gpt-oss 20B on a Jetson Orin Nano Super (8GB)
No. Even Q3_K_M needs 10.7 GiB against 6.0 GiB usable.
Verdict
Does not fit
6.0 GiB usable of 8 GB
Weights (Q4_K_M)
10.8 GiB
12.8 GiB at FP16 · 21.5B params · 6 of 6 quant rows are measured files
KV cache
48.0 KiB per token at FP16
24 layers × 8 KV heads × 64 dimensions
Speed ceiling
38.2 tok/s
offloaded — nothing fits in device memory
Every quant, against this card
Weight bytes are the size of the real published file wherever one exists — 6 of these6 rows are measured from unsloth/gpt-oss-20b-GGUF, the rest computed from the parameter count. Max context is what the KV cache can grow to in whatever memory the weights leave behind, capped at the 128k tokens this model was trained to address.
| Precision | Bits/weight | Weights | Source | Fits | Max context | Ceiling |
|---|---|---|---|---|---|---|
| As released | 5.12 | 12.8 GiB | measured file | −6.8 GiB | — | 32.1 tok/s offloaded |
| Q8_0 | 4.50 | 11.3 GiB | measured file | −5.3 GiB | — | 36.3 tok/s offloaded |
| Q6_K | 4.48 | 11.2 GiB | measured file | −5.2 GiB | — | 36.5 tok/s offloaded |
| Q5_K_M | 4.36 | 10.9 GiB | measured file | −4.9 GiB | — | 37.5 tok/s offloaded |
| Q4_K_M | 4.32 | 10.8 GiB | measured file | −4.8 GiB | — | 37.8 tok/s offloaded |
| Q3_K_M | 4.28 | 10.7 GiB | measured file | −4.7 GiB | — | 38.2 tok/s offloaded |
What the context actually costs
The KV cache is 2 · layers · kv_heads · head_dim bytes per token per element — 2 · 24 · 8 · 64 · 2 B = 48.0 KiB. It depends on the key/value head count, not the attention-head count and not the parameter count. This model shares 8 KV heads across 64 query heads, which divides the cache by 8 against multi-head attention.
Sliding-window attention. 12 of this model's 24 layers attend to a 128-token window and stop growing there; only the remaining 12 keep scaling with context. That is why the cache figures below flatten out — and why a calculator that ignores the layer pattern over-states this model's long-context footprint several times over.
| Context | KV cache (FP16) | KV cache (8-bit) | Plus Q4_K_M weights |
|---|---|---|---|
| 4,096 | 99 MiB | 50 MiB | 10.9 GiB |
| 8,192 | 195 MiB | 98 MiB | 11.0 GiB |
| 32,768 | 771 MiB | 386 MiB | 11.6 GiB |
| 131,072 | 3.0 GiB | 1.5 GiB | 13.8 GiB |
What running it anyway would cost
Nothing here fits, so the weights would have to be split with part of the model in host RAM. At Q3_K_M that is 4.7 GiB on the host side, and assuming 90 GB/s of host memory bandwidth — a dual-channel DDR5 desktop — the ceiling falls to 38.2 tok/s, against 40.7 tok/s if the same weights were resident — 102 GB/s over the 2.51 GB one token reads, which is the weights the model routes through plus one pass over the cache, against a 11.51 GB weight file. That host bandwidth is an assumption and it is the number to change first if your machine differs; the ratio is the part that generalises.
Where these numbers come from
The model
- Parameters
- 21,511,953,984
- Layers
- 24
- Attention / KV heads
- 64 / 8
- Head dimension
- 64
- Trained context
- 131,072
- Checkpoint as published
- 12.8 GiB
Read from openai/gpt-oss-20b. The parameter count is the Hub's own total over the tensor shapes, not a figure taken from the model's name. This release is already quantized — its published checkpoint is 12.8 GiB, far below the 40.1 GiB that two bytes per parameter would imply, so the FP16 row above describes a file nobody ships.
The accelerator
- Memory
- 8 GB LPDDR5 unified
- Bandwidth
- 102 GB/s
- Assumed usable
- 75% → 6.0 GiB
Capacity and bandwidth from the vendor's specification. The usable fraction is an assumption, not a spec: unified memory is shared with the OS and the display, and the GPU working-set cap is raisable on Apple silicon with `sudo sysctl iogpu.wired_limit_mb`.
Questions this pairing answers
How much VRAM does gpt-oss 20B need?
12.8 GiB for the weights at FP16 — 21.5B parameters at two bytes each — and 10.8 GiB at Q4_K_M. The KV cache is on top of that and is not a fixed number: this model spends 48.0 KiB per token of context on its full-attention layers, so 8,192 tokens costs a further 195 MiB. A Jetson Orin Nano Super (8GB) makes 6.0 GiB of its 8 GB available on the assumption below.
Can a Jetson Orin Nano Super (8GB) run gpt-oss 20B?
No. Even Q3_K_M needs 10.7 GiB against 6.0 GiB usable. Holding the smallest quant here would need a card with about 15 GB. Splitting the model across the PCIe bus is possible and slow — see the offload figure on this page.
How fast will gpt-oss 20B run on a Jetson Orin Nano Super (8GB)?
It cannot run in this card's memory alone, so the speed is set by whatever bus the offloaded part is read across, not by the 102 GB/s of the card. At an assumed 90 GB/s of host memory bandwidth the ceiling is 38.2 tok/s at Q3_K_M, against 40.7 tok/s if it were resident — 102 GB/s over the 2.51 GB one token reads at Q3_K_M, priced at the 8k reference context. That is the weights the model routes through plus one pass over the cache, against a 11.51 GB weight file.
Why does the context length change how much memory gpt-oss 20B needs?
Because the KV cache holds one key and one value vector per token, per layer, for the whole conversation, and it is allocated separately from the weights. This model has 24 layers and 8 key/value heads of 64 dimensions, shared across 64 query heads — grouped-query attention, which divides the cache by 8. That works out at 48.0 KiB per token, except that 12 of the 24 layers use a 128-token sliding window and stop growing there. Parameter count tells you nothing about this number.
The same model on a different card
- GeForce RTX 3060 12GBQ5_K_M
- GeForce RTX 5080FP16 / BF16
- GeForce RTX 5070 TiFP16 / BF16
- GeForce RTX 4080 SUPERFP16 / BF16
- GeForce RTX 4070 Ti SUPERFP16 / BF16
- GeForce RTX 4060 Ti 16GBFP16 / BF16
- GeForce RTX 4090FP16 / BF16
- GeForce RTX 3090FP16 / BF16
A different model on the same card
All 62 models on Jetson Orin Nano Super (8GB) →- Mistral Small 24B Instructno fit
- Dolphin Mistral 24B Venice Editionno fit
- Gemma 2 27B Instructno fit
- Gemma 3 27B Instructno fit
- DeepSeek-R1-Distill-Qwen 14Bno fit
- Qwen2.5 14B Instructno fit
- Qwen3 14Bno fit
- Phi-4 14Bno fit
Method and limits. Weight bytes are the byte size of the real published file wherever one exists, and the model's exact parameter count times the published llama.cpp bits-per-weight where it does not. That distinction is on every row above and it matters at both ends: a sub-1B model's Q4_K_M file runs a third larger than the nominal figure because k-quants keep its embedding tables at higher precision, and an already-4-bit release cannot be quantized upward at all. The KV cache is 2 · layers · kv_heads · head_dim · bytes per token, summed over layers with each sliding-window layer capped at its window. The speed figure is a roofline bound, not a benchmark: bandwidth divided by bytes read per token, which no runtime exceeds and every runtime falls short of. The usable fraction of card memory is an assumption: 75% here, which is what this lane assumes for unified memory, where the OS and window server share the same pool — the other class assumes 92%, so it is not one number applied to every device. Because nothing here fits, a second assumed input is in play — the 90 GB/s of host memory bandwidth the offload ceiling is priced at, stated above. Those two are the assumed inputs; every other figure is computed from the model config and the card's published specification. Nothing on this page is written by a language model. Architecture from the model's published config, fetched 2026-08-07.