GPU

NVIDIA GeForce RTX 5090

Edit@2 months ago

Intergrated Memory(VRAM)
Capacity

32 GB

(GDDR7 512-bit)

Bandwidth

1792 GB/s

256 Token/s

Vector Compute
FP64
1.64 T
FP32
104.80 T
FP16
104.80 T
BF16
104.80 T
INT32
104.80 T
INT8
X

NVIDIA GeForce RTX 5090 General-Purpose Float-Point performance (Vector Performance / Scalar Performance)

FP64: 1.64 TFLOPS

FP32: 104.80 TFLOPS

FP16: 104.80 TFLOPS

BF16: 104.80 TFLOPS

INT32: 104.80 TOPS

Matirx Compute
FP64
X
FP32
X
FP16
209.51 T
419.01 T
FP8
419.01 T
838.02 T
TF32
104.75 T
209.51 T
BF16
209.51 T
419.01 T
INT16
X
INT8
838.02 T
1676.04 T
INT4
X

NVIDIA GeForce RTX 5090 AI performance (Tensor Performance / Matrix Performance)

FP16: 209.51 TFLOPS, with sparsity: 419.01 TFLOPS

FP8: 419.01 TFLOPS, with sparsity: 838.02 TFLOPS

TF32: 104.75 TFLOPS, with sparsity: 209.51 TFLOPS

BF16: 209.51 TFLOPS, with sparsity: 419.01 TFLOPS

INT8: 838.02 TOPS, with sparsity: 1676.04 TOPS

Hardware Specs
NVIDIA GeForce RTX 5090 is a 5nm chip, has 92200 million transistors, launched by NVIDIA at 2025. It has 32 GB built-in(On-Board/On-Chip) memory with bandwidth up to 1792 GB/s. It has 21760 general-purpose ALUs(CUDA cores/Shader cores) and 680 matrix cores(Tensor cores) .
Process Node
5 nm
Launch Year
2025
Vector(CUDA) Cores
21760
Matrix(Tensor) Cores
680
Core Frequency
2017 ~ 2407 MHz
Cache
96MB