AI Cloud · GPUs · Bare Metal
Bare Metal GPU Servers
Single-tenant physical servers with direct PCIe access. No virtualization overhead — best for training large models and latency-sensitive HPC workloads.
Whole server, no virtualization — raw performance and full root.
5 configurations available
AMD
AMD GPUs on Bare Metal
CDNA 4
Popular
AMD Instinct MI350X
288GB HBM3e flagship with native FP4/FP6 — the TCO leader for frontier inference
VRAM
288GB HBM3e
Bandwidth
8 TB/s
TDP
1000W
Form factorsBare MetalVMContainer
from£54.40/hr
View AMD Instinct MI350X details
CDNA 3
Popular
AMD Instinct MI300X
192GB HBM3 — proven ROCm performance with the best inference economics
VRAM
192GB HBM3
Bandwidth
5.3 TB/s
TDP
750W
Form factorsBare MetalVMContainer
from£19.60/hr
View AMD Instinct MI300X details
CDNA 3
AMD Instinct MI325X
128GB HBM3 — lowest-cost ROCm inference and AIGC
VRAM
128GB HBM3
Bandwidth
1.6 TB/s
TDP
300W
Form factorsBare MetalVMContainer
from£9.76/hr
View AMD Instinct MI325X details
Why AMD on Technova:
- Full ROCm 6.x stack
- CUDA→ROCm migration support
- Up to 40% lower TCO
NVIDIA
NVIDIA GPUs on Bare Metal
Hopper
Popular
NVIDIA H200
141GB HBM3e Hopper — drop-in CUDA acceleration with first-class tooling
VRAM
141GB HBM3e
Bandwidth
4.8 TB/s
TDP
700W
Form factorsBare MetalVMContainer
from£27.00/hr
View NVIDIA H200 details
Ada / Blackwell
NVIDIA RTX PRO 6000
96GB GDDR7 Blackwell-class workstation GPU — low barrier inference and AIGC
VRAM
96GB GDDR7
Bandwidth
1.8 TB/s
TDP
600W
Form factorsBare MetalVMContainer
from£11.60/hr
View NVIDIA RTX PRO 6000 details
Why NVIDIA on Technova:
- Full CUDA 12.x ecosystem
- NVLink + NCCL
- TensorRT-LLM optimised
Deploy on Bare Metal?
Reserved capacity, custom networking and managed services — all delivered by our UK engineering team.