CLASTR SYSTEMS

CLASTR SYSTEMS - coming soon to cluster near you! Contact [email protected]

[Ian Goodfellow @goodfellow_ian]

CycleGAN turning a horse video into a zebra video ( https://t.co/YYCsVt4rIP )

[. @nivertech]

Are Team Green and Team Red are codenames for NVIDIA and AMD? #gpu

[OpenCL org @OpenCLorg]

Symposium: #Multicore Architectures and Their Effective Operation 2012. 28-29 June 2012, Barcelona, ES. http://t.co/5QtYfHD3 #HPC

[. @nivertech]

New book from former RapidMind guys: "Structured Parallel Programming"

[. @nivertech]

simd pragmas + auto-vectorization is most effective method for TODAY, but it will not stay this way for a long time

[. @nivertech]

MMX - 64 bits wide
SSE - 32
AVX - 256
MIC - MIC

[. @nivertech]

#pragma simd is unsafe

[. @nivertech]

google "SPMD on SIMD" paper from Intel Research

[. @nivertech]

Intel mentioned APL in regard of superior language for vectorization ;)

[. @nivertech]

"Pointers in C and C++ are evil" b/c intel compilers and tools can't see arrays and array slices behind them. Fortran 90 already solved this

[. @nivertech]

So Intel ArrBB is now "array notations",
They also added "SIMD pragmas"

[. @nivertech]

Intel Software guy admits, that Software is always behind Hardware

[. @nivertech]

Intel TBB, Intel Cilk, Intel ArrBB - why they aren't merged already?

[. @nivertech]

@intel claims that TBB is the best solution for Parallelism in C++: @Microsoft C++ AMP guys will be offended;) I think both are not perfect

[. @nivertech]

@nivertech there was a port of Linux kernel compiled with icc, why @intel didn't used it?

[. @nivertech]

Intel ported gcc to Xeon Phi to build a Linux kernel,warns that it shouldn't be used for compute code,since it doesn't support vectorization

[. @nivertech]

Intel Xeon Phi based supercomputer - 118 TFLOPs - #150 on Top500 - much more power-efficient than GPU-based supercomputers

[. @nivertech]

Intel Xeon Phi - "SMP on a chip"

[. @nivertech]

I don't believe it Intel exec said "if you need performance, you can use GPU Compute" ;)