Skip to content
3–4× NVIDIA RTX 6000 Ada Generation

Soika AI Workstation SM 6000

The most capable Ada system we build.

Four RTX 6000 Ada Generation GPUs deliver 1,457 TFLOPS of tensor performance and 960 GB/s of bandwidth per card — the ceiling of what a workstation platform can do before you move to HGX.

WorkstationSOIKASMRTX6000Ada
PWR

3–4× NVIDIA RTX 6000 Ada Generation

GPU memory
48 GB
Bandwidth
960 GB/s
Tensor
1,457 TFLOPS
CUDA cores
18,176

GPU memory

48 GB

Bandwidth

960 GB/s

Tensor

1,457 TFLOPS

CUDA cores

18,176

Key features

The highest-performing Ada configuration in the range.

Four RTX 6000 Ada GPUs, 18,176 CUDA cores each, with the Agent-to-Agent network option enabled.

  • 48 GB GDDR6 per GPU with ECC
  • 960 GB/s of memory bandwidth
  • 18,176 CUDA cores, 568 Tensor Cores, 142 RT Cores
  • 1,457.0 TFLOPS tensor performance
Specifications

The full configuration

Parts
3 years
Live service
3 years, engineer-led
Support tier
Standard 3-year; on-site service and SLA available through partners
Model
SOIKASMRTX6000Ada
Device type
3–4 GPU workstation, Xeon 60-core platform
Graphics
3–4× NVIDIA RTX 6000 Ada Generation
Processor
Intel® Xeon® w9-3595X — 60 cores / 120 threads
Memory
8× 64GB DDR5-4800 2Rx4 ECC RDIMM (512GB)
Storage
8TB NVMe
GPU memory
48 GB GDDR6
Memory interface
384-bit
Memory bandwidth
960 GB/s
Single-precision
91.1 TFLOPS
Tensor performance
1,457.0 TFLOPS
Board power
300 W per GPU
Network
2× 10GbE RJ45 + 1× management LAN
Included software

Licensed, installed and tuned before it ships

Soika hardware exists to remove the setup problem. GPU clustering, model serving and agent design are configured at the factory, not on your time.

Soika Stack

GPU clustering, LLM deployment and inference management, pre-configured and licensed.

Soika Mockingjay

No-code AI agent design and agent-fleet management with the Enterprise licence.

Ubuntu LTS

Hardened Linux base image with the full CUDA and container toolchain already in place.

Pre-loaded model catalogue

  • Qwen
  • Llama
  • DeepSeek
  • Mistral
  • MONAI
  • Meditron
  • BLOOM
  • Calme
  • 200+ optimised models
Where it earns its place

What teams run on these systems

The same box serves very different work depending on who owns it.

Healthcare

Record summarisation, diagnostic support, research and treatment planning.

Government

Citizen-service agents, smart-city systems and in-country model hosting.

Finance

Reporting, risk assessment, trading research and investment analysis.

Legal

Contract drafting, case-law summarisation and legal co-pilots.

Software

Code generation, debugging, application design and internal developer tooling.

Research & academia

Advanced data analysis, reasoning tasks, language and AI research.

E-commerce

Personalised recommendation and supply-chain intelligence.

Edge & autonomous systems

Local inference for robotics, drones and IoT without cloud access.

Order

Configure a SM 6000

Tell us the models you plan to run and the environment it will live in. We will confirm the configuration, power and cooling budget, and introduce the partner who will deliver and commission it.