AI Workstations
Professional multi-GPU workstations for AI research, model training, and high-performance computing, engineered and supported by a team with 30 years of experience in HPC.
Modulus AI workstations are configured for the work we do every day: training and fine-tuning AI models and quantitative finance at scale, running local LLMs, and developing low-latency software, without sending data to the cloud.
Every system is designed, assembled, and tested by in-house engineers. We use only quality components and parts and we provide an ironclad warranty backed by US-based support for all of our systems.
Built for serious AI work
The Modulus Inferno™ AI Workstation is a high performance multi-GPU deep-learning system for training and fine-tuning, high-core-count systems.
Modulus uses only high-quality components, cooling, memory, and storage to keep our systems stable and fast through long training runs and heavy parallel workloads.
- Single- and multi-GPU configurations for deep learning
- High-core-count CPUs for backtesting and simulation
- Large, fast memory and NVMe storage for big datasets
- Quiet, thermally-tuned chassis for sustained workloads
- Local LLM hosting and inference, on your own hardware
- Linux or Windows, configured for your stack
Run bigger models locally
Running models locally means your proprietary data, signals, and model weights never leave your building. For research teams that treat their data as a trade secret, on-premise AI hardware is the difference between control and exposure.
- Processor: AMD Ryzen Threadripper Pro 7000 or 9000, configurable from 24 to 96 cores
- Memory: 64 GB to 1,024 GB DDR5 ECC, eight-channel across 8 DIMM slots
- GPU: 1 to 7 accelerators, mix and match NVIDIA H200 (141 GB), H100 (80 GB), RTX PRO 6000 Blackwell (96 GB), or RTX 5090 (32 GB)
- GPU interconnect: 7× PCIe Gen 5 slots (6× x16, 1× x8)
- Primary storage: PCIe Gen 5 x4 M.2 NVMe (2260, 2280, or 22110)
- Expansion storage: 2× MCIO and 1× SlimSAS connectors for additional NVMe or SATA drives
- Networking: dual 10 GbE LAN (Intel X710-AT2) standard
- Optional networking: 25 GbE or 100 Gbps InfiniBand HCAs
- Remote management: dedicated IPMI port for out-of-band administration
- Cooling: full custom liquid loop covering the CPU and every GPU
- Acoustics: up to 3× quieter than equivalent air-cooled systems
- Power: Titanium-grade PSUs configurable from 1,500 W to 6,000 W
- Input: universal 110-240 V AC, 50-60 Hz
- Chassis: configurable as desktop tower or 3U/4U rack-mount
- Desktop dimensions: 29.5" H × 10.6" W × 24.2" D
- Operating system: Linux or Windows, pre-configured for your workflow
Built and supported by Modulus
Workstation-class builds
Professional components engineered for stability and performance under sustained AI and quant workloads, not consumer hardware.
GPU acceleration
Single- or multi-GPU configurations for model training, fine-tuning, inference, and local large-language-model hosting.
AI & HPC tuning
High-core-count CPUs, large memory, and fast NVMe for backtesting, simulation, and high-throughput data processing.
On-premise control
Keep proprietary data, signals, and model weights entirely in-house, no cloud dependency, no data leaving your firm.
Pre-loaded AI stack
Optionally configured with the Modulus AI development environment, source code, skill files, and MCP servers.
Engineer support
Specified, assembled, tuned, and supported by Modulus in the United States, with a clear upgrade path to newer systems.
Configured for your stack
Let's compute.
Request an instant meeting or schedule a call with our team to discuss your custom AI workstation build.