GPU servers, priced and ready to ship.
HGX B300, B200 and H200 systems, RTX PRO 6000 servers and GB300 NVL72 racks, built and burned in at our own configuration centres. From Options Direct, trading since 2014.
HGX B300
$548,000
6–8 weeks
HGX B200
$468,000
4–6 weeks
HGX H200
$338,000
2–3 weeks
8× RTX PRO 6000 Server
$172,500
1–2 weeks
Indicative, ex-VAT. Demonstration data.
Part of Options Direct, supplying datacentre hardware since 2014.
The same engineers, configuration centres and accounts team that supply AI teams and trading firms in 40+ countries. ISO 9001:2015 certified.
More about the company- ManchesterHeadquarters · Altrincham
- LondonSatellite office
- FrankfurtEMEA logistics
- DubaiDubai Internet City · MEA + APAC
- United StatesOpening soon
Anatomy of an HGX B300 system
Eight GPUs in one 8U chassis
- GPU
- 8× NVIDIA B300, NVLink 5 switch tray
- Form factor
- 8U air-cooled or 4U liquid-cooled
The GPU tray lifts out in one piece
- CPU
- 2× Intel Xeon 6 6960P, 72 cores each
- Memory
- 3 TB DDR5-6400
Each GPU position, tested on its own
- GPU memory
- 2.3 TB HBM3e total
- Burn-in
- 72 hours, all eight GPUs under load
Network and power at the rear
- Network
- 8× ConnectX-8 800G, 2× BlueField-3 DPU
- Power
- Up to 14.3 kW, 6+6 redundant PSUs
Ready to rack
$548,000Indicative, ex-VAT, per system
- Lead time
- 6–8 weeks, subject to allocation
Request allocationDemonstration data
Systems in stock and on allocation
Indicative prices and lead times for the platforms we ship most, updated as allocation moves. Every quote confirms a firm price and a delivery date.
Demonstration data- NVIDIA HGX B3008× B300 288 GB · 2× Xeon 6 · 3 TB DDR5$548,0006–8 weeks
- NVIDIA HGX B2008× B200 180 GB · 2× Xeon 6 · 2 TB DDR5$468,0004–6 weeks
- NVIDIA HGX H2008× H200 141 GB · 2× EPYC 9005 · 2 TB DDR5$338,0002–3 weeks
- 8× RTX PRO 6000 Server8× RTX PRO 6000 Blackwell 96 GB · PCIe 5.0$172,5001–2 weeks
- AMD Instinct MI355X8× MI355X 288 GB · 2× EPYC 9005 · 3 TB DDR5$452,0006–10 weeks
Eight GPUs. One board.
HGX systems arrive as a single tested unit: GPUs, NVLink switch tray, CPUs, memory, storage and networking, configured to your spec and checked as a whole.
GPUs by the card, too.
New and tested cards for your own servers, held in stock in the UK and EU.
- NVIDIA H200 NVL141 GB HBM3e · PCIe 5.0 · dual-slot$33,9001–2 weeks
- RTX PRO 6000 Blackwell Server96 GB GDDR7 · PCIe 5.0 · passive$10,9001 week
- NVIDIA L40S48 GB GDDR6 · PCIe 4.0 · passive$7,4501 week
- NVIDIA H100 SXM, used8-GPU HGX baseboard · 80 GB each · tested$196,0001–2 weeks
- NVIDIA A100 80 GB, used80 GB HBM2e · PCIe 4.0 · tested$8,2001 week
From request to rack
- 01RequestTell us the system, quantity, site and the date you need it live, or send your own bill of materials.
- 02QuoteA firm price, a delivery date and payment terms from a named account manager.
- 03Build and burn-inConfigured in our own centres and run under full load for 72 hours. The report ships with it.
- 04DeliverShipped, imported and installed on site in 40+ countries.
Proforma and bank transfer, with Net-30 terms for approved accounts.
Guides to buying, exporting, deploying and retiring GPU compute
Written by the people who do it every week. Free to read, whether or not you buy from us.
- Buying GPU computeChoosing a platform, reading a quote, allocation and lead times, new or used, and what happens after you order.
Compliance and exportExport controls for AI GPUs, end-user statements, KYC and screening, and the paperwork every order needs.
Deploying GPU clustersPower and cooling, network fabrics, optics and cabling, burn-in, acceptance and site readiness.
Decommissioning and resaleRetiring GPU infrastructure safely: data sanitisation, logistics, WEEE, and what used hardware is worth.
Tell us what you need.
A firm price, a delivery date and payment terms. Or call +44 207 183 3135.