NVIDIA B300 vs NVIDIA H200 NVL

Comparison of NVIDIA B300 with 144 GB HBM3e and 18,944 cores vs NVIDIA H200 NVL with 141 GB HBM3e and 16,896 cores.

Loading...

Performance Rating

NVIDIA B300 outperforms NVIDIA H200 NVL by 13.03% in the overall GPU ARK performance rating

A100 A100
H200 H200
MI325X MI325X

NVIDIA B300

71.9

NVIDIA B300

71.9
RX 7900 XTX RX 7900 XTX
MI250 MI250
Instinct MI300X Instinct MI300X

NVIDIA H200 NVL

63.6

NVIDIA H200 NVL

63.6

Contents:

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

Memory

Memory Size

🔥 +2% 144 ГБ ×2 (288 ГБ)
141 GB

Memory Type

HBM3e HBM3e

Memory Bandwidth

🔥 4.10 TB/s ×2 (8.2 TB/s)
4.89 TB/s

Memory Bus Width

4,096 бит ×2 (8192 бит) 6,144 бит

ML Performance

FP16 (Half Precision)

1.2318 TFLOPS
🔥 +19,489% 241.3 TFLOPS

BF16 (Brain Float)

No No

TF32 (TensorFloat)

No No

Compute Power

FP32 (Single Precision)

🔥 +28% 76.99 TFLOPS
60.32 TFLOPS

FP64 (Double Precision)

1.2029 TFLOPS
🔥 +2,407% 30.16 TFLOPS

CUDA Cores

🔥 +12% 18,944 ×2 (37888)
16,896

RT Cores

No No

Architecture & Compatibility

GPU Architecture

Blackwell Ultra Hopper

SM (Streaming Multiprocessor)

🔥 +12% 148
132

PCIe Version

PCIe 5.0 x16 PCIe 5.0 x16

ML Software Support

CUDA Version

🔥 10.3
9.0

Clocks & Performance

Base Clock

🔥 +22% 1,665
1,365

Boost Clock

🔥 +14% 2,032
1,785

Memory Clock

🔥 +26% 2,000
1,593

Power Consumption

Recommended PSU

1800 W
🔥 -44% 1000 W

Power Connector

No 8-pin EPS

TDP/TGP

1400 W
🔥 -57% 600 W

Rendering

Texture Units (TMU)

🔥 +12% 592 ×2 (1184)
528

ROP

No No

L2 Cache

50 MB 50 MB

Benchmarks

MLPerf, llama2-70b-99.9 (fp4)

7 436 tokens/s

MLPerf, llama3.1-405b (fp4)

130.5 tokens/s

MLPerf, deepseek-r1 (fp4)

2 327 tokens/s

Additional

Slots

🔥 SXM Module
Dual-slot

Release Date

Sept. 11, 2025 Nov. 18, 2024

Display Outputs

No outputs
No outputs

Renting is cheaper than buying