RPW-495: Testing as of May 2025 by AMD. Average tokens per second of three runs, dropping edge cases where the model starts spiraling (more than 2k thinking tokens) to standardize response length. No speculative decode. All tests conducted on LM Studio 0.3.15 (Build 11). Vulkan Llama.cpp 1.28 used for AMD, NVIDIA-recommended CUDA 12 llama.cpp 1.30 with Flash Attention used for NVIDIA. Short Prompt:  "How long would it take for a ball dropped from 10 meter height to hit the ground?“ Long Prompt: “Summarize the following in exactly five lines: [Insert Scene 1 Act 1 of Romeo and Juliet]”,  Models tested: Phi 3.5 MoE Q4 K M, Mistral Small 3.1 24B Instruct 2503 Q8, DeepSeek R1 Distill Qwen 32B Q6, Qwen 32b Q6 System specifications: AMD Ryzen™ 9 7900X, 32GB DDR5 RAM 6000 MT/s, Windows 11 PRO 24H2, AMD Radeon™ AI PRO R9700 32GB using Adrenalin 25.6.1 RC vs AMD Ryzen™ 9 7900X, 32GB DDR5 RAM 6000 MT/s, Windows 11 PRO 24H2 with NVIDIA GeForce RTX 5080 and GeForce 576.4 drivers. Performance may vary. RPW-495.

RPW-496: Testing as of May 2025 by AMD using DeepSeek R1 Distill Qwen 32B Q6, Mistral Small 3.1 24B Instruct 2503 Q8, Flux.1 Schnell, SD 3.5 Medium models. Tested on a System with AMD Ryzen 9 7900X CPU, Radeon AI PRO R9700 GPU, 32GB DDR5 RAM, 1TB Storage, Windows 11 PRO 24H2, Adrenalin 25.6.1 RC drivers, ComfyUI - PyTorch 2.4 on Windows. System configurations may vary yielding different results. RPW-496

GD-239a: Radeon™ PRO W6000 and W7000 Series and Radeon™ AI PRO R9000 Series graphics cards (and later models) are not designed nor recommended for datacenter usage. Use in a datacenter setting may adversely affect manageability, efficiency, reliability, and/or performance.