Open CL vs CUDA performance

I ran some tests on my own and I want verify if my results are normal. Running on a i5 8300H and 1050 TI, rendering a 5 minute video with some fusion and color stuff took 10 minutes on CUDA and 30 minutes on Open CL. Is Open CL really that much worse?

and far worse if you have third party OFX in the mix

I ran some tests on my own and I want verify if my results are normal. Running on a i5 8300H and 1050 TI, rendering a 5 minute video with some fusion and color stuff took 10 minutes on CUDA and 30 minutes on Open CL. Is Open CL really that much worse?

Yes openCL is crippled by NVidia. For Davinci Resolve it doesn't matter if it's OpenCL, Cuda or Metal and leaves it up to the manufacturer (in this case NVidia) to cripple OpenCL in favor of CUDA.
Apple is trying to do the same thing with Metal vs. OpenCL, but has be less succesfull in crippling OpenCL in favor of Metal. It's also quite normal these days that NVidia cripples it's GPU via it's drivers when an AMD platform is detected(Zen, Zen+ and Zen2), they also don't react when questions are asked by review sites like Tomshardware.com.

And they still tell us that competition leads to improvement…

The problem is that nVidia hasn't had any legitimate competition for a long time, so it's been neglecting its OpenCL drivers because it doesn't need to care. Any clod can see how far ahead nVidia is right now... and the fact that several of AMD's GPU designers including the lead GPU architect have left isn't helping. (He's at Intel now.)

Of course, you'd think that the last thing that nVidia would want to do is encourage people to use Intel GPUs... the current ones aren't worth much, but Intel obviously is seeing the writing on the CPU wall...

The problem is that nVidia hasn't had any legitimate competition for a long time, so it's been neglecting its OpenCL drivers because it doesn't need to care. Any clod can see how far ahead nVidia is right now... and the fact that several of AMD's GPU designers including the lead GPU architect have left isn't helping. (He's at Intel now.)

Of course, you'd think that the last thing that nVidia would want to do is encourage people to use Intel GPUs... the current ones aren't worth much, but Intel obviously is seeing the writing on the CPU wall...

I think it helped a lot that the lead GPU architect left the building, some new fresh blood did good. NAVI is promissing, looking at the RX5700xt and it's cheap to produce (small die of only 251mm^2), for about the same price as the RTX 2060 super.
Scott Herckleman has a big grin on his face all day. Ray Tracing has a bright future, the big consols and Google are going to use AMD's solution. Haven't heard much of Jensen lately.

Let's hope intels solution will be as disruptive as they claim, we need more competition.

The architect wasn't the holdup.

Sent from my H1T1000 using Tapatalk

Alright, thanks a bunch mate!

I ran some tests on my own and I want verify if my results are normal. Running on a i5 8300H and 1050 TI, rendering a 5 minute video with some fusion and color stuff took 10 minutes on CUDA and 30 minutes on Open CL. Is Open CL really that much worse?

Yes openCL is crippled by NVidia. For Davinci Resolve it doesn't matter if it's OpenCL, Cuda or Metal and leaves it up to the manufacturer (in this case NVidia) to cripple OpenCL in favor of CUDA.
Apple is trying to do the same thing with Metal vs. OpenCL, but has be less succesfull in crippling OpenCL in favor of Metal. It's also quite normal these days that NVidia cripples it's GPU via it's drivers when an AMD platform is detected(Zen, Zen+ and Zen2), they also don't react when questions are asked by review sites like Tomshardware.com.

This has been proven to be false. Don't spread this FUD. Benchmarks have proven that the performance basically doesn't change when you run those cards on AMD or Intel hardware. They're identical or within margin of error (due to different CPUs, etc.). No different than running AMD on different platforms.

They probably don't respond because the questions are ridiculous (as that would mean they are basically asking for a class action lawsuit that they would lose de facto)...

Just changed my GPU from GTX1070Ti to Radeon 7 using Metal and my renders are much faster, also the timeline is very responsive.
Can reach 30-40FPS now and before was around 10-15...depending on the timeline / export

Any test (s) i can do to see if i am using the full potentiate of the GPU?
I'm on hackintosh running flowless system

Metal is optimal on macOS. OpenCL is deprecated and Nvidia driver support is lackluster. So, it's always better to go AMD & Metal on that platform.

You'd probably get better performance with the 1070 Ti if you ran Resolve on the Windows platform on that machine.

Metal is optimal on macOS. OpenCL is deprecated and Nvidia driver support is lackluster. So, it's always better to go AMD & Metal on that platform.

You'd probably get better performance with the 1070 Ti if you ran Resolve on the Windows platform on that machine.

The 1070ti is no match for the Radeon VII in Resolve.
Since Resolve is hardware agnostic the only things that really matter are: fp32 performance, memory bandwidth and the amount of memory.

1070ti: 8.186 Tflops fp32 peak, 256 GByte/s and 8 GByte.
Radeon VII: 13.44 Tflops fp32 peak, 1,024 Gbyte/s and 16 GByte.

Metal is optimal on macOS. OpenCL is deprecated and Nvidia driver support is lackluster. So, it's always better to go AMD & Metal on that platform.

You'd probably get better performance with the 1070 Ti if you ran Resolve on the Windows platform on that machine.

The 1070ti is no match for the Radeon VII in Resolve.
Since Resolve is hardware agnostic the only things that really matter are: fp32 performance, memory bandwidth and the amount of memory.

1070ti: 8.186 Tflops fp32 peak, 256 GByte/s and 8 GByte.
Radeon VII: 13.44 Tflops fp32 peak, 1,024 Gbyte/s and 16 GByte.

Thanks for this info
I used the GTX 1070 Ti in Windows , i got the same performance as in OS X...

For the record when I said better performance I meant better than in macOS. Not better than the Radeon.

Sent from my SM-N960U using Tapatalk