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

I have the same issue with my rx 480. simple oc just crashes. before had it at 1450 from 1288.

 

Also driver to driver is different sometimes also.
Best are WHQL ones, Stick with this if You can.

I have upped some V to maitain stability even in Betas (don't be Hot head -> this GPU is a Beast)
Learn how to Fly.... then push some more

 

@Mr. Sunshine

At 1.2v, the hot spot will run at around 100c by design for reference models, so temps are fine if you don't intend to go higher on voltages.

However, best to stick to a single driver until you know what your card is capable off.

 

Using 19.3.3 drivers. Card is great. Just trying to find stability is issue. Ex.

P6 1682mhz 1050mv
P7 1717mhz 1120mv
stable for a 5 hour run on Heaven maxed out. Games were soild also, clocks were 1642-1662mhz. But if I try same setting now it crashes.

Its like the card is messing with me. I think I found that its stable then...BOOM crashes.

 

I dont have issue with push power into card. Could card less if it uses 120w or 400w so if maxing voltage makes it stable then so be it. I think I'll grab some thermal grizzly kryonaut and pads and see if I can drop it a few degrees.

 

This damn card has such weird power issues. Ran 3 hour Heaven loop stable at:

P6-1652mhz@1110mv
P7-1702mhz@1150mv
PL-30%
Avg.1655mhz

Was still looking to see if I could get any lower with voltage or higher with clocks. Ended up crashing. So figured was good at last stable levels.....Nope.
Now those same setting wont stay stable for more than 15min. But thing I noticed. First time they were stable I was getting 1653-1658mhz but those same
setting are now make card run 1664-1680mhz?!? I dont get it.

 

^ You'll find the sweet spot soon
Also don't use -% POW when setting the UV (always at 0% or +1)
(- ) tend to lower clock speeds or give fluctuations.

 

A few guidelines:

1. Higher clocks require disproportionately higher voltage.
2. Power and temperature limiters will throttle clocks, so these will fluctuate by design.
3. A stock Vega 64 runs at 1500 MHz on average under moderate load.
4. Most people do not and likely can not run over 1700 MHz on average in most conditions.
5. Mixing under volting and over clocking has limits due to (1), (2), (3) and (4).

Given the above, and considering you're already running 15% higher on core and memory clocks, be reasonable in your expectations and try to work with the technology's general power and feed back design.

 

Yeah stock is probably going to boost to around 1550 - 1600 Mhz if there's headroom for that and then it's usually going to ramp up a bit with voltage requirements for taking it further though AMD's stock is already pretty hefty at 1.2v which is why the undervolt potential is so good if you balance clock speeds and voltage though binning and then differences in bios and the Vega 56 against Vega 64 and Vega 64 water also factor in.

Results seem to mimic what I see with final clock speed usually right between p6 and p7 though it is possible less throttling might push it closer to p7's speeds but with all the sensors and variables here it's kinda tough to get it to stay near or at p7 constantly in fact it's probably not designed for that.

VII also changes the Overdrive API and has less states plus the Radeon Adrenaline 2019 drivers changes and a very high fault tolerance on the memory likely also altered how much people can push these GPU's where some might have to be lowered down a bit and others might find there's more headroom because each GPU is just going to be different and then factors with GPU core binning and coolers and tons of other little details, I'm no expert at all mostly just trying to find compiled info and reading up on how this works and it's a pretty flexible card but it has a lot of restrictions too and the drivers changed it around even more now.

Non-reference designs of the Vega 56 and a few of the Vega 64 GPU's plus PCB's such as the Nano variants also play a role, overall I suppose in the end it's just down to the usual meaning a ton of testing and trial and error finding what's stable and what might cause errors or at worst crash outright and then dialing down and testing again or finding some throttle coming in effect like the HBM2 temperature or hotspot and then the power play table though soft-modding can push this thing even higher though I guess once you're past the 50% default it's going to drastically drop off in just how much you gain and even at 50% extra it might already be a bit much unless the GPU is under water or similar ensuring there's no other limiters coming into effct.

Still need to find better solid info on the Vega HBM2 degradation rumor though it might just be other issues or drivers changing it around or users having pushed these past safe levels early on and then the memory error tolerance and such has kept it going at least for a while until it hit stability problems. (They are heavily overvolted too over the stock specs and also AMD's initial specs for voltage and speed.)


EDIT: And then for both Vega and VII the Watman bugs, issues and misreported values and other problems though AMD has slowly fixed up the Adrenaline 2019 driver release but I think it still has a few snags.

Guess nothing's here really new though, early adopters of Vega likely had it rougher before the first few drivers started shaping it up though later changes also modified GPU behavior and overclocking results so yeah it's fairly flexible in how it scales but it does require tinkering and hitting the limits can take some trial and error figuring out what's limiting the GPU now and how far you can push it before it errors, throttles or just causes a crash.

Nothing new but the way it fluctuates and has a dynamic clock speed can make it a bit of a process.
(Stable at p6 and then a boost speed for p7 and the GPU will probably try hitting that but yeah overall it likes to be somewhere mid the two even if you underclock the core speed where it should be fine at p7 constantly no stress at all.)


EDIT: But as a Vega 56 user and Hynix type HBM2 modules my own situation is a bit different, Pulse has a few differences as well though it works the same as the 64 but hits different limits and then changes in PCB also factor in a bit more when everything's all close together due to shortened length.
(Around 1400 - 1450 Mhz and lower than that performance starts reducing more noticeably, up to 1500 and above and the gains are minimal outside of benchmarks so 1550 and a reduced voltage for power usage and temps with barely a difference in performance is where it's at currently though I am also quite aware CPU and RAM are a limiting factor among others, system is showing it's age now quite clearly.)

 

@JonasBeckman
1550MHz is enough for every game
Pesky unoptimised ones needs more than 1600MHz.... or even 1702MHz (but those are few, uff)

Note:
Now Gaming 'Machine' is about of New APIs & Optimisation
HitMan 2 now is DX12 Title with +40% Perf. Uplift
Some PS4 Games heading towards PC using VLK API..... list goes on, and on

And because of simple Fact that Next-Gen Consoles will have ZEN + Vega, we can be assured that our Vega will live much longer than it's successors.

 

Yeah for benchmarking squeezing everything out is useful and taking the card to it's max but for regular GPU and CPU workloads it seems to peak fairly close after the 1500 Mhz range though HBM2 memory speeds can affect the results somewhat if you can hit 1 Ghz or a bit above though the 2 stack design and bandwidth is also a factor over VII and it's 4 stack design where this is improved thanks to that design.

And yes DirectX 12 and Vulkan and certain advantages and improvements over DirectX 11 though AMD updating and fixing the D3D11 drivers would be interesting to support things like the driver command lists from D3D11.1 which if you were to use GPU Caps Viewer it should show as unsupported. Part of the multi-threading design and also something NVIDIA spent a year or two for with their drivers though pair the GPU with a strong Intel CPU and it can usually force it pretty well enough and newer AMD CPU's are also doing well now that games are actually starting to utilize above 4 cores and the expanded use of threading and more.

Not the whole story but one part of the performance difference between AMD and NVIDIA and AMD's gains under D3D12 and VLK though just because the game is using X feature doesn't mean X feature might be used to it's fullest so even if AMD updated their drivers there would still likely be a varied gain from one title to another and some would likely need driver compatibility work to get around new problems and issues that might pop up.

Also even with this implemented D3D12 or VLK would do it better than D3D11 and you'd still need the CPU to be able to feed the GPU so it has to be fairly fast and hexa core or better can't hurt though it'll still be a while before it's the new standard I guess. Yet another thing that's taken a ton of time to try and dig up more about and read up on but I still have a lot to learn or at least try to get some understanding on the subject.


EDIT: Some sources via AMD's reddit for this info or at least what I've found for it.
https://www.reddit.com/r/Amd/commen...to/ddjncl6/?utm_source=share&utm_medium=web2x

https://www.reddit.com/r/Amd/commen...ir/cwyv3fm/?utm_source=share&utm_medium=web2x


EDIT: Though a deep level knowledge isn't really important for the average GPU user or anything, I'm mostly just curious about any flaws or shortcomings or limitations and have too much free time.
It's interesting though but quite a ways beyond my own knowledge level and skill in this matter although I still find it enjoyable to read up on and get some insight into thee matters though overall it doesn't really *do* much as such.

In the end comparisons are mostly going to be performance and how well the hardware stacks up against the competition with software (Drivers) and features closely following this.
(Well for my own opinion at least, hampered by software issues can be a real hassle though this can be the game, driver or even the OS at times.)

 

For me all the new drivers after 19.1.1 are crap. Clocks jump around and performance is noticabely worse.
Mr.Sunshine try 19.1.1 for OC/UV tests.
I am still on 19.1.1 myself.

 

As for D3D11.1 MT -> look at BFV
Yes, ATI Driver supporting it since BF (back then better was Mantle), then was introduced in new BF 1.
We have MT DX11 Driver... Not every game has code in it (Thats why it's called Lazy Coding)

Compare AC III with new Enchanced AC III (Remaster is fully DX11.1 Game, not best Opti. but good enough)