Best AI Models for NVIDIA GeForce RTX 5090 (32.0GB)
32 GB positions this hardware in the professional tier for local AI. Most popular open-source models run comfortably, and even large 70B parameter models are accessible at lower quantization levels.
This memory amount is a sweet spot for enthusiasts and professionals. You can run 13B–30B models like DeepSeek R1 Distill at Q5 or Q6 quality with smooth token generation, and 7B models at near-lossless precision. The 70B class of models (Llama 3 70B, Qwen 72B) becomes possible at Q2–Q3 quantization, though with some quality trade-off. For day-to-day use with coding assistants, chat models, and reasoning tasks, this tier delivers an excellent experience.
Runs Well
- 7B–13B models at Q6–Q8 quality
- 14B–30B models at Q4–Q5 quality
- Small models (3B–7B) at FP16 precision
- Vision-language models at good quality
Challenging
- 70B models only at Q2–Q3 (noticeable quality loss)
- Large context windows with 30B+ models
What LLMs Can NVIDIA GeForce RTX 5090 Run?
108 models · 2 excellent · 22 good
Showing compatibility for NVIDIA GeForce RTX 5090
| Model | Quant | VRAM | Speed | Context | Status | Grade |
|---|---|---|---|---|---|---|
Q4_K_M·56.8 t/s tok/s·33K ctx·GOOD FIT | Q4_K_M | 20.5 GB | 56.8 t/s | 33K | GOOD FIT | A81 |
Q4_K_M·64.3 t/s tok/s·131K ctx·GOOD FIT | Q4_K_M | 18.1 GB | 64.3 t/s | 131K | GOOD FIT | A72 |
Q4_K_M·62.2 t/s tok/s·262K ctx·GOOD FIT | Q4_K_M | 18.7 GB | 62.2 t/s | 262K | GOOD FIT | A75 |
Q4_K_M·66.9 t/s tok/s·262K ctx·GOOD FIT | Q4_K_M | 17.4 GB | 66.9 t/s | 262K | GOOD FIT | A69 |
Q4_K_M·70.3 t/s tok/s·262K ctx·GOOD FIT | Q4_K_M | 16.6 GB | 70.3 t/s | 262K | GOOD FIT | A67 |
Q4_K_M·64.8 t/s tok/s·8K ctx·GOOD FIT | Q4_K_M | 18.0 GB | 64.8 t/s | 8K | GOOD FIT | A71 |
Q4_K_M·44.1 t/s tok/s·GOOD FIT | Q4_K_M | 26.4 GB | 44.1 t/s | — | GOOD FIT | A77 |
Q4_K_M·77.0 t/s tok/s·131K ctx·FAIR FIT | Q4_K_M | 15.1 GB | 77.0 t/s | 131K | FAIR FIT | B62 |
Q4_K_M·87.7 t/s tok/s·131K ctx·FAIR FIT | Q4_K_M | 13.3 GB | 87.7 t/s | 131K | FAIR FIT | B57 |
BF16·75.6 t/s tok/s·4K ctx·FAIR FIT | BF16 | 15.4 GB | 75.6 t/s | 4K | FAIR FIT | B63 |
Q4_K_M·42.2 t/s tok/s·GOOD FIT | Q4_K_M | 27.6 GB | 42.2 t/s | — | GOOD FIT | A67 |
BF16·75.9 t/s tok/s·8K ctx·FAIR FIT | BF16 | 15.3 GB | 75.9 t/s | 8K | FAIR FIT | B63 |
Q4_K_M·78.4 t/s tok/s·33K ctx·FAIR FIT | Q4_K_M | 14.9 GB | 78.4 t/s | 33K | FAIR FIT | B61 |
Q4_K_M·78.4 t/s tok/s·41K ctx·FAIR FIT | Q4_K_M | 14.9 GB | 78.4 t/s | 41K | FAIR FIT | B61 |
FP16·75.6 t/s tok/s·2K ctx·FAIR FIT | FP16 | 15.4 GB | 75.6 t/s | 2K | FAIR FIT | B63 |
Q4_K_M·40.7 t/s tok/s·33K ctx·FAIR FIT | Q4_K_M | 28.6 GB | 40.7 t/s | 33K | FAIR FIT | B56 |
NVIDIA GeForce RTX 5090 Specifications
- Brand
- NVIDIA
- Architecture
- Blackwell
- Compute Capability
- 10.0 (CUDA SM version)
- VRAM
- 32.0 GB GDDR7
- Memory Bandwidth
- 1792.0 GB/s
- CUDA Cores
- 21,760
- Tensor Cores
- 680
- FP16 Performance
- 209.20 TFLOPS
- TDP
- 575W
- Release Date
- 2025-01-30
- MSRP
- $1,999
Get Started
GPUs to Consider Over NVIDIA GeForce RTX 5090
Similar GPUs and upgrades with more VRAM or higher bandwidth for AI
AMD Instinct MI210
AMD · CDNA 2
NVIDIA A100 40GB PCIe
NVIDIA · Ampere
NVIDIA RTX 6000 Ada Generation
NVIDIA · Ada Lovelace
AMD Radeon PRO W7900
AMD · RDNA 3
NVIDIA L40
NVIDIA · Ada Lovelace
NVIDIA L40S
NVIDIA · Ada Lovelace
Frequently Asked Questions
- Can NVIDIA GeForce RTX 5090 run Qwen3.6 35B A3B?
Yes, the NVIDIA GeForce RTX 5090 with 32 GB can run Qwen3.6 35B A3B, Gemma 4 31B IT, C4ai Command R V01, and 1313 other models. 23 models run at excellent quality, and 182 at good quality. Check the compatibility table above for the full list with VRAM usage and estimated speed.
- Is NVIDIA GeForce RTX 5090 good for AI?
The NVIDIA GeForce RTX 5090 has 32 GB of GDDR7, making it excellent for running local AI models. It supports 205 models at good quality or better. With 1792.0 GB/s memory bandwidth, it delivers fast token generation speeds. This is an enthusiast-grade GPU that handles most popular open-source LLMs.
- How many parameters can NVIDIA GeForce RTX 5090 handle?
With 32 GB, the NVIDIA GeForce RTX 5090 supports models from 3B to 30B parameters depending on quantization level. At Q4_K_M (the recommended sweet spot), you can fit roughly 53B parameters. This means 7B models at high quality (Q6/Q8) or 30B+ models at Q4.
- What quantization should I use on NVIDIA GeForce RTX 5090?
For the best balance of quality and speed on the NVIDIA GeForce RTX 5090, start with Q4_K_M — it preserves ~85% of the original model quality while keeping VRAM usage reasonable. With 24+ GB, you have the headroom to run 7B models at Q5_K_M or even Q6_K for noticeably better output quality. For larger 30B models, Q4_K_M remains the sweet spot.
- How fast is NVIDIA GeForce RTX 5090 for AI inference?
With 1792.0 GB/s memory bandwidth, the NVIDIA GeForce RTX 5090 achieves approximately 259 tokens/sec on a 7B model at Q4_K_M — that's very fast, well above conversational speed. A 14B model runs at ~129 tok/s. Token generation speed scales inversely with model size — smaller models are significantly faster.
tok/s = (1792 GB/s ÷ model GB) × efficiency
Smaller models = faster inference. Memory bandwidth is the main bottleneck for token generation speed.
Estimated speed on NVIDIA GeForce RTX 5090
~53 tok/s~55 tok/s~50 tok/s~57 tok/sReal-world results typically within ±20%. Speed depends on quantization kernel, batch size, and software stack.
- What's the best model for NVIDIA GeForce RTX 5090?
The top-rated models for the NVIDIA GeForce RTX 5090 are Qwen3.6 35B A3B, Gemma 4 31B IT, C4ai Command R V01. The best choice depends on your use case: coding assistants benefit from code-tuned models, while general chat works well with instruction-tuned models like Llama or Qwen.
- What power supply and cooling does NVIDIA GeForce RTX 5090 need?
The NVIDIA GeForce RTX 5090 has a TDP of 575 W. A good rule of thumb is to provide at least double the GPU's TDP to cover the rest of the system — that means a 1200 W PSU or larger. At this power level, a high-airflow case matters: aim for at least two front intake fans and one rear exhaust, with tidy cabling so hot air isn't trapped around the card. LLM inference sustains full GPU load continuously — longer and more consistently than most gaming workloads — so also make sure your CPU cooler can keep up under combined load.