HD 6970 + Dedicated GT 430 Hybrid Physx Mini Review

There seem to be two camps as far as Arkham City goes... For some, hybrid PhysX works great, for others, it causes an instant BSoD.

I'm in the "instant BSoD" camp, where starting Batman: Arkham City with hardware PhysX enabled instantly dumps me to a BSoD with an Nvidia DLL listed as the cause. Software PhysX runs normally.

A few people have said they had to uninstall and reinstall Nvidia's drivers (and re-run the hybrid physX patch) after installing Arkham City, and that doing so fixed the BSoD issue. I'm going to give that a whirl now.

Recommended Nvidia driver versions for Arkham City:
Forceware: 285.62
PhysX System Software: 9.11.0621
Hybrid PhysX: 1.05ff


Thanks for the reply. I'm currently 55% done with the game on normal. I plan to play it again on the harest level when I'm done and I'd like to use the hybrid physX. Please post back and let me know if the uninstall/reinstall worked.
 
Thanks for the reply. I'm currently 55% done with the game on normal. I plan to play it again on the harest level when I'm done and I'd like to use the hybrid physX. Please post back and let me know if the uninstall/reinstall worked.
Yup, re-installation of Nvidia's drivers fixed the BSoD. I'm now running at 5040x1050, "Very High" quality preset, High PhysX.

Do newer Catalyst drivers help this game performance-wise, at all? DirectX 9 mode is pretty quick (even with high PhysX), but DirectX 11 mode runs like crap (and doesn't seem to look any different aside from tessellation).
 
Last edited:
It's using tessellation to a point beyond where it's noticeable but severely impacts performance on AMD hardware. This is one of Nvidia's tricks to prop up the GTX 580 versus Cayman.


THe 7970 has an insane improvement...something like 70% over the 6970 in Arkham City.
 
It's using tessellation to a point beyond where it's noticeable but severely impacts performance on AMD hardware. This is one of Nvidia's tricks to prop up the GTX 580 versus Cayman.

Uh, there's no trickery going on here... there are multiple levels of tessellation available from within Arkham City's settings, so you can turn it down (or off completely) while still leaving the game in DX 11 mode.

And lets not forget, AMD's drivers override in-game tessellation settings by default, so AMD is in complete control of how much tessellation is being run on their cards. Even if Nvidia were asking developers to increase tessellation to absurd levels, it wouldn't matter. You can also uncheck "AMD Optimized" and lower the slider yourself if you want some additional performance (at the cost of image quality).

Ngubh.png


The "AMD Optimized" setting very noticeably degrades tessellation quality in a lot of DX11 games in order to keep performance up. AMD's cards (those from before the HD 7000 series) are simply weak in the tessellation department, even at nominal levels of tessellation, and AMD obviously knows it.

All of this is beside the point, though... even with tessellation off, DX11 mode still runs poorly compared to DX9 mode, and really doesn't look any different. I think the developers just did a poor job with the DX11 implementation.
 
Last edited:
got the GD80 installed, downloading all the drivers and etc now, then will install the GT 440 with my 6970 crossfire setup
 
PhysX, your not doing it right....

i7 2600k @ 4.5GHz
CF 6970 12.1a preview drivers

Batman Arkham Asylum
All settings maxed 1920x1080 no PhysX

min 105
max 423
Avg 248

PhysX High
min 19
max 73
avg 32

Installed Nvidia 285.62, used the default PhysX software the first time that came with nvidia drivers, ran 1.05ff patch, said it was successful

Ran Batman Arkham Asylym, set physX to high, ran internal benchmark. Can see the banners, flags and steam / smoke / fog... on part with inmates and paper, I dont see the paper. Benchmarks low.

Tried enabling the fake display just for the heck of it and connected anyway, extended to it. Re-ran batman bench same results

Updated PhysX to 9.11.1107 re-ran patch, rebenchmarked, same results.

when I set physx in batman I get a message sayign that physx is not supported, thus tellign me the GPU possibly is not seen.... I tried to go to the game foler and rename the only physx file I had seen as per the instructions but no luck.

This is the Steam version of Batman, so dont know if that makes a diffrence. I also have alice maddness returns, mirrors edge and batman arkham city to test with.

Fluid mark loads and works...? does that mean physx is working?
 
when I set physx in batman I get a message sayign that physx is not supported, thus tellign me the GPU possibly is not seen.... I tried to go to the game foler and rename the only physx file I had seen as per the instructions but no luck.
Yeah, batman is using your CPU for PhysX rather than your GPU. Simple fix (you'll need to do this for most PhysX enabled applications and games)


For Batman Arkham City:
Open your "..\batman2\Binaries\Win32" folder.
Find "PhysXDevice.dll" and rename it to "PhysXDevice.bak"


For Alice Madness Returns:
Open your "..\Alice Madness Returns\Alice2\Binaries\Win32" folder.
Find "PhysXDevice.dll" and rename it to "PhysXDevice.bak"


For Mirrors Edge:
Open your "..\Mirrors Edge\Binaries\Win32" folder.
Find "PhysXDevice.dll" and rename it to "PhysXDevice.bak"
Find "PhysXCore.dll" and rename it to "PhysXCore.bak"


This will force these games to use the patched PhysX dll's that the Hybrid PhysX Patch created for you, rather than their own local unpatched copies.
If you upgraded your PhysX system software, downgrade back to 9.11.0621 and re-run the Hybrid PhysX patch.
 
Last edited:
Arkham Asylum doesnt have that file only PhysXExtensions.dll

However I just ran Arkham City bench and it shows same low results... but I will try it with that file renamed

arkham City everything high dx9 1920x1080


EDIT: Sorry had to delete these I realized I had v-sync on at the time so the numbers were wrong.... either way I didnt have physX working
 
Last edited:
I don't think Arkham Asylum needs any modification (assuming the hybrid PhysX patch is installed correctly). It should "just work."

Again, make sure you're using the following software versions:
Forceware: 285.62
PhysX System Software: 9.11.0621
Hybrid PhysX: 1.05ff

Newer drivers and PhysX system software may or may not work correctly with the Hybrid PhysX Patch.
 
reverted back, no luck :(

DANG IT! lol

Just out of curiocity, does anyone have the Nvidia Control panel in the right click context menu on the desktop? I don't...

Just ran fluid mark and got this result:

fluidmark.jpg


which to me tells me I still dont have something setup right.... should show the nvidia card shouldnt it?
 
You need to open your FluidMark folder and rename "PhysXDevice.dll" to "PhysXDevice.bak"

And you need to re-run the Hybrid PhysX Patch after downgrading your Nvidia/PhysX drivers.
 
Yes I have rerun patch after downgrade

Fluidmark was with the file renamed
 
Then something isn't installed correctly... lets try this:

- Open the Windows control panel, uninstall the PhysX System Software, then uninstall Nvidia's video driver and reboot the system.

- Install Forceware 285.62, and when doing so, make sure you click "custom install" and then check the "clean install" box. This will give you a clean slate.

- Forceware 285.62 includes PhysX System Software 9.11.0621, so you don't need to install any PhysX System Software yourself.

- Run the Hybrid PhysX Patch 1.05ff, see if things start working...
 
retrying now. It was clean to start just FYI, I don't mess with this system other than to play games. No experimenting, driver changes unless some major fix or improvement etc etc. Only reason I tried to upgrade the PhysX software was it didnt work the first time, so since the page said that it worked with newer I tried to upgrade and re-patch.

Following this page here BTW:

http://www..com/graphic-cards/17706-hybrid-physx-mod-v1-03-v1-05ff.html

Out of curiocity, do you get the option for PhysX GPU in fluidmark? Mine is grayed out.

If you dont have fluidmark, When you select PhysX on batman arkham city do you get the error of no PhysX device?

Also if I remember correctly on my system that has an nVidia card isnt there a PhysX control panel applet?
 
SOAB....

didn't look at the obvious. In the device manager it says "Windows has stopped this device because it has reported problems. (Code 43)"

hrmmmm......
 
Please excuse a change of direction for a moment, curiosity has me. Is there any difference in PhysX performance related to quantity or quality of memory on the card being used.
From a novice standpoint it would seem a cars using GDDR would do better than one running GDDR2 or GDDR3.
Hope it is not a silly question.
There are GT430 cards with GDDR2 or GDDR3 AND 512 or 1000 GB. GT240 cards with GDDR3 and GDDR5.
I would think the faster ram would benefit play most. Which might explain some of the differing performance experiences.
Also some PCIe slots x4 "lose" their multiplier if other PCIe X1 are occupied.
Anyone have any knowledge or opinions?
I have not found numbers on 64 bit vs 128 bit for low end cards either.
 
Last edited:
Ok so I did what any [H]'er would do.... I pulled the GTX 470 out of my HTPC to see if it was indeed the card (GT 440) that was bad

Batman Arkham Asylum
All settings maxed 1920x1080

no PhysX
min 105
max 423
Avg 248

PhysX High
min 54
max 148
avg 103

arkham City everything high dx9 1920x1080

no PhysX
min 67
max 253
avg 154

PhysX high
min 21
max 73
avg 48

Batman AC - This is whith the PhysXDevice.dll renamed.... numbers seem really off to me, almost like there is a framerate limiter or something as it pegs out at the max framerate and stays there the entire benchmark it never changes..... there are a couple spots where the framerate peaks out on every other run... but with PhysX it pegs out imediately. Oh well..
 
Last edited:
RenamedPhysXDevice.dll in alice madness returns and got physX but it was like batman arkham city.... seemed to be CPU physX, was very slow once I got to a certain point (started a new game because I know in the past when I tried to use software PhysX this part got slow, it is in the first area when the ground falls apart and you have to jump over some lava to a platform then you make your way to where the train station is)
 
Tested out the GT 440 in my HTPC and it works fine..... not sure whats up

For S&G I put the GTX 470 back in to the HTPC to see what numbers I got same settings in both batman games. This system is a AMD Phenom II 955, 8GB mem, 120GB SSD for OS, 1TB HDD for programs on board sound (since I mostly use it hooked to receiver anyway)


Batman AA

No PhysX
min 82
max 160
avg 121

PhysX High
Min 39
max 95
Avg 67

Batman AC

No PhysX
min 38
max 122
avg 83

PhysX High
min 16
max 61
avg 35

One Thing I did notice in all these benchmark runs..... Batman Arkham Asylum.... the only way to get the cloth banners is to enable physX, Batman Arkham City.... The cloth banners are always there...... hmmmmmmmm
 
Dunno whats up with this 440... it works fine in the HTPC, but will not work in my rig for hybrid PhysX... looks like I'll have to find another card :(
 
Out of curiocity, do you get the option for PhysX GPU in fluidmark? Mine is grayed out.

Yup, I get the option to use GPU PhysX in Fluidmark:

xyqKq.png


I used the 720p benchmark preset (it's the button directly above [Latest Fluidmark] on the Fluidmark main window). These results are from a single Radeon HD 6970 doing the rendering + a GeForce GTX260 as the dedicated PhysX card.
Notice the first line says "(GPU PhysX)" at the end.

If I were you, I'd make sure FluidMark works first (If FluidMark doesn't work, nothing will). Once FluidMark works, then start checking your games.

Tested out the GT 440 in my HTPC and it works fine..... not sure whats up
What version of Nvidia's ForceWare driver did you use with the GT440 while it was installed in your HTPC? Was it newer than 285.62?

I wonder if 285.62 doesn't work with the 440...

Please excuse a change of direction for a moment, curiosity has me. Is there any difference in PhysX performance related to quantity or quality of memory on the card being used.
From a novice standpoint it would seem a cars using GDDR would do better than one running GDDR2 or GDDR3.
Hope it is not a silly question.
There are GT430 cards with GDDR2 or GDDR3 AND 512 or 1000 GB. GT240 cards with GDDR3 and GDDR5.
I would think the faster ram would benefit play most. Which might explain some of the differing performance experiences.
Also some PCIe slots x4 "lose" their multiplier if other PCIe X1 are occupied.
Anyone have any knowledge or opinions?
I have not found numbers on 64 bit vs 128 bit for low end cards either.
I'm curious about this as well. Has anybody tested to see what effect PCIe bandwidth, memory bus width, and memory type/capacity has on PhysX performance?
 
Last edited:
I got the GPU option once I put a card that worked in to my rig... but it's interesting... I got a lower score using GPU PhysX (1436) than the Radeons (1783 I think) this was at 1080 benchmark

I did notice in batman Arkham Asylum I got a lower score than the OP who was using a single Radeon 6970 AND two lower Nvidia cards.... but it was higher than straight CPU PhysX...???

285.62 for the GT 440 in the HTPC


At that forum post that I posted earlier they said they didnt notice any performance improvements with above 256MB memory... the more important factor was core speed and shaders / cuda cores.

here it is again just in case: http://www..com/graphic-cards/17706-hybrid-physx-mod-v1-03-v1-05ff.html
 
So... wait... did you get Hybrid PhysX working? You're rendering with an AMD cad and running PhysX on your GT440?
 
no I could only get the GTX 470 working with Hybrid PhysX, the 440 constantly showed an error in the device manager, dont know why. Once I installed the 470 in the GPU PhysX option finally popped up in Fluidmark, but the score was lower than CPU. I read that Fluidmark can give a false positive though so I didnt take it to heart. However the results I got with alice madness returns was a bit odd, still acted like CPU physX not GPU. I was reading again though that this seems to be typical that some games work well, others wont... for example Batman arkham Asylum saw a benefit finally.

I might check in to this GT 450 I have in one of our systems at work (nice being the systems admin sometimes lol) and swap it with the 440 since the 440 works fine standalone, maybe see if it craps out too.
 
Well I was mistaken, it was a GTX 550 Ti that I had here :) same cores as a 450, just a hair faster

I'll try it out tonight in my system possibly
 
Out of curiosity, what did you score with the 720p benchmark preset in Fluidmark? (Screenshot of the results window would be great).
 
Only had a minute so did a quick Arkham Asylum run with the GTX 550 Ti

Batman Arkham Asylum
All settings maxed 1920x1080
Crossifre 6970

no PhysX
min 105
max 423
Avg 248

PhysX High GTX 470
min 54
max 148
avg 103

PhysX High GTX 550 Ti
min 75
max 212
avg 155

Batman AC all settings maxed FSAA DX9
6970 CF

No PhysX
min 38
max 122
avg 83

PhysX High GTX 470
min 16
max 61
avg 35

PhysX High GTX 550 Ti
min 27
max 99
avg 64


Something isn;t right with those numbers.... I may go back and stick the 470 back in to see what the hell was going on. The first set is right, the 550 Ti is right but the 470 seems low.... maybe crossfire was disabled by accident, I thought I made absolute sure..... Dunno if maybe because the GD80 mobo last PCIe slot is only x4 not x8 or x16... maybe something was goign on there to limit the GTX 470?


Will run fluidmark later when I get back home
 
Well... it's doing something in Arkham Asylum. lol :p

I'm interested to see where your FluidMark results land... it's not the most accurate benchmark, but it does tell you if you ballpark figures on weather or not your card is performing as it should.

Something else I'm curious about... those old Agea PhysX cards were PCI, right? I happened across a PCI (not PCIe) GT430 from Zotac: http://www.newegg.com/Product/Product.aspx?Item=N82E16814500221
Wonder how that works as a PhysX card o_O
 
Last edited:
When I speak of all things maxed out in the way of settings I have not modified any of the default settings in the catalyst control panel settings other than to turn VSync off unless application specifies. I typically leave Vsync off in game

System Specs:
Corsair 650D
120mm rear low speed off mobo (exhaust)
200mm front low speed off mobo (intake)
Corsair 800GS
MSI P67A-GD80
i7 2600k @ 4.5GHz right now (was at 4.8GHz but backed down to reduce power consumption lol)
Corsair H100 4x Coolermaster R4 LED Fans running at low speed p/p config (intake)
Fans on H100 attatched to 4 fan controller
4x 4GB Corsair Vengance DDR3
2x MSI refrence 6970 crossfire 12.1a Preview Driver
MSI GTX 550 Ti 285.62 driver 1.5ff PhysX Patch
SB X-Fi Titanium
Hauppauge PCIe (1250 I think) tuner
2x Corsair Force 3 120GB SSD RAID 0 (OS important programs etc)
2x 1TB Seagate 1TB LP RAID 0 (majority of programs downloads etc)
Logitech G110 Keyboard
Razer Death Adder Mouse
Xbox 360 wireless adapter
Xbox 360 play and charge USB (for charging obviously)
MS Remote reciever
Blue LED Lighting (car kit 4 sets of 2 bright blue LEDs)
Windows 7 x64 Enterprise
Samsung S23A750D monitor hooked up via DP




Alice Madness Returns 1920x1080 all set to high

PhysX to high is finally playable and smooth! trying to find a way to get an average FPS on this one! Grat Showcase of PhysX in some areas like goop left behing from Ruin that you can run through and it moves, extra particles flying all over, When you stomp on the chess board it actually braks in to pieces and you can move the pieces around, lots of little stuff.


Metro 2033

Settings:
metro2033settings.jpg




PhysX off

Min: 9.46
Avg: 53.33
Max: 182.45


Advanced PhysX On

Min: 9.43
Avg: 53.00
Max: 182.62

DX 11 was tooooo low FPS So used DX10
Not sure of this one is working right..... FPS are about the same?? Try as I might I couldn't see much of a diffrence on this benchmark except maybe some extra particles in the explosions. I promise I checked the box for advanced physX... it just doesn't look like I did. From wht I see so far PhysX not worth it






Mafia II Everyhting maxed that could be 1920x0180

No PhysX
Avg: 65.1


PhysX High
Avg: 44.1

Some obvious extra particles, when car explodes there are parts left over, when walls are shot particles fall from it to the ground.


Fluid Mark

1280x720 CPU
fluidmark720pCPUjpg.jpg


1280x720 GPU
fluidmark720p-1.jpg




1920x1080 CPU
fluidmark1080pCPU.jpg



1920x1080 GPU
fluidmark1080p.jpg
 
Well, GPU PhysX is working in FluidMark for sure... though I expected to see a MUCH higher score. You barely scored any higher than my GTX 260 (192 SP model). :confused:

Anyway, make sure you've renamed the appropriate PhysX DLL's for each game. The exact DLL's the need to be renamed differ on a game-by-game basis.
 
I wonder why fluidmark sucks at 1920x1080 though, there is virtually no difference with physX GPU

I more than likely wont be able to do it this week, but I am curious to try to track what exactly is going on when physx is set to high, track main render GPUs use, PhysX GPU use, and CPU use.

I'm also next goign to try my 3D monitor out with PhysX enabled as I was curious to see if introducing an Nvidia card disables it, or gets rid of frame sequential and only does side by side. As it is going 3D prettymuch halves the framerates too lol. My goal was to have a system that I could run pretty much anything out there with ALL the eye candy turned up in 3D and still have acceptible framerates.


What I am curious to see is how dual 580s do in SLI with physX turned to high
 
Was messing around with my rig today a bit and got my SSDs back up to speed, then defrayed all my games in steam, and that drive array. I reran the metro bench and came up with much higher numbers for the average and max..... Too many inconsistencies here and not sure what the heck is goingnon, surely just file system performance wasn't killing me that much

I did play around with over clocking the core on the 550 and didn't see a difference. However I did monitor GPU use with metro 2033 and there was hardly any gpu use at all.... When I get a chance I'll look at batman and Alice to see if the same happens.
 
Are you kidding me this is definitely worthwhile! These are great results that show off how good these cards are. I had dual GTX 460's and whenever I put in that second one, boom my benchmark scores increased around trifold. You are right about the power consumption being a bit high though, along with the heat and kind of the sound of the fan when it cranks up all the way. But who doesn't like to hear a nice hum coming from their computer every once in a while?
 
You should post some numbers of your 2 460s without physx on in sli and then 460s in sli with physx turned on then a run with one 460 dedicated only to physx

:)

In my above post I said there were some inconsistancies

Batman aa and ac remained the same, metro 2033 showed higher max and avg by 30fps on both, mafia 2 showed 54 avg up higher than previously, fluidmark showed 2498 for 1920x1080

My ssds used to average 1GB/s but I was down to 380MB/s cause I forgot to move steam off of them, so I moved steam but never regained the performance due to no trim right now for raid 0. I made a backup in win7 broke the array secure erased them and redid the array then restored. Then updated intel chipset drivers and storage drivers. I got back up to 962MB/s

Next I seen a amd dual core optimizer running so I uninstalled it

Finally I defragged my hdd array, defragged my steam cache files and defragged the hdd array again

When I did all this that is how I started getting better numbers? Didn't think it would make that much of a difference


When I monitor batman aa and ac on the 550ti I see it gets about 53% use max, fluidmark goes up to 100% but metro 2033 barely goes up...

Just some quick notes and findings to share hopefully tomorrow night I can play with the 470 again
 
Just a Side note :) I finally got to sit down and really enjoy this setup... Played Batman Arkham Asylum and Alice Madness Returns, with everything turned up, PhysX set to high.... in 3D :D

YAY Eyecandy and Smooth Framerates!

*disclaimer actual framerates not known with physX high and 3D, I just know it is no longer a slide show! Might be time to load FRAPS
 
Uh, there's no trickery going on here... there are multiple levels of tessellation available from within Arkham City's settings, so you can turn it down (or off completely) while still leaving the game in DX 11 mode.

And lets not forget, AMD's drivers override in-game tessellation settings by default, so AMD is in complete control of how much tessellation is being run on their cards. Even if Nvidia were asking developers to increase tessellation to absurd levels, it wouldn't matter. You can also uncheck "AMD Optimized" and lower the slider yourself if you want some additional performance (at the cost of image quality).

Ngubh.png


The "AMD Optimized" setting very noticeably degrades tessellation quality in a lot of DX11 games in order to keep performance up. AMD's cards (those from before the HD 7000 series) are simply weak in the tessellation department, even at nominal levels of tessellation, and AMD obviously knows it.

All of this is beside the point, though... even with tessellation off, DX11 mode still runs poorly compared to DX9 mode, and really doesn't look any different. I think the developers just did a poor job with the DX11 implementation.

I'll quote HardOCP here, this destroys your argument.

Batman: Arkham City : In BM: AA, the XFX R7970 Black Edition easily destroyed the AMD Radeon HD 6970. We were able to run at a higher resolution with the highest in-game settings. We feel this is mostly due to the fact that Tessellation is much faster on the new Radeon HD 7970. The MSI N580GTX Lightning XE 3GB was not able to play at 2560x1600 with decent performance, so we had to lower it down to 1920x1200 with 8X CSAA. The MSI N580GTX Lightning XE was faster than the Radeon HD 6970, but the new XFX R7970 killed it in performance.
 
I installed a GTS 450 as a dedicated PhysX card to go with my 6850 OC (820/1050) and with PhysX on I get a decrease in FPS than without it. I don't know if going to a GTX 550 Ti would help or just scrap the idea and just go eyefinity?
 
Did you by any chance measure the GT430 GPU load during your PhysX testing?


My reasoning is, if it was pinned at 100% then maybe a faster card would be better for PhysX?

Before reading this I was considering like a 768Mb GTX460, but if a GT430 is sufficent, then that's great!
 
Zarathustra[H];1038423559 said:
Did you by any chance measure the GT430 GPU load during your PhysX testing?


My reasoning is, if it was pinned at 100% then maybe a faster card would be better for PhysX?

Before reading this I was considering like a 768Mb GTX460, but if a GT430 is sufficent, then that's great!

According to what I could tell the GPU was using around 50% load and often less.
 
Back
Top