NVIDIA H100 CNX vs AMD Radeon Instinct MI8
GPU · Specs Compare
Graphics Card
Release Date
Mar 2023
Dec 2016
Generation
Tesla Hopper
Radeon Instinct
Type
AI GPU
Professional
Bus Interface
PCIe 5.0 x16
PCIe 3.0 x16
Clock Speeds
Base Clock
690 MHz
—
Boost Clock
1845 MHz
—
Memory Clock
1593 MHz
500 MHz
Memory
Memory Size
80GB
4GB
Memory Type
HBM2e
HBM
Memory Bus
5120bit
4096bit
Bandwidth
2039GB/s
512.0GB/s
Render Config
SM Count
114
—
Shading Units
14592
4096
TMUs
456
256
ROPs
24
64
Tensor Cores
456
—
L1 Cache
256 KB (per SM)
16 KB (per CU)
L2 Cache
50 MB
2 MB
Compute Units
—
64
Theoretical Performance
Pixel Rate
44.28 GPixel/s
64.00 GPixel/s
Texture Rate
841.3 GTexel/s
256.0 GTexel/s
FP16 (half)
215.4 TFLOPS
8.192 TFLOPS
FP32 (float)
53.84 TFLOPS
8.192 TFLOPS
FP64 (double)
26.92 TFLOPS
512.0 GFLOPS
Graphics Processor
GPU Name
GH100
Fiji
Architecture
Hopper
GCN 3.0
Foundry
TSMC
TSMC
Process Size
4 nm
28 nm
Transistors
80 billion
8.9 billion
Die Size
814 mm²
596 mm²
GPU Variant
—
Fiji XT CA(215-0862120)
Board Design
TDP
350W
175W
Suggested PSU
750 W
450 W
Outputs
No outputs
No outputs
Power Connectors
8-pin EPS
1x 8-pin
Graphics Features
DirectX
N/A
12 (12_0)
OpenGL
N/A
4.6
OpenCL
3.0
2.1
Vulkan
N/A
1.2.170
CUDA
9.0
—
Shader Model
N/A
6.5
▲ Specs Compare
Data sourced from official specifications and benchmark tests.