GPU

NVIDIA A100 PCIe

Edit@1 months ago

Intergrated Memory(VRAM)
Capacity

80 GB

(HBM2e 5120-bit)

Bandwidth

1935 GB/s

276 Token/s

Vector Compute
FP64
9.75 T
FP32
19.49 T
FP16
77.97 T
BF16
38.98 T
INT32
19.49 T
INT8
X

NVIDIA A100 PCIe General-Purpose Float-Point performance (Vector Performance / Scalar Performance)

FP64: 9.75 TFLOPS

FP32: 19.49 TFLOPS

FP16: 77.97 TFLOPS

BF16: 38.98 TFLOPS

INT32: 19.49 TOPS

Matirx Compute
FP64
19.49 T
38.98 T
FP32
X
FP16
311.87 T
623.74 T
FP8
X
TF32
155.93 T
311.87 T
BF16
311.87 T
623.74 T
INT16
X
INT8
623.74 T
1247.48 T
INT4
1247.48 T
2494.96 T

NVIDIA A100 PCIe AI performance (Tensor Performance / Matrix Performance)

FP64: 19.49 TFLOPS, with sparsity: 38.98 TFLOPS

FP16: 311.87 TFLOPS, with sparsity: 623.74 TFLOPS

TF32: 155.93 TFLOPS, with sparsity: 311.87 TFLOPS

BF16: 311.87 TFLOPS, with sparsity: 623.74 TFLOPS

INT8: 623.74 TOPS, with sparsity: 1247.48 TOPS

INT4: 1247.48 TOPS, with sparsity: 2494.96 TOPS

Hardware Specs
NVIDIA A100 PCIe is a 7nm chip, has 54200 million transistors, launched by NVIDIA at 2021. It has 80 GB built-in(On-Board/On-Chip) memory with bandwidth up to 1935 GB/s. It has 6912 general-purpose ALUs(CUDA cores/Shader cores) and 432 matrix cores(Tensor cores) .
Process Node
7 nm
Launch Year
2021
Vector(CUDA) Cores
6912
Matrix(Tensor) Cores
432
Core Frequency
1065 ~ 1410 MHz
Cache
80MB

Comment without registration

Share your experience with NVIDIA A100 PCIe / Found an Error? Help Us Improve!