Is Blas, or OpenCL really the right way to go

[AdminX]

Recently compiled Lc0 0.21.2 with the three different blas variants on my mac and run a small test on network 11258-32x4-se.pb.gz.

My MacBook Air with Intel Core i5 from 2015 shows after "go nodes 50000" about

• 2400 nps with Apple vecLib, thread=1
  3500 nps with Apple vecLib, threads=2
  
  2350 nps with openBLAS 0.3.6 from Homebrew, thread=1
  4000 nps with openBLAS 0.3.6 from Homebrew, threads=2
  
  2750 nps with MKL 2019u4 from Intel, thread=1
  4600 nps with MKL 2019u4 from Intel, threads=2

With network netT40.T8.610 only ~50 nps are possible with BLAS.

On my MacBook Air the Intel HD6000 gpu acts slower with openCL. Not to mention the 35+ watts are "killing" the notebook within seconds. Using Intel Power Gadget for this scenario seems an absolute must.

I think I will give a shot at compiling a MKL Dev built sometime later today.

[Ovyron]

I hope they drop OpenCL and switch to something that doesn't crash when doing clGetPlatformIDs, so I can run GPU Leela at all.

[Gian-Carlo Pascutto]

I hope they drop OpenCL and switch to something that doesn't crash when doing clGetPlatformIDs, so I can run GPU Leela at all.

If your drivers are broken then switching APIs isn't going to fix it. I assume you're not on an NVIDIA card or you'd use the CUDA build, so what hardware is it, and what drivers are you using?

For AMD the only other option is ROCm and I'm not even sure it supports Windows at all.

[Ovyron]

I assume you're not on an NVIDIA card or you'd use the CUDA build, so what hardware is it, and what drivers are you using?

Latest drivers on a NVIDIA GeForce 7300 SE / 7200 GS.

It seems CUDA was introduced with GeForce 8800, so I ran out of luck.

[Rebel]

Recently compiled Lc0 0.21.2 with the three different blas variants on my mac and run a small test on network 11258-32x4-se.pb.gz.

My MacBook Air with Intel Core i5 from 2015 shows after "go nodes 50000" about

• 2400 nps with Apple vecLib, thread=1
  3500 nps with Apple vecLib, threads=2
  
  2350 nps with openBLAS 0.3.6 from Homebrew, thread=1
  4000 nps with openBLAS 0.3.6 from Homebrew, threads=2
  
  2750 nps with MKL 2019u4 from Intel, thread=1
  4600 nps with MKL 2019u4 from Intel, threads=2

With network netT40.T8.610 only ~50 nps are possible with BLAS.

On my MacBook Air the Intel HD6000 gpu acts slower with openCL. Not to mention the 35+ watts are "killing" the notebook within seconds. Using Intel Power Gadget for this scenario seems an absolute must.

I haven't loaded LZ for over a year, where can I download these red networks?

Can't find them here - https://lczero.org/networks/

[Max]

You find all the distilled networks at https://github.com/dkappe/leela-chess-w ... d-Networks

The net T40.T8.610 was playing in the superfinal TCEC 15. Don't remember where the download was hostet. Maybe google helps

[AdminX]

Recently compiled Lc0 0.21.2 with the three different blas variants on my mac and run a small test on network 11258-32x4-se.pb.gz.

My MacBook Air with Intel Core i5 from 2015 shows after "go nodes 50000" about

• 2400 nps with Apple vecLib, thread=1
  3500 nps with Apple vecLib, threads=2
  
  2350 nps with openBLAS 0.3.6 from Homebrew, thread=1
  4000 nps with openBLAS 0.3.6 from Homebrew, threads=2
  
  2750 nps with MKL 2019u4 from Intel, thread=1
  4600 nps with MKL 2019u4 from Intel, threads=2

With network netT40.T8.610 only ~50 nps are possible with BLAS.

On my MacBook Air the Intel HD6000 gpu acts slower with openCL. Not to mention the 35+ watts are "killing" the notebook within seconds. Using Intel Power Gadget for this scenario seems an absolute must.

I haven't loaded LZ for over a year, where can I download these red networks?

Can't find them here - https://lczero.org/networks/

You can locate it here at the Network Download Tab (#27): https://docs.google.com/spreadsheets/d/ ... 0/htmlview#
Direct Link: https://www.dropbox.com/s/rmcf0lf1hes10 ... 10000?dl=0