CPU

Fujitsu A64FX

Edit@3 months ago

Intergrated Memory
Capacity

32 GB

(HBM2 4096-bit)

Bandwidth

1024 GB/s

146 Token/s

Socketed Memory
Capacity

NA

Channels x

Bandwidth

GB/s

Vector Compute
FP64
3.38 T
FP32
6.76 T
FP16
13.52 T
BF16
X
INT32
6.76 T
INT8
27.03 T
Theoretical peak performance: 48 Cores @ 2200 MHz with 512-bit SVE and Dual 512-bit FMAs (Support INT8, FP16)

Fujitsu A64FX General-Purpose Float-Point performance (Vector Performance / Scalar Performance)

FP64: 3.38 TFLOPS

FP32: 6.76 TFLOPS

FP16: 13.52 TFLOPS

INT32: 6.76 TOPS

INT8: 27.03 TOPS

Matirx Compute
No dedicated Tensor(Matrix) Units for AI

Fujitsu A64FX AI performance (Tensor Performance / Matrix Performance)

Hardware Specs
Fujitsu A64FX is a 7nm chip, launched by Fujitsu at 2019. It has 32 GB built-in(On-Board/On-Chip) memory with bandwidth up to 1024 GB/s. It has 48 cores, can run up to 48 threads.
Process Node
7 nm
Launch Year
2019
Max SMP
Cache
Unknown
Threads
48
Num of Cores
48
Base Frequency
1800MHz
Boost Frequency
2200MHz
All Core Turbo Freq
Unknown
SIMD Extensions