Unverified50% confidenceFactExact time
The DGX Spark's Grace Blackwell architecture uses a unified memory pool shared between CPU and GPU, so model weights do not need to be copied between system memory and VRAM, but two large models cannot reside in memory simultaneously.
1
Sources
50%
Confidence
Long-term
Relevance
9/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDGX Spark基于NVIDIA的Grace Blackwell架构，采用Grace CPU与Blackwell GPU的紧密耦合设计78% similarVerifiedThe CPU and GPU of DGX Spark share up to 128 GB of LPDDR5X memory, achieving memory-coherent access via NVLink-C2C high-speed interconnect.71% similarUnverifiedThe DGX Spark is equipped with the GB10 Grace Blackwell Superchip, 128 GB of unified memory (LPDDR5x), and officially delivers up to 1 PetaFLOP (FP4 precision) of AI compute.67% similarVerifiedllama.cpp supports hybrid CPU and GPU inference (layer offloading), allowing partial model layers to be loaded into GPU VRAM while the remaining layers are kept in memory67% similarUnverifiedGB10 Grace Blackwell超级芯片将ARM架构的Grace CPU与Blackwell GPU通过NVLink-C2C片间互联技术融合在同一封装中66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/837464API
curl https://kongchang.com/api/v1/knowledge/claims/837464MCP
get_claim(id=837464)