Vulkan's lauch - 16.15.1009

Decided to give these a try, comparing the atiapfxx.exe blob dump showed quite a few changes since 16.1.1hf1

Driver version for these is.
16.15.1009_20160213_1641
(For 16.1.1hf1 and hf2 it's 15.301.1801.1001_20160203_1023 )

These were added to OpenGL "defines"
<property Type="String">PowerLimit</property>
<property Type="String">UpHysteresis</property>
<property Type="String">DownHysteresis</property>
<property Type="String">SclkThreshold</property>

These Crossfire profiles were updated:
3DMarkVantage
World Of Warcraft
3DMark11
3DMark NEXT
Tomb Raider (2013)
Alien Isolation
Sniper Elite 3
3DMark SkyDiver
Dirt Rally
XCom2
Ashes Of Singularity
Star Wars Battle Front
Fallout 4

These Crossfire profiles were added:
Far Cry Primal
DX9 Sample - Pick (some test program?)
Rise of Tomb Raider
The Park

These profiles were changed:
Internet Explorer
Tomb Raider (2013)
Rise of Tomb Raider (Was added)
Grand Theft Auto 5
Monster Hunter Online Benchmark
Rainbow Six (2015)
Shadow of Mordor
Assassins Creed Syndicate
Call of Duty Black Ops 3
Microsoft Edge
Star Wars Battle Front
Fallout 4
AutoCAD
vrProfile (Some test program?)
Just Cause 3
Unreal Paris
Elite Dangerous

And these profiles for Open GL:
Borderlands 2 pre-sequel
Metro Last Light
Modo
Unigine
Civilization Beyond Earth
X3 Reunion
XCom (Was added)
DOTA2
This War of Mine (Was removed?)
DIRT

OpenCL:
BuildOptLegacy

 

URL:http://support.amd.com/en-us/kb-articles/Pages/Radeon-Vulkan-Beta.aspx

>This driver package has been updated on 2/17/2016 to address the items listed under Highlights

Version:AMD Radeon™ Software Beta for Vulkan™ Version 16.15.1009 Beta 4
(16.150.1009.0000 Feb 16)

Highlights
VulkanSDK v1.0.3.1 - Compatibility issues addressed
The Talos Principle - SSAO Fixes

 

None of the vulkan demos work for me still with this new driver either

 

I compiled the Vulkan demos from https://github.com/SaschaWillems/Vulkan yesterday, and they ran fine on my Fury X, with two exceptions which had some assertion fails.

 

Wonder what they changed in the rainbow six crossfire profile. It was disabled before, and it's still disabled. Did they double dog disable it?

 

That explains why i see i difference in Rise of the Tomb Raider while using CrossFireX

 

difference good? or difference bad?

 

Are you talking to me?

 

I don't know about BF4 (I don't own the game).

While I agree that Thief ran significantly better with Mantle, it strictly required >=4Gb GPU RAM (the texture streaming was hard-coded for 4Gb, 3Gb wasn't enough -- I am not sure if a later patch fixed that or not) and it was an old UE3 engine (thus the DirectX renderer of that engine was rather old in comparison to newer game engines with better DX11.x renderers).
So, I think a 280X with 3Gb GPU RAM performed rather bad in that game with Mantle (compared to DirectX).

But I played a lot with DAI (I carried out a "competitionist" play-through throughout the second half of 2015 and I started a quick run-though in early 2016 to see some alternative choices - which is still in progress) and I often tried to switch back and forth between DX11 and Mantle.
I often took some time and re-benchmarked them after almost every game and driver updates or after I unlocked some new areas in the game where my machine couldn't keep up a constant 60fps.
After the last game patch was out, I re-benchmarked some with some old and new Mantle DLL files.
My conclusion is that DX11 is better:

1: Loading times are significantly worse with Mantle.
The Mantle shader compiler is painfully slow. It feels like the game has crashed when it has to rebuild everything (whenever the mantle shader cache gets automatically invalidated after you change the Mantle driver version or you manually delete it just in case before benchmarking).
Even after you theoretically have everything you need in the mantle shader cache (you visit areas you have already seen before - but after the last cache purge) the level loading is still significantly longer (I estimate it to be ~2x in average, although I never used a timer but perception is what is important for me in this case...).
This is especially true since the driver can maintain a system-wide persistent cache for DirectX/OpenGL shaders which also solved the problem with the stuttering cut-scenes (which was a nice aspect of the mantle shader cache before DX11 shaders could be cached).

2: You have to use a Mantle version <= 9.1.10.75 to have tripple-buffered V-sync on Windows 10 (you had to find the right version and you have to manually copy the right DLL files to the game's folder if you wish to keep the system drivers up to date...).
I think this is ridiculous how they decided not to fix this even though the Mantle driver version was incremented since this issue has been widely discussed (so, they did some changes [it's not like the code has been frozen forever, never to be touched again], yet they didn't fix a known bug which seems to be trivial to fix because older versions work fine while new versions don't seem to perform better or offer new features which could interfere).

3: Random graphics glitches -while not limited to but- seem to happen distinctly more often with Mantle (most of these are very subtle and I only ever notice them on screenshots when I happen to search through them).
The most notable is the "anchor" effect (or more so the lack of it) on the Inquisitor's hand (because sometimes it can be confusing both combat-, and story-wise) or the visual effects in the old tevinter building where the time has been completely frozen (you get to see a "time paused" deamon<->mage battle with lots of magical and other effects in a "halted" state around you - or not when some of them are randomly missing...).

4: If you care to benchmark various areas and scenes, you might find that performance differences are not only small but also random.
During the most demanding scenes in the game where you might see the lowest fps numbers (like some areas in the Raw Fade or the Deep Roads) Mantle might show some small but distinct improvement (something like ~53 vs. 51 fps which could be appreciated, given these are the only areas/scenes in the game where the fps is often below 60 with my settings - disregarding the rare random bumps caused by background processes or "the unknown") ...but... it's not like you can't find some other areas/scenes where DX11 provides a smoother gameplay experience overall (areas where you don't expect reproducible fps bumps, yet it can randomly happen).
Yet again, this is more prominent since the introduction of the generic DX/OGL shder cache (thus Mantle no longer have the upper hand granted by the unique, game engine based implementation of a shader cache but DX11 can now also benefit from a similar caching method).

5: It seems to keep my CPU utilization and thus the power-draw slightly higher on my i5-2500K.
This is the opposite of what I would expect since Mantle (like all the low-level APIs) should allow for "offloading" the CPU. Yet, it seems like this particular implementation cares more about utilizing the capabilities of the CPU than optimizing the power-draw. I guess it runs a "busy" thread (which doesn't care to sleep as much as possible and thus let the CPU cores do the same and enter into various sleep states) on every CPU cores/threads regardless of the actual computation needs.
This is far from a real concern of mine. I wouldn't care to pay the price in electricity, had I got some distinct advantage (but, well, see above...).

---

So, all in all, I usually play with DX11.

In the older days I tried to "force" Mantle from time to time because it had a slight advantage in some cases and was theoretically better but since the first "Crimson" driver, I cemented the settings to DX11. There is simply no distinct benefit but only possible disadvantage of using Mantle in DAI.

And I assume other Frostbite 3 games are similar, unless you have a relatively weak CPU with relatively fast GPU(s). But even when I had a 290X CrossFire (well, a 290X and an overclocked 290) for testing, I couldn't see any improvements (so I think I would be fine even after disabling 2 of the 4 cores of my CPU with a single card, thus it would take a really slow CPU to see the benefit...).

 

So Vulkan is out, i have a simple question,

AMD will release a driver (Crimson) let's say 16.2/16.3 with vulkan support correct?

 

Will the sun rise tomorrow morning?
Nobody can guarantee you anything but these look like sure things (both the sun and that AMD will sooner or later release a stable/WHQL driver which has Vulkan 1.0.x support).

 

yes.

 

It's good different. It didn't magically fix the whole stutter problem, but it's much smoother now in some areas/situations. There is still some annoying stutter every once and a while.

Not everything has improved, but nothing is worse. The biggest difference i noticed was in the soviet installation.

 

Crossfire frametimes in mantle are way better (its basically a flat line), max and min fps are also higher. load times in DA:I are insanely long, but not in BF4, so I think that was a game specific issue with mantle, not necessarily a API issue. I have a fairly decent CPU, and it is a noticeable difference between DX11 and Mantle.

 

What The F*-ing Driver! I just did my time to time LuxMark...

Spoiler

Code:

Device		driver			score		utilization

Test 1		262K Trianlges		score		utilization
Fury X		15.150.0.0 W8=>10	29370		93%
Fury X		15.200.1040 Tahiti	29790		94%
Fury X		15.200.1023.10 Fiji	29869		96%
Fury X		15.300.1025.0		31391		95%
[B]Fury X		16.15.1009-160213a	33070		100%[/B]

Test 2		488K Triangles
Fury X		15.150.0.0 W8=>10	3474		95%
Fury X		15.200.1040 Tahiti	3506		95%
Fury X		15.300.1025.0		3620		94%
[B]Fury X		16.15.1009-160213a	4013		100%[/B]

Test 3		2016K Triangles
Fury X		15.150.0.0 W8=>10	1753		100%
Fury X		15.200.1040 Tahiti	1788		100%
Fury X		15.300.1025.0		1779		98%
[B]Fury X		16.15.1009-160213a	1841		100%[/B]

Now I have to update my Fury X benchmarking post!!!

Edit: No real improvements in D3D9~D3D11 and definitely not in 3DMark overhead test.

But none the less, this compute improvement over initial performance is impressive.

 

What is that test? Just for computing power and overal usage %?

 

It is benchmark for LuxRenderer. I still use old 2.0 version, to have comparison for Old HD7970 scores. And utilization is what GPUz shows as GPU Load.
(That's because many drivers had poor OpenCL utilization with HD7970.)

 

Is there a way to mod the crossfire profiles from this driver into 16.1.1? Currently, this driver causes crossfire to break when I alt-tab from games, but I'd still like the updated Xfire profiles.

 

Final, stupid post from me. This driver doesn't have Radeon Settings, right? From extracting the DLL I can tell it has everything else. So installing this, and then Radeon Settings from 16.1.1-Hotfix should make it ok, right? I saw Fox232's results and now I have to try them

 

Any chances that sweetfx/reshade will be working with vulkan?