GPU

NVIDIA RTX A6000

Edit@true years ago

集成高速内存(显存)
容量

48 GB

(GDDR6 384-bit)

带宽

768 GB/s

109 Token/s

向量计算
FP64
604.80 G
FP32
38.71 T
FP16
38.71 T
BF16
38.71 T
INT32
19.36 T
INT8
X

NVIDIA RTX A6000 General-Purpose Float-Point performance (Vector Performance / Scalar Performance)

FP64: 604.80 GFLOPS

FP32: 38.71 TFLOPS

FP16: 38.71 TFLOPS

BF16: 38.71 TFLOPS

INT32: 19.36 TOPS

矩阵计算
FP64
X
FP32
X
FP16
154.83 T
309.66 T
FP8
X
TF32
77.41 T
154.83 T
BF16
154.83 T
309.66 T
INT16
X
INT8
309.66 T
619.32 T
INT4
619.32 T
1238.63 T

NVIDIA RTX A6000 AI performance (Tensor Performance / Matrix Performance)

FP16: 154.83 TFLOPS, with sparsity: 309.66 TFLOPS

TF32: 77.41 TFLOPS, with sparsity: 154.83 TFLOPS

BF16: 154.83 TFLOPS, with sparsity: 309.66 TFLOPS

INT8: 309.66 TOPS, with sparsity: 619.32 TOPS

INT4: 619.32 TOPS, with sparsity: 1238.63 TOPS

一般参数
NVIDIA RTX A6000 is a 8nm chip, has 28300 million transistors, launched by NVIDIA at 2020. It has 48 GB built-in(On-Board/On-Chip) memory with bandwidth up to 768 GB/s. It has 10752 general-purpose ALUs(CUDA cores/Shader cores) and 336 matrix cores(Tensor cores) .
制程工艺
8 nm
发布年份
2020
向量(CUDA)核心数
10752
矩阵(Tensor)核心数
336
核心频率
1410 ~ 1800 MHz
缓存
6MB