Hardware shortlist

Compare the range without hiding the unknowns.

Choose two to four systems. Compare physical capacity, price status, facility burden and model evidence before requesting a specification.

Build a shortlist Read the claims policy

Three-quarter supplier render of a 4U OEM multi-GPU rack server — OEM platform reference render. It is not evidence of a completed customer build or final specification.

11-system range

Build a two-to-four system shortlist.

Compare what is recorded, including the gaps. Indicative prices remain planning values, aggregate GPU memory is not a universal pool, and a model candidate is not a supported deployment.

Default four-system shortlist. Use the controls above to compare another two to four systems.
Decision field	SME workstation Team 32	SME workstation Company 64	SME workstation Studio 96	SME workstation Studio 192	Enterprise PCIe rack Value Rack 128	Enterprise PCIe rack Value Rack 256	Enterprise PCIe rack Enterprise 384	Enterprise PCIe rack Enterprise 768	Enterprise PCIe rack H200 1.1TB	Frontier platform Frontier Native 2.3TB	Frontier platform Frontier Rack 20TB
Price status	£7,500 ex VAT Published package figure Price qualification GPU RIGS package figure for the stated reference configuration. The final written quotation confirms exact parts, availability, delivery, warranty and accepted workload scope.	£18,500 ex VAT Published package figure Price qualification GPU RIGS package figure for the stated reference configuration. The final written quotation confirms exact parts, availability, delivery, warranty and accepted workload scope.	£20,000 ex VAT Published package figure Price qualification GPU RIGS package figure for the stated reference configuration. The final written quotation confirms exact parts, availability, delivery, warranty and accepted workload scope.	£42,000 ex VAT Published package figure Price qualification GPU RIGS package figure for the stated reference configuration. The final written quotation confirms exact parts, availability, delivery, warranty and accepted workload scope.	£45,765 ex VAT Published package figure Price qualification GPU RIGS package figure for the stated reference configuration. The final written quotation confirms exact parts, availability, delivery, warranty and accepted workload scope.	£82,675 ex VAT Published package figure Price qualification GPU RIGS package figure for the stated reference configuration. The final written quotation confirms exact parts, availability, delivery, warranty and accepted workload scope.	£99,999 ex VAT Published package figure Price qualification GPU RIGS package figure for the stated reference configuration. The final written quotation confirms exact parts, availability, delivery, warranty and accepted workload scope.	£189,999 ex VAT Published package figure Price qualification GPU RIGS package figure for the stated reference configuration. The final written quotation confirms exact parts, availability, delivery, warranty and accepted workload scope.	Quote only £349,000 ex VAT indicative budget Quote only Price qualification Final quotation required. Exact configuration, availability, delivery, facility requirements, warranty, validation and support scope determine the customer price.	Quote only Quote only Price qualification Final quotation required. Exact configuration, availability, delivery, facility requirements, warranty, validation and support scope determine the customer price.	Quote only Quote only Price qualification Final quotation required. Exact configuration, availability, delivery, facility requirements, warranty, validation and support scope determine the customer price.
GPU configuration	1 × NVIDIA GeForce RTX 5090 32GB 1 physical GPU; 32GB per GPU	2 × NVIDIA GeForce RTX 5090 32GB 2 physical GPUs; 32GB per GPU	1 × NVIDIA RTX PRO 6000 Blackwell 96GB 1 physical GPU; 96GB per GPU	2 × NVIDIA RTX PRO 6000 Blackwell 96GB 2 physical GPUs; 96GB per GPU	4 × NVIDIA RTX 5090 AI 32GB 4 physical GPUs; 32GB per GPU	8 × NVIDIA RTX 5090 AI 32GB 8 physical GPUs; 32GB per GPU	4 × NVIDIA RTX PRO 6000 Blackwell 96GB 4 physical GPUs; 96GB per GPU	8 × NVIDIA RTX PRO 6000 Blackwell 96GB 8 physical GPUs; 96GB per GPU	8 × NVIDIA H200 NVL 141GB 8 physical GPUs; 141GB per GPU	8 × NVIDIA B300 288GB 8 physical GPUs; 288GB per GPU	72 × NVIDIA GB300 288GB 72 physical GPUs; 288GB per GPU
Aggregate GPU memory Physical total, not a universal pooled or contiguous memory space.	32GB 32GB is available on one GPU as one physical memory space.	64GB 64GB is the physical total across 2 separate 32GB GPUs. It is not automatically pooled; supported software may divide a model or workload across them, subject to topology and testing.	96GB 96GB is available on one GPU as one physical memory space.	192GB 192GB is the physical total across 2 separate 96GB GPUs. It is not automatically pooled; supported software may divide a model or workload across them, subject to topology and testing.	128GB 128GB is the physical total across 4 separate 32GB GPUs. It is not automatically pooled; supported software may divide a model or workload across them, subject to topology and testing.	256GB 256GB is the physical total across 8 separate 32GB GPUs. It is not automatically pooled; supported software may divide a model or workload across them, subject to topology and testing.	384GB 384GB is the physical total across 4 separate 96GB GPUs. It is not automatically pooled; supported software may divide a model or workload across them, subject to topology and testing.	768GB 768GB is the physical total across 8 separate 96GB GPUs. It is not automatically pooled; supported software may divide a model or workload across them, subject to topology and testing.	1,128GB (1.1TB) 1,128GB is the physical total across eight H200 GPUs. It is not automatically one pooled memory space; the exact NVLink pairing, PCIe topology and supported parallel runtime must be confirmed in the quotation.	2,304GB (2.25TB) 2,304GB is distributed across eight B300 GPUs connected by HGX NVLink and NVSwitch. The fabric enables high-bandwidth multi-GPU work, but software support and the exact workload still determine usable capacity.	20,736GB (20.25TB) 20,736GB is distributed across a rack-scale NVLink and NVSwitch fabric. It is specialist shared infrastructure, not one ordinary GPU memory space.
Form and facility	Workstation A ventilated floor or desk-side position with clear intake and exhaust paths A suitable UK circuit and an agreed UPS decision after measured-load review An acoustic and heat check under the accepted workload A named owner for accounts, updates, backups and incident response	Workstation A ventilated floor or desk-side position with clear intake and exhaust paths A suitable UK circuit and an agreed UPS decision after measured-load review An acoustic and heat check under the accepted workload A named owner for accounts, updates, backups and incident response	Workstation A ventilated floor or desk-side position with clear intake and exhaust paths A suitable UK circuit and an agreed UPS decision after measured-load review An acoustic and heat check under the accepted workload A named owner for accounts, updates, backups and incident response	Workstation A ventilated floor or desk-side position with clear intake and exhaust paths A suitable UK circuit and an agreed UPS decision after measured-load review An acoustic and heat check under the accepted workload A named owner for accounts, updates, backups and incident response	Rack server A suitable 19-inch rack, rail depth, handling route and secure operating location A qualified electrical design based on the final PSU population and measured load Cooling and heat-rejection capacity for sustained accelerator operation Appropriate switching, cabling, remote management and network segmentation A named operational owner or contracted support route	Rack server A suitable 19-inch rack, rail depth, handling route and secure operating location A qualified electrical design based on the final PSU population and measured load Cooling and heat-rejection capacity for sustained accelerator operation Appropriate switching, cabling, remote management and network segmentation A named operational owner or contracted support route	Rack server A suitable 19-inch rack, rail depth, handling route and secure operating location A qualified electrical design based on the final PSU population and measured load Cooling and heat-rejection capacity for sustained accelerator operation Appropriate switching, cabling, remote management and network segmentation A named operational owner or contracted support route	Rack server A suitable 19-inch rack, rail depth, handling route and secure operating location A qualified electrical design based on the final PSU population and measured load Cooling and heat-rejection capacity for sustained accelerator operation Appropriate switching, cabling, remote management and network segmentation A named operational owner or contracted support route	Rack server A suitable 19-inch rack, rail depth, handling route and secure operating location A qualified electrical design based on the final PSU population and measured load Cooling and heat-rejection capacity for sustained accelerator operation Appropriate switching, cabling, remote management and network segmentation A named operational owner or contracted support route	Rack server A suitable 19-inch rack, rail depth, handling route and secure operating location A qualified electrical design based on the final PSU population and measured load Cooling and heat-rejection capacity for sustained accelerator operation Appropriate switching, cabling, remote management and network segmentation A named operational owner or contracted support route	Rack server A suitable 19-inch rack, rail depth, handling route and secure operating location A qualified electrical design based on the final PSU population and measured load Cooling and heat-rejection capacity for sustained accelerator operation Appropriate switching, cabling, remote management and network segmentation A named operational owner or contracted support route
Platform	Single-GPU mid-tower AI workstation	Dual-GPU full-tower AI workstation	Single-GPU professional AI workstation	Dual-GPU professional AI workstation	GENOAX2 PCIe Gen5 4U rack platform	GENOAX2 PCIe Gen5 4U rack platform	GENOAX2 PCIe Gen5 4U enterprise rack platform	GENOAX2 PCIe Gen5 4U enterprise rack platform	Eight-GPU H200 NVL PCIe rack platform	SYS-822GS-NB3RT 8U HGX B300 platform	GB300 NVL72 full-rack architecture
System RAM	64GB	128GB	128GB	128GB	Exact system-memory capacity and population pending supplier BOM	Exact system-memory capacity and population pending supplier BOM	Exact system-memory capacity and population pending supplier BOM	Exact system-memory capacity and population pending supplier BOM	1024GB	Exact system-memory capacity and population pending supplier BOM	Exact system-memory capacity and population pending supplier BOM
Storage	2TB NVMe	2TB NVMe	2TB NVMe	2TB NVMe	Exact NVMe capacity, model, endurance and layout pending supplier BOM	Exact NVMe capacity, model, endurance and layout pending supplier BOM	Exact NVMe capacity, model, endurance and layout pending supplier BOM	Exact NVMe capacity, model, endurance and layout pending supplier BOM	11.68TB NVMe	Exact NVMe capacity, model, endurance and layout pending supplier BOM	Exact NVMe capacity, model, endurance and layout pending supplier BOM
Network	Exact network interfaces and fabric pending workload and supplier confirmation	10GbE and 2.5GbE Ethernet	Exact network interfaces and fabric pending workload and supplier confirmation	10GbE and 2.5GbE Ethernet	Exact network interfaces and fabric pending workload and supplier confirmation	Exact network interfaces and fabric pending workload and supplier confirmation	Exact network interfaces and fabric pending workload and supplier confirmation	Exact network interfaces and fabric pending workload and supplier confirmation	200Gb ConnectX-6 plus management/base networking	Exact network interfaces and fabric pending workload and supplier confirmation	Exact network interfaces and fabric pending workload and supplier confirmation
Power	Exact PSU, connector and measured wall-power requirement pending	2050W PSU; measured wall power and site circuit requirement pending	Exact GPU edition, PSU and measured wall power pending	Exact dual-GPU PSU, circuit and measured wall-power requirement pending	Exact four-GPU input, PSU population and circuit requirement pending	High-capacity dedicated power design required; exact measured system input pending	Exact PSU population, input requirement and measured wall power pending	Dedicated high-capacity power required; exact input and measured wall power pending	At least 4.8kW GPU nameplate; exact full-system input pending	Approximately 14kW system envelope; exact quoted input pending	Up to 142kW full-rack power
Model evidence Candidate never means compatible or supported.	Not evaluated No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. No tested or calculated model-fit candidate is listed for this system.	Not evaluated No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. No tested or calculated model-fit candidate is listed for this system.	Not evaluated No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. No tested or calculated model-fit candidate is listed for this system.	Not evaluated No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. No tested or calculated model-fit candidate is listed for this system.	Not evaluated No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. No tested or calculated model-fit candidate is listed for this system.	Not evaluated No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. No tested or calculated model-fit candidate is listed for this system.	Not evaluated No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. No tested or calculated model-fit candidate is listed for this system.	Not evaluated No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. No tested or calculated model-fit candidate is listed for this system.	Not evaluated No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. No tested or calculated model-fit candidate is listed for this system.	Candidate evidence only No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. Moonshot AI moonshotai/Kimi-K3 · Calculated capacity candidate HGX B300's documented 2.30TB aggregate HBM3e exceeds the dated 1.5609TB repository weight total. This arithmetic establishes only a memory-capacity candidate, not a successful load, usable context, throughput, latency or concurrency.	Candidate evidence only No model checkpoint is yet classed as measured-supported for this revision. Capacity candidates and supplier statements are not presented as support. Moonshot AI moonshotai/Kimi-K3 · Calculated capacity candidate GB300 NVL72's documented 20TB aggregate GPU memory is a capacity candidate for this checkpoint. Its commercial purpose is rack-scale throughput and concurrency, neither of which is measured here; checkpoint fit alone does not justify the platform.
Known limitations	Review 8 recorded limitations The accepted model or context needs more than 32GB Several heavy services need guaranteed simultaneous capacity The buyer needs redundant power or enterprise remote management 32GB is one GPU's physical memory; context, cache, batching and concurrency reduce the usable model envelope. No tokens-per-second, user-count, latency, image or video performance has been measured for this revision. A workstation is not presented as a redundant datacentre appliance. No model, speed, context, concurrency or output-quality promise exists without a controlled fit profile. Public supplier price and availability observations are dated inputs, not reserved stock.	Review 9 recorded limitations One model needs a guaranteed contiguous 64GB space The site cannot accept the measured dual-GPU load Rack serviceability or redundancy is required 64GB is aggregate capacity across two 32GB GPUs, not one universal memory pool. No multi-GPU model fit, thermals, noise or wall power has been measured. 64GB is aggregate capacity across two independent 32GB GPUs; the selected supplier lists no NVLink. A workstation is not presented as a redundant datacentre appliance. No model, speed, context, concurrency or output-quality promise exists without a controlled fit profile. Public supplier price and availability observations are dated inputs, not reserved stock.	Review 9 recorded limitations A 32GB route passes the acceptance test Several independent GPU services are required The buyer needs rack resilience or service density The exact RTX PRO 6000 Workstation or Max-Q selection remains a pre-quote decision. 96GB capacity does not prove a named model, context, speed or concurrency. The exact RTX PRO Workstation or Max-Q edition must be confirmed in the quote. A workstation is not presented as a redundant datacentre appliance. No model, speed, context, concurrency or output-quality promise exists without a controlled fit profile. Public supplier price and availability observations are dated inputs, not reserved stock.	Review 9 recorded limitations One 96GB GPU passes the workload acceptance test Enterprise rack serviceability or density is required The office cannot accept the final electrical, heat or noise envelope 192GB is aggregate capacity across two 96GB GPUs, not one universal memory pool. No dual-GPU topology, model fit, cooling, acoustic or performance result has been measured. 192GB is aggregate capacity across two 96GB GPUs; no NVLink or universal pooling claim is made. A workstation is not presented as a redundant datacentre appliance. No model, speed, context, concurrency or output-quality promise exists without a controlled fit profile. Public supplier price and availability observations are dated inputs, not reserved stock.	Review 8 recorded limitations One model needs more than 32GB per GPU without proven sharding A quieter tower meets demand The site cannot support 4U forced-air operation 128GB is aggregate capacity across four 32GB GPUs. The custom passive GPU, exact lane map, thermals and warranty remain supplier gates. Aggregate GPU memory is not one universal memory pool. No model, speed, context, concurrency or multi-GPU efficiency promise exists without a controlled fit profile. Rack, electrical work, UPS, cooling, cabling, colocation and on-site installation require a separate scope.	Review 9 recorded limitations Four workers meet demand The model requires NVLink or NVSwitch The site cannot support the final electrical and cooling design 256GB is aggregate capacity across eight 32GB GPUs. Power, thermal behaviour, PCIe topology and GPU warranty remain unverified. No single-model speed, concurrency or multi-GPU efficiency has been measured. Aggregate GPU memory is not one universal memory pool. No model, speed, context, concurrency or multi-GPU efficiency promise exists without a controlled fit profile. Rack, electrical work, UPS, cooling, cabling, colocation and on-site installation require a separate scope.	Review 9 recorded limitations The workload requires HGX NVSwitch Two or fewer workers meet demand The buyer lacks an enterprise rack operating route 384GB is aggregate capacity across four 96GB GPUs. This is a PCIe system; no NVLink, NVSwitch or universal pooled-memory claim is made. The old Max-Q rack wording is invalid and no exact Server Edition BOM has yet been quoted. Aggregate GPU memory is not one universal memory pool. No model, speed, context, concurrency or multi-GPU efficiency promise exists without a controlled fit profile. Rack, electrical work, UPS, cooling, cabling, colocation and on-site installation require a separate scope.	Review 9 recorded limitations Four workers meet demand The workload requires an HGX fabric Power, cooling or support is unresolved 768GB is aggregate capacity across eight 96GB GPUs. This is a PCIe architecture, not an HGX system and not one pooled 768GB memory space. No Kimi K3 fit, context, throughput, concurrency or fine-tuning result has been measured. Aggregate GPU memory is not one universal memory pool. No model, speed, context, concurrency or multi-GPU efficiency promise exists without a controlled fit profile. Rack, electrical work, UPS, cooling, cabling, colocation and on-site installation require a separate scope.	Review 11 recorded limitations A PCIe RTX PRO worker platform meets demand The purchase depends on unverified speed or concurrency The exact supplier and support route is unresolved 1,128GB aggregate HBM is below the current 1.5609TB official Kimi K3 weight-file total. Kimi K3 is only an aggressive-quantisation candidate on this route until an exact converted checkpoint is tested. No native load, full context, tokens-per-second, concurrency or quality result is claimed. The current £349,000 ex-VAT planning figure is indicative; the system remains quote-only until the supplier BOM, topology, landed cost and support route are accepted. Kimi K3 is a calculated capacity candidate, not a supported or benchmarked claim. Aggregate GPU memory is not one universal memory pool. No model, speed, context, concurrency or multi-GPU efficiency promise exists without a controlled fit profile. Rack, electrical work, UPS, cooling, cabling, colocation and on-site installation require a separate scope.	Review 10 recorded limitations Independent PCIe workers meet the workload The decision depends only on aggregate VRAM Facility or software compatibility is not yet accepted 2.304TB HBM is a calculated capacity envelope, not proof that the official Kimi K3 checkpoint loads or serves one-million-token context. No Kimi K3 throughput, TTFT, concurrency, wall power or quality result has been measured. Pricing is quote-only. Model-fit entries remain candidates until reproducible tests exist. Aggregate GPU memory is not one universal memory pool. No model, speed, context, concurrency or multi-GPU efficiency promise exists without a controlled fit profile. Rack, electrical work, UPS, cooling, cabling, colocation and on-site installation require a separate scope.	Review 10 recorded limitations A single HGX system meets demand The programme lacks a named technical operator Budget or value depends on unverified public performance claims This is a complete 72-GPU liquid-cooled rack, not an the selected OEM 4U product or ordinary office delivery. 20TB is the rounded marketing class; 72 × 288GB equals 20,736GB aggregate GPU memory. Pricing is quote-only. 20TB is aggregate rack accelerator memory and not a generic one-process promise. Aggregate GPU memory is not one universal memory pool. No model, speed, context, concurrency or multi-GPU efficiency promise exists without a controlled fit profile. Rack, electrical work, UPS, cooling, cabling, colocation and on-site installation require a separate scope.
Next evidence	Review Team 32 Evidence reviewed 2026-07-28.	Review Company 64 Evidence reviewed 2026-07-28.	Review Studio 96 Evidence reviewed 2026-07-28.	Review Studio 192 Evidence reviewed 2026-07-28.	Review Value Rack 128 Evidence reviewed 2026-07-28.	Review Value Rack 256 Evidence reviewed 2026-07-28.	Review Enterprise 384 Evidence reviewed 2026-07-28.	Review Enterprise 768 Evidence reviewed 2026-07-28.	Review H200 1.1TB Evidence reviewed 2026-07-28.	Review Frontier Native 2.3TB Evidence reviewed 2026-07-28.	Review Frontier Rack 20TB Evidence reviewed 2026-07-28.

Unknown values remain visibly unknown. Final specification, price, availability, facility fit, warranty and model acceptance require a current written record.

Open 4U OEM GPU server chassis showing passive GPUs, cooling fans, processors and memory slots — OEM supplier render showing one possible internal layout. Components vary with the ordered build. OEM supplier reference image.

Diagram combining model weights, context, cache and active requests into a memory headroom check — Memory fit depends on the workload, context, cache and simultaneous demand, then needs a representative test. GPU Servers technical illustration.

Exploded supplier render of passive GPUs arranged above an open 4U rack chassis — Supplier layout render used to explain GPU density and airflow. It does not represent a confirmed package configuration. OEM supplier reference image.

After the shortlist

Turn candidates into an acceptance plan.

The table narrows the range; it does not select a build. The next step records the exact model revision, runtime, workload, concurrency, site conditions, budget and evidence required before an order.

Send us your shortlist Discuss the shortlist