Engineering notes

Technical deep dives from the systems we build. Local inference, GPU benchmarks, architecture trade-offs, and field reports.

Quantization Explained: VRAM, Quality, and the RAM Performance Cliff

Apr 4, 2026

What quantization does to model quality and speed, how Q4/Q5/Q8 compare on real hardware, and why running a model split between GPU memory and system RAM can be 5-30x slower. Third-party benchmarks, practical recommendations.

BenchmarksquantizationGGUFQ4_K_MVRAMRAM offloadinglocal LLM performanceGPU memory8 min read

Qwen 3.5 Benchmarks: 0.8B to 14B on RDNA4 Vulkan

Mar 31, 2026

Qwen 3.5 benchmarks across 80 prompts on AMD RDNA4 16GB with Vulkan. Unconstrained (2048-token) results show qwen3:14b leading at 0.594, with 9B pulling ahead of 4B when not artificially truncated. Includes analysis of how token budget constraints distort benchmark rankings.

BenchmarksQwen 3.5qwen 3.5 benchmarksqwen 3.5 performanceRDNA4 LLM benchmarksVulkanOllamaRDNA4LLMlocal LLM benchmarks10 min read

LLM Benchmark Comparison: 12-Source Rankings

Mar 26, 2026

Cross-referenced LLM benchmark data from 12+ sources. Consensus rankings for coding, math, reasoning, tool calling. Interactive charts, cost vs. quality scatter plots, and full methodology.

BenchmarksLLM benchmarksLLM benchmark comparisonmulti-model routingClaudeGeminiGPTmodel rankingbest LLM for coding12 min read

How We Price: Aggregate Component Costs by Scrape

Mar 26, 2026

We publish the exact component costs, eBay search queries, retail sources, and markup formula for every Foundry GPU server. Prices are scraped from live market data and fed into an auto-pricing engine. This is the full breakdown.

ArchitecturePricingFoundryTransparencyGPUHardware6 min read

ROCm HIP Idle GPU Bug on RDNA4: Vulkan Workaround

Mar 22, 2026

When loading Qwen 3/3.5 models via Ollama using ROCm/HIP on AMD RDNA4 GPUs, the GPU becomes permanently stuck at 100% utilization. We trace the root cause to HSA runtime teardown and benchmark the Vulkan backend as a drop-in fix with comparable performance.

ToolsAMDROCmVulkanOllamaRDNA4GPU8 min read