Destello Tech · AI & ML Hardware

Run AI locally. Cut the cloud bill.

Professional GPU hardware for inference, LLM hosting, and GPU-accelerated compute. UK-stocked, next day dispatch, at a fraction of Nvidia enterprise pricing.

View GPU Range Get a Quote
48GB
Max VRAM on a single board
UK
Stock, next day dispatch
3 SKUs
Arc Pro range in stock
Direct
Maxsun UK brand partner
Cloud GPU bills spiralling? A single on-premise GPU can replace thousands in monthly cloud compute spend. The payback period on most AI workloads is under 6 months.
Data that can't leave your infrastructure? GDPR, client confidentiality, IP protection. Local inference means your data never touches a third-party server. Full control, zero exposure.
Enterprise Nvidia pricing out of reach? The Intel Arc Pro range delivers serious inference VRAM at a price point that makes on-premise deployment viable for startups and research labs, not just big enterprise.
The Intel Arc Pro Range

Purpose-built for AI inference
and professional compute.

All three SKUs are in UK stock now. Maxsun hardware, supplied direct through our brand partnership.

Arc Pro B60 24GB GDDR6 VRAM
  • Memory typeGDDR6
  • InterfacePCIe 5.0 x16
  • ArchitectureBattlemage Xe2
  • XMX AI CoresYes
  • Best for7B–13B models
Runs Llama 3.1 8B, Mistral 7B, Phi-3, Gemma 2 9B comfortably at full precision. £909.99 from View B60 24GB →
Mid-range sweet spot
Arc Pro B70 32GB GDDR6 VRAM
  • Memory typeGDDR6
  • InterfacePCIe 5.0 x16
  • ArchitectureBattlemage Xe2
  • XMX AI CoresYes
  • Best for13B–34B models
Handles Llama 3.1 70B (quantised), Mixtral 8x7B, CodeLlama 34B, multi-instance inference. £1,399.99 from View B70 32GB →
Arc Pro B60 Dual 48GB GDDR6 VRAM · Dual GPU board
  • Memory typeGDDR6
  • InterfacePCIe 5.0 x16
  • ArchitectureDual Battlemage Xe2
  • XMX AI CoresYes (×2)
  • Best for70B+ models
Full precision Llama 3.1 70B, multi-model hosting, serious throughput for production inference workloads. £1,699.99 from View B60 Dual 48GB →
What Can You Run?

VRAM requirements at a glance.

The single most important spec for local LLM inference is VRAM. Here's what each GPU handles.

ModelVRAM neededB60 24GBB70 32GBB60 Dual 48GBNotes
Llama 3.1 8B~16GB (FP16)Full precisionFull precisionFull precisionRuns comfortably on all three
Mistral 7B~14GB (FP16)Full precisionFull precisionFull precisionExcellent for RAG pipelines
Llama 3.1 70B~40GB (Q4) / 140GB (FP16)Too largeQuantised onlyQuantised (Q4)Q4 quantisation gives ~90% quality
Mixtral 8×7B~26GB (Q4)Quantised onlyFull Q4Full Q4Strong for multi-task workloads
CodeLlama 34B~20GB (Q4)Tight fitComfortableComfortablePopular for code generation
Phi-3 Mini 3.8B~8GB (FP16)Full precisionFull precisionFull precisionIdeal for edge / embedded use
Gemma 2 27B~16GB (Q4)ComfortableFull precisionFull precisionStrong reasoning performance
VRAM estimates are approximate. FP16 = full precision. Q4 = 4-bit quantisation. Arc Pro XMX cores accelerate INT4/INT8 inference natively via OpenVINO and IPEX-LLM.
Why Destello Tech

Direct pricing.
No runaround.

🔌 Maxsun UK Brand Partner Direct trade access to the full Maxsun Arc Pro range. Pricing that most UK resellers cannot access — and no stock sourced from overseas.
🚚 UK Stock, Next Day Dispatch Everything is held in the UK and ships from our Harlow workshop. No overseas lead times, no customs delays, no waiting weeks for a card to clear customs.
📞 Direct to the Founder You deal with Alex directly — no sales team, no ticket queue. Multi-unit orders, workstation builds, trade pricing — straight answer, fast turnaround.
Get in Touch

Tell us the workload.
We'll spec the hardware.

Tell us what you're running, how many instances, and your budget. We'll come back with a GPU recommendation and pricing within 24 hours.

Want to talk it through? Book a quick call below.

Book a 15min Call
Alex Overend · Founder, Destello Tech

Email: alex@destellotech.com

Phone: +44 7756 296228

Harlow, Essex, UK