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

Hello guys, i just want to thank everyone that contributed to this thread with valuable information.

I have a reference blower design V64 and i'm having trouble finding the best possible settings for undervolting it.
My only limit is the fan, which i don't want to surpass 3500RPM.

Which settings should i be using for the best possible performance while keeping temps at 75C?

 

The design itself isn't too bad sucking in air and blowing it out the back of the case but it does get a bit louder and the single fan design has problems shifting the heat generated by the GPU once it's at full load.
And then different software can have varied impact on the GPU's overall performance including total heat and power draw and compensating for these without the fan kicking up to max speed to lower thermals back down.

Temperature limit can be lowered down but that might also see the fan kick into full speed early as you get closer to the thermal target and that might then exceed the desired max RPM value.
For voltage I'd lower by 20mv or so at a time and see how the GPU takes to these changes during testing. Under ideal conditions the GPU should compensate for total power draw by reducing core clock speeds but if it hits a spike it might exceed stable speeds and run into a GPU reset or driver crash so you don't want to lower it too much to begin with although with Vega 64's they're usually binned well but the core clock speed is also higher and running at a higher voltage for core and memory at up to I think it's 1.25v core instead of 1.2v on Vega 56 although it's not as highly clocked as the water cooled version.

1.150 to 1.100 shouldn't be impossible for voltage and you might consider enabling AMD chill and test how that impacts GPU performance comparably to power draw and overall temperature. Power limit to 20% instead of the full 50% should work but it could also be possible to set a negative value to lower it down but this might impact performance if it has to throttle GPU core clocks if it needs the additional power draw.
(-20% seems popular, forcing a lower power draw at a small performance hit.)

Core voltage might also work at even lower values although I recommend testing as you reduce it further, concern is mainly Wattman in this case occasionally pushing either core clocks or voltage to defaults speeds or above so that's going to impact thermals and performance or stability at worst if it returns to 1.2v or higher although I think AMD has improved that in the later drivers although with higher display resolutions or VSR I think I remember reading about the GPU voltage hitting 1.2v for some users even with Wattman set to use lower which is going to generate more heat and might see the fan speeds rev up again.

For total power draw down from 300w - 330w or so as it boosts to 280w or ideally even lower would make it easier on the fan and total heat, chill helps here which is why I recommend testing that in addition to lowering core voltage little by little at a time or even using the low power Wattman setting instead of custom and that should draw even less but without being as customizable so performance will be slightly impacted as it reduces clock speeds.

 

HBM_2 in the SpotLight
Also included with OP.

HBM_2 Good Performance Gains (My Own experience, lot of testing)
HBM_2 works similar to HBM_1 [ in 5MHz Clocks Steps e.g. 950,955,960 etc. ]

For Vega 64 & 56 as well i suppose

950-1000MHz Great "first" Gains
1000-1050MHz Some additional UpLift noted
1050-1100MHz Another Great Bump (Noticable, Best All around, not too Hot)

Finally some Creazy OC:
1150, 1175, 1185 & 1200MHz Nice for 3DMark or Heavy Games like AC: O (Yes the gain is here, it affects Lows 0.1% & 0.01% -> Best for Chill and FRC)

1150 needs *962-968mV (Depends on Game & Chip OC 1717-1732MHz)
1175 needs *981mV
1180 needs *993mV
1200 needs *1000mV and more.

On my Vega is pointless to cross 1230MHz (Not stable enough, too much V needed for stabilize)
Max (OC) for anything is 1210MHz to 1225MHz @ 1000-1025mV

Note 1:
It is for the Best to have normalized OC, you don't need more than 1702MHz for any game (in non XTX Vega)
But you can have some really good results when combined GPU+HBM OC

Make 1585-1647MHz /1100MHz HBM_2 or
1575-1632MHz / 1150MHz HBM_2
in Another words try to find some Golden Spot for your GPU (Mine is 1707/1717/1732 GPU with 1150-1175 HBM for Heavy Games)

Note 2:
Remember to check MHz in HWiNFO64 Just to be sure tou don't give too much of juice.
E.g. Forza MS7 have OC profile of 1702MHz but Game don't use more than 1568MHz ! so you don't need those mV & MHz at all for this game to play smooth.

Note 3:
Remember to check actual MHz in Adrenalin WattMan, just to be sure you get clocks as it should be.
e.g. 1700/1702/1707 Yes there is non 1701, 1703, 1704, 1705 !! -> It is important for OC to work FlawLess.

Note 4:
Make sure your Big Daddy is cooled properly (I mean DIY)
With an little effort you can end up with some better OC with no throtling.
Addtional Fan for the Back and maby one more for the Cooler Air for GPU to suck-in.

Note 5:
Im using OverdriveN Tool v. 2.7.0. Beta4 with my Custom RX_VEGA_64_AIO_Soft
Adjusted for Low tW and long Gaming sessions.

OverdriveN Tool 2.0.7 beta 4
-> https://www.dropbox.com/s/equ297p3otqu28n/OverdriveNTool 0.2.7beta4.7z?dl=1

Spoiler: OC Tool OverdriveN Tool 2.0.7 beta 4 screen

* -> It is Infinity Fabric Planar Voltage Floor (Interconector mV)
Actual HBM_2 Voltage for Vega 64 is 1.356mV (Similar to my HBM_1v-Mod for Fury)

In theory we can Hit up to 1300MHz OC (When Cooled properly, i will conduct some testing in Winter Time )

 

Just try to add Some Fan/s to the Rig Air flow
e.g one Fan to cool it's back (It can be done like DIY )
and another to Blow some cool air for GPU blower.
It can be some old 8-9cm Fan or 12-14cm in the Rig
It is the best possible way to keep this Monster Runing like a charm

 

Those memory speed values for voltage are measured or set in wattman?

Because my HBM behaves in a different way. It could thanks to me using the air blower version, but it goes like this:

950-1025MHz works with 950mV on the voltage floor, but anything past 1025MHz will only work with 956mV.

If i set like 1050MHz with 950mV, it downclocks straight to 800MHz. I'm currently running 1100MHz/956mV along with p6 1557/1012mV and p7 1652/1062mV. Might drop core voltages a bit if the card gets too hot.

 

UPD. Here is the response from our @Unwinder

"Vulkan OSD should now work properly in Vulkan applications when frames are being presented from non-graphics queue. The only currently existing Vulkan application behaving this way is DOOM (only on AMD graphics cards), because it is asynchronously presenting frames form async compute queue in AMD codepath.
Such presentation implementation makes Vulkan OSD rendering rather expensive in performance hit terms, because it requires forcible graphics and async compute queues synchronization.
So now rendering OSD in DOOM/Vulkan on AMD cards may reduce performance by approximately 5% (the same applies to Steam FPS counter in DOOM/Vulkan on AMD), but I believe that there is some space left for optimization and it can be improved in future versions."

Second, my apologies on delaying 4.5.0 release a bit. Sadly I'm having some issues in my real life now, so currently I cannot invest too much time into MSI AB. We planned to release 4.5.0 right after Win10 Spring Creators Update but it is delated due to issues I mentioned above. I hope that I'll have a bit more free time on the next week so I'll be able to post 4.5.0 RC in this thread, then we'll put MSI AB 4.5.0 and RTSS 7.1.0 in Guru3D downloads and make them visible to update servers. It won't be much different to previously posted 4.5.0 beta 5, just some minor GUI fixes and one new feature:

 

After Primitive Shaders not being available in any API, NGG is cancelled too for Vega...

https://www.mail-archive.com/amd-gfx@lists.freedesktop.org/msg24458.html

 

Guessing the initial implementation never worked correctly and now it's been made official that it's being cut at least for this generation of hardware. I remember reading about some of these features alternatively working on a per-software basis via API instead of globally but no-one heard anything about said API or SDK from AMD so not too surprising of a development that it's cancelled although it would have been nice for these features to have seen support.

Focus is probably shifting to Vega 7nm then for it's compute and AI learning capabilities and use in workstations and next year we'll hopefully hear about Navi and whatever form this will take.
Last of the GCN cards perhaps? Sounds like it can technically scale above it's current limitations but it will be interesting to hear more about it once AMD has something to show.

Which might not be very much to begin with but who knows, I don't expect too much for the first half of 2019 myself.

EDIT: Which means with desktop GPU's NVIDIA is now unchallenged for up to half a year with the potential for the 2000 series graphics cards to push ahead even further with more than just the extreme or enthusiast class of the 1080Ti maybe.
Depends on where the 2080 and 2070 land and pricing for these if they can maintain MSRP pricing or if it will be higher though Vega is already troubled a bit from this although it's coming down now a bit.

EDIT: Well there's popularity as well to consider and the age of the Pascal GPU's and people wanting to upgrade, various factors as usual I suppose.


EDIT: Hmm well say there's a up to 30% difference with the 1080Ti and the Vega 64 if not more whereas it competes fairly well with the 1080 non-Ti outside of a few titles where it either really pulls ahead or falls behind.

That does give NVIDIA a good bit of room for the 2070 and 2080 allowing them to be faster than older Pascal GPU's while also leaving the Vega GPU's from AMD behind although it can't be too close to the 2080Ti's performance level either of course.

Need to check the hardware specs again but if the 2080Ti will be faster by a nice 15 - 20% or maybe more then that does also give room for the 2080 to be around or a bit below the performance level of the 1080Ti while still faster than Vega and then the 2070 a bit below that.


Although until benchmarks are available (Non-official ones especially.) it could also end up with them being similar in performance to their Pascal versions with the 2080Ti having the hardware to pull ahead but it's also costlier although with a 2 year period now and people probably ready to upgrade or sell off existing GPU's then that might bridge the cost a bit although it's still likely to be more of a GPU more for the enthusiast segment of course.

Will be interesting to see how it all plays out, nothing against Vega and it's a nice GPU but it might be in some problems and the price fluctuations are going to be a even bigger issue if NVIDIA has even more cards out that can exceed it especially if they can also be cheaper to purchase. Price and performance I suppose plus NVIDIA's existing market lead and popularity as GPU of choice and then also hardware like G-Sync monitors although these too have somewhat of a price premium.
(Especially the newer models with the additional HDR support and other features from that new chip.)

 

Regarding this specific video of HW-U, Vega64 vs 1080, I'm having some objections, and need a 2nd (and even more) opinion.

First of all, when I got my V56, I made a short comparison with my oc'd 290, on my current 8320@4500, both at 1080 and 1440p.

What really made an impression, was the difference in fps at DeusEx:MD, between DX11/12, at 1440p.

The difference (using the exact settings from Guru3d GPU comparison at this bench), is quite impressive, especially after HBM oc (950MHz, core@stock, -130mV, +20%power, custom fan profile), as seen below.

DE:MD DX11 Hi, Guru3d Settings
1st Run 2nd Run
Min 47,4 52,1
Max 94,5 95,2
Av. 71,7 71,2

DE:MD DX12 Hi, Guru3d Settings
1st Run 2nd Run
Min 44,8 37,4
Max 83,7 84,2
Av. 61,5 61,3

I've run the bench multiple times, and the results are constant. Now, the video at HW-U, draws a completely different picture, one I honestly cannot comprehend.

In addition, I'm quite skeptical about the use of Frostpunk, as it clearly messes up all the statistics, resting "comfortably" way outside the normal SD range.

The use of multiple Unreal titles, Hairworks on@Witcher3 (I guess with tesselation=auto, not optimized), Battlefront 2@1440p showing a cleat nVidia lead when every single previous bench was a clear Vega win, only makes me question the methods used.

A final note: Every single bench, without nV control panel/Adrenaline settings visible, lacks credibility (personal opinion). Overclocking the cards, without showing the final settings, is also a dark spot (especially Vega oc, without proper uv...just stating max core freq, doesn't mean anything).

 

^^ For me Deus MkD working a lot better (Smooth*) in DX12.
Maby i will test some more in time (2nd Playthrough) and use Reshade w/EyeCandies at DX11.2

Im Using Ultra Preset with everything at Ultra, no-Blurrrr 1440p 10Bit 70Hz with Chill 64-70

Editi:
* -> more constant

 

It might be that fps are more constant@DX12 (not sure though), but the point was that the overall score @DX11 is >10% better, while the comparison video shows a 3% lead@DX12.

 

Monster Hunter: World PC Performance Vega 56 (only, no OC, no Ultra settings)



Side note:
It's imposible to have average 60FPS 1080p on GTX1080 at Ultra settings preset !
That's why some WWW testing this game with Custom settings (some settings are at Low -> SH Diffuse: Low ! more info here)
On Vega 64 is possible to have ~60FPS in 1440p, in this game Radeon 56 is better than GTX1080.

 

Some addtional Vega tweaking for Best UV tW Long Gaming sessions.
Performance/Thermal/Power - in that order.
Game AC: O with Ultra Settings AA + Reshade + SMAA

177tW (was ~210tW) at 1440p is a nice Gain (No significant FPS Average drops only ~1FPS)
Game still shows Great Performance 64FPS average - Chill is set to 64-70

==

Spoiler: HWinfo sample readings

 

From reddit (FreeSync Gaming on Radeon)

Spoiler

If you buy an Nvidia card + Gsync monitor it cost *** more than the alternative.

So if I buy a 1440p 144hz IPS HDR Freesync panel its $400
1440p 144hz IPS non hdr Gsync panel cost over $750

Vega 56 + Pixio PX277h = better gaming experience than exists on anything Nvidia has to offer.
Nvidia has no HDR GSYNC 1440p 144hz IPS panels except a single monitor that cost 2,000 which price is a joke.

 

When GSync was introduced, the modules were $100 to the OEMs to integrate into their monitors. Now like the drug pushers they are, they charge up to $500 for the modules. This is why those GSync monitors cost so damn much, and why the OEMs would love to get NV to support Open Adaptive Sync.

 

Yup, Radeon Users are Clear winners in 2017-2020 period thanks to Great Value without sacrificing Image Quality at all
One can have Great Gaming PC build around ZEN 6/12 or 8/16 (non-X are the best #Bang4Buck IMO) with Vega 56 or 64 (prices are Great now) + FreeSync 2 Monitor (You can have really Top Dog IPS HDR for less than 480€)

Spoiler

That is if you don't have Bias imprinted into the Head by 'marketing' bla.blah (That Bias leads to even more absurd prices )
And we are Happily standing on the (Normal) other side and watch them struggle to buy even mid-range now (Yes it's their fault, we don't support this kind of 'business' #WeAreNotTheBusiness #WeAreGamers !!!!)


PS as a side note:
You know when i was reading all that WWW Vega reviews i was worry about it actual performance, then i watch MindBlank Tech and other guys in OC forums + read about the specs (and knowing that ATI always do toooo much mV on their GPUs, yes i've had UV for R280X for Fiji and now Vega lol -> It's a Radeons Tweak Pattern)

Now after all that fuss, im happy with the Vega XTX, It is really great GPU after all.
Always think for yourself, that's my motto

If you want to really test your GPU then run DeusEx MkD in DX12 Mode at All Ultra or ME: Andromeda

Note:
Im not counting AC: O cuz' that game lacks good optimization, it will change with AC: Odyssey thanks to AMD Gaming Evolved,
AC: O is good for tW Testing in extreme conditions (therefore 8/16 CPU is strongly recomended for this one)

Spoiler: Buyers Guide, #BeSmart

1440p Gaming PC for Ultra-High AA Gaming?
Zen + Good Mobo + 16GB 3466MHz CL14 + Vega 64 + FreeSync + SB-AE or Z = 1200-1500€

As you know me im not into flame, but more into One have a reason, and think about Gaming future.

 

Hi, which bios is better for Vega 64 reference model?. Sorry my language.

P.D - I have 016.001.001.000.008730 bios. I have installed MSI AB and this is my settings:

 

If you planing on Liquid Mod (You need to WC Vega of course) then you need XTX one.
If you want to remove Stock blower and put Morpheus one on it, then also XTX is best option for you.

PS.
Just remember to have Backup of your original BIOS
How is the upgrade? Are you happy?