Can I Run GPT-OSS 20B on a NVIDIA RTX PRO 6000 Blackwell?
Runs at full precision (fp16). Zero quality loss.
12 quantizations fit your 96.0GB
| Quant | Min VRAM | Recommended | File size | Headroom |
|---|---|---|---|---|
| fp16BEST | 44.0 GB | 45.5 GB | 13.8 GB | +52.0 GB |
| Q8_0 | 23.8 GB | 25.3 GB | 12.1 GB | +72.2 GB |
| Q6_K | 18.7 GB | 20.2 GB | 12.0 GB | +77.3 GB |
| Q5_K_M | 16.3 GB | 17.8 GB | 11.7 GB | +79.7 GB |
| Q5_K_S | 15.8 GB | 17.3 GB | 11.7 GB | +80.2 GB |
| Q4_1 | 14.4 GB | 15.9 GB | 11.6 GB | +81.6 GB |
| Q4_K_M | 14.0 GB | 15.5 GB | 11.6 GB | +82.0 GB |
| Q4_K_S | 13.3 GB | 14.8 GB | 11.6 GB | +82.7 GB |
| Q4_0 | 13.1 GB | 14.6 GB | 11.5 GB | +82.9 GB |
| Q3_K_M | 10.0 GB | 11.5 GB | 11.5 GB | +86.0 GB |
| Q3_K_S | 9.3 GB | 10.8 GB | 11.5 GB | +86.7 GB |
| Q2_K | 8.1 GB | 9.6 GB | 11.5 GB | +87.9 GB |
Try it in the cloud first
Don't want to download GPT-OSS 20B just to try it? Use a hosted API or rent a GPU by the second.
Affiliate links — we earn a commission at no cost to you.
All quant variants, benchmark scores, and use-case tags.
Top-ranked open-source models that fit in 96.0GB.
FAQ
Can the NVIDIA RTX PRO 6000 Blackwell run GPT-OSS 20B?
Yes. The NVIDIA RTX PRO 6000 Blackwell's 96.0GB of VRAM is enough to run GPT-OSS 20B at fp16 quantization (44.0GB required).
What's the best quantization to use?
fp16 is the highest-precision quantization that fits in your 96.0GB. It uses about 44.0GB of memory and 45.5GB recommended for comfortable inference.
What if I need more headroom for context length?
KV cache memory grows with context length. The numbers above assume a baseline 2K-4K context. For long-context use (32K+), add another 2-6GB depending on the model architecture.