Dedicated Bare Metal Rental

"Stop fighting for shared instances. Get a dedicated node all to yourself."

Perfect for: Developers and ML engineers who know what they're doing.

Node Specifications

  • GPU: NVIDIA GeForce RTX 5090 (32GB GDDR7 VRAM)
  • Performance: 3,300+ AI TOPS (Int8)
  • Memory Bandwidth: 1TB/s+
  • System RAM: 64GB DDR7
  • Storage: 2TB NVMe SSD (7,000 MB/s)
  • Connectivity: High-Speed Fiber Uplink

Pre-Configured For

  • PyTorch & TensorFlow Training
  • LLaMA / Mistral / Falcon / Qwen Inference
  • 70B-parameter model inference (70GB+ context)
  • Stable Diffusion image generation
  • Custom CUDA workloads
$0.60 – $0.80 / hour

Privacy Promise: Single-tenant. No logging. Your model weights stay yours. Complete data isolation.

Custom LLM Fine-Tuning

"Turn your messy company data into a smart, private AI assistant."

Perfect for: Businesses with proprietary data but limited ML engineering resources.

We Handle

  • Data Prep: Clean & format PDFs, emails, docs for training
  • Training: LoRA / QLoRA fine-tuning on your chosen model
  • Model Selection: Llama-3, Mistral, Qwen, or Falcon
  • Delivery: Hand you `.gguf` or `.safetensors` files to run offline
  • Support: Direct access to data scientist for optimization

Use Cases

  • Legal contract analysis & review
  • Medical coding & clinical documentation
  • Internal HR & employee onboarding bots
  • Custom documentation Q&A systems
  • Domain-specific chatbots
Custom Quote (Contact us for details)

Data Guarantee: Your training data is never retained. Models are yours alone.

Why Troy Inference?

We don't compete on scale. We compete on expertise, privacy, and personal relationships.

🇺🇸 US-Based Sovereignty

Physically located in Troy, Michigan. Your data never leaves the United States. Perfect for ITAR, GDPR, and HIPAA-conscious workflows.

💾 The 32GB Advantage

Most rental cards max out at 24GB. Our RTX 5090 with 32GB allows larger batch sizes and 70B-parameter model inference that competitors can't handle.

🧑‍🔬 Scientist-Led

We aren't just a server farm. We're led by active data science practitioners who understand gradient accumulation, learning rate scheduling, and model optimization.

🔒 Zero Logging Policy

Single-tenant nodes. No shared resources. No monitoring dashboards. Your model weights, datasets, and prompts remain completely private.

Infrastructure Specs

Current Availability: Node-01 (RTX 5090 Premium Inference Node)

Hardware Configuration

  • GPU: GIGABYTE GeForce RTX 5090 Gaming OC (32GB GDDR7 VRAM)
  • CPU: High-core-count processor optimized for GPU workloads
  • System Memory: 64GB DDR7 System RAM
  • Storage: 2TB NVMe SSD (7,000 MB/s sequential read)
  • Connectivity: High-Speed Fiber Uplink (dedicated bandwidth)
  • Environment: Ubuntu Linux 22.04 LTS + Docker + PyTorch (pre-configured)

Get Started Today

Ready to scale your AI workloads with confidence?

Email us: inference@troyinference.com

For rental bookings, fine-tuning inquiries, or custom GPU workload optimization.