650.0   ,            925.0   ,        1107.0   ,               55.0   ,                          68.0   ,                  57.0   ,                 170.0   , 0.9875   ,
2019-03-05 14:16:02 ,        1650.0   ,            925.0   ,        1107.0   ,               55.0   ,                          68.0   ,                  58.0   ,                 170.0   , 0.9938   ,
2019-03-05 14:16:03 ,        1650.0   ,            925.0   ,        1107.0   ,               55.0   ,                          68.0   ,                  57.0   ,                 171.0   , 0.9938   ,
2019-03-05 14:16:04 ,        1649.0   ,            925.0   ,        1107.0   ,               56.0   ,                          68.0   ,                  58.0   ,                 173.0   , 0.9875   ,
2019-03-05 14:16:05 ,        1648.0   ,            925.0   ,        1107.0   ,               55.0   ,                          68.0   ,                  57.0   ,                 175.0   , 1.0000   ,
2019-03-05 14:16:06 ,        1647.0   ,            925.0   ,        1107.0   ,               55.0   ,                          69.0   ,                  58.0   ,                 180.0   , 0.9938   ,

 

Got the card, came with bios F4, flashed to the latest F5. Sadly it has Hynix memory. Almost ready reinstalling Windows and maybe going for Linux. Perhaps going for Mint.

 

Can anyone give me any advice on how to tweak better my Vega64?

I read already some of the page of this topic and try messing around with OverdriveNTool, it is set:

P6: 1647 @ 1106
P7: 1732 @ 1150
Mem: 1180 @ 1012 (I`m afraid I`m pushing it too much)
PowerTarget: 30% (in the image below is 50% for testing only)

I mostly play R6 Siege but I do play other games like Metro, SOTTR, Wonfenstein, etc..

On R6 Siege using configs above, the maximum Core clock was 1680Mhz (I never been able to achieved +1700, is ther any trick to do so?) but during the game it sits on 1650~1670 and Chip Power on 250~260W but maxed out at 290W (again, the image below was after playing R6 Siege and running Superposition Benchmark, so thats why the Chip Power was +300W all the time)

I must say that I bought this card as an used one (reference one with ekwb block) and the owner probably change its bios (I`m not sure) but its running under a Custom Closed Water Loop.

So, what can I try to get the most out of the card? Do I need such a high HBM clock? should I lower a little bit more the voltage? I think the card is having great temps as its pretty hot here in my city, in fact i`m using HWiNFO + RivaTuner to monitor and sometimes VRM Temps show up and sometimes not.

Spoiler

https://i.**********/nncGqtTR/TEST.png

sorry for any misunderstanding, english it's not my native language.. greeting from Brazil. thanks!

EDIT: BTW, I`m using Adrenalin 18.12.1 as any newer driver I had some Delay while ALT-TABing from a game to desktop.

 

Keitosha
Congrants for the new toy
Enjoy the power of Vega
Great price

 

@Thiago Diniz

Do not surpass 1150MHz on HBM2 @975v.
I have daily setting at 1120-1140 @ 931v | 943v
GPU never surpass 50deg ! (65deg is Hotspot lol)
1680MHz is very good OC for heavy games - IMO keep it that way.

 

8GB HBM vs 6GB GDDR

 

what is the best way to test oveclock/undervolt stability?

I'm going crazy here... I can play/bench fine one day, the next day, same clocks/volts, same game/bench = crash! It seems to crash only on cold boot, after a while I can play fine again...

Like 10x Metro Exodus Benchmarks runs on will do the job?

 

Yup, Unoptimised heavy game is good for testing (Metro or AC)
or Time Spy or FS.

Always give at finish notch higher.
Like me 1680MHz at 1.081v is upped to 1.087v etc.

 

>29k ('road' to 30k begins)
-> https://www.3dmark.com/fs/18270986

 

I tried the 19.2.1 drivers, but still the clocks jump between 1200mhz-1650mhz when benching Metro Exodus. So was with the 19.2.0 which is even gone from official AMD support page.
Returned to pre Metro Exodus driver, 19.1.1 and clock hold stable.
Is somebody else experiencing that kind of problems?
And I alwayd DDU (clean install) when updating drivers.

 

Damn just pulled the trigger on a Sapphire Vega 56 Pulse!! I could not resist anymore @ 289 EUR!!!
Should be an upgrade from my actual 290X from Sapphire that I bought back in 2014. The 290x start showing it's age due to mainly 4 Gb VRAM Limitation.

I'm all over the place on making a tweaking concept (voltage, clocks and so on....)

 

To add to that vega https://www.ebay.co.uk/p/Bykski-A-x...er-Edition/5022817957?iid=113210649115&chn=ps