Intel Gaudi 2 vs NVIDIA A100 SXM Full Specs

Summary

6,144 Shaders 550MHz	6,912 Shaders 1.41GHz
96GB HBM2e2.45TB/s	80GB HBM2e2.04TB/s
Peak·FP32·FP16 -	Peak·FP32·FP16 19.49 TFLOPS
Form Factor OAM Module	Form Factor SXM4
TDP 600W	TDP 400W

Highlights

Peak OPS

Peak OPS FP32 FP16 Tensor FP8-32 Tensor FP16-16 FP16-16 Tensor Sparse Tensor FP16-32 FP16-32 Tensor Sparse FP64 Tensor FP64 BF16 Tensor BF16 BF16 Tensor Sparse Tensor TF32 Tensor INT8 Tensor INT4 Memory Bandwidth TDP

Gaudi 2865.1 TFLOPSFP8-32 Tensor

x1

A100 SXM2.5 POPSINT4 Tensor Sparse

x2.88

Show Top Score

Theoretical Performance

Clock Speed550MHz UC·Boost·OC	Clock Speed1410MHz UC·Base·Boost·OC
Peak OPS 865.1 TFLOPSFP8-32 Tensor	Peak OPS 2.5 POPSINT4 Tensor Sparse
Tensor FP8-32 865.1 TFLOPS	-
-	Tensor FP16-16 311.9 TFLOPS FP16-16 Tensor Sparse 623.7 TFLOPS Tensor FP16-32 311.9 TFLOPS FP16-32 Tensor Sparse 623.7 TFLOPS
BF16 - Tensor BF16 432.5 TFLOPS BF16 Tensor Sparse -	BF16 38.98 TFLOPS Tensor BF16 311.9 TFLOPS BF16 Tensor Sparse 623.7 TFLOPS
Tensor TF32 -	Tensor TF32 155.9 TFLOPS
-	FP32 19.49 TFLOPS
-	FP64 9.75 TFLOPS Tensor FP64 19.49 TFLOPS
Tensor INT4 -	Tensor INT4 1.25 POPS
-	Tensor INT8 623.7 TOPS
Pixel Rate -	Pixel Rate 225.6 GPixel/s
Texture Rate -	Texture Rate 609.1 GTexel/s

Core Config

Shaders 6,144 Shaders	Shaders 6,912 Shaders
TMUs -	TMUs 432 TMUs
ROPs -	ROPs 160 ROPs
Tensor Cores 24 T-Cores	Tensor Cores 432 T-Cores
CUs 2 CUs	SMs 108 SMs

Clocks

Base Clock -	Base Clock 1.27GHz
Boost Clock 550MHz	Boost Clock 1.41GHz

Cache

L2 Cache -

L2 Cache 41MB shared

Memory

96GB HBM2e	80GB HBM2e
Memory Bus 6144-bit	Memory Bus 5120-bit
Memory Speed 3.2GT/s	Memory Speed 3.2GT/s
Memory Bandwidth 2.45TB/s	Memory Bandwidth 2.04TB/s
ECC Yes	ECC -
ESRAM 48MBSize 6400GB/sBandwidth	ESRAM -

Power & Thermals

TDP 600W

TDP 400W

Decoder

Decoder Model 8x Gaudi

Decoder Model 5x NVDEC 4

Card

Form Factor OAM Module	Form Factor SXM4
Cooling Passive	Cooling Passive

Chip

Manufacturer Intel	Manufacturer NVIDIA
Chip Designer Intel	Chip Designer NVIDIA
Architecture Gaudi 2	Architecture Ampere
Family Gaudi	Family Server
Branding	Branding
Codename Gaudi 2	Codename NV170
Chip Variant HL-2080	Chip Variant -
Market Segment Server	Market Segment Server
Release Date May 1, 2022	Release Date Jun 28, 2021

Fabrication

Foundry TSMC	Foundry TSMC
Fabrication Node N7	Fabrication Node 7N
Die Size -	Die Size 826mm²
Transistor Count -	Transistor Count 54.2 Billion
Transistor Density -	Transistor Density 65.62 MTr/mm²

Images

No images available

No images available

Competitors

IntelGaudi 2 vs GPU NVIDIAA800 IntelGaudi 2 vs GPU NVIDIAA100 IntelGaudi 2 vs GPU NVIDIAA100X Show 12 more

NVIDIAA100 SXM vs GPU NVIDIAA100 SXM 40GB NVIDIAA100 SXM vs GPU NVIDIAA100X NVIDIAA100 SXM vs GPU NVIDIAA800 Show 12 more