Can RTX 5080 16GB run gpt-oss 20B?
Measured results for gpt-oss 20B (20.91B MoE, 3.6B active) on RTX 5080 16GB, plus what fits and the commands to run it.
Short answer
Yes. gpt-oss 20B Q8_0 fits fully on RTX 5080 16GB at 16k context, using 12.5 of 15.0 GB usable, and generates about 238.8 tokens/s in our own run. As the context fills it slows down: 211.8 tok/s with 16k tokens already in context (89% of the empty-context speed), measured. For Q8_0, our bandwidth-only estimate was ~253 tok/s against a measured median of ~239 (6% too optimistic).
Measured generation speed by quantization
| Quant | Generation tok/s | Range | Prompt tok/s | Runs | Origin |
|---|---|---|---|---|---|
| Q8 class (median) | 238.8 | 238.8–238.8 | 11106 | 1 | includes our runs |
| Q4 class (median) | 187.7 | 178.3–253.5 | 9608 | 11 | includes our runs |
| Q4_K_M (our run, depth 0) | 253.5 | 3 repeats | 10826 | 3 | own measurement |
| Q8_0 (our run, depth 0) | 238.8 | 3 repeats | 11106 | 3 | own measurement |
Generation = tg128 (128 generated tokens, short context). Public rows are medians of community llama.cpp results (sources); our own runs use llama-bench b11658 on a rented GPU.
Speed vs context depth measured
Our own llama-bench runs on GeForce RTX 5080 (flash attention on, 3 repeats): generation and prompt speed with the given number of tokens already in the context.
| Quant | Context depth | Generation tok/s | vs depth 0 | Prompt tok/s |
|---|---|---|---|---|
| Q4_K_M | 0 | 253.5 | 100% | 10826 |
| Q4_K_M | 4k | 234.1 | 92% | 10032 |
| Q4_K_M | 16k | 221.8 | 87% | 8156 |
| Q8_0 | 0 | 238.8 | 100% | 11106 |
| Q8_0 | 4k | 221.8 | 93% | 10297 |
| Q8_0 | 16k | 211.8 | 89% | 8197 |
Alternatives
Faster for gpt-oss 20B
- RTX 5090 32GB: Q8_0, 31 GB usable, ~375 tok/s measured
- RTX PRO 6000 96GB: Q8_0, 95 GB usable, ~376 tok/s measured
- Mac Studio M5 Ultra 256GB: Q8_0, 179 GB usable, ~281 tok/s estimated
Other models on RTX 5080 16GB, similar size
- LFM2.5 24B-A2B (hardware page): Q4_K_M, fits fully, ~263 tok/s (scaled from measured)
- Gemma 4 12B (hardware page): Q8_0, fits fully, ~62 tok/s (measured)
- Qwen3.5 9B (hardware page): Q8_0, fits fully, ~92 tok/s (measured)
- Ornith 1.5 9B (hardware page): Q8_0, fits fully, ~92 tok/s (scaled from measured)
Memory and context limits
Q8_0 weights are 12.1 GB; the 16k f16 KV cache adds 0.41 GB. Largest context that fits in 15.0 GB usable:
| KV cache type | Max context |
|---|---|
| f16 | 64k |
| q8_0 | 128k |
| q4_0 | 128k |
Native limit 128k. Full KV cache table for gpt-oss 20B.
Copy-paste commands
The advisor's pick for this hardware: Q8_0, 16k context.
llama-server -hf unsloth/gpt-oss-20b-GGUF:Q8_0 -c 16384 -fa on --jinja -ngl 99
OLLAMA_CONTEXT_LENGTH=16384 ollama serve ollama run hf.co/unsloth/gpt-oss-20b-GGUF:Q8_0
FAQ
How fast is gpt-oss 20B on RTX 5080 16GB?
About 238.8 tokens/s for generation (Q8_0 on llama.cpp CUDA, 128-token generation), 11106 tokens/s for prompt processing. Speed drops as the context fills; see the depth table.
How much memory does gpt-oss 20B need on RTX 5080 16GB?
The advisor's pick, Q8_0, is 12.1 GB of weights plus 0.41 GB of KV cache at 16k context (12.5 GB), against 15.0 GB usable.
Is the gpt-oss 20B speed on RTX 5080 16GB measured or estimated?
Measured: our own llama-bench run on a rented GeForce RTX 5080, including runs with 4k to 16k tokens already in context.
What can I use instead of RTX 5080 16GB for gpt-oss 20B?
Nothing with less memory in our list fits it fully. Faster: RTX 5090 32GB (~375 tok/s), RTX PRO 6000 96GB (~376 tok/s), Mac Studio M5 Ultra 256GB (~281 tok/s).
What is the maximum context for gpt-oss 20B on RTX 5080 16GB?
With an f16 KV cache about 64k tokens, 128k with q8_0 and 128k with q4_0, bounded by the model's own limit.
Related
All models on RTX 5080 16GB · gpt-oss 20B on all hardware · VRAM tier · all measured pages · leaderboard
Compare: RTX 5080 16GB vs RTX 4090 24GB