NVIDIA-Certified · 10U · air cooled · available to order
An NVIDIA-Certified, air-cooled 10U server with eight Blackwell Ultra GPUs, 3 TB of DDR5 and an 800G port per GPU: our standard building block for an HGX B300 AI factory.
The SGX30-2 is Compal's HGX B300 system, listed by NVIDIA as an NVIDIA-Certified System for the HGX B300 8-GPU platform. We offer it in one complete, balanced configuration so you are not left choosing memory, storage and networking part by part: eight Blackwell Ultra GPUs, two Intel Xeon 6 processors, 3 TB of DDR5-6400 and a full set of NVMe drives.
What makes it a good AI factory node is the networking. Eight ConnectX-8 SuperNICs are built onto the board, each with its own 800G OSFP port at the front, so every GPU gets a dedicated 800 Gb/s path into an InfiniBand XDR or Spectrum-X fabric with no extra cards to fit. A BlueField-3 B3240 DPU looks after storage and service traffic separately.
It is air cooled. If your hall can supply high-density power and airflow but has no liquid-cooling loop, the SGX30-2 brings Blackwell Ultra in without a facilities project. Four of them make the 32-GPU rack described on our HGX B300 page.
Onboard ConnectX-8 SuperNICs remove eight add-in cards and their cabling, and line up neatly with a rail-optimised fabric.
32 × 96 GB DDR5-6400 fills every channel on both CPUs, giving 3 TB of system memory, more than the 2.1 TB of GPU memory, for data loading and offload.
Eight 3.84 TB PCIe Gen5 E1.S NVMe drives (about 30 TB) for datasets, caches and checkpoints, plus mirrored M.2 boot drives.
NVIDIA-Certified for HGX B300, 6+6 redundant 3200 W Titanium power supplies and a 3-year return-to-base warranty arranged through us.
The configuration we supply. Alternatives (CPU model, memory or storage) can be quoted.
| Chassis | Compal SGX30-2, 10U, air cooled, 447 × 800 × 442 mm |
|---|---|
| GPU | NVIDIA HGX B300 NVL8 baseboard with 8 Blackwell Ultra GPUs, 288 GB HBM3e each |
| AI 性能 | FP4 144 | 108 PFLOPS; FP8 72 PFLOPS (sparse), NVIDIA HGX B300 figures |
| CPU | 2 × Intel Xeon 6 6767P, 6776P or 6768P |
| 内存 | 32 × 96 GB DDR5-6400 ECC RDIMM (3 TB) |
| Boot | 2 × 960 GB M.2 NVMe |
| Data | 8 × 3.84 TB NVMe E1.S, PCIe Gen5 |
| GPU fabric | 8 × 800G OSFP via onboard ConnectX-8 (InfiniBand XDR 800G or 2 × 400GbE per NIC) |
| DPU | NVIDIA BlueField-3 B3240 |
| Expansion | 2 × FHFL PCIe Gen5 x16 |
| 功耗 | 3200 W 6+6 redundant, 80 PLUS Titanium |
| Warranty | 3-year return-to-base |
Figures are NVIDIA's published specifications for the platform. Where NVIDIA quotes two Tensor Core figures, the first assumes sparsity and the second is dense. The exact build (CPUs, memory, storage, cooling, networking) is agreed with you and confirmed on a written quote.
What one SGX30-2 needs from your site.
| 功耗 | In the region of 14 kW per 8-GPU Blackwell Ultra system; twelve 3200 W supplies on redundant feeds |
|---|---|
| 冷却 | High-density air with hot-aisle containment |
| Rack space | 10U per server; 800 mm chassis depth, so a deep rack |
| 网络 | Eight 800G OSFP links per server to the GPU fabric, plus BlueField-3 and management connections |
One accountable UK team from sizing to day-two operations
We confirm allocation, lead time and any export-control requirements before a firm quote.
GPU, memory, NVLink and network stress tests before shipping, with results supplied.
Racking, cabling to your fabric, OS, drivers and NVIDIA software, and acceptance tests.
Return-to-base warranty handled through us, with UK first-line support.
NVIDIA data-centre GPUs are allocated by NVIDIA and the server makers, and lead times are measured in months rather than weeks. Complete, tested systems usually get allocation sooner than loose parts, so we quote these platforms as built systems. Shipments outside the UK, and many advanced GPUs within it, need an end-user and end-use check under UK and US export rules before we can confirm supply.
Trade members get our engineers' time on sourcing and design questions, early stock and waitlist priority, and free technical support on every order.
No. It is available to order only. Lead time depends on HGX B300 allocation and is confirmed on a written quote.
Yes. The listed build is our standard configuration; alternative CPUs, memory sizes or storage can be quoted.
A faulty unit or part is returned for repair or replacement under warranty. If you need on-site cover or spares held locally, ask us to quote it.
Tell us what you want it to produce, how many people will use it and where it would live. An engineer will come back with a sized recommendation and an honest view on whether owning makes sense for you.