GIGABYTE GTX 660 WindForce 2X OC vs NVIDIA H200 SXM 141 GB

Comparison of GIGABYTE GTX 660 WindForce 2X OC with 2 GB GDDR5 and 960 cores vs NVIDIA H200 SXM 141 GB with 141 GB HBM3e and 16,896 cores.

Loading...

Performance Rating

H200 H200
MI325X MI325X
A100 A100

GIGABYTE GTX 660 WindForce 2X OC

GIGABYTE GTX 660 WindForce 2X OC

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

NVIDIA H200 SXM 141 GB

NVIDIA H200 SXM 141 GB

Contents:

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

Memory

Memory Size

2 GB 141 GB

Memory Type

GDDR5 HBM3e

Memory Bandwidth

144.2 GB/s 4.89 TB/s

Memory Bus Width

192 бит 6,144 бит

ML Performance

FP16 (Half Precision)

No 267.6 TFLOPS

BF16 (Brain Float)

No No

TF32 (TensorFloat)

No No

Compute Power

FP32 (Single Precision)

2.108 TFLOPS 66.91 TFLOPS

FP64 (Double Precision)

0.0878 TFLOPS 33.45 TFLOPS

CUDA Cores

960 16,896

RT Cores

No No

Architecture & Compatibility

GPU Architecture

Kepler Hopper

SM (Streaming Multiprocessor)

No 132

PCIe Version

PCIe 3.0 x16 PCIe 5.0 x16

ML Software Support

CUDA Version

3.0
🔥 9.0

Clocks & Performance

Base Clock

1,032 1,500

Boost Clock

1,098 1,980

Memory Clock

1,502 1,593

Power Consumption

TDP/TGP

🔥 -80% 140 W
700 W

Recommended PSU

🔥 -73% 300 W
1100 W

Power Connector

1x 6-pin 8-pin EPS

Rendering

Texture Units (TMU)

80 528

ROP

No No

L2 Cache

384 KB 50 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

Dual-slot SXM Module

Release Date

Sept. 6, 2012 Nov. 18, 2024

Display Outputs

2x DVI
1x HDMI 1.4a
1x DisplayPort 1.2
No outputs

Renting is cheaper than buying