Generated Computers v1 (GC-v1) is our flagship, bespoke workstation engineered specifically for running state-of-the-art AI models locally without the latency or privacy concerns of cloud APIs.

Unlike traditional gaming PCs or standard enterprise servers, the GC-v1 architecture treats memory bandwidth as the primary bottleneck for Large Language Models. By utilizing a custom motherboard topology, we are able to pool up to 192GB of unified VRAM across multiple discrete accelerators, acting as a single logical device.

Core Specifications

• Quad-Neural Processing Units (NPUs) running in parallel.
• Proprietary ultra-high-bandwidth interconnect bridge (1.2 TB/s).
• Whisper-quiet thermals: Custom phase-change cooling loop allowing 100% sustained utilization under 35dB.
• Out-of-the-box support for PyTorch, MLX, and our proprietary Waverider inference engine.

The GC-v1 isn't just hardware; it's a paradigm shift in how developers and enterprises deploy localized intelligence.

Waverider is our proprietary, bare-metal inference engine designed to squeeze every ounce of performance out of local hardware. It is the software soul of Generated Computers.

Most open-source inference engines rely on layers of abstraction that introduce latency. Waverider bypasses the standard OS scheduler, communicating directly with the silicon. This allows for dynamic KV-cache management and zero-copy memory transfers between the CPU and NPU.

Performance Metrics

When running Llama-3 70B (4-bit quantized) on the GC-v1 using Waverider, we consistently observe:

• Time To First Token (TTFT): < 45ms
• Generation Speed: 215 tokens per second
• Context Processing: 128k tokens in under 4 seconds via FlashAttention-3 integration.

Waverider comes pre-installed on all GC-v1 workstations and includes a seamless API drop-in replacement for OpenAI endpoints, meaning your existing applications work instantly on local hardware.