Local AI Advisor

Can Desktop PC · GeForce RTX 3090 · 64 GB RAM run Recursive Flint (40B-A2.5B)?

24 GB VRAM + 64 GB RAM · 936 GB/s memory bandwidth · Recursive Flint (40B-A2.5B) has 39.8B parameters (2.5B active per token).

Short answer

Yes, excellent for chat; borderline for 64K-token documents.

Best quantization for chat: MXFP4 (20 GB of weights) on Ollama. Generation: 280–370 tok/s, leaving about 2.1 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 WebUIExcellentMXFP4280–370 tok/sunder 1 s8KDetails
Coding questionswith ContinueExcellentMXFP4260–350 tok/sunder 1 s16KDetails
Coding in a repositorywith AiderComfortableMXFP4210–290 tok/s≈2 s32KDetails
Agentic codingwith OpenCodeComfortableMXFP4200–270 tok/s≈3 s32KDetails
Long documents (64K)with Open WebUIBorderlineMXFP4140–180 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.