# Gemma 3 27B Instruct on a Jetson Orin Nano Super (8GB)

No. Even Q3_K_M needs 12.5 GiB against 6.0 GiB usable.

Canonical page: https://makerportal.ai/lab/llm-vram/gemma-3-27b-it/jetson-orin-nano-super-8gb
Page title: Gemma 3 27B Instruct on Jetson Orin Nano 8GB — does not fit

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:** Does not fit — 6.0 GiB usable of 8 GB
- **Usable memory:** 6.0 GiB — 75% of the card's 8 GB — an assumption for unified memory, not a single fraction applied to every device
- **Weights at FP16:** 51.1 GiB — 27.4B parameters
- **Weights (Q4_K_M):** 15.4 GiB — 51.1 GiB at FP16 · 27.4B params · 1 of 6 quant rows are measured files
- **KV cache:** 496.0 KiB per token at FP16 — 62 layers × 16 KV heads × 128 dimensions
- **Speed ceiling:** 6.5 tok/s — offloaded — nothing fits in device memory
- **Largest context that fits:** none — the weights do not fit
- **Trained context:** 131,072 — the cap the KV-cache figures are held to
- **Card bandwidth:** 102 GB/s — 8 GB LPDDR5 unified

## Every quant, against this card

Every quantization of Gemma 3 27B Instruct against the 6.0 GiB usable on a Jetson Orin Nano Super (8GB). "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 |
|---|---|---|---|---|---|---|
| FP16 / BF16 | 16.00 | 51.1 GiB | measured file | short by 45.1 GiB | — | 1.6 tok/s offloaded |
| Q8_0 | 8.50 | 27.1 GiB | nominal | short by 21.1 GiB | — | 3.1 tok/s offloaded |
| Q6_K | 6.56 | 20.9 GiB | nominal | short by 14.9 GiB | — | 3.9 tok/s offloaded |
| Q5_K_M | 5.67 | 18.1 GiB | nominal | short by 12.1 GiB | — | 4.5 tok/s offloaded |
| Q4_K_M | 4.83 | 15.4 GiB | nominal | short by 9.4 GiB | — | 5.3 tok/s offloaded |
| Q3_K_M | 3.91 | 12.5 GiB | nominal | short by 6.5 GiB | — | 6.5 tok/s offloaded |

## Questions this page answers

### How much VRAM does Gemma 3 27B Instruct need?

51.1 GiB for the weights at FP16 — 27.4B parameters at two bytes each — and 15.4 GiB at Q4_K_M. The KV cache is on top of that and is not a fixed number: this model spends 496.0 KiB per token of context on its full-attention layers, so 8,192 tokens costs a further 1.0 GiB. 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 Gemma 3 27B Instruct?

No. Even Q3_K_M needs 12.5 GiB against 6.0 GiB usable. Holding the smallest quant here would need a card with about 17 GB. Splitting the model across the PCIe bus is possible and slow — see the offload figure on this page.

### How fast will Gemma 3 27B Instruct 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 6.5 tok/s at Q3_K_M, against 7.0 tok/s if it were resident — 102 GB/s over the 14.51 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 13.41 GB weight file.

### Why does the context length change how much memory Gemma 3 27B Instruct 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 62 layers and 16 key/value heads of 128 dimensions, shared across 32 query heads — grouped-query attention, which divides the cache by 2. That works out at 496.0 KiB per token, except that 52 of the 62 layers use a 1024-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. 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.

## 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/gemma-3-27b-it/jetson-orin-nano-super-8gb. Free to quote and cite with attribution and a link to the canonical page.