# Qwen2.5 72B Instruct on a GeForce RTX 4060 Ti 16GB

No. Even Q3_K_M needs 35.1 GiB against 14.7 GiB usable.

Canonical page: https://makerportal.ai/lab/llm-vram/qwen2-5-72b-instruct/rtx-4060-ti-16gb
Page title: Qwen2.5 72B Instruct VRAM on RTX 4060 Ti 16GB — 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 — 14.7 GiB usable of 16 GB
- **Usable memory:** 14.7 GiB — 92% of the card's 16 GB — an assumption for a dedicated card, not a single fraction applied to every device
- **Weights at FP16:** 135.4 GiB — 72.7B parameters
- **Weights (Q4_K_M):** 44.2 GiB — 135.4 GiB at FP16 · 72.7B params · 6 of 6 quant rows are measured files
- **KV cache:** 320.0 KiB per token at FP16 — 80 layers × 8 KV heads × 128 dimensions
- **Speed ceiling:** 3.0 tok/s — offloaded — nothing fits in device memory
- **Largest context that fits:** none — the weights do not fit
- **Trained context:** 32,768 — the cap the KV-cache figures are held to
- **Card bandwidth:** 288 GB/s — 16 GB GDDR6

## Every quant, against this card

Every quantization of Qwen2.5 72B Instruct against the 14.7 GiB usable on a GeForce RTX 4060 Ti 16GB. "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 | 135.4 GiB | measured file | short by 120.7 GiB | — | 0.7 tok/s offloaded |
| Q8_0 | 8.50 | 72.0 GiB | measured file | short by 57.2 GiB | — | 1.3 tok/s offloaded |
| Q6_K | 7.08 | 59.9 GiB | measured file | short by 45.2 GiB | — | 1.6 tok/s offloaded |
| Q5_K_M | 5.99 | 50.7 GiB | measured file | short by 36.0 GiB | — | 1.9 tok/s offloaded |
| Q4_K_M | 5.22 | 44.2 GiB | measured file | short by 29.4 GiB | — | 2.3 tok/s offloaded |
| Q3_K_M | 4.15 | 35.1 GiB | measured file | short by 20.4 GiB | — | 3.0 tok/s offloaded |

## Questions this page answers

### How much VRAM does Qwen2.5 72B Instruct need?

135.4 GiB for the weights at FP16 — 72.7B parameters at two bytes each — and 44.2 GiB at Q4_K_M. The KV cache is on top of that and is not a fixed number: this model spends 320.0 KiB per token of context, so 8,192 tokens costs a further 2.5 GiB. A GeForce RTX 4060 Ti 16GB makes 14.7 GiB of its 16 GB available on the assumption below.

### Can a GeForce RTX 4060 Ti 16GB run Qwen2.5 72B Instruct?

No. Even Q3_K_M needs 35.1 GiB against 14.7 GiB usable. Holding the smallest quant here would need a card with about 39 GB. Splitting the model across the PCIe bus is possible and slow — see the offload figure on this page.

### How fast will Qwen2.5 72B Instruct run on a GeForce RTX 4060 Ti 16GB?

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 288 GB/s of the card. At an assumed 90 GB/s of host memory bandwidth the ceiling is 3.0 tok/s at Q3_K_M, against 7.1 tok/s if it were resident — 288 GB/s over the 40.38 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 37.70 GB weight file.

### Why does the context length change how much memory Qwen2.5 72B 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 80 layers and 8 key/value heads of 128 dimensions, shared across 64 query heads — grouped-query attention, which divides the cache by 8. That works out at 320.0 KiB per token. 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: 92% here, which is what this lane assumes for a dedicated card — the other class assumes 75%, 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/qwen2-5-72b-instruct/rtx-4060-ti-16gb. Free to quote and cite with attribution and a link to the canonical page.