Workstations

High-performance AI compute that scales with you

From RTX 5090 workstations to GB10 / DGX Spark systems that chain to frontier scale. Engineered for sustained load, specced to your models, built to grow.

Request a build

Frontier scale on your desk

Chain GB10 units to run 405B-class models

A single GB10 / DGX Spark unit delivers up to 1 PetaFLOP of FP4 compute and 128 GB of coherent unified memory. Link two units over ConnectX-7 at 200 Gbps and run models up to 405 billion parameters. Near-silent, 240W per unit, desktop-sized.

Buy the capability you need now. Add units as models grow. No forklift upgrade, no re-buy.

Two chained workstation nodes with interconnect

One GB10 unit - up to 200B

Start with one GB10 unit, chain a second over ConnectX-7 for 405B-class models, then scale the cluster. Add hardware as the models grow.

Close-up of matte black chassis cooling and ports

Built for sustained load

Engineering over theatre

Panel gaps, airflow, and power delivery tuned for long runs. Status LEDs, not RGB. Specced to the model; stress-tested before it ships.

Specifications

Match hardware to workload.

Workload	GPU	VRAM	Memory
Local LLM inference 7B-32B, quantized	RTX 5090	24 GB	64-128 GB
Agentic workflows 24/7 automation	RTX 5090	24 GB	64-128 GB
Large-model inference up to 200B params	GB10 / DGX Spark	128 GB unified	128 GB
Frontier models up to 405B, chained	GB10 x2 (ConnectX-7)	256 GB unified	256 GB
Training / fine-tune multi-GPU	RTX 6000-class / multi	48 GB+	128-512 GB

Guidance only. Frontier configs chain GB10 / DGX Spark units over ConnectX-7. Final spec depends on model size, quantization, context length, throughput, cooling, and availability.