| ||||||||||
2013/2/19 Teddy Gage <teddygage@gmail.com> Depends on overclocking, because the "speed boost" of the 6xx series actually limits maximum OC, which isn't true on the 580, which you can push farther. Anyway I'm dying to see some AE benchmarks on the GTX Titan, it's basically a consumer tesla K20, I hope they didn't gimp the CUDA performance like they did with the 6xx series. Have you had anyone test a k20 on your render benchmarks? Vray has a really nice gpu renderer too. Do you say that nvidia gimp the CUDA performance on the 6xx cards compared with the pro cards? I have the Quadro K5000 that is based on the same processors as the 680 and have the same amount, but it is clocked a bit lower so is actually slower than the gamer card. As i understand the Kepler architechure, is that it is just not made for double precision floating point performance like the older Fermi archtechure was. And that is what makes the newer cards slower in compute applications (eg raytracing in AE) than the 5xx series.
The tesla K20 should have better double precision floating point performance and i would be surprised if nvidia has put that version into this Titan gamer card, but if true it would make it a very interesting card to have :-) maybe a cheap "telsa" solution for the apps you could "hack" to see it as an extra compute card? AE maybe could?
- Robert |
