NVIDIA H200 SXM 141 GB vs NVIDIA Tesla D870

Comparison of NVIDIA H200 SXM 141 GB with 141 GB HBM3e and 16,896 cores vs NVIDIA Tesla D870 with 1,536 GB GDDR3 and 128 cores.

Loading...

Performance Rating

A100 A100
H200 H200
MI325X MI325X

NVIDIA H200 SXM 141 GB

67.4

NVIDIA H200 SXM 141 GB

67.4
RX 7900 XTX RX 7900 XTX
MI250 MI250
Instinct MI300X Instinct MI300X

NVIDIA Tesla D870

NVIDIA Tesla D870

Contents:

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

Memory

Memory Size

🔥 141 ГБ
1,536 MB ×2 (3072 ГБ)

Memory Type

HBM3e GDDR3

Memory Bandwidth

🔥 4.89 TB/s
76.80 GB/s ×2 (153.6 GB/s)

Memory Bus Width

6,144 бит 384 бит ×2 (768 бит)

ML Performance

FP16 (Half Precision)

🔥 267.6 TFLOPS
No

BF16 (Brain Float)

No No

TF32 (TensorFloat)

No No

Compute Power

FP32 (Single Precision)

🔥 +19,261% 66.91 TFLOPS
0.3456 TFLOPS

FP64 (Double Precision)

🔥 33.45 TFLOPS
No

CUDA Cores

🔥 +13,100% 16,896
128 ×2 (256)

RT Cores

No No

Architecture & Compatibility

GPU Architecture

Hopper Tesla

SM (Streaming Multiprocessor)

🔥 +725% 132
16

PCIe Version

PCIe 5.0 x16 PCIe 1.0 x16

ML Software Support

CUDA Version

🔥 9.0
1.0

Clocks & Performance

Base Clock

🔥 1,500
No

Boost Clock

🔥 1,980
No

Memory Clock

🔥 +99% 1,593
800

Power Consumption

TDP/TGP

700 W
🔥 -26% 520 W

Recommended PSU

1100 W
🔥 -18% 900 W

Power Connector

8-pin EPS No

Rendering

Texture Units (TMU)

🔥 +1,550% 528
32 ×2 (64)

ROP

No No

L2 Cache

🔥 50 MB
96 KB

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
Dual-slot

Release Date

Nov. 18, 2024 May 2, 2007

Display Outputs

No outputs
No outputs

Renting is cheaper than buying