NVIDIA DRIVE AGX Thor Blackwell Full Specs

Summary

2,560 Shaders 1.57GHz

Shared Memory273GB/sLPDDR5X

Peak·FP32·FP16 8.06 TFLOPS

Form Factor iGPU

TDP Shared

Highlights

Peak OPS

Peak OPS FP32 FP16 Tensor FP6 FP6 Tensor Sparse Tensor NVFP4 NVFP4 Tensor Sparse Tensor FP4 FP4 Tensor Sparse Tensor FP8-32 FP8-32 Tensor Sparse Tensor FP16-32 FP16-32 Tensor Sparse FP64 BF16 Tensor BF16 BF16 Tensor Sparse Tensor TF32 Tensor INT8 Memory Bandwidth

DRIVE AGX Thor Blackwell2.06 PFLOPSNVFP4 Tensor Sparse

x1

Top Score3.35 PFLOPS FP4 Tensor Sparse

x1.62

Desktop1xGeForce RTX 5090

Theoretical Performance

Clock Speed1575MHz UC·Boost·OC

Peak OPS 2.06 PFLOPSNVFP4 Tensor Sparse

Tensor FP4 1.03 PFLOPS FP4 Tensor Sparse 2.06 PFLOPS NVFP4 Tensor 1.03 PFLOPS NVFP4 Tensor Sparse 2.06 PFLOPS

FP6 Tensor 516.1 TFLOPS FP6 Tensor Sparse 1.03 PFLOPS

Tensor FP8-32 516.1 TFLOPS FP8-32 Tensor Sparse 1.03 PFLOPS

FP16 16.13 TFLOPS Tensor FP16-32 258 TFLOPS FP16-32 Tensor Sparse 516.1 TFLOPS

BF16 16.13 TFLOPS Tensor BF16 258 TFLOPS BF16 Tensor Sparse 516.1 TFLOPS

Tensor TF32 129 TFLOPS

FP32 8.06 TFLOPS

FP64 126 GFLOPS

Tensor INT8 516.1 TOPS

Texture Rate 126 GTexel/s

Core Config

Shaders 2,560 Shaders

TMUs 80 TMUs

Tensor Cores 80 T-Cores

SMs 20 SMs

Clocks

Boost Clock 1.57GHz

Cache

L1 192KB/SM Tex

L2 Cache 32MB shared

Memory

Shared MemoryLPDDR5X

Memory Bus 256-bit

Memory Speed 8.5GT/s

Memory Bandwidth 273GB/s

API Support

OpenGL 4.6 Vulkan 1.3

CUDA 11

Encoder

Encoder 2x NVENC 9.1

Codec AVC (H.264) HEVC (H.265)

Decoder

Decoder 2x NVDEC 6.1

Codec MPEG-1 MPEG-2 MPEG-4 VC-1 VP8 VP9 AVC (H.264) HEVC (H.265) AV1

Card

Form Factor iGPU

Chip

Manufacturer NVIDIA

Chip Designer NVIDIA

Architecture Blackwell

Family Blackwell

Branding

Codename GB10B

Market Segment Automotive

Release Date Aug 25, 2025

Fabrication

Foundry TSMC

Fabrication Node 4N

Die Size 402mm²

Images