github aleibovici/ollama-gpu-calculator v0.1.0
v0.1.0 (CALIB 2026.09)

latest release: v0.2.0
3 hours ago

First versioned release. It covers all work up to this date.

Added

  • CALIB 2026.09 update: expanded GPU catalog (99 GPUs), 29 Ollama model presets including MoE and cloud-only tags, KV cache type selection (f16, q8_0, q4_0) and context lengths up to 1M tokens (#12).
  • Light and dark theme toggle with a saved preference (#10).
  • Instrument-style UI redesign with a searchable GPU picker (May 2026).
  • Calculator pipeline extracted into testable modules with baseline regression tests (May 2026).
  • GPUs added over time: NVIDIA RTX 3060 8GB, RTX 3050, Tesla P4, Tesla M10, A2, L40S, L4, Tesla P100 and Apple M3 family; AMD RDNA4 consumer and workstation cards and Instinct CDNA2–CDNA4 (#6, #7).
  • GitHub Actions workflow that builds and deploys to GitHub Pages (March 2026).
  • MIT license (March 2026).
  • Power consumption estimates, system RAM checks and configuration warnings (2024).

Changed

  • Migrated from Create React App to Bun, Vite and Vitest (#8, February 2026).

Fixed

  • Google Analytics tag now ships with the GitHub Pages build (#13).
  • Single-GPU overflow and power calculation when the model fits on one GPU (#12).
  • VRAM, KV cache and tokens/sec now use standard transformer math (April 2026).
  • Apple unified memory handling, quantization-aware minimum RAM and power calculation (April 2026).
  • GPU TFLOPS values and accessibility issues found by Lighthouse (March 2026).

Don't miss a new ollama-gpu-calculator release

NewReleases is sending notifications on new releases.