“I have always been told that Nvidia cards are superior for video editing but I think that is just in regard to Adobe products.” Which never looks good if the other guy isn’t doing that. Could be chip yield, could just be software holding the Titan back.
I wouldn’t be surprised if the Tesla did better on those benchmarks anyway. No telling if it’s intentional or not, but it would sure be interesting to see a Tesla 20X run those same benchmarks… same chip as the Titan, the Kepler GK110. Hitting a wall like that usually tells you there’s a problem of some kind. They’re not even testing the nVidia GTX5xx or GTX6xx devices form that era. Sure, the HD5870 and HD6970 are most of those benchmarks, but keep in mind, the HD6970 is from 2011 it’s a contemporary of the GTX5xx series. But look at any of the other OpenCL benchmarks - nVidia just hits a wall. And sure, there’s that double precision Folding Home run that has the Titan pull way out in front of everyone else… the Titan does have the same chip as the Tesla 20X, which is sold specifically for “compute” applications. What you’d expect to see is something like the DirectCompute benchmarks (first and last in that linked article)… gradual dropping of performance as you go to older or lesser cards. But it’s not… the HD6970 is actually beating it on the Sony benchmark.īut here’s the weird part: nVidia hitting the wall. Look at the latest stuff… the nVidia Titan should, by all rights, be wiping the floor with my 2011-vintage HD6970, like the R9 290 is. Ok, sure, you’d expect the newer AMDs and nVidias to do better - both nVidia’s Kepler architecture and AMD’s GCN architecture were designed from the ground up to do this General Purpose GPU computing, as well as the usual stuff.
They cost the same, so this was a no-brainer.īut it’s not just that. As I’ve mentioned before, I bought both the nVidia GTX570 and the AMD HD6970 right after Vegas 11 came out, did lots of benchmarks, and found the AMD faster at everything, and bugs in some of the OpenCL stuff (non-Vegas) on the nVidia.
nVidia was also faster at gaming for the most part, and maybe had an edge in OpenGL, at least from time to time, so no one really questioned this idea.īut looking at least at OpenCL benchmarks, and in particular at Vegas benchmarks, that’s no longer the case.
AMD has this thing called Streams that no one supported, and once Adobe figured out how to have the GPU help you out, it made nVidia a clear leader. When GPGPU computing was young (oh so much younger than today), it was pretty much just CUDA. One problem with memes and other things “everyone knows”… they die a hard death.
0 kommentar(er)
