Benchmarks of 5870 and 5970 with Nvidia PhysX

[busydude]

Tweaktown has just posted benchmarks of 5870 and 5970 coupled with an Nvidia card for Physx using 257.17 beta drivers.

Link

PhysX is no doubt a cool technology and as we see more and more games make use of it, why wouldn't you want to use it? - The problem at the moment for NVIDIA is that no one with a high end ATI card is going to move over to a high end NVIDIA one. At the same time, if people want the fastest single card on the market, you can't look past the HD 5970.


PhysX for most people isn't going to be a big enough swaying point to pick one brand over another. Yes, the list of games that support the technology is big, but once you cut that list down to games you want to play and then cut that list down to ones that actually play better with PhysX, you're not going to have a large list of titles in front of you.



On the other hand, if you're using a HD 5800 or HD 5900 series card and for a few hundred dollars could add PhysX ability, the option becomes quite attractive. I could easily see people picking up a GTX 465 to go alongside a HD 5800 series like we have here today.

The min FPS did not change, which is the most important factor for gameplay, but the averages were a little higher while using Nvidia card. The conclusion was no different though, more good games + awesome effects = win for PhysX.

 

[3DVagabond]

So, do you think they were just bored, or were they purposely poking nV in the ribs? I laughed, personally, when I read that review (well scanned it, actually).

Just to show though that no experiment is a total waist of time, I was surprised to see the fps be higher, using the same ATI card for rendering, when a better nVidia card was used for PhysX. I would have figured that a GTX-260 would be more than capable to pass all of the PhysX info and there'd be no difference with a GTX-285/465 or even higher.

 

[zerocool84]

The conclusion was no different though, more good games + awesome effects = win for PhysX.

Once that happens let us know. Us consumers are still waiting for Physx to deliver.

 

[busydude]

Once that happens let us know. Us consumers are still waiting for Physx to deliver.

Yeah, been waiting for too long. No ground breaking games atm and other than Mafia II nothing substantial in the coming months.

 

[konakona]

If anything, in case you wanted physx add-on card for these games you don't need anything better than a 260. Wonder how low you can go on that (not that I am interested in physx yet)

 

[StrangerGuy]

Fragmenting an already fragmented PC gaming market. That's some great business sense from Nvidia.

BTW I don't care how good the PhysX technically is, if the games aren't fun I simply won't give a shit period.

 

[RaistlinZ]

Could an 8800GT be used as a PhysX card with a 5800 series card?

 

[darckhart]

i havent done any real benching, but i have a gtx260 and a gt240 and honestly, i think the gt240 gets it done pretty well. sure, you get more particles for like smoke and such with the 260, but when i'm running around, i don't think i really notice. and the gt240 uses a hell of a lot less power.

 

[taltamir]

The min FPS did not change, which is the most important factor for gameplay, but the averages were a little higher while using Nvidia card. The conclusion was no different though, more good games + awesome effects = win for PhysX.

AFAIK the way physX works is that it would provide better quality graphics when enabled, rather then push for improved FPS. So I would be interested to know if it actually does that or not. (afaik, not enough to matter)

 

[Scali]

I think it's a shame they didn't include GeForces doing video AND PhysX, to compare.
Eg, if I look here: http://www.tweaktown.com/reviews/3232/gigabyte_geforce_gtx_480_video_card/index4.html
The GTX480 scores about as well as a 5870 + GTX465 combo. But the GTX480 would be cheaper and use less power...

In Darkest of Days, the GTX480 seems to perform considerably better than the Radeon + GeForce combo's:
http://www.tweaktown.com/reviews/3232/gigabyte_geforce_gtx_480_video_card/index10.html

But in Batman it seems like a separate PhysX card pays off:
http://www.tweaktown.com/reviews/3232/gigabyte_geforce_gtx_480_video_card/index12.html

Same with Dark Void:
http://www.tweaktown.com/reviews/3232/gigabyte_geforce_gtx_480_video_card/index13.html

 

[Scali]

AFAIK the way physX works is that it would provide better quality graphics when enabled, rather then push for improved FPS. So I would be interested to know if it actually does that or not. (afaik, not enough to matter)

PhysX can go either way, it just depends on how a game is designed.
Some games may have very heavy physics load on the CPU, so the game becomes very CPU-limited. In that case, offloading physics to the GPU can improve performance.

Other games use relatively light physics in CPU-mode, and throw in the eyecandy when the GPU is enabled.

In theory it's even possible to hit a sweet-spot where your game will both increase minimum FPS AND give more eyecandy.
In practice however, it's often the other way around. CPU-physics are generally VERY simplistic (with the possible exception of Crysis, but since it doesn't support GPU-acceleration, there's no opportunity to improve on the minimum FPS), so there's not much of a CPU-bottleneck to begin with. And when GPU-physics are enabled, they overdo it, so the average framerate takes a big hit.

 

[taltamir]

In theory it's even possible to hit a sweet-spot where your game will both increase minimum FPS AND give more eyecandy.
In practice however, it's often the other way around. CPU-physics are generally VERY simplistic (with the possible exception of Crysis, but since it doesn't support GPU-acceleration, there's no opportunity to improve on the minimum FPS), so there's not much of a CPU-bottleneck to begin with. And when GPU-physics are enabled, they overdo it, so the average framerate takes a big hit.

All good points... Which still begs the question of why look only at FPS but neglect IQ

 

[GaiaHunter]

I think it's a shame they didn't include GeForces doing video AND PhysX, to compare.
Eg, if I look here: http://www.tweaktown.com/reviews/3232/gigabyte_geforce_gtx_480_video_card/index4.html
The GTX480 scores about as well as a 5870 + GTX465 combo. But the GTX480 would be cheaper and use less power...

In Darkest of Days, the GTX480 seems to perform considerably better than the Radeon + GeForce combo's:
http://www.tweaktown.com/reviews/3232/gigabyte_geforce_gtx_480_video_card/index10.html

Sincerely I don't understand those numbers and graphics.

All of them are set as physX low to start with and the article states:

Note: With the PhysX set to Medium or High Darkest of Days take advantage of the NVIDIA PhysX abilities. For that reason we will test ATI cards at the Low preset, NVIDIA based cards though will be tested at Low and High.

The Radeon cards score exactly the same with a physX card or without it.

They also state that only in medium or high physX will it be accelerated by the GPU and that they are testing radeons with physx low only and nvidias at low and high, on the 480/470 article, but they only present 1 chart for the geforces that is labeled at physx low.

I think there is some problem with that article.

 

[taltamir]

The Radeon cards score exactly the same with a physX card or without it.

They also state that only in medium or high physX will it be accelerated by the GPU and that they are testing radeons with physx low only and nvidias at low and high, on the 480/470 article, but they only present 1 chart for the geforces that is labeled at physx low.

I think there is some problem with that article.

Excellent points, so in addition to not showing IQ comparisons they also have all the issues you described.

 

[busydude]

AFAIK the way physX works is that it would provide better quality graphics when enabled, rather then push for improved FPS. So I would be interested to know if it actually does that or not. (afaik, not enough to matter)

I guess it would be hard for reviewers to do IQ analysis with PhysX. Cloth movements and glass shattering are the two common things PhyxS implements that can be easily discerned. In case of PhysX IMO, a video comparison is the best way than the regular static images. These movements cannot be analyzed using a single image.

One more interesting thing I observed in the benchmarks is although the FPS went up when comparing GTX 260 to GTX 465, The averages were less than the normal. I mean the average FPS of a single 5970 is greater than the averages we see while using PhysX even with a GTX 465. Can any one answer where the bottleneck is being created? With GTX 465 + PhysX avg = 109, without Physx avg = 167 a performance hit of 53%.

 

[GaiaHunter]

I guess it would be hard for reviewers to do IQ analysis with PhysX. Cloth movements and glass shattering are the two common things PhyxS implements that can be easily discerned. In case of PhysX IMO, a video comparison is the best way than the regular static images. These movements cannot be analyzed using a single image.

One more interesting thing I observed in the benchmarks is although the FPS went up when comparing GTX 260 to GTX 465, The averages were less than the normal. I mean the average FPS of a single 5970 is greater than the averages we see while using PhysX even with a GTX 465. Can any one answer where the bottleneck is being created? With GTX 465 + PhysX avg = 109, without Physx avg = 167 a performance hit of 53%.

http://www.tweaktown.com/reviews/3232/gigabyte_geforce_gtx_480_video_card/index12.html

If you see a GTX480 also lose about half its framerate.

 

[taltamir]

I guess it would be hard for reviewers to do IQ analysis with PhysX. Cloth movements and glass shattering are the two common things PhyxS implements that can be easily discerned. In case of PhysX IMO, a video comparison is the best way than the regular static images. These movements cannot be analyzed using a single image.

Yea, the thing about those is:
1. prescripted glass shattering can work out very well and doesn't really need replacing by physX
2. physX cloth effect suffer from horrible clipping (where the cloth passes through itself)... that is because physX seems to have a "fluid" setting (that is not very fluid BTW) which it reuses for oil, water, fabric, etc... and its not quite right for all of them... especially for cloth because cloth is a solid and does not pass through itself, although, fluid does not pass through solids either so I am thinking maybe its just some physX bugs rather then them reusing the same basic functions of liquid to make clothe... I mean, I am just making a guess here.

if it is a bug then it needs fixing. PhysX has been around for a while and about the only impressive things anyone has ever been able to point out are clothe, glass, and liquid. And glass is the only one of the three that isn't made of fail (clipping for clothe, unrealistic motion for liquid) while glass is unnecessary.

 

[3DVagabond]

One more interesting thing I observed in the benchmarks is although the FPS went up when comparing GTX 260 to GTX 465, The averages were less than the normal. I mean the average FPS of a single 5970 is greater than the averages we see while using PhysX even with a GTX 465. Can any one answer where the bottleneck is being created? With GTX 465 + PhysX avg = 109, without Physx avg = 167 a performance hit of 53%.

PhysX is a pig? :\ (j/k)

Apparently how fast these cards can process the PhysX information is the bottleneck.

 

[Scali]

PhysX is a pig? :\ (j/k)

Apparently how fast these cards can process the PhysX information is the bottleneck.

It's not entirely PhysX alone...
With more debris, smoke particles and other effects, there are also more graphics to render.

 

[3DVagabond]

It's not entirely PhysX alone...
With more debris, smoke particles and other effects, there are also more graphics to render.

That actually makes more sense.

 

[busydude]

It's not entirely PhysX alone...
With more debris, smoke particles and other effects, there are also more graphics to render.

So, are all those extra effects enabled with PhysX?