AI Briefing6900 XT ROCm vs Vulkan Comparison
Key point
A llama.cpp benchmark comparing ROCm 6.4.2 and Vulkan on the 6900 XT was published.
Details
On the AMD Radeon RX 6900 XT, llama.cpp was built with ROCm 6.4.2 and the latest Vulkan to compare speeds.
For Gemma 4 E2B Q4_K, pp512 was faster with Vulkan at ubatch 64/128/512/1024, while ROCm was ahead at 32/256. For tg128, Vulkan led across the entire range, recording 173.35~174.59 tok/s, while ROCm was at around 151.41~151.92 tok/s.
For Qwen 3.5 4B Q8_0, pp512 favored ROCm across all ubatch values. The gap was large: 1368.32 vs 706.18 at ubatch 32, and 3019.70 vs 2386.97 at ubatch 1024.
For tg128, Vulkan also led across the entire range on Qwen 3.5. Vulkan ranged from 88.46~88.58 tok/s, while ROCm ranged from 77.57~77.97 tok/s.
In summary, on this 6900 XT setup, Vulkan was consistently strong in generation speed, while ROCm showed higher throughput in prompt processing for Qwen 3.5.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Why LLM Research Agents Generalize Without Benchmark Overfitting: 'Compressibility' Is Key
ArtificialAnalysis's LLM Intelligence-Cost Graph Misleads with Log Scale and Data Center Pricing