Systems

Built, tested and handed over working

Every system arrives with its software stack installed and a burn-in report showing 72 hours at full load without fault. Clusters are commissioned and tested on your site, not delivered as a stack of boxes. Any of these can be opened in the configurator and altered.

Go in a rack. Two power supplies so one can fail, and air pulled front to back.

SUPERCOMPUTERSSystemNo photograph on file

gpu serverNew build48h burn-in36mo warranty

TF-L8 Edge Inference

Processors: AMD EPYC 9124
Accelerators: 8 x NVIDIA L4 24GB
Cores: 16 total
Memory: 256GB DDR5
Storage: 4x 1.92TB U.2 NVMe
Fabric: 2x 25GbE
Rack space: 1U
Peak power: 0.9kW · 4A at 230V
Heat rejected: 3,071 BTU/hr

Eight L4s in 1U on 900W total
No auxiliary GPU power cables anywhere in the box
Fits a colocation cabinet with a normal power allocation

Ships with Ubuntu 22.04 LTS · Triton Inference Server · TensorRT · Prometheus

Quoted per order

30 working days

Request a quote Full spec

SUPERCOMPUTERSSystemNo photograph on file

gpu serverRefurbished48h burn-in12mo warranty

TF-V2 VDI Node

Processors: 2x Intel Xeon Gold 6548Y+
Accelerators: 4 x NVIDIA L40 48GB
Cores: 64 total
Memory: 1TB DDR5
Storage: 8x 3.84TB U.2 NVMe
Fabric: 2x 25GbE
Rack space: 2U
Peak power: 2.4kW · 10A at 230V
Heat rejected: 8,189 BTU/hr

Up to 64 virtual workstations on four L40s
vGPU licensing quoted separately, it is per-user and annual
Designed for engineering firms consolidating CAD seats

Ships with VMware ESXi or Proxmox · NVIDIA vGPU host driver · Horizon or Omnissa

Quoted per order

30 working days

Request a quote Full spec

SUPERCOMPUTERSSystemNo photograph on file

gpu serverRefurbished48h burn-in12mo warranty

TF-4A100 Refurb Trainer

Processors: 2x AMD EPYC 7543
Accelerators: 4 x NVIDIA A100 PCIe 80GB
Cores: 64 total
Memory: 512GB DDR4
Storage: 4x 3.84TB U.2 NVMe
Fabric: 1x ConnectX-6 HDR200
Rack space: 4U
Peak power: 2.6kW · 11A at 230V
Heat rejected: 8,871 BTU/hr

320GB HBM2e for well under half the price of a new node
Every GPU load-tested 48 hours before it ships
The best PKR-per-training-hour we sell

Ships with Ubuntu 22.04 LTS · CUDA 12 · PyTorch 2 · Slurm · NCCL tuned

Quoted per order

30 working days

Request a quote Full spec

SUPERCOMPUTERSSystemNo photograph on file

gpu serverNew build48h burn-in36mo warranty

TF-V2 VDI Node

Processors: 2x Intel Xeon Gold 6548Y+
Accelerators: 4 x NVIDIA L40 48GB
Cores: 64 total
Memory: 1TB DDR5
Storage: 8x 3.84TB U.2 NVMe
Fabric: 2x 25GbE
Rack space: 2U
Peak power: 2.4kW · 10A at 230V
Heat rejected: 8,189 BTU/hr

Up to 64 virtual workstations on four L40s
vGPU licensing quoted separately, it is per-user and annual
Designed for engineering firms consolidating CAD seats

Ships with VMware ESXi or Proxmox · NVIDIA vGPU host driver · Horizon or Omnissa

Quoted per order

30 working days

Request a quote Full spec

SUPERCOMPUTERSSystemNo photograph on file

gpu serverRefurbished48h burn-in12mo warranty

TF-4A100 Refurb Trainer

Processors: 2x AMD EPYC 7543
Accelerators: 4 x NVIDIA A100 PCIe 80GB
Cores: 64 total
Memory: 512GB DDR4
Storage: 4x 3.84TB U.2 NVMe
Fabric: 1x ConnectX-6 HDR200
Rack space: 4U
Peak power: 2.6kW · 11A at 230V
Heat rejected: 8,871 BTU/hr

320GB HBM2e for well under half the price of a new node
Every GPU load-tested 48 hours before it ships
The best PKR-per-training-hour we sell

Ships with Ubuntu 22.04 LTS · CUDA 12 · PyTorch 2 · Slurm · NCCL tuned

Quoted per order

30 working days

Request a quote Full spec

SUPERCOMPUTERSSystemNo photograph on file

gpu serverRefurbished48h burn-in12mo warranty

TF-8L40S Inference Node

Processors: 2x AMD EPYC 9354
Accelerators: 8 x NVIDIA L40S 48GB
Cores: 64 total
Memory: 1TB DDR5
Storage: 10x 3.84TB U.2 NVMe
Fabric: 2x ConnectX-6 HDR200 + 2x 25GbE
Rack space: 4U
Peak power: 4.8kW · 21A at 230V
Heat rejected: 16,378 BTU/hr

384GB of GDDR6 ECC across eight cards
Air-cooled in 4U, no facility water needed
The sensible first production inference box

Ships with Ubuntu 22.04 LTS · CUDA 12 · vLLM · Triton Inference Server · Prometheus + Grafana

Quoted per order

60 working days

Request a quote Full spec

SUPERCOMPUTERSSystemNo photograph on file

gpu serverNew build48h burn-in36mo warranty

TF-8L40S Inference Node

Processors: 2x AMD EPYC 9354
Accelerators: 8 x NVIDIA L40S 48GB
Cores: 64 total
Memory: 1TB DDR5
Storage: 10x 3.84TB U.2 NVMe
Fabric: 2x ConnectX-6 HDR200 + 2x 25GbE
Rack space: 4U
Peak power: 4.8kW · 21A at 230V
Heat rejected: 16,378 BTU/hr

384GB of GDDR6 ECC across eight cards
Air-cooled in 4U, no facility water needed
The sensible first production inference box

Ships with Ubuntu 22.04 LTS · CUDA 12 · vLLM · Triton Inference Server · Prometheus + Grafana

Quoted per order

60 working days

Request a quote Full spec

SUPERCOMPUTERSSystemNo photograph on file

gpu serverRefurbished72h burn-in12mo warranty

TF-8H100 HGX Node

Processors: 2x Intel Xeon Platinum 8480+
Accelerators: 8 x NVIDIA H100 SXM5 80GB
Cores: 112 total
Memory: 2TB DDR5
Storage: 8x 7.68TB U.2 NVMe
Fabric: 8x ConnectX-7 NDR400 + 2x 100GbE
Rack space: 8U
Peak power: 10.2kW · 44A at 230V
Heat rejected: 34,802 BTU/hr

8-way NVLink all-to-all, 900 GB/s per GPU
Delivered racked, cabled and burned in for 72 hours
10.2kW peak, three-phase and real cooling are prerequisites, not options

Ships with Ubuntu 22.04 LTS · NVIDIA driver + CUDA 12 · NVIDIA Container Toolkit · Slurm · NCCL tuned

Quoted per order

60 working days

Request a quote Full spec

SUPERCOMPUTERSSystemNo photograph on file

gpu serverNew build72h burn-in36mo warranty

TF-8H100 HGX Node

Processors: 2x Intel Xeon Platinum 8480+
Accelerators: 8 x NVIDIA H100 SXM5 80GB
Cores: 112 total
Memory: 2TB DDR5
Storage: 8x 7.68TB U.2 NVMe
Fabric: 8x ConnectX-7 NDR400 + 2x 100GbE
Rack space: 8U
Peak power: 10.2kW · 44A at 230V
Heat rejected: 34,802 BTU/hr

8-way NVLink all-to-all, 900 GB/s per GPU
Delivered racked, cabled and burned in for 72 hours
10.2kW peak, three-phase and real cooling are prerequisites, not options

Ships with Ubuntu 22.04 LTS · NVIDIA driver + CUDA 12 · NVIDIA Container Toolkit · Slurm · NCCL tuned

Quoted per order

60 working days

Request a quote Full spec

None of these quite fits?

These are starting points, not a fixed menu. Open the configurator and change anything you like; the compatibility engine will hold you to what those changes cost in power, thermal budget and rack space.

Configure from scratch