Which local LLMs can the GeForce RTX 5090 run?
Assumes 32 GB of DDR5-5600 system RAM, Windows with this GPU driving the display, 8k context and an f16 KV cache.
Specs
- VRAM
- 32 GB
- Memory bandwidth
- 1,792.0 GB/s
- FP16 compute
- 104.8 TFLOPS
- Launch year
- 2025
- Price
- $4,200 street (as of 2026-08-09)
Verdicts at a glance
At Q4_K_M (or the closest available quant) with 8k context.
- Runs great24 modelsRuns well0 modelsRuns slowly1 modelWon't run4 models1 more model runs at a lower quant.
How to read the verdicts
- Runs great
- Fully on the GPU at 20 tok/s or more
- Runs well
- Fully on the GPU at 8–20 tok/s; MoE experts in system RAM at 20 tok/s or more; or at least 90% on the GPU at 8 tok/s or more
- Runs slowly
- 2–8 tok/s; CPU-only; less than 90% on the GPU; or MoE experts in system RAM below 20 tok/s
- Won't run
- Does not fit, or under 2 tok/s
Every model on the GeForce RTX 5090
Dense models split between GPU and CPU slow down sharply — the CPU side sets the pace. MoE models that keep only their experts in system RAM degrade far more gently.
Scroll sideways to see every column.
| Model | Verdict | Speed | Quant | Memory | Runs as | Context | Notes |
|---|---|---|---|---|---|---|---|
| EXAONE 4.0 1.2B | Runs great | est. 955.1 tok/s840.5–1,069.8calibrated estimate ±12% | Q4_K_M | 2.5 / 32.0 GB | Full GPU | up to 64k | — |
| HyperCLOVA X SEED 1.5B | Runs great | est. 691.0 tok/s608.1–773.9calibrated estimate ±12% | Q4_K_M | 3.0 / 32.0 GB | Full GPU | up to 16k | — |
| gpt-oss-20b | Runs greatDetails | 282.3 tok/s8k estimate 235.0 tok/s (188.0–282.0, calibrated estimate ±20%)measured (1 run, 2k context) | MXFP4 | 13.5 / 32.0 GB | Full GPU | up to 128k |
|
| Gemma 4 26B-A4B | Runs great | est. 235.0 tok/s188.0–282.0calibrated estimate ±20% | Q4_K_M | 18.5 / 32.0 GB | Full GPU | up to 256k | — |
| Kanana 1.5 15.7B-A3B | Runs great | est. 235.0 tok/s188.0–282.0calibrated estimate ±20% | Q4_K_M | 12.7 / 32.0 GB | Full GPU | up to 32k | — |
| Qwen3 30B-A3B (2507) | Runs greatDetails | est. 235.0 tok/s188.0–282.0calibrated estimate ±20% | Q4_K_M | 20.5 / 32.0 GB | Full GPU | up to 64k | — |
| Qwen3.5 35B-A3B | Runs great | est. 235.0 tok/s188.0–282.0calibrated estimate ±20% | Q4_K_M | 24.0 / 32.0 GB | Full GPU | up to 256k | — |
| DeepSeek R1 Distill Llama 8B | Runs great | est. 209.3 tok/s184.2–234.4calibrated estimate ±12% | Q4_K_M | 7.2 / 32.0 GB | Full GPU | up to 128k |
|
| Kanana 1.5 8B | Runs great | est. 209.3 tok/s184.2–234.4calibrated estimate ±12% | Q4_K_M | 7.2 / 32.0 GB | Full GPU | up to 32k | — |
| Llama 3.1 8B | Runs greatDetails | est. 209.3 tok/s184.2–234.4calibrated estimate ±12% | Q4_K_M | 7.2 / 32.0 GB | Full GPU | up to 128k | — |
| Qwen3.5 9B | Runs great | est. 204.4 tok/s179.8–228.9calibrated estimate ±12% | Q4_K_M | 7.3 / 32.0 GB | Full GPU | up to 256k | — |
| Qwen3 8B | Runs great | est. 201.1 tok/s177.0–225.2calibrated estimate ±12% | Q4_K_M | 7.4 / 32.0 GB | Full GPU | up to 32k | — |
| Gemma 4 12B | Runs great | est. 163.8 tok/s144.2–183.5calibrated estimate ±12% | Q4_K_M | 8.8 / 32.0 GB | Full GPU | up to 256k | — |
| Gemma 3 12B | Runs greatDetails | est. 160.1 tok/s140.9–179.3calibrated estimate ±12% | Q4_K_M | 9.0 / 32.0 GB | Full GPU | up to 128k | — |
| HyperCLOVA X SEED Think 14B | Runs great | est. 123.0 tok/s86.1–160.0theoretical estimate ±30% | Q4_K_M | 11.4 / 32.0 GB | Full GPU | up to 128k |
|
| Qwen3 14B | Runs greatDetails | est. 121.3 tok/s106.7–135.8calibrated estimate ±12% | Q4_K_M | 11.5 / 32.0 GB | Full GPU | up to 32k | — |
| Phi-4 | Runs great | est. 116.9 tok/s102.9–131.0calibrated estimate ±12% | Q4_K_M | 11.9 / 32.0 GB | Full GPU | up to 16k |
|
| Mistral Small 3.2 24B | Runs great | est. 80.0 tok/s70.4–89.6calibrated estimate ±12% | Q4_K_M | 16.8 / 32.0 GB | Full GPU | up to 64k | — |
| Gemma 3 27B | Runs great | est. 72.8 tok/s64.1–81.6calibrated estimate ±12% | Q4_K_M | 18.4 / 32.0 GB | Full GPU | up to 128k | — |
| Qwen3.5 27B | Runs great | est. 71.1 tok/s62.6–79.6calibrated estimate ±12% | Q4_K_M | 18.8 / 32.0 GB | Full GPU | up to 128k | — |
| EXAONE 4.0 32B | Runs great | est. 63.1 tok/s55.5–70.7calibrated estimate ±12% | Q4_K_M | 21.1 / 32.0 GB | Full GPU | up to 128k |
|
| EXAONE 4.5 33B | Runs great | est. 60.9 tok/s53.6–68.2calibrated estimate ±12% | Q4_K_M | 21.8 / 32.0 GB | Full GPU | up to 128k |
|
| Qwen3 32B | Runs great | est. 57.3 tok/s50.4–64.1calibrated estimate ±12% | Q4_K_M | 23.1 / 32.0 GB | Full GPU | up to 32k | — |
| DeepSeek R1 Distill Qwen 32B | Runs great | est. 57.0 tok/s50.2–63.9calibrated estimate ±12% | Q4_K_M | 23.2 / 32.0 GB | Full GPU | up to 32k |
|
| Llama 3.3 70B | Runs slowly | est. 2.7 tok/s2.2–3.3calibrated estimate ±20% | Q4_K_M | 32.0 / 32.0 GB + 14.4 GB RAM | Partial offload | up to 16k |
|
| Qwen3.5 122B-A10B | Won't runTry UD-Q2_K_XL (heavy quality loss): Runs slowly | est. 33.3 tok/s23.3–43.2theoretical estimate ±30% | UD-Q2_K_XL | 32.0 / 32.0 GB + 12.2 GB RAM | Partial offload | up to 256k |
|
| gpt-oss-120b | Won't runDetails | — | MXFP4 | needs 64.3 GB | — | — |
|
| Solar Open 100B | Won't run | — | Q4_K_M | needs 64.4 GB | — | — |
|
| Solar Open 2 250B | Won't run | — | IQ4_XS | needs 137.2 GB | — | — |
|
Reasoning models spend extra tokens thinking, so their speed thresholds are 1.5× stricter (30 / 12 / 3 tok/s).
Measured results on the GeForce RTX 5090
Public benchmarks we calibrate against. Their conditions (context, backend, flags) can differ from the estimates above.
| Model | Quant | Backend | Context | Prompt (tok/s) | Generation (tok/s) | Flags | Source | Measured |
|---|---|---|---|---|---|---|---|---|
| llama-2-7b | Q4_0 | llama.cpp | 512 | 14,073.0 | 290.0 | — | github.com | 2025-08-01 |
| gpt-oss-20b | MXFP4 | llama.cpp | 2k | 9,841.0 | 282.3 | — | github.com | 2025-08-15 |
| Qwen3 8B | Q4_K_XL | llama.cpp | 16k | — | 145.3 | — | hardware-corner.net | 2026-08-09 |
| Qwen3 30B-A3B (2507) | Q4_K_XL | llama.cpp | 16k | — | 141.6 | — | hardware-corner.net | 2026-08-09 |
| gpt-oss-120b | MXFP4 | llama.cpp | 4k | — | 9.6 | --n-cpu-moe 21, DDR4 server | hardware-corner.net | 2026-08-09 |
Frequently asked questions
What is the largest model that runs entirely on the GeForce RTX 5090?
Qwen3.5 35B-A3B at Q4_K_M (a 22.63 GB file) fits entirely in 32 GB with 8k context, at est. 235.0 tok/s (188.0–282.0, calibrated estimate ±20%).
How many local LLMs run well on the GeForce RTX 5090?
At Q4_K_M with 8k context, out of 29 tracked models: 24 run great, 0 run well, 1 runs slowly and 4 won't run.
Can the GeForce RTX 5090 run a 70B model like Llama 3.3 70B?
Runs slowly — Partial offload, Q4_K_M: est. 2.7 tok/s (2.2–3.3, calibrated estimate ±20%).
Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.