One GPU can NOT handle physx + video: benchmarks

Status
Not open for further replies.
We can all agree that the GPU is too busy doing graphics at high resolutions to do physics at the same time. Let's see what happens when physics is run in parallel on a gpu as opposed to a cpu with SLI. I think THAT'S what everyone wants to see.

This *is* with only one GPU doing physics and graphics at the same time, no SLI.
 
This *is* with only one GPU doing physics and graphics at the same time, no SLI.

I understand this. Common sense says that the gpu is busy doing graphics at high resolution. What I'm saying is the real numbers will come in with a gpu dedicated to physics and another doing graphics. I predict these numbers will blow away the cpu/gpu and the gpu/ppu solution.
 
Personally I think the most interesting test would involve scaling of several GPUs running physX on them. In other words, run the high res test on the 9800GTX again but add a 8600GT or similar to the system to offload physX. Test with 8400GS, 8800GS, etc.
 
I understand this. Common sense says that the gpu is busy doing graphics at high resolution. What I'm saying is the real numbers will come in with a gpu dedicated to physics and another doing graphics. I predict these numbers will blow away the cpu/gpu and the gpu/ppu solution.

I don't see why you would dedicate an entire GPU though. Just a small amount of processing power will already blow the CPU and PPU out of the water, you don't need an entire GPU.
 
I don't see why you would dedicate an entire GPU though. Just a small amount of processing power will already blow the CPU and PPU out of the water, you don't need an entire GPU.

Why not? Everyone knows at high resolutions the GPU is the bottleneck. Why place more of a load on where the bottleneck is? To address this issue, there needs to be an additional GPU for physics, yes?
 
Why not? Everyone knows at high resolutions the GPU is the bottleneck. Why place more of a load on where the bottleneck is? To address this issue, there needs to be an additional GPU for physics, yes? CPU's aren't parallel in nature and will never be able to calculate physics in the required manner - they just aren't built for execution of a massively parallel problem. Physics calculations are calculated on many multiple objects at once, which is the perfect algorithm for the GPU. CPU's are linear.

I'm just saying that if you have two GPUs in SLI, it'd be a waste to only do physics on one of them. That would leave plenty of graphics power untapped. So it's better to not dedicate the entire GPU, but just allocate a portion of the resources for physics and use the rest for graphics, just as with a single GPU.
 
I'm just saying that if you have two GPUs in SLI, it'd be a waste to only do physics on one of them. That would leave plenty of graphics power untapped. So it's better to not dedicate the entire GPU, but just allocate a portion of the resources for physics and use the rest for graphics, just as with a single GPU.

You don't know how much % of a GPU need to be allocated to physics. No one will know until we see this configuration benchmarked as I stated previously. Beuller? Anyone with gtx280 sli and ut3? Run it, we need to know. : )
 
Hey, maybe we can do sli and say 50% of the 2nd gpu be allocated to physics. I bet someone is working on it now.
 
You don't know how much % of a GPU need to be allocated to physics.

It will be scheduled dynamically.
nVidia already does that anyway, because their GPUs can run vertex shaders, pixelshaders, geometry shaders and CUDA at the same time.
 
Really, I'm wanting two good GPUs for video, plus a lower/mid range one for physics, and suddenly I don't have enough PCI-e slots!
 
I am just wondering what the big whoop is. Hell as it is most video cards can't run games in that resolution, let alone with PhysX being run on top of it.
 
interesting test they ran...just curious why they didn't benchmark the other two maps that came out from ageia - tornado and the lighthouse, the latter being extremely physx-heavy. i think that would give a better indication of how the gpu would perform.
 
Did you even read the conclusions ?!
Obviously that @ such a high and demanding resolution, the GPU is too busy with the graphics load. Also, the 9800 GTX is not exactly the most powerful GPU around, much less for such a high resolution. We need tests with a GTX 260 and especially GTX 280.


the OP is right once you load up the gpu with a game that demands a lot from the card, it will not have time to do anything else. UT is pretty damned fast even on a mid to low end card.....so it is easy to show that it can do physics calcs with a game that is easy on video.......
 
the OP is right once you load up the gpu with a game that demands a lot from the card, it will not have time to do anything else. UT is pretty damned fast even on a mid to low end card.....so it is easy to show that it can do physics calcs with a game that is easy on video.......

Except that it was faster when done on the GPU than when they did it on the CPU.
 
the OP is right once you load up the gpu with a game that demands a lot from the card, it will not have time to do anything else. UT is pretty damned fast even on a mid to low end card.....so it is easy to show that it can do physics calcs with a game that is easy on video.......

Except that it was faster when done on the GPU than when they did it on the CPU.
 
Except that it was faster when done on the GPU than when they did it on the CPU.


only on the lower resolution where the card was not taxed see what happened when the card was takes at max resolution? frame rate was identical for CPU and GPU
 
The 280 remember was designed to preform better at higher resolutions, AA, and cuda applications.


If that is the case it has failed miserably since a 4870 can out run it when the IQ is with AA and AFcranked up in some cases
 
only on the lower resolution where the card was not taxed see what happened when the card was takes at max resolution? frame rate was identical for CPU and GPU

So clearly the OPs conclusion is dead wrong. Best case: GPU PhysX is faster. Worst case: GPU PhysX is the same as a high end quad core. Since most people still have dual core CPUs (especially those buying a $200 video card), the worst case is that GPU PhysX will be faster than the CPU at 2560x1600, so how the hell does that equal "One GPU can NOT handle physx + video"? It is extremely obvious that even at 2560x1600 one GPU very much CAN handle PhysX + video, and will do it better than most dual-cores out there.
 
yes but that will change for each GPU once the games really start taxing it so it is a short term gain at best
 
A point people seem to be missing is that this could potentially extend the longevity of your graphics cards.

Imagine you have a current gen card, a couple of years later you decide that you want to upgrade to the latest and greatest GTX 12000 series. Rather than chucking your once very expensive GTX 280, you keep it as a dedicated PPU to boost the physics processing performance of your rig.

Now that would be seriously cool, assuming that Nvidia decides not to be bastards by forcing users to have an nforce chipset to take advantage of dual GPU's.
 
At this point it makes more sense to just buy the damn PhysX PPU card for $120 bucks and call it a day.
 
Another caveat: There are how many games right now that utilize physx? No, 3D vantage does not count.

There will be how many games in the future that will support physx? Who knows. Maybe everyone will just move to Havok, or maybe combine the two into something different.

I currently see almost nothing to get excited or unexcited about.
 
the OP is right once you load up the gpu with a game that demands a lot from the card, it will not have time to do anything else. UT is pretty damned fast even on a mid to low end card.....so it is easy to show that it can do physics calcs with a game that is easy on video.......

They ran it at an insane resolution, so the game still demanded a lot from the card.
 
yes but that will change for each GPU once the games really start taxing it so it is a short term gain at best

Not at all. GPUs are so powerful that it's much better to take 5-10% of the GPU resources for physics than to run the physics on the CPU. On the CPU it would take much more than 5-10%, and you'll get CPU-limited. Which is why we haven't even *seen* physics like these PhysX levels yet in most games: even with a high-end quadcore you still only get 30 fps at most. That's why physics have been extremely basic in games thusfar.
Clearly, even on a mid-end card like the 9800GTX, you have a lot more physics potential than on your average CPU. This advantage will only increase over time, because GPUs evolve quicker than CPUs.
 
Not at all. GPUs are so powerful that it's much better to take 5-10% of the GPU resources for physics than to run the physics on the CPU. On the CPU it would take much more than 5-10%, and you'll get CPU-limited. Which is why we haven't even *seen* physics like these PhysX levels yet in most games: even with a high-end quadcore you still only get 30 fps at most. That's why physics have been extremely basic in games thusfar.
Clearly, even on a mid-end card like the 9800GTX, you have a lot more physics potential than on your average CPU. This advantage will only increase over time, because GPUs evolve quicker than CPUs.

all of you guys who think that GPU physics is going to help with game play are just grasping at straws.........do you honestly think that if Crysis supported Physix, you would get acceptable framerates?

GPU under heavy load is gonna have ZERO resources to handle anything else.....there will be trade offs either you get max performance or you get less performance + physics...

Nvidia has been stupid with their aquisition of Agea up to this point.....they should have continued to develop their product as a stand alone and THAT WAY people like me who will never buy an nvidia graphics card, could have bought one and used with with a card of my choosing the benchmarks bear this out........
 
lol forgot add as newer cards come out that support physics the older titles at the time will run well with it but most likely not cutting edge games released at the time.
 
all of you guys who think that GPU physics is going to help with game play are just grasping at straws.........do you honestly think that if Crysis supported Physix, you would get acceptable framerates?

This is such a big fallacy...
What makes you think that a game like Crysis, where physics are generally not the bottleneck, would fun faster when offloading the physics to another processor?
Nobody made any such claims.

GPU under heavy load is gonna have ZERO resources to handle anything else.....there will be trade offs either you get max performance or you get less performance + physics...

The same goes for CPUs, which this test clearly demonstrates. With the heavy physics in these PhysX levels, even a high-end quadcore can only reach 30 fps, and limits the videocard. So because of the physics you get less performance.
Offloading the physics to a separate PPU is ofcourse the best solution (currently... since PPUs will not evolve and GPUs will, at some point a PPU will limit a game just as much as a CPU does now, and GPU physics will be the best way to go).
GPU-physics and CPU-physics are both tradeoffs, but GPU-physics is still the better tradeoff of the two.
This test clearly demonstrates that a $200 videocard outperforms a much more expensive CPU in 'low' resolutions (still much higher resolution than what I use, I'm still at 1280x1024 myself), and performs equally in high resolutions.
So if you ask me, GPU physics are better bang-for-the-buck than CPU physics.

For example, I currently have an E6600 with 8800GTS320. I can't play these levels at all, because my CPU isn't powerful enough.
Now, I could buy a new CPU, which makes the physics go faster, but doesn't do much for all my other games, because they're all GPU-limited. I could get a 9800GTX... Not only is it cheaper than a fast quadcore, it also makes ALL my games faster, with or without PhysX.
 
The whole situation is bogus if you ask me.

Video cards can't just magic up more power to run loads of complex physics along side video, you have to squeeze the 2 onto the same GPU which means video needs to take a hit in performance.

CPU's maybe less efficient at calculating physics but they've also got loads of untapped potential now, there are many quad cores around, the Q6600 style chips are really cheap and fit into the 775 socket type, the Q9450 style chips are a bit more expensive but again fit into socket 775. Pretty much no games are properly utilising 2 cores much less all 4, theres so much untapped potential there, we'll see that come to light with Alan Wake I feel.
 
all of you guys who think that GPU physics is going to help with game play are just grasping at straws.........do you honestly think that if Crysis supported Physix, you would get acceptable framerates?

GPU under heavy load is gonna have ZERO resources to handle anything else.....there will be trade offs either you get max performance or you get less performance + physics...

Nvidia has been stupid with their aquisition of Agea up to this point.....they should have continued to develop their product as a stand alone and THAT WAY people like me who will never buy an nvidia graphics card, could have bought one and used with with a card of my choosing the benchmarks bear this out........

Well first of all, physix is an added feature list as in ut3. No one ever said it would improve framerates by itself per se.

As far as the ageia purchase.. it was an excellent fit. What are you talking about develop as a stand alone product? Nv makes gpu's, physics was made to be emulated on gpus. I guess you never heard of maybe buying another gpu to do physics instead of a stand alone card. I know what I would rather have, Hmm.. two graphics cards or a graphics and ppu that performs worse. I wonder.

The benchmarks are flawed - comprende? probably not.
 
Video cards can't just magic up more power to run loads of complex physics along side video, you have to squeeze the 2 onto the same GPU which means video needs to take a hit in performance.

GPU takes less of a hit than a CPU, that's the whole point.
It's a better tradeoff.

CPU's maybe less efficient at calculating physics but they've also got loads of untapped potential now, there are many quad cores around, the Q6600 style chips are really cheap and fit into the 775 socket type, the Q9450 style chips are a bit more expensive but again fit into socket 775. Pretty much no games are properly utilising 2 cores much less all 4, theres so much untapped potential there, we'll see that come to light with Alan Wake I feel.

Wrong, PhysX and Havok are already optimized for multithreading and can easily use 4 or more cores. What you're seeing here is about as good as it gets. CPUs are just really poor at physics (which is why most games have such simple physics. This test already uses special levels with significantly more detailed physics than the regular game, to demonstrate the point of how GPUs or PPUs can enhance games beyond the level of even the fastest of CPUs).
 
Nvidia has been stupid with their aquisition of Agea up to this point.....they should have continued to develop their product as a stand alone and THAT WAY people like me who will never buy an nvidia graphics card, could have bought one and used with with a card of my choosing the benchmarks bear this out........

Someone just showed their true colours...
 
hence the point of this thread

What point ? That a 9800 GTX cannot handle 2560x1600 properly in most games, even when NOT using PhysX ?

Did you really need a thread to say that ?

Because the link you posted clearly shows the advantages of having physics calculations done on the GPU instead of CPU or a PPU. So your title is either wrong or you tried to prove a point that is implicit i.e. a 9800 GTX cannot handle 2560x1600 properly.
 
that just proves my point GPU under heavy load = ZERO benefit....

Again, a 9800 GTX is not exactly the most powerful GPU at this point. Show me results with a GTX 280 and if the numbers are similar, then I may agree that under heavy graphics loads, the GPU may not handle physics properly or show benefits over CPUs or PPUs. Until then, this link pretty much proved how much better physics on the GPU are, especially when compared with CPUs.
 
Status
Not open for further replies.
Back
Top