NVIDIA GeForce RTX 4090 vs NVIDIA H100 SXM5 96 GB

Comparison of NVIDIA GeForce RTX 4090 with 24 GB GDDR6X and 16,384 cores vs NVIDIA H100 SXM5 96 GB with 96 GB HBM3 and 16,896 cores.

Loading...

Performance Rating

NVIDIA H100 SXM5 96 GB outperforms NVIDIA GeForce RTX 4090 by 80.72% in the overall GPU ARK performance rating

A100 A100
H200 H200
MI325X MI325X

NVIDIA GeForce RTX 4090

32.1

NVIDIA GeForce RTX 4090

32.1
RX 7900 XTX RX 7900 XTX
MI250 MI250
Instinct MI300X Instinct MI300X

NVIDIA H100 SXM5 96 GB

57.9

NVIDIA H100 SXM5 96 GB

57.9

Expert Comparison

NVIDIA GeForce RTX 4090 более подходящий выбор для игр и средней интенсивности задач благодаря своей более низкой потребляемой мощности и более высокому GPI (32.1 против 57.9 у H100). NVIDIA H100 SXM5 96 GB лучше для вычислительной мощности и обработки больших объемов данных из-за огромной памяти (96 ГБ) и высокой пропускной способности (3.36 TB/s), что делает его идеальным для научных вычислений и AI-приложений.

Contents:

Memory ML Performance Compute Power Architecture & Compatibility ML Software Support Clocks & Performance Power Consumption Rendering Benchmarks Additional

Memory

Memory Size

24 ГБ
🔥 +300% 96 ГБ

Memory Type

GDDR6X HBM3

Memory Bandwidth

1.01 TB/s
🔥 +233% 3.36 TB/s

Memory Bus Width

384 бит 5,120 бит

ML Performance

FP16 (Half Precision)

82.58 TFLOPS
🔥 +224% 267.6 TFLOPS

BF16 (Brain Float)

No No

TF32 (TensorFloat)

No No

Compute Power

FP32 (Single Precision)

🔥 +23% 82.58 TFLOPS
66.91 TFLOPS

FP64 (Double Precision)

No
🔥 33.45 TFLOPS

CUDA Cores

16,384
🔥 +3% 16,896

RT Cores

🔥 128
No

Architecture & Compatibility

GPU Architecture

Ada Lovelace Hopper

SM (Streaming Multiprocessor)

128
🔥 +3% 132

PCIe Version

PCIe 4.0 x16 PCIe 5.0 x16

ML Software Support

CUDA Version

8.9
🔥 9.0

Clocks & Performance

Base Clock

🔥 +66% 2,235
1,350

Boost Clock

🔥 +27% 2,520
1,980

Memory Clock

1,313 1,313

Power Consumption

TDP/TGP

🔥 -36% 450 W
700 W

Recommended PSU

🔥 -23% 850 W
1100 W

Power Connector

1x 16-pin 8-pin EPS

Rendering

Texture Units (TMU)

512
🔥 +3% 528

ROP

🔥 128
No

L2 Cache

🔥 +44% 72 MB
50 MB

Benchmarks

MLPerf, llama3.1-8b-edge (fp32)

44.7 tokens/s

llama.cpp, llama 7B Q4_0

154.7 tokens/s
🔥 +61% 248.8 tokens/s

llama.cpp, llama-2-7b-Q4_0

189.0 tokens/s
🔥 +49% 280.7 tokens/s

Geekbench AI, FP16

53 496 points

Geekbench AI, INT8

29 155 points

Geekbench AI, FP32

39 033 points

Additional

Slots

Triple-slot
🔥 SXM Module

Release Date

Sept. 20, 2022 March 21, 2023

Display Outputs

1x HDMI 2.1
3x DisplayPort 1.4a
No outputs

Renting is cheaper than buying