NVIDIA H200 SXM 141 GB vs Palit RTX 3080 Ti GameRock OC

Comparison of NVIDIA H200 SXM 141 GB with 141 GB HBM3e and 16,896 cores vs Palit RTX 3080 Ti GameRock OC with 12 GB GDDR6X and 10,240 cores.

Loading...

Performance Rating

H200 H200
MI325X MI325X
A100 A100

NVIDIA H200 SXM 141 GB

NVIDIA H200 SXM 141 GB

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

Palit RTX 3080 Ti GameRock OC

Palit RTX 3080 Ti GameRock OC

Contents:

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

Memory

Memory Size

141 GB 12 GB

Memory Type

HBM3e GDDR6X

Memory Bandwidth

4.89 TB/s 912.4 GB/s

Memory Bus Width

6,144 бит 384 бит

ML Performance

FP16 (Half Precision)

267.6 TFLOPS 35.33 TFLOPS

BF16 (Brain Float)

No No

TF32 (TensorFloat)

No No

Compute Power

FP32 (Single Precision)

66.91 TFLOPS 35.33 TFLOPS

FP64 (Double Precision)

33.45 TFLOPS 0.552 TFLOPS

CUDA Cores

16,896 10,240

RT Cores

No 80

Architecture & Compatibility

GPU Architecture

Hopper Ampere

SM (Streaming Multiprocessor)

132 80

PCIe Version

PCIe 5.0 x16 PCIe 4.0 x16

ML Software Support

CUDA Version

🔥 9.0
8.6

Clocks & Performance

Base Clock

1,500 1,365

Boost Clock

1,980 1,725

Memory Clock

1,593 1,188

Power Consumption

TDP/TGP

700 W
🔥 -43% 400 W

Recommended PSU

1100 W
🔥 -27% 800 W

Power Connector

8-pin EPS 3x 8-pin

Rendering

Texture Units (TMU)

528 320

ROP

No 80

L2 Cache

50 MB 6 MB

Benchmarks

MLPerf, llama2-70b-99.9 (UNSET)

3 534 tokens/s

MLPerf, llama2-70b-99.9 (fp16)

3 553 tokens/s

MLPerf, llama2-70b-99.9 (fp8)

2 444 tokens/s

MLPerf, llama3.1-405b (fp16)

40.8 tokens/s

MLPerf, llama3.1-405b (fp8)

25.3 tokens/s

MLPerf, llama3.1-8b (fp8)

5 161 tokens/s

MLPerf, deepseek-r1 (fp8)

1 113 tokens/s

MLPerf, mixtral-8x7b (fp8)

7 132 tokens/s

Additional

Slots

SXM Module Triple-slot

Release Date

Nov. 18, 2024 May 31, 2021

Display Outputs

No outputs
1x HDMI 2.1
3x DisplayPort 1.4a

Renting is cheaper than buying