Skip to main content

Docs

God shall bless us; and all the ends of the earth shall fear him.

도메인에 맞는 아키텍처

computer-architecture

AI 가속기 디자인 스펙트럼

slide 21

가장 유연한 CPU GPU Soft DPU(FPGA): filed programmable Hard DPU(NPU): TPU, ASICs(Arch Specific I C) 가장 효율적

Hard DPU: Google's Tensor Processign Unit

inference, training 두 종류로 나누어지고 있는 중 추론용은 lifetime이 짧고 메모리 footprint가 적음 training은 backpropagation으로 tensor의 lifetime이 김 v3(HBM 32GB) -> v4(8GB)

v4i(inference)기준 7nm two tensor core(mat mul) BF16 with FP32 accumulation support int8

systolic array: 1970년대 optimized for throughput

mat mul(compute intensive)

tpu v1-inference only, alphago

gemini(tpu v5-inference only) v5e(efficiency) 추론 위주지만, 학습도 할 수 있다! v5p(performance)

tco: Total cost of ownership? performance / TCO

Soft DPU: Microsoft BrainWave

performance / scalability

Fast, Flexible, Friendly

model parallelism(scale) => hight network bandwith => compiler(software) is most important

a custom FP format (ms-fp8/9) TOPS goes up with negligible accuracy loss 추론엔 좋을지라도 학습엔 의문

Soft DPU: Xilinx Versal Serires

DSP(일반적인 int, floating point 연산?)

ASIC: A3 - Attention Accelerators

iso-FLOPS comparison

20-157x speedup over GPU

  1. a novel approximate self-attention
  2. a specialized hardware architecture

결론

The historic end of Moore’s Law and Dennard Scaling ○ Requires innovations from architecture to improve cost-energy-performance.

Domain-specific architectures are booming for AI applications. ○ To get the most out of existing transistors via specialization ○ Delivers higher performance/Watt/mm2/$$ to meet the application demands

SW interface design and optimization are as equally (or even more) important as HW design. ○ e.g., support from OS, compiler, runtime system, and other tools ○ Cross-layer optimization (from applications to hardware) is required to balance efficiency, programmability, and cost for DSAs.