GPU

NVIDIA GeForce RTX 3090

Edit@5 days ago

Intergrated Memory
Capacity

24 GB

(GDDR6X 384-bit)

Bandwidth

936 GB/s

133 Token/s

Vector Compute
FP64
556 G
FP32
35.58 T
FP16
35.58 T
BF16
35.58 T
INT32
17.79 T
INT8
X

NVIDIA GeForce RTX 3090 General-Purpose Float-Point performance (Vector Performance / Scalar Performance)

FP64: 556 GFLOPS

FP32: 35.58 TFLOPS

FP16: 35.58 TFLOPS

BF16: 35.58 TFLOPS

INT32: 17.79 TOPS

Matirx Compute
FP64
X
FP32
X
FP16
71.16 T
142.33 T
FP8
X
TF32
35.58 T
71.16 T
BF16
71.16 T
142.33 T
INT16
X
INT8
284.65 T
569.30 T
INT4
569.30 T
1138.61 T

NVIDIA GeForce RTX 3090 AI performance (Tensor Performance / Matrix Performance)

FP16: 71.16 TFLOPS, with sparsity: 142.33 TFLOPS

TF32: 35.58 TFLOPS, with sparsity: 71.16 TFLOPS

BF16: 71.16 TFLOPS, with sparsity: 142.33 TFLOPS

INT8: 284.65 TOPS, with sparsity: 569.30 TOPS

INT4: 569.30 TOPS, with sparsity: 1138.61 TOPS

Hardware Specs
NVIDIA GeForce RTX 3090 is a 8nm chip, has 28300 million transistors, launched by NVIDIA at 2020. It has 24 GB built-in(On-Board/On-Chip) memory with bandwidth up to 936 GB/s. It has 10496 general-purpose ALUs(CUDA cores/Shader cores) and 328 matrix cores(Tensor cores) .
Process Node
8 nm
Launch Year
2020
Vector(CUDA) Cores
10496
Matrix(Tensor) Cores
328
Core Frequency
1395 ~ 1695 MHz
Cache
6MB

Comment without registration

Share your experience with NVIDIA GeForce RTX 3090 / Found an Error? Help Us Improve!