Local AI Advisor

Can Desktop PC · Radeon AI PRO R9700 · 64 GB RAM run Recursive Flint (40B-A2.5B)?

32 GB VRAM + 64 GB RAM · 640 GB/s memory bandwidth · Recursive Flint (40B-A2.5B) has 39.8B parameters (2.5B active per token).

Short answer

Yes, excellent for chat and comfortable or better for every workload we test.

Best quantization for chat: MXFP4 (20 GB of weights) on Lemonade. Generation: 250–340 tok/s, leaving about 10 GB free for your other apps.

How it rates for each workload

Same model and machine, different jobs. Agents and long documents send far bigger prompts than chat, so they need much more speed and memory.

WorkloadVerdictQuantGenerationFirst replyContext
Chatwith Open WebUIExcellentMXFP4250–340 tok/sunder 1 s8KDetails
Coding questionswith ContinueExcellentMXFP4240–320 tok/sunder 1 s16KDetails
Coding in a repositorywith AiderComfortableMXFP4200–270 tok/s≈2 s32KDetails
Agentic codingwith OpenCodeComfortableMXFP4180–250 tok/s≈3 s32KDetails
Long documents (64K)with Open WebUIExcellentMXFP4130–170 tok/s≈8 s64KDetails

Speeds are estimates calibrated against measured benchmarks on similar hardware. “First reply” is the time to read the workload’s opening prompt. How we calculate

Different tool, context size or workload?

These pages use typical settings. The full check models your exact setup.