• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

PhysX Benchmark on a hybrid system

Soldato
Joined
10 Oct 2012
Posts
4,842
Location
Denmark
Well i have been searching Google dry forever to answer several things before i went out and bought a nvidia card to be used as a dedicated physX Card with my Sapphire 7950(for games like Borderlands 2 and Batman). Would it even work without issues in a hybrid setup? how good would the card have to be to be able to handle the latest and greatest with as little performance loss as possible? How does overclocking the card affect the physX performance? What about the amount cuda cores?

While there are plenty of benchmarks, i feel most of them are rather useless or does only test in a pure nvidia enviroment. So for that reason i went out and purchased a gt 640 and are going to bench it against a XFX gtx 260 BE. The bench will including OC of memory, core and shader clock and most of all in a hybrid system using a 7950 as the main render. It will also contain, if i can make it happen, clock for clock test between the 2 nvidia cards.

The GT 640 should arrive tomorrow and the gtx 260 benchmarks are 80% complete as of this post. With a bit of luck if time allowes it i will post everything in this thread tomorrow night.

Feel free to include any request or idea's or thoughts. There seem to be a lot of misinformation on this very subject. The point of this thread is NOT to determine if its worth the money but more if its even reliable and to contribute to this awwsome forum i've been enjoying for years as a passive guest.

UPDATED 11 October 2012
PhysX Benchmark Spreadsheet on GOOGLE
 
Last edited:
I don't think a 640 was the best choice.
The 6xx series are known to be worse at compute/cuda then the 4xx/5xx range, although I am unsure if this affects physx or not...

This will be very interesting :)
 
The reason behind why i chose the 640 gt is based on the number of cuda cores alone and the core speed. So fare using fluidmark the OCing on my gtx 260 showed a rather unimpressive increase in physx SPSon both shader clock and memory clock and while Core Clock OC yielded not much more it was still the best place to gain physX performance i fluidmark.
 
I found raw clock speed makes negligible difference. I have a custom profile which runs the 480 undervolted at 500mhz core. The difference between that and stock was only about 5fps obviously there will be a cliff somewhere but I've not found it with my limited testing.
 
I found raw clock speed makes negligible difference. I have a custom profile which runs the 480 undervolted at 500mhz core. The difference between that and stock was only about 5fps obviously there will be a cliff somewhere but I've not found it with my limited testing.

Interesting. I to did underclock but i did not lower the volt. Results were as i expected less SPS (FPS)
 
Would it even work without issues in a hybrid setup?

Yes, although release day titles could(have done in the past) lack working Physx if Nvidia changes the code on a new driver, it's possible, you may have to wait for a new hack.

how good would the card have to be to be able to handle the latest and greatest with as little performance loss as possible?

Up to B2(I have only played the first level so I can't confirm this title) everything's ran fine on my 9800GT(using 5870/6950/7950/7970), but, I'm suspicious of this title that it could possibly be slowing down the 7970.

Your data will be welcome on this title.

Afaik, somewhere in the sewers(?) later on in the game, there is an awful lot of physX going on, that would be the place to try it.

I asked in another thread for a save for me to try it out(as I won't get to play it until I complete other titles), but no one replied.

Will you be comparing PhysX Off, AMD/Cpu PhysX(ini hack) enabled and Hybrid PhysX?

How does overclocking the card affect the physX performance?

Your not going to notice anything I imagine as PhysX doesn't use much of the core at all.

What about the amount cuda cores?

The more the better in regards to PhysX performance, but larger returns may be minimal-having only used a 9800GT, this is guesswork ;).

7970/9800GT Metro 2033 Physx Comparison,

7970 1000/1400 PhysX Disabled:

1aaaa4f54b94fe44b435d92ab54756c9.jpg


7970 1000/1400/9800GT PhysX Enabled:

e988b0018f9d6d650b69e51d05ba70dc.jpg


7970 Metro-1165/1600 PhysX Disabled:

9a29fc40ae3f0ec1fef17d56154b0509.jpg



7970 Metro-1165/1600/9800GT PhysX Enabled:

5f3d38b383633ae5d199b5e5f48b3127.jpg
 
True, my 9800GT was lying about doing nothing, and by the figures above, it has added +FPS to my rig while running Hybrid PhysX.

It has ran every PhysX title out there, while giving a zero hit on the 5870/6950>70/7950/7970 gpus that I've used it with over the years.:)

I'm sure the newer series gpu's could be better, but as I said, PhysX doesn't use much of the 9800GT.

Nvidia would like you to use a 460 to use as an independent Physx, as I presumed it would be the same-hence an overpowered gpu just to use up to ~30%.

It may get you +5FPS on your 79* in Hybrid mode, but, is it worth the extra outlay if it doesn't push a gfx card much that is only rendering PhysX effects?

I don't know if the newer gpus use any more resources to run as an independent PhysX renderer, so if they can do a comparison, as one of them is a 480, then we'll find out.

If Phixsator/layte has Metro 2033 and can do their setups comparisons it would shed some light on the matter, if you are willing to try it and don't have the title-I may be able to get you one(if my mate hasn't activated the one I gave him).
 
Last edited:
Yer I mentioned it a while back, Not sure if I created a thread or just commented lol
It was something about bench tests anyway am not sure what GPU I will pick maybe this thread will shed some light on this :)
 
I can't remember which review site I read it, but the 8800gt did lower fps when paired with a 580(?) I believe. It was actually better to leave out the 8800gt and have it run on the rendering gpu. So in your case it could be that you could gain more even from a modern architectue but similar core count nvidia card.
 
If any AMD users has an old Nvidia PhysX lying about doing nothing, you can try it out for FREE too and join in.

You can get Hybridiz here:

http://www.ngohq.com/graphic-cards/...with-latest-physx-and-geforce-285-solved.html

Don't just jump in, read up about it on the page and see what you need to remove in each games dir, or it won't work.

For B2, you need to run the reg entry 'PhysX-HwSelection-set-GPU' found in the 'extras' folder to enable gpu PhysX, if you don't it will run off your cpu!

It just works.:D

@layte,

What Fluidmark are you using and do you have a link?

I noticed your pics are gone, as stated you can resize them, but, if you want to use large pics, you can post them this way:

7970 1000/1400 PhysX Disabled:
1aaaa4f54b94fe44b435d92ab54756c9.jpg



:)
 
Last edited:
What are you waiting for?

You can't get any better than free, that's for sure.

It's not that much slower than a 9800GT, besides you can set gpu PhysX from low to high.

What's not to like?
:)
 
Ok will pop it in tomorrow. Problem I dont own Borderlands 2 lol what other games use physX? any demos I can download
edit
One game I have witch is Metro 2033 lol I know I can download Batman AC demo
 
Last edited:
I do own Metro 2033 and are planning to test it as well. During testing in Batman Arkham City a gtx 260 would be enough as a dedicated physX Card but not when it comes to borderlands 2, got horrible low fps in that game if PhysX was set to medium or above.

Im just sitting here yelling at the postman i cant see to come and deliver my package so i can start some serious benchmarking hehe. I want that gt 640 NOW!
 
Back
Top Bottom