NVIDIA GA10B 28SM vs NVIDIA GA10B 8SM Full Specs

Summary

1,792 Shaders 930MHz	512 Shaders 625MHz
Shared Memory204.8GB/sLPDDR5	Shared Memory34.1GB/sLPDDR5
Peak·FP32·FP16 3.33 TFLOPS	Peak·FP32·FP16 640 GFLOPS
Form Factor iGPU	Form Factor iGPU
TDP Shared	TDP Shared

Highlights

Peak OPS

Peak OPS FP32 FP16 Tensor FP16-16 FP16-16 Tensor Sparse Tensor FP16-32 FP16-32 Tensor Sparse Tensor BF16 BF16 Tensor Sparse Tensor TF32 Tensor INT8 Tensor INT4 Memory Bandwidth

GA10B 28SM213.3 TOPSINT4 Tensor Sparse

x5.21

GA10B 8SM40.96 TOPSINT4 Tensor Sparse

x1

Show Top Score

Theoretical Performance

Clock Speed930MHz UC·Boost·OC	Clock Speed625MHz UC·Boost·OC
Peak OPS 213.3 TOPSINT4 Tensor Sparse	Peak OPS 40.96 TOPSINT4 Tensor Sparse
Tensor FP16-16 26.66 TFLOPS FP16-16 Tensor Sparse 53.33 TFLOPS Tensor FP16-32 26.66 TFLOPS FP16-32 Tensor Sparse 53.33 TFLOPS	Tensor FP16-16 5.12 TFLOPS FP16-16 Tensor Sparse 10.24 TFLOPS Tensor FP16-32 5.12 TFLOPS FP16-32 Tensor Sparse 10.24 TFLOPS
Tensor BF16 26.66 TFLOPS BF16 Tensor Sparse 53.33 TFLOPS	Tensor BF16 5.12 TFLOPS BF16 Tensor Sparse 10.24 TFLOPS
Tensor TF32 13.33 TFLOPS	Tensor TF32 2.56 TFLOPS
FP32 3.33 TFLOPS	FP32 640 GFLOPS
Tensor INT4 106.7 TOPS	Tensor INT4 20.48 TOPS
Tensor INT8 53.33 TOPS	Tensor INT8 10.24 TOPS
Pixel Rate 22.3 GPixel/s	Pixel Rate 5 GPixel/s
Texture Rate 13 GTexel/s	Texture Rate 2.5 GTexel/s

Core Config

Shaders 1,792 Shaders	Shaders 512 Shaders
TMUs 14 TMUs	TMUs 4 TMUs
ROPs 24 ROPs	ROPs 8 ROPs
Tensor Cores 56 T-Cores	Tensor Cores 16 T-Cores
SMs 14 SMs	SMs 4 SMs

Clocks

Boost Clock 930MHz

Boost Clock 625MHz

Cache

L2 Cache 4.1MB shared

L2 Cache 1MB shared

Memory

Shared MemoryLPDDR5	Shared MemoryLPDDR5
Memory Bus 256-bit	Memory Bus 64-bit
Memory Speed 6.4GT/s	Memory Speed 4.3GT/s
Memory Bandwidth 204.8GB/s	Memory Bandwidth 34.1GB/s

Encoder

Encoder Model NVENC 7

Encoder Model NVENC 7

Decoder

Decoder Model NVDEC 5

Decoder Model NVDEC 5

Card

Form Factor iGPU

Form Factor iGPU

Chip

Manufacturer NVIDIA	Manufacturer NVIDIA
Chip Designer NVIDIA	Chip Designer NVIDIA
Architecture Ampere	Architecture Ampere
Family Jetson	Family Jetson
Branding	Branding
Codename Hercules	Codename Hercules
Market Segment Embedded	Market Segment Embedded
Release Date Oct 1, 2022	Release Date Jan 1, 2023

Fabrication

Foundry Samsung	Foundry Samsung
Fabrication Node 8N	Fabrication Node 8N
Transistor Count 17 Billion	Transistor Count 17 Billion

Images

No images available

No images available

Used In

CPUs AGX Orin 32GB

CPUs Orin Nano 4GB

Competitors

NVIDIAGA10B 28SM vs GPU QualcommAdreno 840 NVIDIAGA10B 28SM vs GPU QualcommAdreno 830 NVIDIAGA10B 28SM vs GPU ARMImmortalis-G925 MP12 1300MHz Show 12 more

NVIDIAGA10B 8SM vs GPU QualcommAdreno 722 NVIDIAGA10B 8SM vs GPU Apple5-Core A17 Pro NVIDIAGA10B 8SM vs GPU QualcommAdreno 810 Show 9 more