Local AI Advisor

Can Desktop PC · GeForce RTX 4090 · 64 GB RAM run Recursive Flint (40B-A2.5B)?

24 GB VRAM + 64 GB RAM · 1008 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: 320–440 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 WebUIExcellentMXFP4320–440 tok/sunder 1 s8KDetails
Coding questionswith ContinueExcellentMXFP4300–410 tok/sunder 1 s16KDetails
Coding in a repositorywith AiderComfortableMXFP4250–340 tok/s≈1 s32KDetails
Agentic codingwith OpenCodeComfortableMXFP4230–320 tok/s≈2 s32KDetails
Long documents (64K)with Open WebUIBorderlineMXFP4160–220 tok/s≈4 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.