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

So the Fury GPU had this issue:
https://community.amd.com/thread/188642
(290 / 290X suffered from a occasional black screen crash before this GPU.)

And for Vega it's this:
https://community.amd.com/thread/223844

Hmm I wasn't even aware one could lock states to minimum via Wattman, that's a interesting little discovery and a pretty useful one too potentially.
Since the GPU in the thread is a liquid cooled Vega 64 that means reference so not just a issue with custom card models though I guess it's not too frequent but it seems this workaround of locking the GPU to a higher p state is a decent workaround until AMD can resolve it in software via newer drivers. (For those who do get black screen crashes with Vega when it downclocks.)

 

The Crew Beta 2 test GPU

Spoiler: Antigraviator test

Spoiler: Agony test

i7 5960X 4.6GHz OC is better than ZEN 1800X 3.6-4GHz in this game...

 

Some guys on YT (Hardware Unboxed) compared Custom 1070Ti MSI with reference Vega56 lol
I will still go for V56 + Good Quality 1440p Freesync (no Brainer here.... still this is #Better choice for Gaming)

-> here:

"I have a Sapphire Nitro + Vega 64 and I dont think it runs slower in any way, but it runs definetly cooler than a reference card.
The comparison between a custom nvidia card and a reference amd card is not really fair because both chips (AMD and Nvidia) are very temperature sensitive, so the AMD card loses a lot of Performance due to the poor reference cooler and the MSI cooler on the Nvidia card is among the best coolers in the current generation.
I think the difference between custom and reference AMD cards is more than 5%. While OCing my card i could get at least 15% more performance out of the card while still running below 70° and at moderate sound levels."

 

Yeah, I can't for the life of me understand why a reviewer would do this. It only makes the reference card look weak. And yet we have seen it happen in so many reviews. Just not good enough. Reference vs reference and custom vs custom for a fair comparison.

 

Just remember to Adjust UE3/4 Ini for any of UE Engine game on Radeon (nV has it in-Driver so no bottlenecks there)

in GameEngine.ini (MyDocuments)

[TextureStreaming]
PoolSize=4096 or 8192 for Vega
MemoryMargin=20
MemoryLoss=0
HysteresisLimit=30

 

Something is wrong there. Vega 56 and Vega 64 are equal. Must be a driver issue.

 

It's Game side issue, driver has nothing to do with it IMO, UE3/4 needs to be well polished for Radeon to work properly (or you need to Tweak Ini as i do)
FuryX is ~980Ti/1070 in this game.

-> http://gamegpu.com/rpg/ролевые/vampyr-test-gpu-cpu
Here test with some OC'ed 5960X 4.6Ghz

Im on ZEN so no 4.6Ghz for me
We're close to this numbers with some good Mem @3466 CL14 or better with Zen @~4Ghz

 

Fury-X is still holding its own in that bench. It was a card i tried to buy when they launched.

 

Vega 64 is doing great. But by God, the 1080Ti is a beast.

 

^^ Yup, this round nV is a clear winner @750€ Price point.
But in 500-600€ bracket (Yes MSRP in mind) Radeons are the Best Buy IMO (FreeSync in Mind)
So everybody should be happy....
not counting nV owners that not have G-sync Monitor -> majority of 1060/1070/1080 users -> lot's of FPS...
but nowhere near the Radeons Fluid Smooth FreeSync Gameplay

Spoiler: Freesync is still better

Ask @gupsterg He had nV 1080 without G-sync and Gameplay was worst that Fury+ Fresync and Yes those not familiar with it will not comprehend it. (now he is equiped with Vega64 WC)

-> http://www.overclock.net/forum/67-a...ro-duo-fiji-owners-club-286.html#post26318727
-> http://www.overclock.net/forum/67-a...ro-duo-fiji-owners-club-287.html#post26363886

"An update as an ex Fury X owner.

Last night was 1st time I did some actual gaming on MSI GTX 1080 Sea Hawk EK X. Prior time has just been spent benching/tweaking ThreadRipper setup.

I have ~150hrs+ clocked up on SWBF with Fury X, so I'd say experience of how it performances capped at 88FPS and using FreeSync I have alot.

I have not OC'd GTX 1080, due to it being on WC and how nVidia Boost 3.0 works I see flat clocks of ~1965MHz in benches.

So SWBF same Ultra 1440P setup but no V-Sync and FreeSync, I get range of say ~120-144FPS. The gameplay at times does not seem as smooth as Fury X capped at 88 FPS with FreeSync
Next I tried Lords of the fallen, I get somewhere around ~90FPS+.
Again the experience was not as smooth as Fury X IMO
I then thought let's cap FPS in games to say 90, use same refresh rate, with V-Sync, to my suprise I found no such option in nVidia driver panelI called it a night then and plan to try other things today to see if I can smooth out gameplay 'experience'.
On another note my Fury X sold a day or so ago. Packed and ready to ship today. Some what some regret has crept in on swapping GPUs.
Not welching on sale though.I have sneaky feeling that variable refresh rate tech has spoiled me
I may have to dispose of GTX 1080 sooner, rather than later"

..... let's us hope that those mind $hare will soon end and some of nV users will embrace what Radeon have to offer....

 

and some Messed Game (NP because it's Early Access )

 

Yup, I will have my RX VEGA XTX 64 LC in friday
Yup -> (622€ in total)

I have to do it, 622€ ! w/3 Years Warranty -> Just a steal -> 1440p Ultra + AA + reshade AA for me.

==

 

Very nice. I would love one of those LC cooled Vega 64's.

 

Yeah guaranteed Samsung HBM2 memory and no need to flash or overclock anything. Downvolting can still reduce the power draw total but the cooler can handle the heat though Vega can scale down really well to lower voltages.
Although the 64 has the full 4096 cores and a higher base clock plus an additional p state and the HBM states differ a bit too though it's still very tweakable.

Wattman is a bit different on a Vega but it's not too difficult to get the hang of how things work although the HBM voltage slider should either be removed on Vega or renamed to what it actually does for voltage floor / minimum amount.
Better pricing and availability and Vega could have competed nicely with the 1070's and 1080's though I suppose AMD's GPU's are still selling out but they also ship less of them compared to NVIDIA.

The dud with primitive shaders and that other thing is a bit of a miss but the GPU is still fast and exactly how much of a gain this could be if it had worked without developers having to manually add support for it (Through a API that's not even available yet.) well outside of benchmarks I wouldn't expect too much from it in normal gaming scenarios since there's other things and bottlenecks and it depends heavily on the game as well.

 

OK, THX for the Info

Now we can do some business
Tell me what i can do for it? (I need to process data before arrival)
I will use Our OverdriveN_Tool (No MSI AB will be needed any longer)
JonasBeckman -> Do some math and gimme screen what i can get.

 

It varies a lot from GPU to GPU though the V 64 GPU's eliminate the issue of having Hynix HBM2 memory modules.

Now these aren't terrible for the chosen clock speeds but even with a compatible V 64 bios to flash for the memory voltage (1.250v upped to 1.350v) the max clock speeds will likely not be much higher than 900 Mhz or maybe 950 Mhz.
For Samsung 1000+ is possible and the GPU sees a pretty sizeable performance gain up to about this amount and then a small gain going higher.

For the GPU core itself the stock models are high quality and the water cooled versions retain the higher binning since they run at a higher frequency through a additional 8th power state as I recall, 1700 Mhz or thereabouts.
Also uses a higher voltage for these at up to 1.250v against the max cap at 1.200v on the air and V 56 models.

GPU Z and Wattman should be able to confirm speeds, voltages and once re-enabled also the hot-spot sensor. This will be a bit of a hotter area on the card and might see throttling if it goes above 85 degrees Celsius I believe it was, HBM starts downclocking around 75 degrees although timings might loosen earlier than that around 65 from what I've read.
(Water cooling keeps things nice though for the liquid Vega 64's.)


For power draw the stock is around 300w I belive and probably a bit higher for the liquid version and their additional power state, with some tuning this can be reduced pretty noticeably to around 200w or lower and further still if you downclock the GPU a bit.

This needs a lot of testing and measuring, GPU works through a combination of power total and clock speeds maintained and balanced through temperature and other parameters so on average the speed tends to be between the two highest clock states and voltage will fluctuate as a result, GPU can adjust itself if needed if it requires more power but it can also see GPU driver crashes if you go too hard on the downvolting and some games will run hotter and a higher speed than others which is the games I tend to test with since if these are stable then that's a good indication the current settings will hold.

For my own V 56 Pulse GPU this is not the first batch so it does use Hynix memory and clock speeds are kept at 850 Mhz up from 800 without any errors or indicators like graphical glitches, higher might be doable but would ideally require a bios flash.
For core I'm using 1500 and 1550Mhz down from 1536 and 1590Mhz but like I said the clock speeds tends to be in-between the two highest states so 1520 - 1540 when measured and retaining a similar performance level.

But at a pretty good 100mv reduction in core clock voltage since it's re-adjusted from 1150v and 1200v to 1050v and 1100v and it might be possible to do even higher with tuning, 1100mv is already what the p5 state is using though and I don't think the card can go below that without registry editing the power play values.

65 - 70 degrees load temperature in most games so I can't complain, fan speeds are upped a bit but the fan only applies in higher clock states so it's running at next to nothing during desktop and idle periods and such.


For Vega 64 liquid though I would probably start with the stock settings and see how it performs and how it does in terms of temps and what it's clocking to, Wattman, GPU Z or a in-game OSD like from Afterburner should all work to check on this.
Not really too sure how close to the edge AMD is pushing these since they're clocked higher than 64 air models and have a higher voltage though some reduction is likely possible before even going for adjusting the clock speeds although it also has the additional cores though being higher binned it could still be possible to downvolt these nicely.

The turbo setting in Wattman is best just ignored, you don't gain much at all for pushing the GPU voltage and such to max, balanced is better for the stock and then custom for fine tuning the settings further.
(Even the power saving setting has been shown to not have too big of a impact on performance while maintaining a much lower power draw and as a result less temperature on the GPU too allowing it to operate without throttling.)

It's a bit of a readjustment with how the sensors and modules on the GPU maintain clock speeds rather than working on absolute values or settings but it's also not too difficult to learn and check how well it all works and what your own GPU is comfortable and stable with, if a certain stock binning quality and memory chip quality could have been upheld it would have been nice to see a more balanced default setup for core clock speeds, memory and voltage instead of AMD pushing the defaults pretty close to max for what, a 5% gain or so perhaps on average? 10% at best? It's performance but not much of a gain for how much additional voltage and heat it causes.

Or can cause, again the liquid models probably hold up really well since the stock PCB and configuration they're using is very high quality. (From the cooler to the actual hardware such as the PCB and components on it.)


EDIT: Still a lot to learn on these GPU models, I like working with them though as long as the driver is co-operating which so far it's been no real problems though at launch there were some quirks but I guess that happens for new GPU releases even if GCN itself is quite mature by now although AMD has tweaked it a little bit.



EDIT: Also between the V 56 and Fury there's about a 30% gain or so in performance, it's good but it entirely depends on cost. I put the budget for a GPU upgrade at 500 - 550 EUR which for GPU prices here and how that fluctuated that was leaning to a 1070Ti depending on availability and pricing, maybe a 1080 since these are often on sale.
(Vega 56's hitting up to almost 1000 Euro during the worst of it, then usually hovering around 700 - 800 Euro meaning there was no contest between these and the similarly priced 1080Ti which was also faster.)

Anyways the 64 should be another 10% or so or maybe a little bit more, I mainly upgraded due to concerns for when the next generation of cards would actually be released and the 4 GB VRAM which well since framerate nearly doubled in games like Watch_Dogs 2 and Wildlands I'd say that a really severe bottleneck although how much will vary and turning down settings can still see the Fury deliver good performance.


And now we'll see if AMD and NVIDIA has something for 2019 I guess, from the sounds of it AMD's 7nm GPU could be similar performance to Vega but more efficient and probably with some tweaks and tuning that makes it pull ahead a bit more.
(Kinda like the 500 series against the Fury, weaker hardware but it uses it really well.)

Of course the Pascal or Volta or whatever the 1100's from NVIDIA will be should also be a interesting comparison. Doesn't seem like either of the two are in a hurry for now though.
(If AMD is working on a successor to GCN that's going to involve a lot of work but it should remove some of the limitations too such as ROP's against the cores and such.)

 

Proud (and recent) owner of a V56 Pulse here (Samsung mem), and I've been tampering with ocing via Wattman for the best part of the last days.

I need posts like this one form @JonasBeckman , if I'm going to gain any extra fps from the card while lowering the wattage.

So far, I've managed to go down to 1070mV for p7 and 1060mV for p6 (was stable even lower, at 1050mV but had lower scores/fps at FS Ultra -used for testing), while adding a bit o Power (+20%). P6=1537, P7=1592 (the "stock" values). Next step, HBM oc...will start at 900MHz and see what I can get.

@OnnA: Have fan with your new toy!