
Product image for illustration; configuration may differ.
Compal SGX30-2 HGX B300 8-GPU Server (10U, air cooled)
NVIDIA-Certified 10U air-cooled HGX B300 server: 8 Blackwell Ultra GPUs, 3 TB DDR5-6400, 8 × 800G ConnectX-8 OSFP and BlueField-3. Available to order.
Members pay £616,299.00 for this (ex VAT; £739,558.80 inc VAT). Join for £75 + VAT/month and take £500 off today — membership's more than half paid for before it ships.
- 60-day price lock for members
- Engineer on the phone, Mon–Fri 9–6
- 30-day returns (60 for members)
Ex VAT, at basket checkout. Terms
| Availability | To order only (not stocked) |
|---|---|
| AI performance | FP4 144 PFLOPS sparse / 108 PFLOPS dense; FP8 72 PFLOPS sparse (NVIDIA HGX B300 figures) |
| Boot storage | 2 × 960 GB M.2 NVMe |
| Memory | 32 × 96 GB DDR5-6400 ECC RDIMM (3 TB) |
| Data storage | 8 × 3.84 TB NVMe E1.S (PCIe Gen5) |
| DPU | NVIDIA BlueField-3 B3240 |
| GPUs | NVIDIA HGX B300 NVL8 baseboard: 8× Blackwell Ultra SXM, 288 GB HBM3e each (2.1 TB total) |
| Form factor | 10U rackmount, air cooled (447 × 800 × 442 mm) |
| Power | 3200 W 6+6 redundant 80 PLUS Titanium PSUs |
| GPU networking | 8 × 800G OSFP ports via onboard NVIDIA ConnectX-8 SuperNICs (InfiniBand XDR 800G or 2 × 400GbE per NIC) |
| GPU interconnect | Fifth-generation NVLink, 1.8 TB/s per GPU |
| Certification | NVIDIA-Certified System (HGX B300 8-GPU) |
| Expansion | 2 × FHFL PCIe Gen5 x16 |
| CPUs | 2× Intel Xeon 6 6767P, 6776P or 6768P |
| Warranty | 3-year return-to-base |
Your build
estimate — final price on enquiryCheck a model against our machines before you buy.
Pick the model and quantisation you plan to run. We'll show whether it fits in memory, how much room is left for context, and the throughput to expect. Estimates come from memory bandwidth and are replaced by our own bench figures where we have them.
Machine
Quantisation
Comfortable. 70 GB spare — room for 128k context, a second small model, or a RAG index in memory. Expect roughly 3.8 tok/s single-stream.
Machines that run this configuration
Weights 42.8 GB at 4.85 bits/param · KV cache 5.0 GB at 16k (fp16, approximate) · 10 GB reserved for OS and runtime. Speed is single-stream decode, memory-bound; prompt processing and batching behave differently. Members: £5,099 on this machine. Or WhatsApp an engineer on +44 7515 356 146.
An NVIDIA-Certified HGX B300 system from Compal, configured as a complete node for a Blackwell Ultra AI factory. Eight Blackwell Ultra GPUs on the HGX B300 NVL8 baseboard share 2.1 TB of HBM3e over fifth-generation NVLink, and each GPU gets its own 800 Gb/s ConnectX-8 SuperNIC port for the cluster fabric, so the server drops straight into an InfiniBand or Spectrum-X rail design.
It is air cooled, which means it can go into a high-density air-cooled hall without a liquid-cooling loop. Plan for a 10U slot and power and airflow for a system in the region of 14 kW.
- 8× NVIDIA Blackwell Ultra GPUs (HGX B300 NVL8), 288 GB HBM3e each
- 2× Intel Xeon 6 (6767P, 6776P or 6768P)
- 3 TB DDR5-6400 ECC RDIMM (32 × 96 GB)
- 8 × 800G OSFP via onboard ConnectX-8, plus NVIDIA BlueField-3 B3240
- 8 × 3.84 TB NVMe E1.S data + 2 × 960 GB M.2 boot
- 3200 W 6+6 redundant Titanium power supplies
- 3-year return-to-base warranty
Available to order only: not held in stock. Price is indicative and confirmed on a written quote after an allocation and export-control check.