MSI XpertStation WS300

● HIGHLY RATED
Site Score
4.4 / 5.0
Buyer Guidance: An uncompromising deskside supercomputer for enterprise research teams requiring local trillion-parameter AI execution without immediate data center infrastructure investments.
Mizex Audit Breakdown
Engineering & Core Performance (30%) 4.8 / 5.0
Build Quality & Physical Design (20%) 4.4 / 5.0
Thermal, Power & Acoustics (20%) 4.2 / 5.0
Reliability & Stability (20%) 4.4 / 5.0
Buyer Value (10%) 3.7 / 5.0
Score Rationale
The composite score of 4.3 reflects pioneering engineering and architecture (4.8) integrating the GB300 Grace Blackwell Ultra Superchip, 748GB unified coherent memory, and 800 Gbps ConnectX-8 optical fabric. Build quality (4.4) is characterized by an expansive, reinforced steel chassis and precision-mounted liquid cooling components. Thermal and acoustic management (4.2) is backed by dual 360mm radiators and seven total fans rated to dissipate 1,400W with a 100W thermal safety buffer over maximum compute draw, though local office heat rejection remains substantial. Reliability (4.4) benefits from an enterprise 1,600W Titanium power supply and direct cold-plate coverage across all critical silicon, while buyer value (3.7) balances an imposing $85,000 to $100,000 price point against the capability to run massive frontier models without dedicated datacenter floor space.
Audited yesterday
ⓘ Multi-Dimensional Index evaluated across 5 core engineering pillars. Refresh requests open every 60 days.
👥 Community Score
Community Score
— / 5.0
No community ratings submitted yet. Be the first to record your experience.
Your Community Audit
— / 5.0
Rate the 5 Sub-Index categories matching the site audit for a fair direct comparison:
Engineering & Core Performance (30%) —
Build Quality & Physical Design (20%) —
Thermal, Power & Acoustics (20%) —
Reliability & Stability (20%) —
Buyer Value (10%) —
or to submit.
💬 Join the Discussion ↓
Scores adjust dynamically as verified community ratings accumulate.

Overview

The MSI XpertStation WS300 brings datacenter-class artificial intelligence training and inference into a standalone deskside form factor, built around the NVIDIA GB300 Grace Blackwell Ultra Superchip. Architecturally, the workstation bridges a 72-core Arm Neoverse V2 Grace processor with a Blackwell Ultra accelerator across a high-speed NVLink-C2C bus operating at 900 GB/s bidirectional bandwidth. This tightly coupled computing fabric is fed by a massive 748GB coherent memory pool consisting of 252GB of HBM3e delivering 7,100 GB/s of bandwidth alongside 496GB of high-density LPDDR5X across four modular SOCAMM sockets, accompanied by an integrated NVIDIA ConnectX-8 SuperNIC offering dual 400GbE optical interfaces for 800 Gbps clustering.

To manage the immense thermal envelope of the 1,300W Superchip and high-speed networking subsystems, MSI implements an industrial-grade closed-loop liquid cooling assembly rated for 1,400W of total thermal dissipation. Dedicated copper cold plates make direct contact across the CPU, GPU, LPDDR5X SOCAMM assemblies, and ConnectX-8 controller, circulating coolant through dual 360mm radiator arrays driven by six 120mm radiator fans and an additional 120mm exhaust fan. Power delivery is anchored by a high-efficiency 1,600W Titanium-rated power supply, housed inside an expansive 247.8 x 567.7 x 527.9 mm chassis engineered to provide structural stability and controlled internal airflow under continuous enterprise compute workloads.

In practical enterprise deployments running Ubuntu and NVIDIA AI Developer Tools, the workstation eliminates data center rack infrastructure requirements, allowing research teams to host trillion-parameter models locally. While continuous sustained loads will exhaust noticeable ambient heat into office spaces and command significant floor area, the robust industrial build, integrated optical fabric, and 100W thermal overhead headroom establish a stable deskside platform for cutting-edge generative AI development.

Technical Specifications

Superchip NVIDIA GB300 Grace Blackwell Ultra
Cpu NVIDIA Grace 72-core Arm Neoverse V2
Gpu NVIDIA Blackwell Ultra
Ai Compute Fp4 20 PetaFLOPS
Unified Memory Capacity 748GB
Hbm3e Memory 252GB (7100 GB/s bandwidth)
Lpddr5x Memory 496GB across four SOCAMM modules
Interconnect NVLink-C2C at 900 GB/s bidirectional
Networking NVIDIA ConnectX-8 SuperNIC with 2x 400GbE optical ports (800 Gbps fabric)
Cooling System Liquid loop with cold plates on CPU, GPU, LPDDR5X, and ConnectX-8, dual 360mm radiators, 6x 120mm radiator fans, 1x 120mm chassis fan (1,400W rated dissipation)
Power Supply 1600W Titanium-rated PSU
Chassis Dimensions Mm 247.8 x 567.7 x 527.9
Rear Io 4x USB 3.1 Type-A, 1x Micro USB COM, 2x 400GbE QSFP
Operating System Ubuntu with NVIDIA AI Developer Tools
✔ Pros
  • Delivers 20 PetaFLOPS of FP4 AI compute in a deskside form factor without datacenter rack infrastructure
  • Massive 748GB coherent memory pool allows trillion-parameter models to run entirely resident in local memory
  • Custom liquid cooling assembly rated for 1,400W provides a 100W buffer over the 1,300W Superchip limit to curb thermal throttling
  • Integrated NVIDIA ConnectX-8 SuperNIC provides 800 Gbps optical networking for multi-node clustering
✖ Cons
  • Substantial enterprise investment ranging between $85,000 and $100,000
  • High continuous power draw requiring up to 1,300W on the Superchip and a 1,600W dedicated supply
  • Large, heavy chassis footprint measuring nearly 57 cm in depth and 53 cm in height
📋

Primary Sources & Audited Citations

Where To Buy

Newegg
Check Retailer

Disclosure: Mizex is an independent hardware database & audit index. When you purchase through retailer links on our site, we may earn an affiliate commission at no additional cost to you.

100% Sponsor-Free Platform
Zero paid placement • Zero brand bias

Support Independent Hardware Research

Mizex is an independent hardware intelligence platform. We accept zero manufacturer sponsorships, paid product reviews, or promotional placements. Every hardware profile, specification breakdown, and reliability rating is researched objectively from public documentation and verified community telemetry. If our database helped guide your build or purchase, consider supporting our research.

✓ Community-Driven ✓ Open Research ✓ 100% Objective Data
Request a Profile →

Community Discussion