Real servers.
Real prices.
No fluff.

A right-sized server for a single-owner firm. Host the practice website, an intake form, and a scheduling page on one tidy box that costs less than a phone line.

• 1× IPv4 + IPv6
• Snapshot included
• Root access

Same shape as Starter, with extra RAM headroom for the days your contact form blows up after a referral. Built for firms running both a marketing site and a small client portal.

• 1× IPv4 + IPv6
• Daily snapshot
• Root access

The everyday server for a small firm running a real client app — case-management dashboards, billing tools, the Postgres + Redis + API trio that actually drives the practice.

• Dedicated 1 Gbps port
• Auto backups
• DDoS protection

For a firm whose internal tools have outgrown a single VM. Postgres, Elasticsearch, Redis, and a Node API can all stay warm without elbowing each other.

• Dedicated 1 Gbps
• Auto backups
• Optional managed Postgres

Sixteen cores and sixty-four gigs of RAM — the "I never want to think about CPU on demo day" tier. Good fit for a multi-tenant SaaS the firm sells alongside its services.

• Dedicated 1 Gbps
• Hardware RAID
• IPMI access

The top of the shared-compute ladder before moving to bare metal. Built for the firm that runs a real product alongside its core practice and needs predictable throughput under load.

• Dedicated 1 Gbps
• Hot snapshots
• Priority support

Cheap, durable bulk SSD for the firm’s off-site backups, closed-matter archives, and document warehouses. Restic-friendly, snapshot-rollback ready, costs less than a filing cabinet.

• SFTP / SMB / NFS
• Snapshot rollback
• Auto-expand option

Step up when the firm’s archives cross 300 GB — practice document libraries, scanned client correspondence, and shared-team Nextcloud all fit comfortably.

• SFTP / SMB / NFS
• Snapshot rollback
• Versioned backups

A full terabyte of SSD for the firm that retains everything: matter files, recorded depositions, CCTV from the office. Veeam + Restic friendly, gigabit uplink for fast restores.

• 1 Gbps port
• NFS / SMB / SFTP
• Restic-friendly

Storage plus enough RAM to actually do something with it — deduplicate scanned files, run OCR over old depositions, process video evidence on the same machine that holds it.

• 1 Gbps port
• CPU headroom
• Daily snapshots

The top storage tier with serious compute alongside it. For the firm running forensic image analysis, multi-camera ingest, or dedicating one box to its discovery pipeline.

• 1 Gbps port
• 50 GB RAM
• Hardware RAID

Your firm’s own physical machine, nothing shared. Twelve cores at 3.7 GHz hosting a self-managed Postgres, a trading sandbox, or an internal dev cluster the partners actually trust.

• Bare metal
• IPMI included
• Dual NVMe option

Server-grade EPYC with ECC RAM and RAID-1 SSD. The "grown-up" tier for firms with compliance commitments or production databases that cannot lose a transaction.

• ECC memory
• RAID 1 included
• IPMI / remote console

Thirty-two cores of pure 4.2 GHz boost. Compile the firm’s internal apps and serve them from the same box without anyone noticing. NVMe RAID handles peak I/O.

• NVMe RAID
• 4.2 GHz boost
• Hot-swap drives

Enterprise bare metal: 64 cores, 192 GB RAM, dual NVMe. For the firm running its own data warehouse, multi-tenant SaaS, or heavy nightly ETL on closed-matter data.

• Hardware firewall option
• Dedicated 10 Gbps upgrade
• Compliance-ready

The cheapest GPU-attached box we sell. For the firm associate learning CUDA on the clock, a hardware video encoder for marketing reels, or a headless display server for a remote desk.

• CUDA-capable
• Bare hypervisor allocation
• Snapshot included

Newest-gen consumer GPU slice for firms experimenting with AI workflows on a budget. Entry image generation, small Q4 LLMs, in-house chat assistants for partners.

• Blackwell-gen
• GDDR7
• Virtualized GPU slice

Entry-class dedicated GPU for the firm’s first real AI workload. Run an internal image generator, batch up speech-to-text for old depositions, or stand up a small detection model.

• CUDA + cuDNN preinstalled
• Docker ready
• Jupyter optional

Modern 8 GB GPU on a 24-core host, bare-metal dedicated. For the firm putting an AI feature into production where shared slices won’t cut it.

• Blackwell + GDDR7
• 24-core Platinum host
• Bare-metal dedicated

Entry-tier Pascal GPU on dedicated metal. Light CUDA acceleration for visualization plugins, video transcode, low-power lab work.

• Pascal architecture
• Low-power footprint
• Bare-metal dedicated

Quadro-class entry GPU, perfect for the firm running a CAD viewport, a small inference loop, or an image preprocessor next to its main practice apps.

• Quadro drivers
• Bare-metal dedicated
• OpenGL + CUDA

A pair of older datacenter dies giving the firm 24 GB combined VRAM at a budget price. Great for batch CUDA jobs, legacy ML pipelines, bulk transcoding overnight.

• Dual K80 dies
• 24 GB combined VRAM
• Bare-metal dedicated

Turing-gen workstation GPU with a modern driver stack. Low TDP, virtualization-friendly, suited for a firm running multi-monitor compute or a workstation VM.

• Turing architecture
• Low TDP
• Quadro drivers

Affordable consumer Turing card on dedicated metal. For the principal who wants a real CUDA lab outside the firm’s production environment.

• Turing architecture
• CUDA + RTX optional
• Bare-metal dedicated

Six gigs of GDDR5 paired with a dual-Xeon host — a solid entry-image-gen + 7B Q4 inference rig for the firm’s in-house experiments.

• 6 GB GDDR5
• Dual-Xeon host
• Bare-metal dedicated

RTX-class GPU on a 128 GB host. RT + Tensor cores give the firm a real entry-level production card for refiner pipelines and small LLM serving.

• RT + Tensor cores
• 128 GB host RAM
• Bare-metal dedicated

Datacenter Pascal P100 with 16 GB HBM2 and serious FP64 throughput. For firms doing scientific compute, structural simulation, or reference training runs.

• HBM2 memory
• FP64 strong
• Datacenter pedigree

Modern Ada architecture with DLSS 3 — a solid all-rounder for the firm running production inference, render workloads, and modern compute jobs on the same box.

• Ada architecture
• DLSS 3
• Bare-metal dedicated

RTX 2060 paired with a 40-core host. For the firm running hybrid CPU + GPU pipelines — batch ML preprocessing followed by accelerated inference on one machine.

• 40-core host
• RT + Tensor cores
• Bare-metal dedicated

Datacenter V100 with 32 GB HBM2 and Tensor Cores. Still a workhorse for mid-size firm training jobs and 13B LLM inference at predictable cost.

• 32 GB HBM2
• Tensor Cores Gen1
• NVLink-ready

Ampere RTX 3060 Ti with strong tensor throughput per dollar. Solid for an in-house image generator, a 7B-class LLM endpoint, or one node in a small render farm.

• Ampere architecture
• Tensor Cores Gen3
• 2 TB storage

Ampere RTX A5000 pro card with 24 GB ECC and workstation-class drivers. For the firm running production inference workloads that have to behave.

• 24 GB GDDR6 ECC
• NVLink
• Quadro-class drivers

RTX A4000 on dedicated metal — 16 GB ECC VRAM, NVLink, full pro driver stack. The firm’s "always-on" image-gen or LLM endpoint.

• 16 GB GDDR6 ECC
• NVLink
• Bare-metal dedicated

Modern 16 GB workstation card with ECC VRAM and Quadro-class drivers. Built for firms running 3D rendering, CAD/CAM, and light inference next to their core practice apps.

• ECC VRAM
• Quadro-class drivers
• OpenGL + Vulkan

16 GB VRAM puts the firm into real ML territory: fine-tune small image-gen models, run 7B LLMs at usable speed, or process video diffusion in batches.

• 16 GB GDDR6
• PyTorch / TF / JAX ready
• NVLink option

Twenty-four gigs of pro-grade VRAM and 56 GB system RAM — the sweet spot for firms running a production inference endpoint with predictable SLA expectations.

• Quadro-class driver
• 24/7 SLA option
• Tier-1 datacenter

Forty-eight gigs of modern workstation VRAM. For professional design firms running CAD + BIM + 3D modeling for client work, with an in-house AI assistant warm on the same box.

• 24 GB GDDR6
• NVLink-ready
• Multi-instance GPU

Latest-gen consumer flagship in a workstation chassis. Brutal throughput for the firm running heavy image-gen pipelines or real-time video stylization.

• 32 GB GDDR7
• 1 Gbps uplink
• Liquid-cooled chassis

Dedicated bare-metal 24 GB GPU with 256 GB system RAM and triple-tier storage. For the firm running a private LLM endpoint or a long-running internal training job.

• Dedicated GPU
• 256 GB system RAM
• Triple-tier storage

Forty-eight gigs of ECC VRAM on a dedicated chassis. The firm’s honest tier for "chat + media warm at the same time" — multi-modal inference without the elbowing.

• 48 GB GDDR6 ECC
• NVLink
• 24/7 dedicated

Ninety-six gigs of modern ECC VRAM — the workstation flagship for the firm running multi-modal pipelines, AI video generation, or in-house fine-tunes.

• 96 GB GDDR7 ECC
• 1 Gbps uplink
• Workstation drivers

Bare-metal flagship GPU with 256 GB host RAM and triple-tier storage. For firms standing up an "always-on" AI service for their own product or internal teams.

• Latest-gen flagship
• 256 GB host RAM
• Triple-tier storage

Datacenter-class card with ECC + RAS and MIG-partition support. For firms running an inference farm or carving one box into multiple tenant slices.

• Datacenter card
• ECC + RAS
• Up to 7 MIG instances

Datacenter A100 — the reference card every ML team learns on. For firms running serious training, 13-30B LLM inference, or diffusion at scale on dedicated metal.

• Tensor Cores Gen3
• HBM2e memory
• 600 GB/s NVLink

Eighty gigs of HBM2e — fits the largest open-weights LLMs unquantized and keeps the firm’s entire media stack warm together. Datacenter SLA-grade.

• 80 GB HBM2e
• Full A100 architecture
• Datacenter SLA

The flagship. Hopper architecture, Transformer Engine, 3 TB/s HBM3 — best single-GPU performance available. For firms training their own model or running real-time 70B inference.

• Hopper architecture
• Transformer Engine
• 3 TB/s HBM3

Two 24 GB cards side by side — parallel inference or model + media warm on separate cards. For firms running A/B model serving or dev + prod on one box.

• 2× consumer flagship
• Dual PCIe Gen4
• CUDA peer-to-peer

Three datacenter cards giving the firm 96 GB combined HBM2. Built for distributed training, mid-size fine-tunes, or carving the box into multi-tenant inference slots.

• 3× datacenter V100
• HBM2 across all
• NVLink option

Three pro Ampere cards with 72 GB combined ECC VRAM. A great mid-tier training cluster for firms running production fine-tunes or multi-team inference.

• 3× pro Ampere
• 72 GB ECC total
• NVLink-ready

Two newest-gen 32 GB cards on a 44-core host, liquid-cooled. For firms running heavy training, fast 70B-class serving, or dual-tenant isolation under one roof.

• Latest-gen pair
• 44 cores host
• GPUDirect ready

Three pro cards giving the firm 144 GB combined VRAM — more headroom than dual A100 80 GB at lower cost. NVLink-capable, ECC across all cards.

• 3× pro cards
• NVLink possible
• ECC across all

Quad-GPU bare metal with 192 GB VRAM and 512 GB host RAM. For firms running internal foundation training or a multi-team shared compute pool.

• Quad-GPU
• 512 GB host RAM
• 20 TB storage

Training-cluster grade. Four datacenter A100 80 GB cards with 320 GB combined VRAM — enough headroom for a firm to train its own foundation model.

• 4× datacenter A100
• NVLink Bridge
• Optional InfiniBand