Local AI Advisor

Can Desktop PC · Radeon RX 7900 XT · 32 GB RAM run Recursive Flint (40B-A2.5B)?

20 GB VRAM + 32 GB RAM · 800 GB/s memory bandwidth · Recursive Flint (40B-A2.5B) has 39.8B parameters (2.5B active per token).

Short answer

It loads, but it's borderline even for chat.

Best quantization for chat: MXFP4 (20 GB of weights) on Lemonade. Generation: 110–180 tok/s, leaving about 0.0 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 WebUIBorderlineMXFP4110–180 tok/sunder 1 s8KDetails
Coding questionswith ContinueBorderlineMXFP495–160 tok/s≈2 s16KDetails
Coding in a repositorywith AiderBorderlineMXFP470–120 tok/s≈7 s32KDetails
Agentic codingwith OpenCodeBorderlineMXFP465–110 tok/s≈10 s32KDetails
Long documents (64K)with Open WebUIBorderlineMXFP435–60 tok/s≈35 s64KDetails

Speeds are bandwidth-based estimates; no direct benchmark exists for this pair. “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.