Can the GeForce RTX 5090 run Recursive Flint 40B-A2.5B?
Yes. A new GeForce RTX 5090 ($6,500) runs Recursive Flint 40B-A2.5B at Q4_1 with room for about 168K tokens of context (its full 256K with 8-bit KV cache), at about 366 tok/s estimate. At $18 per tok/s it is the cheapest of the 2 priced GPUs that fit it.
Yes, with room for long context. On the NVIDIA GeForce RTX 5090 (32 GB) there is room for about 168K tokens of context with the default FP16 KV cache, or its full 256K with an 8-bit (q8_0) KV cacheestimate: what is left after the weights and a 1 GB runtime reserve. That is 32K tokens or more, enough for long documents and code. The Q4_1 GGUF of Recursive Flint 40B-A2.5B is 25.99 GB (source); with our 1.5 GB allowance for the runtime and a short context it needs about 27.5 GB.
Recursive Flint 40B-A2.5B fits 2 of the 18 GPUs we have a dated price for; the smallest of them has 32 GB. The GeForce RTX 5090's 1792 GB/s of memory bandwidth makes it the fastest of the 2, at ~366 tok/s estimate. Only 2.5B of its 39.8B parameters are active per token, so the estimate reads 1.63 GB per token rather than the whole 25.99 GB file, plus a per-token routing and expert-gather cost calibrated to 45 measured llama.cpp benchmarks (2025–26). On the same card, the mixture-of-experts NVIDIA Nemotron 3.5 Lightning 30B-A3B (3B of 31.58B active, 25.48 GB) estimates ~177 tok/s.
The cheapest way to run it per tok/s. At $6,500 (new) the GeForce RTX 5090 works out to $18 per tok/s; the next best is the RTX A6000 at $34. The cheapest complete build we priced that runs Recursive Flint 40B-A2.5B is our 32 GB: two Radeon RX 9070 build at $3,589.93.
Last updated Oct 16, 2026. Speed is a bandwidth-based estimate (1792 GB/s, NVIDIA), scaled by per-platform constants calibrated to 45 measured llama.cpp benchmarks (2025–26), not a benchmark of this pair. Max context is an estimate from the model's architecture for llama.cpp with flash attention.
Next size up: Ripple-65B Preview (38.8 GB at Q4_K_M) needs ~40.3 GB and does not fit on one NVIDIA GeForce RTX 5090. See what runs Ripple-65B Preview.
Every priced GPU that runs Recursive Flint 40B-A2.5B, by cost per token/sec
| GPU | VRAM | Speed estimate | Max context estimate | Low price | $ per tok/s |
|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 32 GB | ~366 tok/s | 168K · 256K (8-bit) | $6,500 (new) | $18 |
| NVIDIA RTX A6000 | 48 GB | ~243 tok/s | 256K (full) | $8,363 (new) | $34 |
Prices: picksbycard.com, 2026-11-02; see how prices are sourced. Prices move daily. Bar length shows value: longer is cheaper per tok/s.