squall/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Updating readme with relative per chart

Browse Source
This commit is contained in:
dumerrill 2017-12-05 22:40:47 -05:00
parent e2bf51c3fe
commit 0428c89fd5
2 changed files with 8 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
README.md
View File

@ -5,7 +5,7 @@
CUTLASS is a collection of CUDA C++ template abstractions for implementing CUTLASS is a collection of CUDA C++ template abstractions for implementing
high-performance matrix-multiplication (GEMM) at all levels and scales within CUDA. high-performance matrix-multiplication (GEMM) at all levels and scales within CUDA.
It incorporates strategies for hierarchical decomposition and data movement similar It incorporates strategies for hierarchical decomposition and data movement similar
to those used to implement cuBLAS. CUTLASS decomposes these “moving parts” into to those used to implement cuBLAS. CUTLASS decomposes these "moving parts" into
reusable, modular software components abstracted by C++ template classes. These reusable, modular software components abstracted by C++ template classes. These
thread-wide, warp-wide, block-wide, and device-wide primitives can be specialized thread-wide, warp-wide, block-wide, and device-wide primitives can be specialized
and tuned via custom tiling sizes, data types, and other algorithmic policy. The and tuned via custom tiling sizes, data types, and other algorithmic policy. The
@ -20,6 +20,13 @@ point (FP64) types. Furthermore, CUTLASS demonstrates CUDA's WMMA API for targe
the programmable, high-throughput _Tensor Cores_ provided by NVIDIA's Volta architecture the programmable, high-throughput _Tensor Cores_ provided by NVIDIA's Volta architecture
and beyond. and beyond.
![ALT](/media/fig-09-complete-hierarchy.png "Relative performance of CUTLASS and cuBLAS for large matrices")
CUTLASS is very efficient, with performance comparable to cuBLAS for scalar GEMM
computations. The above figure shows CUTLASS performance relative to cuBLAS
compiled with CUDA 9.0 running on an NVIDIA Tesla V100 GPU for large matrix
dimensions (M=10240, N=K=4096).
For more exposition, see our Parallel Forall blog post ["CUTLASS: Fast Linear Algebra For more exposition, see our Parallel Forall blog post ["CUTLASS: Fast Linear Algebra
in CUDA C++"](https://devblogs.nvidia.com/parallelforall/cutlass-linear-algebra-cuda). in CUDA C++"](https://devblogs.nvidia.com/parallelforall/cutlass-linear-algebra-cuda).

BIN
media/cutlass-performance-plot.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 39 KiB