Skip to content

Path

Getting startedGPU → ROCm → llama.cpp → optional harness
StacksDaily driver and fast-agent cards from our test rig
BenchmarksFirst-party tok/s from the R9700 rig — no leaderboard spam
BudgetStarter → Pro Radeon builds, no invented prices
CalculatorSize a model, ask what a Radeon can run, or read a speed floor — Q4+
QuantsQ4–Q8, UD dynamic, KV types, why not below Q4
Stack LabUbuntu companion — same catalog as the YAML

Stack

GPU directoryRadeon cards, gfx IDs, VRAM, official ROCm rows
ROCmWhat it is — stack, gfx IDs, Vulkan vs HIP, versions
ROCm problemsOpen problems and how you can help — cited only
BackendsInference engines compared — llama.cpp first
AI softwarellama.cpp first; vLLM and Strata as tries
AbliteratedHub: what it is, how to vet files, catalog — educational only
HarnessesWhat to plug your local model into — agents, chat UIs, ComfyUI
HardwarePSU, cooling, noise, VRAM vs model size

Practice

TutorialsPractical how-tos for this stack
TroubleshootingGPU not seen, wrong wheel, CPU fallback
Error pasteMatch a log against cited ROCm / Ollama patterns
UpdatesRSS + ROCm × llama.cpp × vLLM 0.31 matrix
FAQShort answers for a first home box
BlogLonger posts and ROCm version notes
ForumThreads, rig badges, solved marks — off unless FORUM_ENABLED
AccountOptional saved rigs and stack-update alerts
PrivacyWhat optional accounts store
DisclosureAmazon Associates, Rakuten, B&H — and what is not affiliate

Stacks Benchmarks Budget Calculator

Getting startedGPU → ROCm → llama.cpp → optional harness
StacksDaily driver and fast-agent cards from our test rig
BenchmarksFirst-party tok/s from the R9700 rig — no leaderboard spam
BudgetStarter → Pro Radeon builds, no invented prices
CalculatorSize a model, ask what a Radeon can run, or read a speed floor — Q4+
QuantsQ4–Q8, UD dynamic, KV types, why not below Q4
GPU directoryRadeon cards, gfx IDs, VRAM, official ROCm rows
ROCmWhat it is — stack, gfx IDs, Vulkan vs HIP, versions
ROCm problemsOpen problems and how you can help — cited only
BackendsInference engines compared — llama.cpp first
AI softwarellama.cpp first; vLLM and Strata as tries
AbliteratedHub: what it is, how to vet files, catalog — educational only
HarnessesWhat to plug your local model into — agents, chat UIs, ComfyUI
HardwarePSU, cooling, noise, VRAM vs model size
Stack LabUbuntu companion — same catalog as the YAML
TutorialsPractical how-tos for this stack
TroubleshootingGPU not seen, wrong wheel, CPU fallback
Error pasteMatch a log against cited ROCm / Ollama patterns
UpdatesRSS + ROCm × llama.cpp × vLLM 0.31 matrix
FAQShort answers for a first home box
BlogLonger posts and ROCm version notes
ForumThreads, rig badges, solved marks — off unless FORUM_ENABLED
AccountOptional saved rigs and stack-update alerts
PrivacyWhat optional accounts store
DisclosureAmazon Associates, Rakuten, B&H — and what is not affiliate

Account

Sign in

Accounts are off on this static build (ACCOUNTS_ENABLED=0). No database is used. See privacy.

amdai

R9700 Lab Notes

Independent living notes for hobbyists. Not an AMD publication. Version pins are dated snapshots. Product names belong to their owners.

Path

Getting started
Stacks
Benchmarks
Budget
Calculator
Quants
Stack Lab

Stack

GPU directory
ROCm
ROCm problems
Backends
AI software
Abliterated
Harnesses
Hardware

Practice

Tutorials
Troubleshooting
Error paste
Updates
FAQ
Blog
Forum
Account
Privacy
Disclosure

Independent, not affiliated with AMD. Refresh src/data/what-works.yaml when the frontier moves.Weekday scrape.Affiliate disclosure.Privacy.As an Amazon Associate I earn from qualifying purchases. Always confirm the live ROCm matrix.

Search the site

esc

Type to filter. Enter opens the first match. Full-text search lives on /search.