Nenu
[H]ardened
- Joined
- Apr 28, 2007
- Messages
- 20,314
assists^ = assets 
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
I was merely stating what happened when NV bought out 3dfx, raising the possibility that something similar may happen again.
Its not the same. NV didnt buy 3DFX they bought some of 3DFXs assists. NV didnt and dont own 3DFX the company. It was 3DFX that chose to stop driver support not NV.
With Ageia NV did buy the company and do own them.
How can NV buying some of 3DFX's assists and not the company be classed as similar to NV buying Ageia the company? NV have to honer the Ageia warrantees e.c.t They didnt with 3DFX as they dont own them.
Saw almost an identical post over Here. I am interested in seeing this take off (Hey more people that get into physics the better). However, I am more interested to see how a $400 video card physics acceleration is going to compare to the $100-$150 Physx card.
I don't think you need to compare a "$100-$150" PhysX card to a "$400" G80 card. Ageia was claiming 58 GFLOPS performance for the PhysX card. Let's start from there.
It's not clear if that 58 GFLOPS figure peak theoretical or actual throughput. To give the PhysX card the benefit here, say that 58 GFLOPS was actual throughput.
Observed throughput per SP in GPGPU loads is about 1.6 FLOPS per SP per shader clock on G8x cards. Applying that to a 8400GS, you get about 23 GFLOPS throughput, which clearly isn't a match and kind of rules out expectations for an integrated video chip matching a PhysX card in this generation. However, a vanilla 8600GT would have over 70 GFLOPS measured throughput in real GPGPU loads. That brings a performance competitive part down to under $90. IMO, the 8500GT is probably the most comparable card for throughput to the PhysX card and is even cheaper.
Why am I treating the PhysX card like a general purpose SIMD engine and not some super-optimized physics engine? See here: http://arstechnica.com/articles/paedia/hardware/physx-hpc.ars
Of course we won't know how it compares until nvidia released the CUDA PhysX driver.
I'm not talking about LINPACK GFLOPS, but actual workload measured performance. The theoretical performance for the G8x SPs is almost double the 1.6 figure I used above. If you read what you quoted, I gave the PPU full credit for the 58 GFLOPS Ageia performance claim even though actual throughput is probably significantly lower.FLOPS are useless here, it's the efficiency and the implemetation in real world usage that counts.
LOL, because it does have some physics performance data. You only exclude it because it's inconvenient.No one has a clue to the preformance of any 8x00 GPU in physics..despite "Quantum".
I didn't say or imply the PhysX PPU chip was a GPU. I said it was a generic SIMD processor, which your link backs up.This is not a GPU...no matter how much PR is spun.
I wouldn't use these as end all examples, but I found some CUDA physics videos.No one has a clue to the preformance of any 8x00 GPU in physics..despite "Quantum".
I don't think you need to compare a "$100-$150" PhysX card to a "$400" G80 card.
I'm not talking about LINPACK GFLOPS, but actual workload measured performance. The theoretical performance for the G8x SPs is almost double the 1.6 figure I used above. If you read what you quoted, I gave the PPU full credit for the 58 GFLOPS Ageia performance claim even though actual throughput is probably significantly lower.
Show me anything like the physcis via the PhysX API..LOL, because it does have some physics performance data. You only exclude it because it's inconvenient.
You did mention it like both the PPU and GPU are godd at SIMD and hance good at physcis.I didn't say or imply the PhysX PPU chip was a GPU. I said it was a generic SIMD processor, which your link backs up.![]()
I have heard this before...my el-cheapo-ass card will be the same as the PPU.And if you want to go on record, i'll predict that the 8500GT will be pretty close in performance to the Ageia PPU once nvidia has optimized PhysX under CUDA. What's your prediction? The branching and switching performance problems nvidia had with the G7x GPUs were fixed in the G8x series and above.
I wouldn't use these as end all examples, but I found some CUDA physics videos.
Colour Particles (8800GTX)
Colour Particles (8800GTS)
Lava Particles
N-Body (8800GTX)
Elastic
I'm thinking that this might be really good news for the 9800GX2 (or not necessarily the 9800GX2 but future GX2's; could be for the 9800GX2 as games running CUDA physics require the gpu to do the physics processing, and then you would need to sacrifice gfx at that point but would at least be able to get the physics going if you wanted them). I think what will be most interesting is how well SLI scales while working on both graphics and physics.
Wow Atech, such pessimism.
More PR, wake me when they get the shit running.What has been said before and what has been realisable have been very different things but now Physics on a GPU is very realisable, especially now NVidia have access to use an established brand and the technology rights.
There was talk in 2006 about GPU-physics...Definitely room for a good discussion on the topic, it doesnt deserve being beaten down.
You've said it all many times already.
You "asked", don't whine when I "answer"...fine, we know your opinion now, no need to troll.