AMD Radeon Instinct MI300X vs NVIDIA PG506-232

Comparison of AMD Radeon Instinct MI300X with 192 GB HBM3 and 19,456 cores vs NVIDIA PG506-232 with 24 GB HBM2 and 3,584 cores.

Loading...

Performance Rating

AMD Radeon Instinct MI300X outperforms NVIDIA PG506-232 by 660.1% in the overall GPU ARK performance rating

A100 A100
H200 H200
MI325X MI325X

AMD Radeon Instinct MI300X

94.1

AMD Radeon Instinct MI300X

94.1
RX 7900 XTX RX 7900 XTX
MI250 MI250
Instinct MI300X Instinct MI300X

NVIDIA PG506-232

12.4

NVIDIA PG506-232

12.4

Contents:

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

Memory

Memory Size

🔥 +700% 192 ГБ
24 ГБ

Memory Type

HBM3 HBM2

Memory Bandwidth

🔥 10.3 TB/s
933.1 GB/s

Memory Bus Width

8,192 бит 3,072 бит

ML Performance

FP16 (Half Precision)

🔥 +6,234% 653.7 TFLOPS
10.32 TFLOPS

BF16 (Brain Float)

No No

TF32 (TensorFloat)

No No

Compute Power

FP32 (Single Precision)

🔥 +692% 81.72 TFLOPS
10.32 TFLOPS

FP64 (Double Precision)

🔥 +1,483% 81.72 TFLOPS
5.161 TFLOPS

CUDA Cores

🔥 +443% 19,456
3,584

RT Cores

No No

Architecture & Compatibility

GPU Architecture

CDNA 3.0 Ampere

SM (Streaming Multiprocessor)

No
🔥 56

PCIe Version

PCIe 5.0 x16 PCIe 4.0 x16

ML Software Support

CUDA Version

No 8.0

Clocks & Performance

Base Clock

🔥 +8% 1,000
930

Boost Clock

🔥 +46% 2,100
1,440

Memory Clock

🔥 +108% 2,525
1,215

Power Consumption

TDP/TGP

750 W
🔥 -78% 165 W

Recommended PSU

1150 W
🔥 -61% 450 W

Power Connector

None 8-pin EPS

Rendering

Texture Units (TMU)

🔥 +443% 1,216
224

ROP

No No

L2 Cache

16 MB
🔥 +50% 24 MB

Benchmarks

MLPerf, llama2-70b-99.9 (UNSET)

1 983 tokens/s

MLPerf, llama2-70b-99.9 (fp16)

1 740 tokens/s

MLPerf, llama2-70b-99.9 (fp8)

1 057 tokens/s

MLPerf, llama3.1-405b (UNSET)

30.4 tokens/s

MLPerf, llama3.1-405b (fp16)

34.8 tokens/s

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

232.9 tokens/s

MLPerf, mixtral-8x7b (fp8)

5 975 tokens/s

Additional

Slots

OAM Module Dual-slot

Release Date

Dec. 6, 2023 April 12, 2021

Display Outputs

No outputs
No outputs

Renting is cheaper than buying