AMD Catalyst (Modified Build)

FXAA suckz it is just an advanced MLAA....

Only SSAA rulez!

 

(sorry for the another spam again)

Sorry ..SSAA in dX11 or DX10 not working..so I guess you will use MLAA..

Using FXAA here, and games looks a bit better.. Games like; Arcania (Gothic 4!)_Risen_Divinity 2 DKS_Witcher 2_Dirt 2 DX11_Dragon Shiitt 2 (Dragon Age 2) DX11 (still haven't tried others..)

Using MLAA on the above really shows **** on the screen, like pix-elated quality onscreen but edges are Anti Aliasing.

(AMD USER HERE..RSGSSAA on ATI/AMD really rocks..but it doesn't work in some games, and MLAA reall sucks ballz)

-Edit-

More info..

MLAA was originally developed and introduced by Intel, for their CPU bound MLAA Anti-Aliasing.

As for FXAA check the following
http://timothylottes.blogspot.com/

-> http://developer.download.nvidia.com/assets/gamedev/files/sdk/11/FXAA_WhitePaper.pdf <-

 

Sorry but with the right program ssaa works perfectly with dx11 or dx10 for amd/ati... but maybe in an other thread

 

@Raiga

Sadly I got to go with @GaBBa-Gandalf on this FXAA thing. It's an new improved method of applying the old already failed MLAA on GPU and it's likely to become only an marketing trick. This is at least the 6th variation of 1 concept.

Also "Note, FXAA presets 0 through 2 require an anisotropic sampler with max anisotropy set to 4" as in 4 pixels.

Currently we are in funny generation of blaming the incorrect direction and of course GPU developers are having field day on this since games are designed so damn bad no hardware can handle them.

@commonly

NOTE! This does nobody any good, but it's a start ;P

Here's an Intel HD Graphics and general Sandy Bridge 12 Shader graphics fix to get OpenCL functional:

Download: Intel OpenCL SDK v1.1 gold (Runtime is enough by system Architecture)

Copy this to CMD and run as admin:

32-bit / x86 - Windows Operating Systems:

Code:

SETX INTELOCLSDKROOT "%ProgramFiles%\Intel\OpenCL SDK\1.1" /M
fsutil hardlink create "%SystemRoot%\SysWOW64\intelocl.dll" "%ProgramFiles%\Intel\OpenCL SDK\1.1\bin\x86\intelocl.dll"

64-bit / x64 - Windows Operating Systems:

Code:

SETX INTELOCLSDKROOT "%ProgramFiles(x86)%\Intel\OpenCL SDK\1.1" /M
fsutil hardlink create "%SystemRoot%\System32\intelocl64.dll" "%ProgramFiles(x86)%\Intel\OpenCL SDK\1.1\bin\x64\intelocl.dll"
fsutil hardlink create "%SystemRoot%\SysWOW64\intelocl.dll" "%ProgramFiles(x86)%\Intel\OpenCL SDK\1.1\bin\x86\intelocl.dll"
REG DELETE "HKLM\SOFTWARE\Khronos\OpenCL\Vendors" /v "intelocl.dll" /f
REG ADD "HKLM\SOFTWARE\Khronos\OpenCL\Vendors" /v "intelocl64.dll" /t REG_DWORD /d 0

Follow this guide:


and add Display to 'Intel HD Graphics' as was added on picture for NVIDIA adapter and you are done. Worst scenario nightmare would look like this:



but then again OpenCL is present:


as for usage well, ther'es a bug in call GPU device on Intel OpenCL 1.1. Still need to figure how to lock the Intel OpenCL to GPU device while or let AMD handle CPU OpenCL device.

 

so funny thing, i saw that you used that OCL and DC benchmark, so i went and downloaded it again for ****s and giggles and ran a test. The DirectCompute part literally took lik 5 seconds and netted a score of D186413.5, OpenCL was at normal speed and got a score of C8126.8.

TwL i was wondering if you encountered the same thing. I have 11.8 drivers installed with the latest OCL drivers that came with it (like 11.8 came with newer OCL drivers then 11.7).

 

that actually is incorrect.. 11.7 comes with newer OpenCL than 11.8 did. (OpenCL 1.1 AMD-APP-SDK-v2.5 (684.212))

Can't much test DirectCompute, atm, as I am combining GPUs and CPUs here trying something out. Current score on DC for DirectCompute is
'D251756.6' I am using NVIDIA, AMD and Intel chips, but only dong the DirectCompute Test with AMD card. However, you got weaker CPU and AMD
driver likes to combine in itself now days and using Sandy Bridge here at 4,7Ghz so ~186.4k sounds just fine for Q9450 + single HD5850 setup.

It is truly sad no one has written any OpenCL or DirectCompute tests supporting this scenario hardware. Combining all 3 methods or <any> card
+ CPU would give one hell of an kick to people start asking this kinda technology for regular applications or games.

 

...that function just with Z68 not with P67, right?

 

LucidLogix, yeah...

Although, heard there might be some MSI board P67 out with iGPU enabled. Perhaps it might be possible in future to flash and fake display as above.

-edit-

Then again this Intel GPU is so weak thing don't even know, if it's any good no matter what the board is.

 

Thanks for posting.
I tested this out and must say it works pretty well.
There is even the option to edit the entire calculations.
Now I can finally force some descent AA that doesn't give me a huge performance drop in some games.

 

You board is with the Lucid® LT22102 Chipset?

 

not that lucky.

 

glad to see your still doing this, I may have moved green but still lurk good job man !

 

I think so far that and one older are the only of their chips that can do multi GPU 3D... but the boards such are on are a bit expensive, like the MSI Big Bang Marshal. :/
Still, would it be worth going for such a board, so that I can combine my 5870 with my 580...
Would also require from me to get a second loop installed.

As for FXAA, it is slightly blurry compared to SSAA, but still better than MLAA and Edge Smothing AA.
Plus it doesn't add much performance penalty.
You should check it out, I bet you will find the additional files interesting.
I can imagine that some kind of control panel GUI to change the settings could be useful.
New version http://hotfile.com/dl/125975700/ef4f...dude_9.7z.html has ability for sharpen.h
Makes it look much closer to SSAA.

 

Actually I was working on green yesterday. The R280.

We have CUDA 4.0.1 + OpenCL 1.1 now officially supported + all the new features. The good news is also that this new '280.19 beta' (NVIDIA cursed
the Geforce to beta while Quadro gets WHQL, btw) will allow hybrid setups + DC/OpenCL/PhysX/CUDA supports all together. So, we are sure looking up for
NVIDIA side driver update next.

Only REALLY bad new son the R280 is that it's slow I mean REALLY slow to process OpenCL/DirectCompute. It's even several times slower than Intel
HD Graphics. Something most be bugging on NVIDIA OpenCL code or they require some kind of own optimization for it.

You mean to tell me that for this buggy system I would actually pay to get some board? No way in hell as far I've looked on this system.

LucidLogix doesn't work. Their optimization of GPU selection, Multi GPU functionality on situations like today world people mixing GPUs is
unfunctional at best as their software crashes / not recognize GPUs correctly and their GPU supports no features like compute which would be the first
thing what comes in mind while speaking of combining multiple technologies to gain power.

As for FXAA...

Well, I did check the tools provided. I dislike the idea, but since people makes such a fuzz about it I'll take another look. Technically even I didn't
see anything really ideal about FXAA, btw, which reminds me I just did test 1 game which had FXAA in it as an option. Funny, can't remember which game it was.

 

Which board do you have?

 

I am using ASUS P8Z68 with the VIRTU software.

LucidLogix Study:
Seems that as I was using/monitor attached to the dGPU/AMD card nothing works, but yesterday I decided to check up if I activate PCIe lane 0 first as
in attach monitor to Intel graphics then install the VIRTU and then activate the monitor seems to be working. However, the processing on secondary
NVIDIA GPU is 0 and processing something like DX11 crashes, DX10 seems to work, but the initializing the AMD graphics card behind the Intel HD
Graphics seems to take a while (All games starts at black screen and when alt-enter couple times AMD GPU activates correctly seems to be the 5
seconds lucidLogix Logo is on screen, lol.).

However, as soon it starts seems performance is pretty 'OK' getting clean +30FPS on Crysis (VSYNC was on). Sadly while you enable this technology
PCI-Express bus speeds decrease allowing only 3.xGB/s throughput as it seems to activate the PCI-Express lane for some reason for Sandy Bridge GPU split with dGPU.

As for NVIDIA PhysX + PCI-Express <-> DMI relation:

Functionality:
Still working seems that the Intel DMI2 (the Southbridge connected DMI interface which is PCI-Express x4 total) -> will through as x1 max
22-30FPS(minimum) on 30K of particles (currently at batman we use around 28K max which is largest PhysX game I know). So, PhysX on x1 lane
might after all be enough for now. (I also downgraded the Geforce from 700/1728/1100 downclocked by my modified BIOS to 350/864/550 which
decreases the performance even further it seems as I was thinking that x1 PCIe lane can't handle more bandwidth anyway, but seems the thing is that
CUDA/OpenCL/DC and PhysX are using the card internal speeds. So, downclocking a card even on x1 will produce lowered bandwidth and process speed.)

The regular for 30K of particles of 1x9800GTX (128 Shader Units) would be around 174-175FPS minimum. Downclocked and at x1 as said is 22-30FPS(minimum) on 30K of particles which is enough.

Which remind me to mention everyone on new PhysX seems that, if there will be a game out RIGHT now using any PhysX engine above what was
released at v9.09.1112 those engines are purposely coded by NVIDIA to work on CPU (while AMD card is detected) no matter what hacks you apply.
(heck though we were already done with this ). So, do not believe the ideas that PhysX would be hardware based, if there's anything really NEW game using new PhysX.

PCI-Express<->DMI(CPU) minus DMI2:
At x1 mode the primary DMI on CPU will be working 6,8GB/s as long Intel HD Graphics is either enabled and LucidLogix is disabled or Intel HD
Graphics are disabled. While x4 mode apply this changes(Seems these are shared also) as the DMI2 takes more bandwidth seems maximum
throughput left for DMI1(CPU)->Primary Graphics(AMD) will be around 5,3GB/s still giving from GPU->CPU around 6,3GB/s.

 

so your going into modified build nvidia driver next? great looking forward my friend

 

@commonly

Perhaps it's time for some 'where are we'-update...

What I was currently working on was to understand the LucidLogix functionality through understand the Intel Z68 DMI interfaces/memory
interfaces and PCI-Express 2.0 Interfaces and started to build new AMD GPU build against all the new crap out on market. (No worries old X38/X58
boards will work 50 times better than LGA1155, lol. This Intel design is truly the most unstable crap I've seen yet not in words literal means as crashing
unstable, but nothing is on sync, 1 side of hardware pulls another as above post explained to @mineria and this platform needs a lot
of work to feel solid, although, it has several advantages on memory speeds/latencies to completely freeze free interface(doesn't get stuck at
high clocks, but locks down to lowest), however, the board bandwidth runs out for it to be anything better than a toilet paper).

* Using OpenCL 1.1 / CUDA 4.0.1 / PhysX 9.11.0621 at R280 NVIDIA driver release. It's just so slow I don't dare to even suggest it as replacement to what we have on first page.
* I added the new clean-up should serve a bit better
* Been building an CrossfireX driver couple days ago to test main build has to be done against 11.8->11.6b (Searching processing through some little bit more stable code than these). (as errors below explained + more issues on utilization)
* Currently looking at 11.7 WHQL release and the 11.7 hotfix releases and couple older smaller code/faster code modules. So, we could have a bit better single card driver which actually could draw correctly(Multiple errors at vertical sync), process faster the errors(Multiple errors and jumps on specific setups) and get at least the OpenGL(Multiple errors at middle level OpenGL (probably v3.0-v3.1 extensions)) and video decoding(Multiple errors at vertical sync) to function correctly.

 

Actually, these new drivers are quite good, but well more I use them more bugs I reveal. Took a while to get hold where is AMD driver team at (a week),
but looks like we are still surfing on same waves just better updated drivers since 11.2.

 

I have OpenCL 1.1 AMD-APP-SDK-v2.5 (709.2), CAL 1.4.1523 (VM) installed and i believe i got it from the preview driver set.