Home Hot News 엔비디아, MLPerf HPC 벤치마크서 최고의 AI 성능 입증

엔비디아, MLPerf HPC 벤치마크서 최고의 AI 성능 입증

2
0
엔비디아, MLPerf HPC 벤치마크서 최고의 AI 성능 입증

(CEO Ȳ) ý ǻ(HPC) ø̼ǿ AI ɿ ġũ MLPerf HPC 1.0 5 ׽Ʈ 4 ߴ.

 

MLPerf 2018 5 ó õ ׿ ġũ ׽Ʈ̴. MLPerf HPC AI Ͽ ǻͿ ùķ̼ ȭϰ ȭϴ ǻ Ÿ ٷ. ֱ о ڵ 罺 AI ä , õ ùķ̼ǿ HPC AI ̷ ִ.

 

MLPerf HPC 1.0 HPC Ϳ Ϲ ũε忡 AI Ʒ ߴ:
ڽ÷ο(CosmoFlow) ̹ ִ ü
ķ(DeepCAM) Ϳ 㸮ΰ (atmospheric river) Ž ׽Ʈ
īŻƮ(OpenCatalyst) ý 󸶳 ϴ

׽Ʈ ý 󸶳 ƷýŰ ϴ ϸ(Strong Scaling) ̿ ϴ ִ ý ó, ־ ð ý Ʒ ִ ϴ ϸ(Weak Scaling) ι Ǿ.

ƴ ۳ MLPerf 0.7 忡 5 ڽ÷ο ´. ķ ׽Ʈ 뷫 7 . η Ŭ (Lawrence Berkeley National Lab) ޸ 1(Perlmutter Phase 1) ý 6,144 A100 ټھ GPU 512 Ͽ īŻƮ ׽Ʈ ϸ ι ˳´.

, ϸ ι ۾ 16 256 ۾ Ͽ ķ ׽Ʈ ֵϿ. ׽Ʈ 系 ý ִ ǻ (Selene) Ǿ.

MLPerf ֽ AI ÷ ٸ ش. ƴ ͼ, Ŭ Ʈũ AI ߷ ϴ MLPerf ġũ 8 ְ ޼Ͽ.

 

忡 8 7 GPU Ͽ. ̴ ǻ (Jülich Supercomputing Center), ǻ (Swiss National Supercomputing Center), ̱ Ƹ η Ŭ (Argonne and Lawrence Berkeley National Laboratories), ǻ ø̼ (National Center for Supercomputing Applications) ػ罺 ǻ (Texas Advanced Computing Center) Ѵ.

MLPerf ġũ ˸ٹ, , , Ÿ( ̽), ̲ ׷ MLĿս(MLCommons) ϰ ִ.

̹ ü Ʈ ϴ ϼ AI ÷ ޼ߴ. MLPerf ƴ ó ȭϴ ޸(DALI) ִ 1,024 ̻ GPU ȿ Ȯϰ, ұԸ ġ ̱ (CUDA) ׷ ̿ ֵ ڵ带 ߴ.

ű׳IO(MagnumIO) ٽ (SHARP) Ǿµ, ̴ Ʈũ ǻ Ͽ ȭϰ ǴϹ(Quantum InfiniBand) ġ ۾ ε ְ ش.

⿡ Ʈ MLPerf ҿ ̿ ϴ. ƴ Ʒõ AI , ø̼ ӿũ, GPU ø̼ Ÿ Ʈ ҽ Ʈ NGC īŻα׿ ̷ ڵ带 ߰ϰ ִ.

%