FitMyGPU

Will It Fit?

Estimate text inference VRAM across Transformers and vLLM with a compact, explainable breakdown.

Model pages and release notes are included here too.

4B dense • 131,072 context • 8 KV heads

vLLM estimates use the selected GPU memory utilization. Transformers stays a fixed 4K single-request baseline.

Fits on selected GPU

Qwen 3 4B · vLLM · Official BF16 checkpoint

Required GPU VRAM (0.9 budget)

11.0 GB

Core estimate: 9.9 GB. Against RTX 4090 24GB, this leaves 14.7 GB of headroom.

At 4,096 tokens, the estimated max concurrency is 34 concurrent requests.

Quick read

Selected GPU

RTX 4090 24GB

Runtime

vLLM

Load dtype

FP16

GPU utilization

0.9

GPU count

1

KV cache dtype

BF16

Context length

4,096

Current concurrency

1

Max concurrency @ context

34

Class

Consumer

Bandwidth

1,008 GB/s

Nominal VRAM

24 GB

Core estimate

9.9 GB

Required GPU VRAM (0.9 budget)

11.0 GB

Headroom

14.7 GB

GPU compare

What fits this setup

GPUStatusHeadroomMax conc.
GB200 NVL72 GPU 186GBFits188.7 GB411
B200 180GBFits182.2 GB397
H200 141GBFits140.4 GB306
RTX PRO 6000 96GBFits92.1 GB202
A100 80GBFits74.9 GB164
H100 80GBFits74.9 GB164
L40 48GBFits40.5 GB90
A100 40GBFits31.9 GB71
RTX 5090 32GBFits23.3 GB53
RTX 3090 24GBFits14.7 GB34
RTX 4090 24GBFits14.7 GB34

Breakdown

Where the memory goes

Weights

4B resident parameters at 2.01 bytes each. Calibrated from the official checkpoint profile.

8.0 GB

KV cache

Concurrency 1, context 4,096, 36 KV-bearing layers, 8 KV heads, BF16 cache storage.

0.4 GB

Runtime / safety overhead

Conservative buffer for allocator fragmentation, kernels, and runtime scratch space.

1.5 GB

Weights = parameter count × bytes per parameter.

KV cache grows with context length, KV-bearing layers, concurrent requests, and the selected KV cache dtype.

Show the substituted formulas

Weights

parameter count × bytes per weight

4B × 2.01 = 8.0 GB

The official Qwen3-4B safetensor weights total about 8.04 GB on Hugging Face.

KV cache

batch × effective KV tokens across attention layers × 2 × KV heads × head dim × bytes per KV element

1 × 147,456 × 2 × 8 × 80 × 2 = 0.4 GB

Longer context, deeper models, and larger batches all expand the cache linearly. BF16 controls the bytes per stored KV element.

Overhead

max(1.5 GB, 10% of weights + KV cache + linear state)

max(1.5 GB, 10% of 8.4 GB) = 1.5 GB

This leaves room for runtime buffers instead of claiming an unrealistically exact fit.

Model

Selected model

Qwen 3 4B

4B dense • 131,072 context • 8 KV heads

About model

Total params

4B

Active params

Dense model

Layers

36

Hidden size

2,560

Attention heads

32

KV heads

8

KV-bearing layers

36