1. Cloud Deep Learning Chips
Training & Inference
Created date:2019.12.07
Updated date : 2019.12.15/17/25
@Vengineer
2. This is a summary of learning and inference
chips for deep learning in the cloud.
Each company's chips, product photos and
images are borrowed from the URL on the same
page.
3. Habana Labs:Goya (DRAM)
Intel Nervana:NNP-I (DRAM)
Google:TPU v1 (SRAM)
Groq (SRAM)
Alibaba:Hanguang(含光) 800 (SRAM)
The inference chip does not have an
interconnect itself.
InferenceTraining
Google:TPU v2/v3 (HBM2)
Intel Nervana:NNP-T (HBM2)
Habana Labs:Gaudi (HBM2)
Alphaics:RAP (HBM2 option)
Huawei:Ascend 910 (HBM2)
Graphcore:GC2 (SRAM)
Cerebras:CS-1 (SRAM)
The training chip has its own
interconnect.
https://vengineer.hatenablog.com/entry/2019/11/05/060000
https://github.com/basicmi/AI-Chip
https://twitter.com/jwangARK/status/1189560904872058880
21. Glow: A community-driven approach to AI
infrastructure specification
https://engineering.fb.com/ml-applications/glow-a-community-driven-approach-to-ai-infrastructure/
https://github.com/pytorch/glow
22. Intel and Baidu Continue Collaboration
across AI, AD and 5G
https://newsroom.intel.com/articles/intel-baidu-continue-collaboration-across-ai-ad-5g/
● BaiduBrain* (Baidu’s AI platform),
● PaddlePaddle* (Baidu’s deep learning platform)
● DuerOS* (Baidu’s AI-powered voice assistant platform)
● Apollo* (Baidu’s autonomous driving platform)
● Intel® Xeon® Scalable platform
● Intel® Optane™ DC Persistent Memory
● Intel® Optane™ DC SSD
● silicon photonics
● Ethernet
● Intel AI accelerators and Intel software stack
23. Training PyTorch models on Cloud TPU
Pods
https://cloud.google.com/tpu/docs/tutorials/pytorch-pod
github : https://github.com/pytorch/xla
24. MICROSOFT AND GRAPHCORE
COLLABORATE TO ACCELERATE
ARTIFICIAL INTELLIGENCE
https://www.graphcore.ai/posts/microsoft-and-graphcore-collaborate-to-accelerate-artificial-intellig
ence
Today we are very excited to share details of our collaboration with Microsoft,
announcing preview of Graphcore® Intelligence Processing Units (IPUs) on
Microsoft Azure.
● Graphcore IPUs with Dell EMC DSS 8440 Server
● Graphcore also delivers a full training runtime for ONNX and is working closely with the ONNX
organisation to include this in the ONNX standard environment. Initial PyTorch support is
available in Q4 2019 with full advanced feature support becoming available in early 2020.
25. Baidu, Facebook and Microsoft work together to
define the OCP Accelerator Module
specification
https://www.opencompute.org/blog/baidu-facebook-and-microsoft-work-together-to-define-the-oc
p-accelerator-module-specification
https://146a55aca6f00848c565-a7635525d40ac1c70300198708936b4e.ssl.cf1.rackcdn.com/images/22fa829b159a4c
ea7b33aa12bc2c61909e52d077.pdf
Other than Apple and Amazon =>
26. I am a computer engineer,
not a deep learning craftsman
ありがとうございました。
Thanks
@Vengineer
ソースコード解析職人
Source code analysis craftsman