Geforce GT220 for PhysX

King of Heroes

[H]ard|Gawd
Joined
Mar 26, 2008
Messages
2,006
Has anyone tried out the new GT220s for PhysX processing? A recent post on Fudzilla suggested these might be good for that.
 
A 9600GSO, even the gimped one, is the same, but cheaper. (the 'good' GSO is way better).

Of course, if you wanted 40nm, low power consumption, then the 220 is the way to go.
 
So what's better as a PhysX card; a 9600GSO 512mb (96sp), a 9600GSO 384mb (96sp), or a 9600GT 512mb (64sp)?
 
Good question. I've seen a couple of people with multi card setup's. From what I've seen it's been a mixed bad to have the dedicated card. I'm sure if you google it, you'll find some info.
 
Googling makes it worse :p

There are all kinds of inconsistencies between reviews. Some show the 9600GT stomping all over the GSO, others show them neck-and-neck, and still others show the 9600GSO stomping all over the GT. The good version of the 9600GSO (G92, 96sp, 192bit GDDR3) seems like it should beat the 9600GT (G94, 64sp, 256bit GDDR3) simply because of the increased number of stream processors, but that doesn't always appear to be the case.

And things get even weirder because the gaming performance doesn't translate into Folding/PhysX performance. Quite a few reviews show the 9600GT better for gaming, but worse at folding, than the 9600GSO.

It's a horrible, horrible mess...makes me just want to get a 9800GT for a PhysX card to leap-frog over it.
 
This is a good discussion.

The lack of correlation between gaming performance and physx performance is because physx and gaming affect different parts of a video card.

Physx is heavily dependent on the GPU itself, not the memory, etc. As an example, the 9600GT has a 256 bit memory bus. So in bandwidth-limited gaming scenarios, it will be faster than a 9600GSO, which has a better GPU but narrorwer bus.
 
Heads up: the 9600GSO needs that extra 6pin. I kinda forgot that today, and am now looking for an adaptor.

The GT220 shouldn't need one, but at that price... (70USD)...

Physx doesn't need the memory bandwidth (as much?)?
 
Yeah, I think it's safe to assume that memory bandwidth doesn't effect PhysX a whole lot.

Here's the breakdown on the 9600 cards.
______________________________________________
| 9600GT . .| G94 . . . | 64sp | 256bit memory | GDDR3 |
| 9600GSO | G92-150 | 96sp | 192bit memory | GDDR3 | Rebadged 8800GS
| 9600GSO | G92 . . . | 96sp | 128bit memory | GDDR2 |
| 9600GSO | G94 . . . | 48sp | 128bit memory | GDDR3 | Crippled 9600GT

The G92-150 version should be the best PhysX card in theory, but they're getting hard to find.
The G92 version is all over the place, but the 128bit DDR2 might be enough of a handicap to slow down PhysX.

Edit:
Looks like the "good" GSO's are priced about the same as a 9800GT, which makes them a bit of a pointless choice (9800GT would be faster for both graphics and PhysX).
So the question is, how does the G92 version of the 9600GSO stack up with 96 stream processors, but only a 128bit memory bus?
 
Last edited:
So what's better as a PhysX card; a 9600GSO 512mb (96sp), a 9600GSO 384mb (96sp), or a 9600GT 512mb (64sp)?

The 9600GT is. 256bit vs. 128 or 192bit. Also the GT's core and shader and memory speeds are way faster then the GSO. Mine for example runs at 700core/1700shader/2000memory. Find me a GSO that can run at those speeds even a OC version...
 
The 9600GT is. 256bit vs. 128 or 192bit. Also the GT's core and shader and memory speeds are way faster then the GSO. Mine for example runs at 700core/1700shader/2000memory. Find me a GSO that can run at those speeds even a OC version...
sorry, but that's completely wrong.

physx (and all CUDA apps that I know of) are almost completely dependent upon the number and speed of the shader processors.

So, if maleficarus' 9600GT's 64 SPs can run at 1700mhz, then a 9600GSO at anything above 1133mhz for its' 96 SPs will be faster for physx.

PhysX does not run faster with more memory, or wider memory bus, or memory speed. Obviously for gaming it's a different story.
 
220... meh. The 240 is actually a tad bit better overall than the "good" 9600GSO. This is the lowest part I'd consider for a dedicated physx card . The GTS 250/9800GTX+ is what people are beginning to give the 'ideal physx card' tag to after EVGA put out that weird GTX card with the built-in GTS 250 core to be dedicated for physx.

*take the above post with a grain of salt. I just woke up and my memory absolutely sucks.
 
220... meh. The 240 is actually a tad bit better overall than the "good" 9600GSO. This is the lowest part I'd consider for a dedicated physx card . The GTS 250/9800GTX+ is what people are beginning to give the 'ideal physx card' tag to after EVGA put out that weird GTX card with the built-in GTS 250 core to be dedicated for physx.

*take the above post with a grain of salt. I just woke up and my memory absolutely sucks.
you're right. In fact, I think a guy on these forums benched a 9800gt vs a 9800gtx and found the gtx provided better performance when used as a physx card. I think he even showed gains when moving from a 9800gtx to a GTX260 in certain scenarios
 
Dear PhysX Phriends,

I FINALLY got Mirror's Edge back from my brothers. I've had a 220 for testing for over a week, but haven't been able to make the trip, and didn't want to make them mail it. (They love this game!) So tonight I have started diligently testing PhysX out. I have Mirror's Edge set with everything at max and 2x anti aliasing. I am using 1920x1200 resolution. Specs are in my sig. I disabled smooth motion and vsync in Mirror's Edge, and the 3 frame rendering ahead in nVidia's control panel in order to get true values in FRAPS. I used the second level checkpoint D (IIRC) for testing with all the lovely realistic broken glass effects.

First I tested using only the GTX 295 as both graphics and physx. Here are the results:

Min FPS: 3 (what the???)
Max FPS: 202
Avg FPS: 98.75

The Max and Avg values are pretty solid, but for a good chunk of time the game literally chugged at 14 FPS or LOWER, then jumped back up again. I ran the test three times to be sure, and it did it every time. I decided to see if things improved while using the 9600GT for physx, but there's no reason it should've dropped that low. I never remember it dropping that low before I had the 9600GT as a physx card when I first beat the game.

Here are the results with the 9600GT enabled for physx:

Min FPS: 3 (double what the ...??)
Max FPS: 199
Avg FPS: 99.30

Those results are virtually identical!! Something is obviously awry. I'm downloading the latest physx system software from nVidia's website (09.1112, I think???) and I will retest. Everything I've seen online shows at least a noticeable improvement with a dedicated card, even if your main card is two cards in SLI. And I cannot explain that huge drop all the way down to 3 fps in the middle of the run. If the latest physx system software doesn't seem to change anything, I am going to reinstall the game and retest again.

I'll keep you all posted!!!

Edit: I've already got that version installed. Going to reinstall the game now.
 
Ok, something's not right. I reinstalled the game. Updated to the latest patch, etc. I even installed it on my SSD this time. Getting the SAME results. And I'm still getting that huge drop when you cross the second bridge with all the glass windows outside on your way to the plaza. And when your underground, running through the pink-lit hall, it suddenly and abruptly jumps from unplayable rates right back up to frame rates of 100 or higher. This is bizarre. There are a few other areas in the game where it suddenly becomes unplayable.

Why the sudden drops and then great performance everywhere else???
Why is the dedicated physx card not seeming to make ONE BIT of difference? Do I have to restart the computer every time I switch the physx processing card in the nVidia control panel? And why would I test the 220 yet, if I can't even get the 9600GT to make a difference? The test would be worthless unless I have solid results from a 9600GT to compare too.

THIS IS MAKING ME MAD! If anyone has any ideas, please let me know.
 
That's honestly bizarre. I saw a fairly big improvement going from a GTX260 doing everything, to a GTX260 doing graphics and a 9800GT doing PhysX. I'm also not seeing slowdowns like you're describing.

You know what...try turning off Antialiasing. See what happens.
 
I have been using a GT220 for physx in batman and mirror's edge without any problems. I've never had issues with mirror's edge, does this happen at any other point in the game?

As for performance, I don't see a big difference in ME going from the GTX260 alone to using both cards, but I do see a big jump in min. frames in batman AA. During the scarecrow sections I see a jump from about 10fps to about 25-30fps using the GT220.
 
Give a reboot a try.

If not, the next thing I'd do is re-install my nvidia drivers.
 
That's honestly bizarre. I saw a fairly big improvement going from a GTX260 doing everything, to a GTX260 doing graphics and a 9800GT doing PhysX. I'm also not seeing slowdowns like you're describing.

You know what...try turning off Antialiasing. See what happens.

Buying a GTX 295 and then having to turn off AA is like buying a Taurus SHO and disabling the turbo ... but I'll give it a try!

Give a reboot a try.

If not, the next thing I'd do is re-install my nvidia drivers.

Tried a reboot. Will try reinstalling drivers. Arg..
 
Ok. Weird. I turned off AA. Didn't help one bit. Turned off PhysX and had some strange results. Max FPS was about the same, but it NEVER dropped below 80, even in the parts where it would crawl before. Something PhysX wise is completely jacked up. It shouldn't take THAT MUCH of a hit, especially with the kind of hardware I have pushing it. Something is completely wrong.

I tried two other PhysX benchmarks that I downloaded from nVidia's website. Both ran for a little while and then crashed. Each crashed about the same time, regardless of settings or resolution. I get the jittery or crashing results regardless of whether I use the 295 or the 9600GT to process PhysX. Something is really wrong, and it definitely looks to be software/driver related. I'm going to start trying different nVidia driver versions (minus the latest of course) to see if that helps.

As soon as I get these problems resolved, I'll start working on some 220 vs 9600 results for you guys. Sorry I'm having so many problems!!! I plan on measuring power consumption as well, as I have a watt meter.
 
stupid question but you have the GT220 set as the physX card in the nvidia control panel right?
 
stupid question but you have the GT220 set as the physX card in the nvidia control panel right?

Just read back one page. It does not good to test the 220 if I can't even get the 9600GT to work. Something is very wrong with my nVidia PhysX drivers. I'll get back on it as soon as I have more time, but the whole point is to compare the 220 to the 9600GT for dedicated physx performance. They both have roughly the same price right now.
 
sorry i knew that ive been following all along.

you have the 9600gt set as the physX card in the nvidia control panel right?
Posted via [H] Mobile Device
 
sorry i knew that ive been following all along.

you have the 9600gt set as the physX card in the nvidia control panel right?
Posted via [H] Mobile Device

Yeah, and what's strange is that the results are IDENTICAL whether the 9600GT is set as the physx card or whether the 295 just does the physx. In fact, they is almost no margin of error in my testing. We're talking one or two frames different - and I'm just running through the level manually, by hand - no macro. The results are remarkably consistent.

I'm downloading some older nVidia drivers to test out right now. I'll let you know if the results are any different. There's no reason turning on physx should make the game tank from high 90 fps to 3 fps. Especially with a 295 powering the graphics, and a 9600 powering the physx. Even the 295 alone should be able to gobble this game up for lunch.
 
Update:

Well, I rolled back to older nVidia drivers, as well as PhysX drivers. With the 9600GT set as my physx card, I'm still dropping to 3 fps in the SAME PLACES as with the previous drivers. This doesn't make any sense. And with these older drivers the new avg is 80 fps, and the max is now 159 fps instead of 200 like it was with the latest drivers. How could this be so screwed up? I am completely puzzled. Maybe I have a faulty 9600. But that doesn't explain why I get the SAME results when using the 295 for graphics and physx. My next test will be to pull the 9600GT and test the 295 by itself.

I have never had such troubles with PhysX before. It's always worked good in my few games that support it. This is very odd.
 
Well, first of all, I have been fighting my setup for so long I didn't even realize that I'd ordered a GT 240 rather than a 220 for my new dedicated physx card. My hopes were that the performance would be on par with the 9600GT, that the power consumption would be less, and that my noise levels would go down, as my Galaxy 9600GT was fairly noisy. To be fair, I'm replacing my Galaxy 9600GT with a Galaxy GT 240. In order to get this all to work, I had to disable SLI. So essentially my only successful (or conclusive) testing has been done with only half of my 295 actually working. Here are the settings and results:

All detail levels on "highest"
2x AA
vsync disabled (in game settings document as well)
PhysX enabled
1920x1200

I'm running the GTX 295 with multi-gpu disabled. Half a 295 is still a fairly strong graphics card by itself - halfway in between a 285 and a 260 (core 216).

With half the 295 running graphics and physx, these are the averages of all tests ran with these particular settings:

Min. FPS: 52
Max: 111
Avg: 78

When the 9600GT is enabled for physx, I get these results:

Min: 55
Max: 116
Avg: 81

Replacing the 9600GT with the GT 240 for physx yielded these results:

Min: 52
Max: 118
Avg: 81

Now, to be fair, this first set of tests was done with the 195 drivers. I decided to test the GT 240 with the latest drivers, and this is how it did with a driver update:

Min: 56
Max: 117
Avg: 81

The GT 240 performs REMARKABLY on par with the 9600GT. That was not unexpected. I was kind of hoping for some improvement, but even with the latest drivers, there is virtually NO improvement in using a GT 240 over a 9600 for dedicated physx. By these numbers, I estimate that at least in Mirror's edge, if you have at least a GTX 275 or better, there is NO reason to get a GT 220 as a dedicated physx card. The 9600GT and the GT 240 are the minimum that you will see a noticeable difference, and this difference is not very big. Obviously I must test more games, but at least with Mirror's Edge, my results echo what many have recommended in this thread - don't get anything less than a 240 or a 9600 as a dedicated physx card.

Now here is the kicker - power consumption. You would think that the 240 would be more efficient than the 9600GT. It is on a new 40nm process, and unlike the 9600, the GT 240 requires no external power.

At least as a dedicated physx card, the idle power consumption is virtually unchanged, but the load power consumption saw a noticeable increase. At load, average and peak power draw figures both increased by TEN WATTS!!! After testing, testing, testing, and retesting, all of my performance numbers and power consumption numbers are extremely similar across the board.

So I'd like to take this moment to congratulate nVidia for GOING BACKWARDS when it comes to dedicated physx hardware. Now as a dedicated graphics card only, the 240 is quite efficient, but if you're going to buy a dedicated physx card, SKIP the GT 240 altogether. Get a 9600GT (preferably a quiet one - I used to have a quiet one ...) or go with the 9800GT/GTS 250, which many have found to be a fair bump up from the 9600GT.

I'm going to retest power consumption numbers one more time, because that simply makes no sense, but I'm extremely disappointed. To members of this forum, I would also recommend skipping the GT 220 as a dedicated physx card, unless you have a primary card that struggles only when physx is enabled. Then it might be worth it, but I'd still spend the extra $20 and get the 9600GT if you really want to see a noticeable boost.

On the bright side, the new GT 240 is smaller and quieter! :rolleyes:

Edit: .... actually, it's not much quieter. Even at 0% the GT 240 is STILL the most audible fan in the case. I JUST CAN'T WIN.
 
Last edited:
thanks for the hard work you've put into this

With SLI disabled on your GTX295, is there a way to force physx to run on the 2nd GPU? That would be a good test to compare to these results to see if a stronger physx GPU would improve performance
 
Yeah, it seems like the 9600GT / GT240 are exactly at the point of equilibrium; they process PhysX just fast enough to hold our gimped GTX295 back at exactly the same framerates.

Do you have a faster card you can try to see if you get get over this hump?
 
Yeah, it seems like the 9600GT / GT240 are exactly at the point of equilibrium; they process PhysX just fast enough to hold our gimped GTX295 back at exactly the same framerates.

Do you have a faster card you can try to see if you get get over this hump?

What? A faster card than the 295? To get much faster than even half the 295 you'd need a 285. That's all there is from the nVidia camp, and it's not going to be a huge difference. Are you suggesting a 5970?? :eek:
 
What? A faster card than the 295? To get much faster than even half the 295 you'd need a 285. That's all there is from the nVidia camp, and it's not going to be a huge difference. Are you suggesting a 5970?? :eek:

I think he means a faster card dedicated for Physx
 
Yeah, a faster PhysX card... because it looks like the 9600GT / GT240 are just fast enough to hold back the GTX295 at almost exactly the same frame rates.
 
If you want to test physx performance with different setups I would recommend another game. Mirror's Edge has very few GPU physx effects and the graphics aren't too demanding.
 
Yeah, a faster PhysX card... because it looks like the 9600GT / GT240 are just fast enough to hold back the GTX295 at almost exactly the same frame rates.

When you're hitting 50's min fps and 100's max, 4-6 fps faster because of a dedicated physx card is actually a noticeable increase. In fact, at the minimum fps parts of the game that's like a 10% increase in fps, just because you have a dedicated card. I just wish I could get physx and SLI to work at the same time. These minimum fps numbers of 3 or 4 in the bridge areas where they shoot a lot of glass doesn't make any sense. I know Mirror's Edge scales well with SLI.

Don't worry though. I'll test some other games. What do you guys recommend for intensive GPU supported physx?
 
Batman Arkham Asylum, with PhysX set to high, is fairly demanding. The developers recommend a GTX260 for graphics with a 9800GTX (also known as the GTS250) dedicated to PhysX
 
Batman Arkham Asylum, with PhysX set to high, is fairly demanding. The developers recommend a GTX260 for graphics with a 9800GTX (also known as the GTS250) dedicated to PhysX
right. the "Scarecrow" levels in Batman AA are the most impressive use of PhysX available IMO. Batman also has a built in benchmarking tool which is impacted by physX performance, in my experience.
 
9600 gt. that's what i run with for physx works great in games like batman aa and mirrors edge.:)
nope

physx is almost entirely dependent on number of SPs and the shader clock

So, go with whichever of those 9600GSOs that has the higher shader clockspeed (which of course can be overclocked)
 
nope

physx is almost entirely dependent on number of SPs and the shader clock

So, go with whichever of those 9600GSOs that has the higher shader clockspeed (which of course can be overclocked)

usually that will be the 9600GT with 64. I couldn't find a GSO available right now with more than 48.
 
Back
Top