NVIDIA OpenCL Benchmarks 6-Way With The 396.45 Linux Driver

[pcxmac]

nvidia has a problem with dicking it's vendors and customers. Some people keep going back for more, so they don't really care to change their attitude. AMD, not nearly as bad, but they don't have the same kind of market share. I vote with my dollars, and appreciate firms who choose to contribute more towards the 'open source' route.

[deppman]

The reason you're saying that (trolling?)

Anyway, Apple is dropping OpenCL, so they clearly don't care about it (or CUDA).

Apple and Nvidia had some bad blood that I don't know too much about. You can surely find much written about it, elsewhere. Besides that, my guess is that AMD probably gives Apple better pricing. Since Apple has a captive customer base, AMD's lower efficiency and performance is less of an issue for them. Plus, that relationship dates back before Maxwell, which is when Nvidia really started to pull a clear lead.

I'm not trolling. I am actually wondering if NV opencl support is now as good as AMD.

As for the efficiency comment, I'd say Nvidia started to pull ahead with Kepler and kept rolling with Maxwell and Pascal as it's mobile first initiate paid off. The differences now are substantial indeed.

I guess apple is dropping openCL in favor of MetalCL these days?

[coder]

I'm not trolling. I am actually wondering if NV opencl support is now as good as AMD.

There's no good reason why it should be bad, exactly - it's not much different than CUDA. The biggest issue with their support is that they're still stuck at v 1.2.

As for the efficiency comment, I'd say Nvidia started to pull ahead with Kepler and kept rolling with Maxwell and Pascal as it's mobile first initiate paid off. The differences now are substantial indeed.

I remember Maxwell being a huge leap, at least in graphics performance. But it seems Kepler had good theoretical compute efficiency, at least.

https://www.karlrupp.net/2013/06/cpu-gpu-and-mic-hardware-characteristics-over-time/