Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

llama.cpp on gfx1030 — Overview

llama.cpp is a fast, low-dependency way to run GGUF LLMs on gfx1030. This section covers stock builds and the community RDNA2-optimized fork.

Where to start

GoalPage
Build stock llama.cpp (ROCm or Vulkan)Building & running
Multi-GPU tensor parallel, MMQ tuning, DFlash2RDNA2 fork overview
Fork benchmarks and PR statusRDNA2 benchmarks
DFlash2, MTP, ngram speculative decodingRDNA2 speculative decoding
Launch commands, Docker, limitsRDNA2 serving
Something brokellama.cpp troubleshooting

Which path?

  • Stock llama.cpp — good starting point, single-GPU. Vulkan without ROCm is possible but #llamacpp prefers HIP for --split-mode tensor (RADV has crashed V620 hosts).
  • edwinbrowwn/llama.cpp-rdna2 — multi-GPU V620 rigs, tensor parallel, RCCL all-reduce, DFlash2. Most #llamacpp performance work happens here. A matched A/B on Qwen3.8-27B Q6_K + MTP reported +57% vs stock with identical outputs — see Benchmarks.

llama.cpp vs vLLM on V620 (#llamacpp / #vllm-rdna)

Community rule of thumb (Aug 2026):

WorkloadPrefer
Getting ROCm + multi-GPU working; GGUF; single-stream / low concurrencyllama.cpp (RDNA2 fork) — more battle-tested on V620
Multi-stream / agentic loads with prefix cachingvLLM -extras — caching + concurrency usually win
Qwen3.8 Flash-Next on 4× V620Prefer vLLM Flash-Next recipe (~60–100+ t/s decode class after Sep 2026 prefill work; earlier ~50–60) over llama.cpp (~15–35 t/s typical; 2-card APEX GGUF ~26 t/s) — see vLLM overview and Flash-Next on llama.cpp
MoE / lighter agenticEither — community sweet spot for these cards; see What fits well

Neither stack is “finished” for every model. New to the cards? Start with RDNA2 serving, then try vLLM when you need concurrency.

Multi-GPU tensor parallel benefits greatly from PCIe P2P.