NVIDIA H100 SXM 80GB vs AMD Instinct MI250X Full Specs

Summary

16,896 Shaders 1.98GHz	14,080 Shaders 1.7GHz
80GB HBM33.35TB/s	128GB HBM2e3.28TB/s
Peak·FP32·FP16 66.91 TFLOPS	Peak·FP32·FP16 47.87 TFLOPS
Form Factor SXM5	Form Factor OAM Module
TDP 700W	TDP 560W

Highlights

Peak OPS

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

H100 SXM 80GB3.96 POPSINT8 Tensor Sparse

x10.33

Instinct MI250X383 TOPSINT4 Tensor

x1

Show Top Score

Theoretical Performance

Clock Speed1980MHz UC·Base·Boost·OC	Clock Speed1700MHz UC·Base·Boost·OC
Peak OPS 3.96 POPSINT8 Tensor Sparse	Peak OPS 383 TOPSINT4 Tensor
Tensor FP8-16 1.98 PFLOPS FP8-16 Tensor Sparse 3.96 PFLOPS Tensor FP8-32 1.98 PFLOPS FP8-32 Tensor Sparse 3.96 PFLOPS	-
Tensor FP16-16 989.4 TFLOPS FP16-16 Tensor Sparse 1.98 PFLOPS Tensor FP16-32 989.4 TFLOPS FP16-32 Tensor Sparse 1.98 PFLOPS	Tensor FP16-16 383 TFLOPS FP16-16 Tensor Sparse - Tensor FP16-32 383 TFLOPS FP16-32 Tensor Sparse -
BF16 133.8 TFLOPS Tensor BF16 989.4 TFLOPS BF16 Tensor Sparse 1.98 PFLOPS	BF16 - Tensor BF16 383 TFLOPS BF16 Tensor Sparse -
Tensor TF32 494.7 TFLOPS	Tensor TF32 -
FP32 66.91 TFLOPS Tensor FP32 -	FP32 47.87 TFLOPS Tensor FP32 95.74 TFLOPS
FP64 33.45 TFLOPS Tensor FP64 61.84 TFLOPS	FP64 47.87 TFLOPS Tensor FP64 95.74 TFLOPS
Tensor INT4 -	Tensor INT4 383 TOPS
Tensor INT8 1.98 POPS	Tensor INT8 383 TOPS
Pixel Rate 47.5 GPixel/s	Pixel Rate -
Texture Rate 1.04 TTexel/s	Texture Rate 1.5 TTexel/s

Core Config

Shaders 16,896 Shaders	Shaders 14,080 Shaders
TMUs 528 TMUs	TMUs 880 TMUs
ROPs 24 ROPs	ROPs -
Tensor Cores 528 T-Cores	Tensor Cores 880 T-Cores
SMs 132 SMs	CUs 220 CUs

Clocks

Base Clock 1.06GHz	Base Clock 1GHz
Boost Clock 1.98GHz	Boost Clock 1.7GHz
Tensor Clock 1.83GHz	Tensor Clock -

Cache

L2 Cache 51.2MB shared

L2 Cache 16.4MB shared

Memory

80GB HBM3	128GB HBM2e
Memory Bus 5120-bit	Memory Bus 8192-bit
Memory Speed 5.2GT/s	Memory Speed 3.2GT/s
Memory Bandwidth 3.35TB/s	Memory Bandwidth 3.28TB/s
ECC -	ECC Yes

Power & Thermals

TDP 700W

TDP 560W

Encoder

-	Encoder Model 2x VCN 2.6

Decoder

Decoder Model 7x NVDEC 5

Decoder Model 2x VCN 2.6

Card

Form Factor SXM5	Form Factor OAM Module
Cooling Passive	Cooling Passive

Chip

Manufacturer NVIDIA	Manufacturer AMD
Chip Designer NVIDIA	Chip Designer AMD
Architecture Hopper	Architecture CDNA 2
Family Server	Family Instinct
Branding	Branding
Codename NV180	Codename Aldebaran
Chip Variant -	Chip Variant Aldebaran XTX
Market Segment Server	Market Segment Server
Release Date Mar 22, 2022	Release Date Nov 8, 2021

Fabrication

Foundry TSMC	Foundry TSMC
Fabrication Node 4N	Fabrication Node N6
Die Size 814mm²	Die Size 1448mm²
Transistor Count 80 Billion	Transistor Count 56 Billion
Transistor Density 98.28 MTr/mm²	Transistor Density 38.67 MTr/mm²

Images

No images available

No images available

Competitors

NVIDIAH100 SXM 80GB vs GPU NVIDIAB100 NVIDIAH100 SXM 80GB vs GPU NVIDIAL40S NVIDIAH100 SXM 80GB vs GPU NVIDIAL40 Show 12 more

AMDInstinct MI250X vs GPU AMDInstinct MI250 AMDInstinct MI250X vs GPU AMDInstinct MI300A AMDInstinct MI250X vs GPU AMDInstinct MI300X Show 12 more