NVIDIA H100 SXM5 96 GB vs SPARKLE Arc A310 OmniView

Comparison of NVIDIA H100 SXM5 96 GB with 96 GB HBM3 and 16,896 cores vs SPARKLE Arc A310 OmniView with 4 GB GDDR6 and 768 cores.

Loading...

Performance Rating

NVIDIA H100 SXM5 96 GB outperforms SPARKLE Arc A310 OmniView by 2,533.64% in the overall GPU ARK performance rating

A100 A100
H200 H200
MI325X MI325X

NVIDIA H100 SXM5 96 GB

57.9

NVIDIA H100 SXM5 96 GB

57.9
RX 7900 XTX RX 7900 XTX
MI250 MI250
Instinct MI300X Instinct MI300X

SPARKLE Arc A310 OmniView

2.2

SPARKLE Arc A310 OmniView

2.2

Contents:

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

Memory

Memory Size

🔥 +2,300% 96 ГБ
4 ГБ

Memory Type

HBM3 GDDR6

Memory Bandwidth

🔥 3.36 TB/s
124.0 GB/s

Memory Bus Width

5,120 бит 64 бит

ML Performance

FP16 (Half Precision)

🔥 +4,255% 267.6 TFLOPS
6.144 TFLOPS

BF16 (Brain Float)

No No

TF32 (TensorFloat)

No No

Compute Power

FP32 (Single Precision)

🔥 +2,078% 66.91 TFLOPS
3.072 TFLOPS

FP64 (Double Precision)

🔥 +4,255% 33.45 TFLOPS
0.768 TFLOPS

CUDA Cores

🔥 +2,100% 16,896
768

RT Cores

No
🔥 6

Architecture & Compatibility

GPU Architecture

Hopper Generation 12.7

SM (Streaming Multiprocessor)

🔥 132
No

PCIe Version

PCIe 5.0 x16 PCIe 4.0 x8

ML Software Support

CUDA Version

9.0 No

Clocks & Performance

Base Clock

1,350
🔥 +48% 2,000

Boost Clock

1,980
🔥 +1% 2,000

Memory Clock

1,313
🔥 +48% 1,937

Power Consumption

TDP/TGP

700 W
🔥 -89% 75 W

Recommended PSU

1100 W
🔥 -77% 250 W

Power Connector

8-pin EPS None

Rendering

Texture Units (TMU)

🔥 +1,550% 528
32

ROP

No
🔥 6

L2 Cache

🔥 +1,150% 50 MB
4 MB

Benchmarks

llama.cpp, llama 7B Q4_0

248.8 tokens/s

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

280.7 tokens/s

Additional

Slots

🔥 SXM Module
Single-slot

Release Date

March 21, 2023 Nov. 18, 2024

Display Outputs

No outputs
4x mini-DisplayPort 2.0
4x HDMI 2.1

Renting is cheaper than buying