GPU

NVIDIA GeForce RTX 4070 Ti

Edit@1 months ago

Intergrated Memory(VRAM)
Capacity

12 GB

(GDDR6X 192-bit)

Bandwidth

504 GB/s

72 Token/s

Vector Compute
FP64
626.40 G
FP32
40.09 T
FP16
40.09 T
BF16
40.09 T
INT32
20.05 T
INT8
X

NVIDIA GeForce RTX 4070 Ti General-Purpose Float-Point performance (Vector Performance / Scalar Performance)

FP64: 626.40 GFLOPS

FP32: 40.09 TFLOPS

FP16: 40.09 TFLOPS

BF16: 40.09 TFLOPS

INT32: 20.05 TOPS

Matirx Compute
FP64
X
FP32
X
FP16
80.18 T
160.36 T
FP8
160.36 T
320.72 T
TF32
40.09 T
80.18 T
BF16
80.18 T
160.36 T
INT16
X
INT8
320.72 T
641.43 T
INT4
641.43 T
1282.87 T

NVIDIA GeForce RTX 4070 Ti AI performance (Tensor Performance / Matrix Performance)

FP16: 80.18 TFLOPS, with sparsity: 160.36 TFLOPS

FP8: 160.36 TFLOPS, with sparsity: 320.72 TFLOPS

TF32: 40.09 TFLOPS, with sparsity: 80.18 TFLOPS

BF16: 80.18 TFLOPS, with sparsity: 160.36 TFLOPS

INT8: 320.72 TOPS, with sparsity: 641.43 TOPS

INT4: 641.43 TOPS, with sparsity: 1282.87 TOPS

Hardware Specs
NVIDIA GeForce RTX 4070 Ti is a 5nm chip, has 35800 million transistors, launched by NVIDIA at 2023. It has 12 GB built-in(On-Board/On-Chip) memory with bandwidth up to 504 GB/s. It has 7680 general-purpose ALUs(CUDA cores/Shader cores) and 240 matrix cores(Tensor cores) .
Process Node
5 nm
Launch Year
2023
Vector(CUDA) Cores
7680
Matrix(Tensor) Cores
240
Core Frequency
2310 ~ 2610 MHz
Cache
48MB