Nvidia flaunts Kepler’s GPU power in video demos
May 16, 2012
Nvidia is flexing its graphics muscle at the 2012 GPU Technology Conference, and the videos below show off Kepler’s new visual tricks:real-time ray tracing, simulation of physical bodies, and cloud gaming powered by its new GeForce Grid system.
Comments (8)
by Matthew
What you are saying Locke makes a lot of business sense. Nvidia does not want its compute business to bleed into its gaming business, since the latter funds the former!
by Locke
Absolutely, it makes perfect business sense.
Sadly it takes the high compute capability out of the hands of DIY’ers and hobbyists they had with Nvidia’s cards up until this generation.
These folks will move to AMD/ATI for performance reasons, which in turn negatively impacts the same communities on the Linux platform, which historically has had much better driver support from Nvidia than from ATI.
Bottomline: I feel the move to more locked in, proprietary solutions alienates a community that is likely to be the forerunners of the technological future that everybody who read this site is dreaming about.
by Locke
I would also like to add to this that Nvidia has other incentives than pure performance for users to buy their Tesla line of cards, such as more on board memory (very important for graphical visualization applications), more robustness (designed to run for days uninterrupted) and professional support.
With the Fermi GPU, a GTX-580/GTX-590 actually had better raw compute performance than most of the Tesla lineup in certain applications, but was lacking the 4-6+ Gb of RAM necessary for other applications.
So to me this means capping the compute performance of new gamer cards for anything non-game related isn’t really buying Nvidia much except maybe a limited segment in the lower end of the VFX industry.
by Matthew Fuller
I would love to see the internal Nvidia stats to back that last paragraph up Locke! I bet many analysts would…
It’s hard to understand why companies do what they do when it seems to violate good business practice.
Just out of curiosity, say I wanted to create really cool, complex 3d objects rotating in space. Do you happen to know the best program for this purpose if I am going for realism (and real time) rendering?
by Locke
I am sure it makes business sense to some, not sure if I agree. I guess numbers do not lie in the end, but I would be curious to see the numbers that say high compute customers buy games hardware and run it 24/7 without meltdowns and live with the memory limitations. This is just speculation from me at this point, more likely it probably comes down to Nvidia not wanting to explain why a piece of hardware that cost a quarter of the price of their commercial product is capable of the same compute speeds. If this is artificially capped at all of course, it could just be the software has not caught up yet.
But too bad for those who use these consumer cards for everything from hobbyist 3D rendering to educational institutions doing low cost, high compute number crunching.
As for real time realistic graphics, I suppose it depends on your expectations of the level of realism. There are a number of packages that could probably live up to current game type graphics which can look fairly realistic. Maybe look into something like UDK or Unity 3D? I believe both are free for non-commercial use (and commercial use up to a certain point of revenue too.)
by Matthew Fuller
I am not exactly sure why they do what they do. But they do rebrand old cards with a higher number. You can be sure every major move is highly calculated to maximize profit. In essence, businesses keep secret what they can to maintain an edge but the secret doesn’t last long. There was once a programable upgrade for intel processors to unlock an extra core! Remember that? Lol.
As for game engines, yeah, the UDK sounds like a good bet. My main issue is importing complex pencil based drawings. I am not much of an artist so I go the abstract route! hehe. thanks, and good luck with everything.
by Matthew Fuller
Oops, I should add that while blocking progress through business secrets is a futile strategy, a very different strategy that works far too well is copyright. But that is an entirely different subject.
by Locke
Too bad Kepler (at least in its GK104 consumer boards) has very disappointing performance in CUDA/OpenCL raytacing compared to last generation’s Fermi chip.
Look at the Blender/Cycles and LuxRender hobbyist communities for reference.
This could be an artificial cap from Nvidia or the raytracing code could need a serious rewrite, but as it stands the future of hobbyist GPU compute performance on the latest Nvidia cards is uncertain.
I did notice they helped Stanford optimize the Folding@Home code to run at comparable speeds to those of Fermi based GPUs, so I guess there is still a chance unless Nvidia tweaked the driver to uncap consumer cards for a very narrow range of applications.