GeekRadar / The comparison desk
COMPARE.
GPU
NVIDIA

NVIDIA H100 SXM5 94 GB

Graphics processor

Video memory
94 GB*
Memory bus
5,120 bit*
Rated memory bandwidth
3,360 GB/s*

Benchmark results

  • LLM Inference: MLPerf: gptj-99 (fp8) · version unreported · score1,693.47* tokens/sPlatform unreported Compare results
  • LLM Inference: MLPerf: gptj-99.9 (fp8) · version unreported · score1,693.47* tokens/sPlatform unreported Compare results
  • LLM Inference: MLPerf: llama2-70b-99 (fp8) · version unreported · score1,668.90* tokens/sPlatform unreported Compare results
  • LLM Inference: MLPerf: llama2-70b-99.9 (fp8) · version unreported · score2,106.08* tokens/sPlatform unreported Compare results
  • LLM Inference: MLPerf: llama2-70b-interactive-99 (fp8) · version unreported · score1,380.19* tokens/sPlatform unreported Compare results
  • LLM Inference: MLPerf: llama2-70b-interactive-99.9 (fp8) · version unreported · score1,380.19* tokens/sPlatform unreported Compare results
  • LLM Inference: MLPerf: mixtral-8x7b (fp8) · version unreported · score4,714.20* tokens/sPlatform unreported Compare results

Processing & platform

Architecture
Hopper*
Process node
5 nm*
GPU cores
16,896*
Graphics interface
PCIe 5.0 x16*
Reported chip die area
814 mm²*
Transistors
80,000 million*
GPU L2 cache
50 MB*
Graphics processor
GH100*
Texture mapping units
528*
Raster output units
24*
Streaming multiprocessors
132*
Tensor cores
528*
Chip foundry
TSMC*
Card or module format
SXM Module*
Declared GPU TDP
700 W*
Suggested system PSU
1,100 W*
External power connectors
8-pin EPS*
OpenCL
3.0*
CUDA compute capability
9.0*

Memory & storage

Memory options
94*
Video memory
94 GB*
Video memory type
HBM3*
Memory bus
5,120 bit*
Rated memory bandwidth
3,360 GB/s*

Power specifications

Thermal design power
700 W*

System configuration

Detailed specifications

Specifications 1 · 6 fields
GPU architecture — H100-SXM5 (94GB) — 94 GB
Hopper*
Native hardware support: FP4 tensor — H100-SXM5 (94GB) — 94 GB
No*
Native hardware support: FP6 tensor — H100-SXM5 (94GB) — 94 GB
No*
Physical configuration — H100-SXM5 (94GB) — 94 GB
SXM5*
Memory capacity per GPU — H100-SXM5 (94GB) — 94 GB
94* GB
GPU model — H100-SXM5 (94GB) — 94 GB
H100-SXM5 (94GB)*
Specifications 2 · 51 fields
proc transistors
80,000 million*
proc die size
814* mm²
cores
16,896*
tensor cores
528*
sm count
132*
base clock
1,350*
boost clock
1,980*
memory clock
1,313*
ram
94*
name
NVIDIA H100 SXM5 94 GB*
test name Graphics benchmark result
LLM Inference: MLPerf: gptj-99 (fp8)*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
test name Graphics benchmark result
LLM Inference: MLPerf: llama2-70b-interactive-99 (fp8)*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
test name Graphics benchmark result
LLM Inference: MLPerf: llama2-70b-interactive-99.9 (fp8)*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
test name Graphics benchmark result
LLM Inference: MLPerf: llama2-70b-99.9 (fp8)*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
test name Graphics benchmark result
LLM Inference: MLPerf: llama2-70b-99 (fp8)*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
test name Graphics benchmark result
LLM Inference: MLPerf: gptj-99.9 (fp8)*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
test name Graphics benchmark result
LLM Inference: MLPerf: mixtral-8x7b (fp8)*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
ram type
HBM3*
bus width
5,120*
bus interface
PCIe 5.0 x16*
ram bandwidth
3.36 TB/s*
fp16 performance
267.6*
fp32 performance
66.91*
fp64 performance
33.45*
score Graphics benchmark result
1,693.4675*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
score Graphics benchmark result
1,693.4675*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
score Graphics benchmark result
2,106.075*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
score Graphics benchmark result
1,668.9*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
score Graphics benchmark result
1,380.1875*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
score Graphics benchmark result
1,380.1875*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
score Graphics benchmark result
4,714.2*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
tdp
700* W
suggested psu
1,100* W
slot width
SXM Module*
power connectors
8-pin EPS*
api cuda
9*
is retail board
No*
gpi value
57.75*
unit Graphics benchmark result
tokens/s*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
unit Graphics benchmark result
tokens/s*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
unit Graphics benchmark result
tokens/s*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
unit Graphics benchmark result
tokens/s*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
unit Graphics benchmark result
tokens/s*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
unit Graphics benchmark result
tokens/s*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
unit Graphics benchmark result
tokens/s*
Scope and conditions
benchmark_context
suite
Per-row third-party test name*
test_configuration_verified
No*
comparable
No*
vendor
nvd*
manufacturer
NVIDIA*
arch name
Hopper*
card release date
2023-03-21*
proc foundry
TSMC*
proc process size
5* nm
Specifications 3 · 35 fields
CUDA compute capability
9.0*
OpenCL
3.0*
Architecture
Hopper*
Graphics interface
PCIe 5.0 x16*
Memory bus
5,120* bit
Reported shader processors
16,896*
Reported theoretical FP16 peak
267.6* TFLOPS
Scope and conditions
Arithmetic density
Not stated*
Reported theoretical FP32 peak
66.91* TFLOPS
Scope and conditions
Arithmetic density
Not stated*
Reported theoretical FP64 peak
33.45* TFLOPS
Scope and conditions
Arithmetic density
Not stated*
Declared L1 cache configuration
256 KB (per SM)*
Declared GPU L2 cache
50 MB*
Reported display outputs
No outputs*
Theoretical pixel fill rate
47.52 GPixel/s*
External power connectors
8-pin EPS*
Reported chip die area
814 mm²*
Chip foundry
TSMC*
Process node declaration
5 nm*
Transistors
80,000 million*
Graphics processor
GH100*
Reported video memory
94* GB
Rated memory bandwidth
3.36 TB/s*
Video memory type
HBM3*
Raster output units
24*
Reported shading units
16,896*
Card or module format
SXM Module*
Streaming multiprocessors
132*
Suggested system PSU
1100 W*
2:4 tensor sparsity
Yes*
Declared GPU TDP
700 W*
Tensor cores
528*
Theoretical texture fill rate
1,045.4 GTexel/s*
Texture mapping units
528*
GPU cores
16,896*
Video memory
94* GB
Video memory type
HBM3*

* Provisional data