cM :: MORLEY|DIGITAL

Fair warning – I’m going to ramble through a few topics here, so please bear with me.

PhysX is the Lazarus of the PC industry after Aegia was acquired by NVIDIA last year.  However, the usefulness of a dedicated piece of hardware to accelerate physics has been debated more and more since its introduction in mid-2006, particularly in light of rival software -solution Havok’s successful install base and support amongst developers.  Demonstrations I saw at Quakecon 2005 showed an impressive array of hundreds of boxes colliding and splintering or water “realistically” flowing down uneven terrain.  Cool, but not worth $250.  So NVIDIA is ditching the dedicated hardware solution and is apply the awesome brute strength of their GPUs, along with CUDA, to enable PhysX processing on the GPU.  However, the fact that recently released NVIDIA demos basically demonstrated the same old water effect accelerated by CUDA-driven PhysX technology gives me pause, especially after seeing little real-world application of PhysX that truly impacts game play, not just accelerate physics “effects.”

It all started when I stumbled upon a very interesting blog post from an HP/Voodoo engineer on their NextBench community forum.  It was interesting in a couple of ways.  The first thing that jumped out at me was the fact that the engineer solely ran tests using FutureMark’s Vantage PC benchmarking software.  One of the tests in the benchmarking suite includes dedicated tests to measure the physics processing capabilities of a PC.  It’s similar to the CPU test in 3D Mark 06 where 3D acceleration was taken out of the picture and software rendering was used to render a 3D scene.  Of course it ran in the 1-4 frames per second range, but then again it wasn’t intended to demonstrate a real world scenario, just remove all other variables from the test as much as possible.  The same goes with the physics test in Vantage.  However, NVIDIA has released drivers that enable the GPU to process physics calculations instead of the CPU which show a remarkable increase in performance on that test.  On July 21^st of this year, FutureMark began to remove GPU PhysX scores from its Hall of Fame, citing that its physics test was meant solely to test the processing power of the CPU and clearly stated that a GPU or a driver could not affect the score.  Now if you want to submit to FutureMark’s Hall of Fame, you must submit with a WHQL driver and not install NVIDIA’s PhysX acceleration software.  I would presume that if PhysX-enabled games were more widespread that FutureMark would have a tougher time convincing the general public that NVIDIA’s use of their GPUs to accelerate PhysX was some sort of cheating, or even violation of the “spirit” of the rules.  Because the fact of the matter is that a GPU’s stream processors are going to be busy banging out wicked 3D graphics and there definitely needs to be more study on how that affects its ability to accelerate PhysX in the real world.  Perhaps this generation’s hardware is powerful enough, perhaps not.  But that just goes back to the type of people that would be interested in shelling out about $1000 in NVIDIA products (which NVIDIA would love-more on that later) just to get physics accelerated effects.  I say effects because to date there has been no significant title that has a compelling reason to add PhysX hardware at additional cost to the user to enable a better gaming experience.  NVIDIA adding PhysX support to GPUs that people are already buying or already have is a good play because there’s a very large install base of 8 and 9-series NVIDIA GPUs-at least when it comes to the mainstream.  This is a good and bold strategy.  But developers aren’t developing games for the 5% enthusiast space that is even interested in GPU enabled PhysX, much less those who can afford it where it seems to be most effective: high-end (read: expensive) multi-GPU solutions.  Developers want to reach the broadest audience possible.  And that’s the key, because according our friendly HP/Voodoo engineer; the widely popular and mainstream 8600GTS was a sore replacement for a CPU when it came to accelerating PhysX in FutureMark’s Vantage.  Perhaps if the 8600GTS were a secondary GPU in the system would it make sense.  But that again goes back to the argument for or against buying dedicated hardware for physics processing, which is what Aegia started out doing. 

I didn’t make it pasts the graphs in the engineer’s post showing the performance deltas before stopping to think about how I would write this article.  Going back to it, I realized I failed to read the last paragraph, which really put the whole thing in perspective for me, which brings me to the other thing I found interesting in the post.
I realized the whole affair was a condensed course of logic to get to get to the conclusion NVIDIA has come to and spent a considerable amount of effort in marketing and PR: that the GPU is slowly making the CPU irrelevant and will possibly replace it in some way someday.  It’s the whole “balanced computing” campaign in a nutshell.  The bottom line is that NVIDIA wants you to spend more money on their products than on a CPU.  That explained why in that blog post the synthetic nature of Vantage vs. the rest of the real world was topically glossed over, and no reference was made to the fact that the Vantage scores were basically irrelevant today in at least FutureMark’s eyes. 

Of course, none of this necessarily invalidates NVIDIA’s points on a balanced computing experience.  There are many compelling reasons today to pay attention to your GPU budget just as closely as your CPU budget: we truly are in the Age of Visual Computing.  I believe that the premise is a good one, and something that some of us in the industry do every day – deliver the performance that a customer wants that is tailored their needs.  And if that means making sure they spend more on a GPU, or more on a CPU, that’s what we try to deliver.

Not to get off on a tangent, but to hopefully wrap up this post, I wish that NVIDIA would spend their PR budget in regards to their balanced PC campaign in the retail space.  It’s always been disheartening to see how specifications are spun to an unsuspecting public and how the big box boys are dictating to the PC manufacturers what should or shouldn’t be included in a BOM.  Honestly, I think NVIDIA could do the mainstream a great service by focusing on retail buyers.   And they don’t even have to talk about PhysX to make a compelling story when PCs are being sold with 6GB of RAM and integrated graphics!  Yeah, that’s the ticket.  Leave the 900 pound gorilla – Intel – alone, and go after those pesky memory manufacturers!