Specialist dual-GPU workstation

Company 64

A specialist two-GPU tower for teams that cannot use a rack - not the next step in the standard range.

A separately evaluated Threadripper workstation for concurrent private AI, development, rendering and batch queues, with 64GB aggregate GPU memory.

Specialist workstation outside the standard range.

This tower is evaluated only when two GPU workers are required and a rack genuinely cannot be used. It is not the next tier after Core 48.

Plan around Company 64 Compare all systems

Price status: Indicative, ex VAT
Commercial model: Reviewed 26 July 2026
Range position: Specialist alternative

Supplier platform / proposed build

Inspect the platform

64GB aggregate VRAM* UK-built full-tower dual-GPU AI workstation proposed physical platform GPU route 2 × NVIDIA GeForce RTX 5090 Founders Edition 32GB

Buyer fit

Start with the reason to own it.

A workstation-form company platform for two useful GPU queues, or for a supported multi-GPU workload proven on the exact model and runtime before purchase.

Indicative purchase price

£37,000

ex VAT · £44,400 inc VAT

Proposed indicative price, excluding VAT. Final price, specification, availability, delivery, warranty and model fit require a current supplier quote and written order confirmation.

A credible fit

Two independent private AI or development workers
A company assistant alongside rendering, transcription or batch work
A technical office that needs more capacity but cannot yet support a rack server

Choose another route when

One guaranteed 64GB model without supported multi-GPU validation
A quiet desk-side role beside occupied workstations
A buyer who needs redundant power, rack serviceability or enterprise GPU support

Black full-tower workstation with a mesh front panel — Supplier reference view of the proposed dual-GPU workstation family. It is not evidence of a completed customer build. UK workstation supplier reference image. Written reuse permission pending.

Side view of a black full-tower workstation chassis — Supplier chassis view used to explain the floor-standing form factor. Internal parts and dimensions remain controlled by the final bill of materials. UK workstation supplier reference image. Written reuse permission pending.

Diagram showing jobs entering a queue, being assigned to independent workers and producing measured outputs — Queued rendering, batch and coding work can be divided between workers, with waiting time and failures measured. Original explanatory plate. It sets out a decision method, not a measured result.

Diagram of an evidence pack containing an asset schedule, burn-in record, health readings, workload test and admin guide — A credible handover records the supplied assets, checks, operating evidence, instructions and unresolved items. Original explanatory plate. It sets out a decision method, not a measured result.

Workload route

Every use needs its own acceptance test.

Model name and aggregate VRAM do not prove a business outcome. Each proposed route below states how the capacity could be used and what must be tested before the order treats it as suitable.

Company assistant plus build queue

Proposed use: Keep one GPU on an approved assistant while the second handles coding, transcription, embeddings or another scheduled service.
Acceptance evidence: Test the interactive service alone and under the agreed mixed workload, recording response time, queue delay and wall power.

Rendering and batch production

Proposed use: Divide independent render, image or analytical jobs between two GPU workers.
Acceptance evidence: Measure completed useful work, temperature, failure recovery and acoustic impact over a representative production run.

Supported model sharding

Proposed use: Evaluate a model across both GPUs only where the model and runtime support the intended execution pattern.
Acceptance evidence: Record per-GPU memory, throughput and latency at the disclosed quantisation, context, batch and concurrency.

Specification certainty ledger

Known, proposed and still to be confirmed.

A proposal should expose missing facts. The final order replaces every confirmation row with an exact part, measured result or named customer decision.

Package proposal Supplier confirmation Customer decision

Proposed Company 64 specification and evidence status
Item	Current value	Status	What the record must show
Form factor	Workstation	Package proposal	The site, delivery and support route is designed around this physical class.
Physical platform	UK-built full-tower dual-GPU AI workstation	Package proposal	Platform family selected for the proposed configuration.
GPU route	2 × NVIDIA GeForce RTX 5090 Founders Edition 32GB	Package proposal	Exact manufacturer, part number, condition and serials belong in the final bill of materials.
Per-GPU memory	32GB	Package proposal	The safer model-fit starting point before any supported multi-GPU test.
Aggregate GPU memory	64GB across 2 GPUs	Package proposal	Not presented as one universal memory pool.
System memory	128GB	Package proposal	Memory population, speed and expansion route require the final platform bill.
Primary storage	2TB NVMe	Package proposal	Drive model, endurance, layout and backup destination are confirmed in the order.
Network	10GbE plus 2.5GbE	Package proposal	Customer switching, cabling, storage traffic and segmentation remain part of site design.
Power planning	2050W PSU; dedicated circuit, heat and acoustic review required	Customer decision	A qualified site review and measured reference build must replace the planning figure.
CPU and motherboard	AMD Ryzen Threadripper 9960X, 24 cores / 48 threads	Package proposal	Current supplier reference value; the exact motherboard, lane layout and part numbers still belong in the final bill.
Power supplies and leads	To be confirmed for the ordered build and UK site	Supplier confirmation	Include PSU count, rating, redundancy position, input requirements and lead specification.
Remote management	Platform route to be confirmed and access policy agreed	Supplier confirmation	The supplier interface shown in the gallery is a reference, not a promise of the final feature set.
Dimensions and weight	Full-tower chassis; exact ordered dimensions and weight required	Supplier confirmation	Placement, handling and the delivery route depend on the exact ordered-system values.
Hardware warranty route	Supplier advertises three years: year one onsite, years two and three return to base	Supplier confirmation	The final order records the whole-system, component, onsite, return and dead-on-arrival boundaries.

32GB

Per GPU is the first sizing boundary.

Company 64 has 2 GPUs and 64GB aggregate VRAM. A runtime may split a supported model or distribute independent jobs, but the page does not claim one physical 64GB memory pool. Quantisation, context, KV cache, batching and concurrency still change fit.

Software and security

The usable product is more than the chassis.

The final stack stays deliberately small. Versions, licences, access and recurring ownership are recorded so the customer is not left with an opaque collection of containers.

Proposed software baseline

Ubuntu LTS on a recorded operating-system version
NVIDIA driver, CUDA components and container support validated for the ordered hardware
Docker Engine and NVIDIA Container Toolkit
One primary model server selected from Ollama, vLLM or llama.cpp for the accepted workload
Open WebUI or another reviewed browser interface
Named authentication, TLS and reverse-proxy approach
GPU, node and service monitoring with an agreed log-retention period
Pinned versions, a software bill of materials and a model source and licence record

Security ownership

GPU RIGS baseline: Initial operating-system state, named administrator handover, host firewall baseline, agreed access route, secrets transfer and documented update state.
Customer or contracted operator: User lifecycle, network and VPN policy, backups, monitoring review, patch approval, incident response, data governance and lawful use.
Shared before acceptance: Model and software licence checks, retention and logging choices, recovery test, acceptance criteria and a named owner for every recurring task.

Local infrastructure can reduce disclosure to external AI APIs. It does not automatically make the service secure, accurate, confidential or UK GDPR compliant.

Testing and acceptance

The evidence pack is part of the machine.

No public benchmark is invented for this page. The accepted workload, exact build and disclosed test conditions decide what can be claimed after the reference system exists.

01
Record the final bill of materials, serial numbers and firmware versions.
02
Run at least 24 hours of GPU, CPU, memory and storage stress testing.
03
Capture temperature, fan, error, health and wall-power evidence under the agreed load.
04
Test cold boot, restart and the available remote-management route.
05
Check drive health, network throughput and the container and GPU runtime.
06
Run model and workload smoke tests against the written acceptance set.
07
Record any measured speed only with the model, quantisation, context, concurrency and runtime disclosed.
08
Test an agreed fault or recovery route and provide the resulting handover record.

What is not claimed today

No fixed users, tokens per second, latency, model size, accuracy, availability, savings or marketplace contribution is stated without the missing configuration and test conditions.

Read the evidence method

Site, power and cooling

The room is part of the specification.

Planning power is not a measured promise. The final pre-flight records the circuit, voltage, loaded wall power, heat route, noise tolerance, rack, network, UPS decision and operating owner.

Read the site guide

Measure

Current loaded-system planning basis

A floor-standing technical workspace with unrestricted airflow and safe physical clearance
An electrical route approved for the final measured load; a dedicated circuit may be required
Cooling and acoustic review for the real duty cycle
10GbE switching and cabling where local data transfer requires it
A named technical owner and a written backup, update and access policy

Delivery, support and warranty

A written boundary before money moves.

The proposed route covers discovery, configuration, evidence and remote handover. It does not quietly absorb facilities work, migration or an always-on managed service.

Included in the proposed baseline

Documented workload and site-fit review
Confirmed bill of materials before procurement
Configuration, burn-in and agreed smoke-test evidence
Asset schedule, admin notes and user quick-start material
Collection or the quoted kerbside or pallet-delivery route
Remote onboarding and 30-day configuration-defect support

Separate scope or customer responsibility

Building electrical work, rack, UPS, cooling or structured cabling
Nationwide on-site installation unless separately quoted
Migration of customer data, every integration or every application
Continuous managed operations, security monitoring or a 24-hour support agreement
Third-party model, API, marketplace or software charges
A compliance certificate, performance guarantee or income guarantee

Supplier and warranty gates

Unresolved items stay visible.

Current trade or delivered price, availability and lead time confirmed before quotation
Exact dual-GPU spacing, motherboard topology, power leads, thermals and firmware compatibility recorded
GPU, whole-system, onsite, dead-on-arrival and return responsibilities confirmed in writing
Written permission or replacement photography obtained before supplier images are publicly released
Reference mixed-workload thermals, wall power and acoustics captured before an acceptance claim

Evidence and change record

Proposal inputs remain dated.

Package model: UK proposal reviewed 25 July 2026
Price state: Calculated from dated public supplier inputs, not a live quotation
Performance state: No public benchmark until the exact reference build is tested
OEM platform record: Exact platform identity and bill of materials provided in the written quotation.

Commercial reality

Tax and spare capacity are supporting questions.

Neither belongs in a guaranteed saving or payback claim. The purchase must stand on the accepted workload, control case and operating plan.

Finance, VAT and capital allowances

The displayed price is an indicative proposal based on dated public supplier inputs. The final bill of materials and supplier quote control the order.
Prices exclude VAT. VAT recovery depends on the buyer, its taxable activities and the normal evidence rules.
Qualifying equipment may be plant and machinery for capital-allowance purposes. The buyer's accountant decides eligibility and timing.
A third-party lease or hire-purchase route may be explored after partner validation and credit approval. GPU RIGS is not presented as a lender.
There is no generic capital-gains advantage and no automatic research and development relief because the equipment supports AI.

Read the guarded UK buyer notes

Optional idle capacity

Marketplace mode is off by default and is excluded from the purchase case.
A separate environment, no customer data mounts, network controls and a local kill switch would be required.
The customer, insurer, supplier warranty and marketplace terms must permit the proposed use.
Vast.ai, Render, Golem and direct batch work do not guarantee acceptance, demand, rate or income.
Any pilot must report achieved utilisation, fees, electricity, cooling, faults and operator time before continuation.

Review the marketplace decision gate

Limits and alternatives

A good specification leaves room for “no”.

The fit check can recommend a smaller system, hosted service, hybrid route, custom build or no purchase. That is preferable to making this package carry a workload it has not proved.

Package boundaries

The two GPUs provide 64GB aggregate VRAM, not one universal 64GB memory pool.
The public configuration states that the GPUs do not use NVLink.
The 2050W PSU, heat, weight and acoustic impact need a pre-order site review.
Each GPU has 32GB of local VRAM. The 64GB figure is aggregate capacity.
No NVLink or one-pool memory claim is made.
No user-count, token-rate, latency, accuracy or energy-cost promise is made before testing.
The proposal is based on a public retail listing, not a reseller appointment or supplier partnership.

Team 32

The lower-cost office route when one 32GB worker and scheduled batch work are enough.

Consider this route Core 48

A lower-priced rack alternative when two 24GB workers and a suitable equipment environment meet the workload.

Consider this route Business 128

Consider when four independent workers, rack density and a server-room operating model are justified.

Consider this route

Questions answered

Company 64 questions that affect the order

These answers preserve the validation boundary. The written quotation and acceptance plan replace proposal assumptions.

Why does Company 64 cost more than Core 48?

It uses two current 32GB RTX 5090 GPUs, a Threadripper platform and a public UK-built workstation whose input cost is materially higher. Core 48 uses refurbished 24GB GPUs and carries a different warranty risk.

Can one model use all 64GB?

Not automatically. The machine has two 32GB GPUs without NVLink. Some runtimes can split supported models, but memory behaviour and speed must be tested.

Can it work in an office?

It avoids the rack requirement, but two high-end GPUs and a 2050W PSU create significant heat and fan noise. Electrical, thermal and acoustic acceptance is mandatory.

Is the £37,000 price final?

No. It is an indicative ex-VAT proposal based on a public UK supplier price reviewed on 26 July 2026. The current quote, exact parts and delivery terms control the order.

Prepare the next decision

Specify the work before the parts.

Record the workload, data boundary, users, site and acceptance test. No confidential documents or credentials are needed for the first brief.

Proposed indicative price, excluding VAT. Final price, specification, availability, delivery, warranty and model fit require a current supplier quote and written order confirmation.

Build the Company 64 brief Read the delivery boundary