# gpt-oss 20B on an Apple M3 Ultra (512GB)

Yes — the unquantized weights fit with room for 128k tokens of context.

Canonical page: https://makerportal.ai/lab/llm-vram/gpt-oss-20b/apple-m3-ultra-512gb
Page title: gpt-oss 20B VRAM on M3 Ultra 512GB — fits as released

This markdown document and the HTML page above are rendered from the same solved values at build time, by the same functions. Nothing here is written by a language model and nothing is fetched at request time.

## Key figures

- **Verdict:** Fits at FP16 — 384.0 GiB usable of 512 GB
- **Usable memory:** 384.0 GiB — 75% of the card's 512 GB — an assumption for unified memory, not a single fraction applied to every device
- **Weights at FP16:** 12.8 GiB — 21.5B parameters
- **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:** 277 tok/s — 819 GB/s ÷ bytes read per token
- **Largest context that fits:** 131,072 tokens at FP16 / BF16
- **Trained context:** 131,072 — the cap the KV-cache figures are held to
- **Card bandwidth:** 819 GB/s — 512 GB LPDDR5 unified

## Every quant, against this card

Every quantization of gpt-oss 20B against the 384.0 GiB usable on an Apple M3 Ultra (512GB). "Measured" rows are the byte size of a real published file; "nominal" rows are the parameter count times the published bits-per-weight.

| Precision | Bits/weight | Weights | Source | Fits | Max context | Ceiling |
|---|---|---|---|---|---|---|
| As released | 5.12 | 12.8 GiB | measured file | yes | 131,072 (model cap) | 277 tok/s |
| Q8_0 | 4.50 | 11.3 GiB | measured file | yes | 131,072 (model cap) | 312 tok/s |
| Q6_K | 4.48 | 11.2 GiB | measured file | yes | 131,072 (model cap) | 313 tok/s |
| Q5_K_M | 4.36 | 10.9 GiB | measured file | yes | 131,072 (model cap) | 321 tok/s |
| Q4_K_M | 4.32 | 10.8 GiB | measured file | yes | 131,072 (model cap) | 323 tok/s |
| Q3_K_M | 4.28 | 10.7 GiB | measured file | yes | 131,072 (model cap) | 327 tok/s |

## Questions this page 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. An Apple M3 Ultra (512GB) makes 384.0 GiB of its 512 GB available on the assumption below.

### Can an Apple M3 Ultra (512GB) run gpt-oss 20B?

Yes — the unquantized weights fit with room for 128k tokens of context. That is the weights and the KV cache together, against 384.0 GiB of usable memory.

### How fast will gpt-oss 20B run on an Apple M3 Ultra (512GB)?

No faster than 277 tokens/second at FP16 / BF16, and in practice below it. Decoding is memory-bound: every token reads the 20% of weights this MoE routes to 2.6 GiB of weights plus the cache, and this card moves 819 GB/s. That division is the ceiling — no kernel, runtime or driver beats it, and a real runtime typically reaches 60–80% of it.

### 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.

## 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. That is the only assumed input on this page; 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.

## Related tool

[LLM VRAM & KV-Cache Footprint Calculator](https://makerportal.ai/lab/llm-vram-kvcache-calculator) — Interactive VRAM planner — change the model geometry, quantization, context length and card capacity and read the weight bytes, KV-cache bytes and the bandwidth ceiling on decode speed.

---

Source: MakerPortal — https://makerportal.ai/lab/llm-vram/gpt-oss-20b/apple-m3-ultra-512gb. Free to quote and cite with attribution and a link to the canonical page.