GPU

AMD Instinct MI300X GPU

Edit@6 days ago

Intergrated Memory(VRAM)
Capacity

192 GB

(HBM3 8192-bit)

Bandwidth

5300 GB/s

757 Token/s

Vector Compute
FP64
81.70 T
FP32
163.40 T
FP16
X
BF16
X
INT32
NI
INT8
X

AMD Instinct MI300X GPU General-Purpose Float-Point performance (Vector Performance / Scalar Performance)

FP64: 81.70 TFLOPS

FP32: 163.40 TFLOPS

Matirx Compute
FP64
163.40 T
326.80 T
FP32
163.40 T
326.80 T
FP16
1307.40 T
2614.80 T
FP8
2614.90 T
5229.80 T
TF32
653.70 T
1307.40 T
BF16
1307.40 T
2614.80 T
INT16
X
INT8
2614.90 T
5229.80 T
INT4
X

AMD Instinct MI300X GPU AI performance (Tensor Performance / Matrix Performance)

FP64: 163.40 TFLOPS, with sparsity: 326.80 TFLOPS

FP32: 163.40 TFLOPS, with sparsity: 326.80 TFLOPS

FP16: 1307.40 TFLOPS, with sparsity: 2614.80 TFLOPS

FP8: 2614.90 TFLOPS, with sparsity: 5229.80 TFLOPS

TF32: 653.70 TFLOPS, with sparsity: 1307.40 TFLOPS

BF16: 1307.40 TFLOPS, with sparsity: 2614.80 TFLOPS

INT8: 2614.90 TOPS, with sparsity: 5229.80 TOPS

Hardware Specs
AMD Instinct MI300X GPU is a 6nm chip, launched by AMD at 2023. It has 192 GB built-in(On-Board/On-Chip) memory with bandwidth up to 5300 GB/s. It has 19456 general-purpose ALUs(CUDA cores/Shader cores) and 1216 matrix cores(Tensor cores) .
Process Node
6 nm
Launch Year
2023
Vector(CUDA) Cores
19456
Matrix(Tensor) Cores
1216
Core Frequency
1000 ~ 2100 MHz
Cache
256MB

Comment without registration

Share your experience with AMD Instinct MI300X GPU / Found an Error? Help Us Improve!