[julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Douglas Bates
I have been working on a package https://github.com/dmbates/ParalllelGLM.jl and noticed some peculiarities in the timings on a couple of shared-memory servers, each with 32 cores. In particular changing from 16 workers to 32 workers actually slowed down the fitting process. So I decided to che

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Viral Shah
> On 05-Dec-2014, at 1:16 am, Douglas Bates wrote: > > Thanks, I'll try that. I'm still curious as to why there is so little > difference between 8 and 16 threads. peakflops() just performs a matrix multiplication to estimate the flops. It uses a 2000x2000 matrix by default, which is good fo

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Douglas Bates
On Thursday, December 4, 2014 1:50:06 PM UTC-6, Viral Shah wrote: > > > On 05-Dec-2014, at 1:16 am, Douglas Bates > wrote: > > > > Thanks, I'll try that. I'm still curious as to why there is so little > difference between 8 and 16 threads. > > peakflops() just performs a matrix multiplicati

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Viral Shah
> On 05-Dec-2014, at 1:32 am, Douglas Bates wrote: > > On Thursday, December 4, 2014 1:50:06 PM UTC-6, Viral Shah wrote: > > On 05-Dec-2014, at 1:16 am, Douglas Bates > > > wrote: > > > > Thanks, I'll try that. I'm still curious as to why there is so little > > difference between 8 and 16 t

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Stefan Karpinski
Hyperthreading? Of the threshold is 16 but you're really only getting 8 cores, you might only get scaling up to 8. > On Dec 4, 2014, at 3:24 PM, Viral Shah wrote: > > >> On 05-Dec-2014, at 1:32 am, Douglas Bates wrote: >> >> On Thursday, December 4, 2014 1:50:06 PM UTC-6, Viral Shah wrote:

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Douglas Bates
On Thursday, December 4, 2014 2:32:01 PM UTC-6, Stefan Karpinski wrote: > > Hyperthreading? Of the threshold is 16 but you're really only getting 8 > cores, you might only get scaling up to 8. > This machine has AMD Opteron processors. I know Intel uses hyperthreading, does AMD also use it? I

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Johan Sigfrids
The new AMD architectures are weird in that they have two integer cores share the same FP hardware so you half the FP cores compared to integer cores. The reported number of cores is based in integer cores. On Thursday, December 4, 2014 11:13:38 PM UTC+2, Douglas Bates wrote: > > On Thursday, D

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Douglas Bates
On Thursday, December 4, 2014 4:17:19 PM UTC-6, Johan Sigfrids wrote: > > The new AMD architectures are weird in that they have two integer cores > share the same FP hardware so you half the FP cores compared to integer > cores. The reported number of cores is based in integer cores. > Thanks

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Stefan Karpinski
Well, that's bizarre. On Thu, Dec 4, 2014 at 5:17 PM, Johan Sigfrids wrote: > The new AMD architectures are weird in that they have two integer cores > share the same FP hardware so you half the FP cores compared to integer > cores. The reported number of cores is based in integer cores. > > On

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread Viral Shah
That is very interesting. Any idea why they are doing this? -viral On Friday, December 5, 2014 3:47:19 AM UTC+5:30, Johan Sigfrids wrote: > > The new AMD architectures are weird in that they have two integer cores > share the same FP hardware so you half the FP cores compared to integer > cores

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread cdm
here is something from the wiki (http://en.wikipedia.org/wiki/Opteron): "The Opteron approach to multi-processing is not the same as standard symmetric multiprocessing ; instead of having one bank of memory for all CPUs, each CPU has its o

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-04 Thread cdm
a touch more on the architecture: http://en.wikipedia.org/wiki/Piledriver_(microarchitecture) i am sure there is much more out there ... later, cdm On Thursday, December 4, 2014 10:58:47 PM UTC-8, cdm wrote: > > > here is something from the wiki (http://en.wikipedia.org/wiki/Opteron): > > "

Re: [julia-users] Decreasing rate of return on number of blas threads

2014-12-05 Thread Johan Sigfrids
This was first released in AMD's Bulldozer architecture in 2011. I believe, and this is speculation on my part, that the long term goal AMD is working toward are heterogeneous CPUs. This would be a CPU where you have a pile of high performance integer cores working similar to cores on CPUs today