Skip to content
Bitpute

Model guide

What GPU do you need for DeepSeek-R1 32B?

By Bitpute · Published 12 July 2026 · Updated 24 July 2026 · How we estimate · Sources · Editorial policy · Version history · Report an error

In short

Same memory as Qwen2.5 32B — about 22.2 GB at Q4_K_M — because DeepSeek-R1 32B is distilled onto the Qwen2.5 32B architecture. It fits a 24 GB card at Q4 with 8K context.

Architecture at a glance

Parameters

32.8B

Layers

64

KV heads

8

Head dim

128

Max context

128K

KV @8K

2.0 GB

The practical difference is behavioral, not architectural: R1 is a reasoning model that writes long chains of thought before answering, so it consumes far more of its context window per question. Budget more context than you would for a chat model — and remember every extra token of thinking sits in the KV cache. On a 24 GB card, that makes the 8K window feel smaller than it does on ordinary models.

VRAM by quantization (8K context)

QuantTotal VRAMCheapest GPU that fits
Q4_K_M22.2 GBRTX 3090
Q5_K_M25.6 GBRTX 5090
Q6_K29.1 GBRTX 5090
Q8_036.8 GBRTX A6000
FP1666.9 GBA100 80GB

Weights and KV cache are exact arithmetic from the model's published config; overhead (0.75 GB + 5% of weights) is a calibrated estimate. "Fits" means at most 95% of the card. Method on the Engineering Center.

Compatible GPUs at Q4_K_M

Green fits comfortably, amber is tight, faded doesn't fit — each links to that card's full page.

RTX 3060 12GB ✗RTX 4060 Ti 16GB ✗RTX 4070 Ti SUPER ✗RTX 3090 ✓RTX 4090 ✓RTX 5090 ✓RTX A6000 ✓L40S ✓A100 40GB ✓A100 80GB ✓H100 80GB ✓H200 141GB ✓

Can an RTX 4090 run DeepSeek-R1 32B?

Yes. At Q4_K_M with an 8K context, DeepSeek-R1 32B needs about 22.2 GB, leaving 1.8 GB of headroom on the 4090's 24 GB.

What about maximum context?

At the full 128K window, the KV cache grows to about 32.0 GB and the total to 52.2 GB. The KV cache is the part that grows — the weights never change. To see the exact split at any context, run this model through the GPU memory calculator, check an Ollama tag in the Ollama calculator, or size a fine-tune in the training memory calculator.

Running it in the cloud

For rented hardware the sensible floor is the A100 40GB — the smallest datacenter card that holds this model comfortably at Q4_K_M. Marketplace clouds also rent consumer cards; anything from the RTX 3090 up works for this model and usually costs less per hour. Hourly prices move weekly, so we don't print them here — the workspace carries the current figures and weighs rental against electricity for this exact model.

Qwen2.5 32B

32.8B params · 22.2 GB at Q4_K_M

Same architecture, different training.

See requirements →
Gemma 2 27B

27.2B params · 19.8 GB at Q4_K_M

Closest size in another family.

See requirements →
← Qwen2.5 32B · All models · Llama 3.3 70B →

Exact memory figures for every quantization: How much VRAM does R1 Distill Qwen 32B R1 need?

Common questions

What is the minimum GPU for DeepSeek-R1 32B?

At the default Q4_K_M quantization with an 8K context, DeepSeek-R1 32B needs about 22.2 GB of VRAM, so the practical minimum is a RTX 3090. Weights and KV cache are exact arithmetic from the model's config; a small runtime overhead estimate is included.

How much VRAM does DeepSeek-R1 32B need at maximum context?

At the full 128K window, the KV cache grows to about 32.0 GB and the total to 52.2 GB.

Can an RTX 4090 run DeepSeek-R1 32B?

Yes. At Q4_K_M with an 8K context, DeepSeek-R1 32B needs about 22.2 GB, leaving 1.8 GB of headroom on the 4090's 24 GB.

Before you buy

Compare the shortlisted cards head to head in GPU Compare, size the exact context you need in the GPU Memory Calculator, and read which quantization to run before committing to a card — dropping one format down often removes the need for the next tier up.