AMD Catalyst (Modified Build)

[TwL]

yep, using one. That is how it should be done.

All you need is basic DVI to VGA and 3 50-75ohm pins (doesn't matter really which) costs some cents. perhaps an duck tape to cover it wouldn't be too bad idea. What do you mean has anyone try'd it, everyone should have one, if you ask me

As what is the 'gain' well you get Memory Usage working without tricks. So kinda empty idea as we got 'PhysX Patch v1.03' which already gives us 'GPU Usage' from NVIDIA, but you do not need the adapter to PhysX to be functional with NVIDIA card you only need the patch applied correctly.

as for the card just an friendly advice. Try to find either "9800 GTX" as in NOT '9800 GT' because old 65nm cores supports usages clean while new does not. Next series which supports the readings are GT200 series. Basically what you want is fast memory in card for PhysX to work 'smooth' so buying an slow card ain't gonna cut it good which is as good as no PhysX card at all. I'd aim between 60Gbit/s - faster even half a 9800GTX style G92 core would handle the physX flawless.

-edit-

as for cooling and space for the card. hahahaha..

Here's something could turn some geek eye brows around. If you got decent cooling in box (AIR). This picture is from my box settled in 1 card slot 9800GTX cooled from front to back and out directly from OCZ GameXStream 850W right below it:



o and fail to mention it's 42c IDLE(with stock NVIDIA cover it's 60c) extreme load card gets to 58c(on stock blasting sh1t it's around 80-90c). So, it ain't always bad to have a bit custom solution plus less noise. (Sorry, for the cable mess took it straight up while testing this s775 board)

[CoreyPL]

Thx TwL for such good clarification on the subject

As for the cooling, my 800W Tagan is mounted in the same position as your OCZ I will definitely change cooling on GF when I get one, as I hate all the noise (already have Scythe Mugen 2 on CPU, Scythe Musashi on HD4890 and few low-rpm fans in case - so air flow is good ).

I can get 9800GTX 512MB used card a little cheaper then GTS250 new. As I recall GTS250 is only a very little refreshment over 9800GTX. So I think I will get 9800GTX.

And off to electronic shop for parts DVI-VGA dongle plus resistors will cost me about $2-$3, so no worries

[TwL]

I'd go with old 9800 GTX over 250 GTS any day. It's exact same card with exception 250 GTS is new and new is always bad when there's old to choose which is known 100% sure to work. Lowers clockied 9800 GTX(525/1295/825 Mhz) is 3 times more of an card than any physX could require even any where near(it dumps 125000 particles on screen with 20-24SPS & FPS which is 6 times more than any, atm, released game has.) only bad concept is that the older than GT200 doesn't have Voltage control so it will take more Power Consumption while IDLE I'd say that's a big minus, tho.

but as an oracle: 'make up your own damn mind' just giving a few pointers why/what to pick for what reason and my opinion.

[TwL]

Here's pretty cool screenshot using FluidMark 1.2.0 with 125000 particles with ATI Catalyst EDGE AA 8x(24x) with SSAA mode selected, AF 16x + HQ, Catalyst AI disabled, VSYNC Disabled & Second set is the same but 'Temporal AA' & VSYNC enabled. hell, it still spins like minimum 50-70FPS. Point was to see what happens when you dump everything you can on screen and it's 100% ATI only 68-70% NVIDIA because NVIDIA drivers has a bug in over 130000 particles they simply doesn't stay stable.


(Temporal AA Disabled)

(Temporal AA + VSYNC Enabled)

PhysX in suggested main page setup now is 30-34FPS faster than before by default which is pretty impressive I'd say and what is funny I have absolutely no clue why it's so as only change was ATI tweaks & NVIDIA Display v196.34beta was updated to 197.57 beta(which alone wouldn't do this change). It's possibly because no attached adapter on NVIDIA card uses less memory on it's own could be one explain.

[iamgod1985]

Originally Posted by iamgod1985

do these work with oem or engineering versions of the ati 5870?

^^^^??!?!?!?!

[TwL]

Actually, now I am not sure about would the 'Experimental' work or not. As I do have one of these cards here and I think it was used while testing the 'experimental' setup.

Originally Posted by perhaps too early statement

nope, I was building an OEM/Engineer version here yesterday, but I failed miserably as 10.5 driver and had to pull them offline has so huge bug I simply cannot make them work, sorry.

If I will find an solution to any working with oem/engineer version. I will note it for sure, but as for now the driver would have to built from so old driver I don't wanna waist your time with it.

Above quoted out. Tested here today not gonna happen on OEM/Engineer cards. Sorry, cannot get them working properly (or as should with any of the, atm, drivers onlined.)

[TwL]

Ok, Final build v2.11 is now up. Looks that this time I managed to put something together which actually works in all aspects, but hell. This is now 6th released edition (where 1 was pulled out) and every single time there's something, but I did my best on testing & installations.

Please follow the installation to the booth. There's some REALLY small things that affects the outcome and as for 'catalyst' Control Panel Apply 3D or Video settings. I do know it's VERY slow when clicking Apply and sorry about that, but functionality of it is working correctly and speed of changes had to be secondary against functionality.

-edit 01-

Well, seems I managed to get some kick in these. Even on an single card these are actually 809 points faster now than the original build was in 3DMark06 and around 200 Points on Vantage. As for NVIDIA side seems improving also we began from 55-70FPS on same 9800GTX card and now we are looking 132-141 FPS and +6 SPS on suggested drivers. Seems even old long loading times on games & something like 'Crash and Burn' test on Vantage are all now gone.

Have yet to find any problems (except a bit slow 'Apply' on Catalyst).

-edit 02-

Please drop an message of issues, if some or other comments & "ask before 'do-it-myself'" gets you much better result on installation progress.

-edit 03-

Seems that catalyst slow down was my personal bug in .NET not common 'bug' on everyone. Basically repaired it works fast now by simply going to 'Programs and Features' in control panel > .NET 4.0 > repair both client and extension.

Still waiting to ehar how is the QuadFire working on this setup. Would be interested to hearon 2 or 3 ATI Card Crossfire setups, if someone has such fun how the setup works with these.

[Hans Dampf]

This build 2 or 5 10 days ago:


This is build 2.11


No comments.

Just great work

[TwL]

yeah, would seem there's bug on running the QuadFire on HD5970 on multi displays still. However, the good news is we did solve the issue in some level. Not much of performance boost on CFX over single actually more of an drop with 3x 1920x1080x60hz.

but I will correlate this bug to next 'Final Build'. Sad news is we have to draw back to either 10.4b or 10.5a where I have to still take a look which actually drops a bit kick as 10.5a seems to suck on performance quite badly.

Final build will be above v2.20 when done as there's now 9 changes I counted we need to make to get QuadFire functional build up.

[TwL]

-UPDATE-

Ok, here is an problem summary of latest drivers on front page now [12.06.2010]:

* Stuttering in 'DirectX 11'-only (Fix found already)
* Issues on running 'Vantage' DX10 benchmark. (Runs very slowly. Unknown status of fix - working on it. Doesn't happen on single cards at all.).

and as for reminder x86/32-bit is still under experimental status even while support included.

-UPDATE 14.06.2010-

I am working to get this mess figured. Testing on crossfire/quadfire take incredible amount of time, but as it looks for the moment. Crossfire or multi-screen setup specifically on fastest possible functional driver combination(v10.4b baseline driver) is not working directly out of box and needs profile editing (which I will not do). Only other solution would be to use 10.5a slower driver and that is no solution when we are speaking of high-end graphics.

as for single display & single (ATI)card there is an incredible fast solution I built today from the scratch. the correct driver on front page would need 2 files from 10.5a (atidlxx32.dll & aticfx32.dll) to work without stuttering on DirectX 11. however, there is faster driver coming as soon possible.

as it looks for now, atm, drivers. There will be multi-screen/CFX driver separated from single (ati)card systems after all to get any sense what works what doesn't.

[TwL]

Ok, entire topic updated.

v2.57 is alive finally. Took a bit mirroring etc etc, but now it's here. Fixed every single issue there was no drivers way or another. Just read the main/first post.

There is no known bugs of any kind in this build and it is tested through QuadFire -> single card setup to be functional this time. It is the fastest ATI driver there is comparing to ANY official builds and yet it has full OpenCL/DirectCompute supports enabled including both Stable & Unstable builds CFX DX9 acceleration/scaling.

Build explainatory is simply:
* Use the Single Card build to any system which doesn't have CFX.
* Use the MultiScreen Unstable build, if you understand how to edit 'Profiles.xml' to get stable clocks for multi display 2D setup on CFX or single card multi display setup.
* Use the MultiScreen Stable if you do not understand above 'profiles.xml'.

[CoreyPL]

TwL, now if I were you I would check if ATI hasn't put you on their employment list, and if there is a paycheck waiting

But seriously, good job and thanks for all the hard work.

[TwL]

OK, x64 single card driver now updated it is good now.

Sorry if someone got the first bad RAR archive there was, but pulled the link as soon as I noticed it. Hopefully fast enough.

Anyway, to subject.. Finally the 2.57 finals are out and this time optimized to 2 kind of systems.

@CoreyPL

Thanks,

hehe, employee list. I don't think they like me. ATI/AMD forum mod actually last I checked banned me to their forums for asking questions about the bugs I am fixing here. ;P

[TwL]

anyway, this time I think we have 'real' tested final now online. Damn, they are fast and finally got around the 'bugs'.

hell, it's been quite some time now I've been thinking to write on 'bugs' section "NONE", but always something came up in either last minute notice or in testing stage. I think this time, if there possibly would be any level of error it will be found on multi display setups. However, even that got now pretty damn huge testing.
(even on equipment ATI/AMD has no clue of support on their own drivers.)


-edit-

If someone can find an error or slowdown on these drivers please let me know.

Always interested...

[MerolaC]

Downloaded and installed the last x64 version. Single card.
One word: Awesome!
Thanks for your time on making this!

[drutort]

dude you rock, found this info on the other forum and it worked, i spent few hours with the basic steps and had issues but your steps are like flawless it explains a lot of the issues that a lot of ppl including myself run into

thx again

[TwL]

thanks, these current versions are insane compared to earlier builds. Seems now everything actually works as it should without bugs from Multiscreens/CFX to Single Card setup.



-edit-

I just included Interesting notion for people with 'Multi-Display'+/-CFX+NVIDIA Card. There is NOW explain on main page HOW you should work with 'NVIDIA display' & Crossfire/Multiscreen setups and NOTE this HAS TO BE DONE. otherwise it won't work.

Good news is no reboots are needed while switching, but it's extremely tricky.

Here's guide[/slideshow of Images] I built in screens (also on first post):

( 9 Images series ) - - ( 4 Images series )

(Exactly same thing would have to be done on single screens for GPU Usage to be functional and OpenCL to display the NVIDIA card)

I still think true solution is everything no matter how hard it is to execute and as long it doesn't change the functionality of anything.

-UPDATE-

and found an bug on 'Single Card'-build. There is an 10.6 driver version OpenGL 4.0 bug with 'Heaven Benchmark 2.1' which will crash the driver sadly. The Multi-Screen builds are not affected as this is an new bug in 10.6 driver.

Bug is now updated on main/first page.

[TwL]

Added solution now on main post for recently found OpenGL bug in 'single card'-driver under OpenGL 4.0 heaven benchmark. Should only be used with Single card build as the Multimonitor/CFX builds actually works flawless and they actually even scale correctly, but anyway.

FIX is under the 'bugs'-section as this is temporal solution to current drivers and doesn't affect installations or other stuff than 'Unigine Heaven benchmark 2.1' it seems.

OR copying here the FIX: Mirror 01 - Mirror 02 [MD5: 56899a4aa0171b827913dab65d6bd70b] (Should only be used with 'Single card'-build)

The bug is in both x86 and x64 systems. fix includes fix for both systems architectures.

[GaBBa-Gandalf]

Can u also use that Driver when I don't have a NIVIDA card (as a sec. card)?

[TwL]

yeah, NVIDIA has nothing to do with this it's just there as an 'Option, if people has NVIDIA card OpenCL & general everything would still be working in all equipment'.

As an single card ATI driver it is still the fastest and most functional driver there is to any setup.

[DualCORE1]

hi iam having trouble getting directcompute to work i have installed latest 257SingleCard version also used 10.5a and stream 2.1 and also shows directcompute as disabled when running gpu-z it worked before format but after i reinstalled drivers theirs no directcompute support;-( nothings changed apart from updating direct x redstictable june 2010 still same setup same drivers etc open cl works though.

[GaBBa-Gandalf]

ok... i will give it a try

installed
genetix_final257_SingleCard
extra functions (reg)
genetix_OpenGL_Fix_for_v2.57.rar

Works fine so far, but the auotrun for the openGL-fix doesn't work.
I expanded the file by myself and copied manuel in my windows\system32 folder

[TwL]

that's weird tested it in here. You sure you ran it as administrator?
(Just don't see anything wrong with the .cmd file which I quickly added with it)

[GaBBa-Gandalf]

Yes I ran it as admin...

Error:

--------------------------------------------------
This is a quick fix for OpenGL for Final Build
v2.57 it will simply expand - overwrite correct
OpenGL file in your system. Please make sure you
run this as Administrator otherwise it cannot
overwrite the file directly.
--------------------------------------------------
Drücken Sie eine beliebige Taste . . .
Microsoft (R) Dateiexpansionsprogramm  Version 6.1.7600.16385
Copyright (C) Microsoft Corporation. Alle Rechte vorbehalten.

Fehler beim Öffnen der Eingabedatei: atioglxx_x86.dl_.

Die Datei kann nicht in sich selbst kopiert werden.
        0 Datei(en) kopiert.

Is is a german OS

[Hans Dampf]

Originally Posted by GaBBa-Gandalf

Yes I ran it as admin...

Error:

--------------------------------------------------
This is a quick fix for OpenGL for Final Build
v2.57 it will simply expand - overwrite correct
OpenGL file in your system. Please make sure you
run this as Administrator otherwise it cannot
overwrite the file directly.
--------------------------------------------------
Drücken Sie eine beliebige Taste . . .
Microsoft (R) Dateiexpansionsprogramm  Version 6.1.7600.16385
Copyright (C) Microsoft Corporation. Alle Rechte vorbehalten.

Fehler beim Öffnen der Eingabedatei: atioglxx_x86.dl_.

Die Datei kann nicht in sich selbst kopiert werden.
        0 Datei(en) kopiert.

Is is a german OS

Short translation:
failed while opening the input file: atioglxx_x86.dl_.
This file cannot be copied into itself.