CONTEXT → CACHE

KV-cache memory calculator

A weight file can fit while a long conversation does not. See the separate cost of keeping keys and values in memory as context or concurrency increases.

By InfoIsland Editorial · Updated · Calculated estimates, not hardware benchmarks

Your configuration

Runs in your browser
Tool limit: 32,768 tokens per sequence; no context scaling assumed.
Runtime and weight assumptions
Planning allowance for compute buffers and runtime allocations; change it to match your measurements.
Overrides the weight estimate for this checkpoint. A downloaded GGUF size is only a proxy; use observed device weight allocation when available.

These presets cover dense text models using grouped-query attention. They do not model the full DeepSeek-R1 MoE, vision encoders, speculative decoding or training.

Estimated KV cache

1.13 GiB

Qwen3-8B

Room in estimate

Model weights
4.58 GiB
KV cache
1.13 GiB
Runtime allowance
1.50 GiB
Usable GPU budget
11.00 GiB

Memory fit is an estimate, not a speed or compatibility guarantee. All sizes use GiB (2³⁰ bytes). A green status leaves at least 10% of usable memory free.

Qwen3-8B memory guide →
Publisher architecture configuration ↗

Context and cache precision

This table follows your selected model and number of parallel sequences. It shows cache allocation only; weights and runtime are separate.

Default: Qwen3-8B, one parallel sequence. Cache estimates in GiB.
Context / sequenceFP16 / BF16Q8_0Q4_0
2,0480.28 GiB0.15 GiB0.08 GiB
4,0960.56 GiB0.30 GiB0.16 GiB
8,1921.13 GiB0.60 GiB0.32 GiB
16,3842.25 GiB1.20 GiB0.63 GiB
32,7684.50 GiB2.39 GiB1.27 GiB

How the estimate works

Start with the stored weights, add the conversation cache, then allow space for the runtime. The display and other applications consume a separate reserve. The result assumes the whole model and its cache stay on one GPU.

Weights (GiB) ≈ parameters × assumed bits per weight ÷ 8 ÷ 2³⁰
KV cache (GiB) = 2 × layers × KV heads × head dimension × bytes per cached value × context tokens × parallel sequences ÷ 2³⁰
Total allocation = weights + KV cache + runtime allowance

Weight presets use 4.8, 5.7, 6.6 and 8.5 effective bits per parameter for Q4_K_M, Q5_K_M, Q6_K and Q8_0. These are planning assumptions, not exact file sizes. Tensor mixtures, unquantized tensors and container metadata vary by conversion. Use the advanced weight override when you have measured the checkpoint allocation.

The cache calculation models a full dense GQA cache at the selected capacity. FP16 uses 2 bytes per value. The Q8_0 and Q4_0 estimates include block overhead (34 and 18 bytes per 32 values). Backend padding and allocation strategies can increase usage. Your runtime must support the chosen cache type and kernels; smaller caches can affect output quality.

What the fit labels mean

Usable memory is GPU capacity minus the display/other-app reserve. “Room in estimate” leaves at least 10% of that usable budget unused. “Tight budget” means the estimate fits but has less headroom. “Exceeds budget” means this full-GPU configuration exceeds the selected allocation. Partial CPU offloading may still work, with different performance.

Check it on your machine

Use the exact quantized file, runtime version and context settings you plan to deploy. Load one sequence first, inspect the runtime's memory report and GPU monitor, and then test a conversation near the intended context length. Record peak use during both prompt processing and generation before increasing concurrency.

In Ollama, ollama ps reports the processor split and allocated context. A model partially offloaded to CPU can run even when it exceeds the GPU budget. Do not interpret successful loading alone as proof of full GPU residency.

Where this tool stops

This is inference planning for the six named dense text checkpoints. It does not estimate training, fine-tuning, image generation, MLA, sliding-window caches, prefix sharing, multi-GPU placement, or Apple unified-memory limits. No tokens-per-second claims are made. Context is bounded to the presets listed in the source notes; fitting memory does not validate model quality at that length.

Sources and review date

Publisher specifications reviewed October 31, 2026. Weight sizes and memory tables are InfoIsland calculations using the assumptions above; they are not measured benchmark results.