Specifications
Model
EdgeXpert MS-C931
SKU
9S6-C9311-60S (4TB configuration); MS-C931-32S variant
Platform
NVIDIA DGX Spark platform
Superchip
NVIDIA GB10 Grace Blackwell Superchip
AI performance
1 PFLOP (petaFLOP) of desktop AI compute
CPU architecture
20-core Arm platform (Grace CPU) integrated on the GB10 superchip
Memory
128 GB LPDDR5X unified memory
Model capacity
Runs cutting-edge models up to 200 billion parameters
Memory model
Unified CPU/GPU memory pool — no PCIe host-to-device copy for model weights
Storage
4 TB NVMe M.2 with self-encryption
Display outputs
3x DisplayPort over USB-C plus 1x HDMI
Networking
10 GbE RJ-45 plus Wi-Fi and Bluetooth
Wireless
Wi-Fi 7 and Bluetooth 5.3 class connectivity
Software stack
Pre-installed NVIDIA AI software stack and DGX OS environment
Form factor
Compact deskside / desktop AI supercomputer
Use cases
Local LLM development, fine-tuning, inference and AI research
Compute class
Same superchip family as the NVIDIA DGX Spark reference system
Power and cooling
Deskside power envelope — no datacenter infrastructure required
Deployment
Runs standalone at a desk, lab bench or edge site
Multi-unit
Can be paired with a second unit for larger model memory pools
RFQ lead time
15-day sample lead time; volume pricing on request
Overview
The MSI EdgeXpert MS-C931 is a compact desktop AI supercomputer built on the NVIDIA DGX Spark platform. Its core is the NVIDIA GB10 Grace Blackwell Superchip, which combines an Arm Grace CPU complex with Blackwell-generation GPU compute to deliver one petaFLOP of AI performance in a deskside form factor — the same silicon family driving NVIDIA's DGX Spark reference system.
The defining architectural feature is memory. The MS-C931 carries 128 GB of LPDDR5X unified memory shared between CPU and GPU, so model weights do not have to be shuttled across a PCIe bus. That, combined with the 1 PFLOP compute envelope, allows models up to roughly 200 billion parameters to be developed, fine-tuned and served locally — workloads that previously required datacenter GPUs or cloud instances.
Storage is a 4 TB self-encrypting NVMe drive, and the unit ships with a pre-installed NVIDIA AI software stack so development can start without environment assembly. Connectivity covers three DisplayPort outputs over USB-C plus HDMI, 10 GbE wired networking and Wi-Fi 7 class wireless. Because it runs standalone on ordinary mains power, the MS-C931 suits research labs, engineering benches and edge sites that need serious local AI capacity without rack infrastructure.
Key Benefits
1 PFLOP on a desk: datacenter-class AI throughput without racks, PDUs or datacenter cooling. 128 GB unified memory: run and fine-tune models up to ~200B parameters locally. DGX Spark platform: the same NVIDIA GB10 Grace Blackwell superchip as NVIDIA's reference system, with a pre-installed AI software stack. No PCIe weight shuffling: unified CPU/GPU memory removes host-to-device copy overhead for large models. Self-encrypting 4TB NVMe: at-rest protection for proprietary models and datasets. 10GbE plus Wi-Fi 7: move checkpoints and datasets at usable speed.
Applications
Local LLM and generative AI development, fine-tuning and inference; AI research and academic labs; model evaluation and red-teaming workstations; private on-premises AI where data cannot leave the building; physical AI and robotics policy research; edge AI model preparation and quantisation workflows; engineering benches needing GPU-class local compute.
Request a Quote — MSI EDGEXPERT MS-C931 — 1 PFLOP DESKTOP AI SUPERCOMPUTER ON NVIDIA DGX SPARK PLATFORM
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →