Best GPU for DeepSeek-R1-Distill-32B Locally
Real-time prices and hardware recommendations updated for August 2026.
DeepSeek-R1-Distill-32B is a dense model: all 32B parameters activate on every token, so generation speed is bound by how fast your card can stream the full weights.
To run DeepSeek-R1-Distill-32B locally you need roughly 22.5 GB of VRAM at Q4_K_M quantization with a 32k token context. The best-value card that fits is the GeForce RTX 3090 (24 GB), which should generate around 10 tokens per second.
Adjust Context Length
Slide to popular sizes or type any value manually.
Q3 weights (~14GB) plus 32k GQA KV cache push past 16GB. 24GB cards are the practical entry point for this 32B model.
Runs comfortably on 24GB cards at Q4. Strong reasoning performance unconstrained by VRAM.
Full precision requires workstation-grade 48GB+ cards for the weight matrix alone.
Other models with the same VRAM requirement
Because DeepSeek-R1-Distill-32B's weights fit a 24 GB card, other models of a similar size run on the same GPU. Generation speed varies — mixture-of-experts models are faster, dense models slower — but any of these load in the same VRAM:
Optimizing Setup for DeepSeek-R1-Distill-32B
Quantization Recommendations
For daily coding and reasoning tasks, Q4_K_M (4-bit quantization) offers the best balance of quality and memory efficiency — it reduces memory requirements by over 70% with minimal quality loss compared to FP16. Q8 and higher presets preserve more fidelity at the cost of significantly higher VRAM usage, which may force layer offloading and hurt throughput.
Recommended Local Software
We recommend using Ollama as the primary runner for local inference due to its automated GPU model splitting and context cache optimizations. For advanced fine-tuning or quantization splits, llama.cpp with Flash Attention compiled natively provides the best granular control.
Running DeepSeek-R1-Distill-32B locally — FAQ
How much VRAM do I need to run DeepSeek-R1-Distill-32B?
At a 32k context with KV cache quantization on, DeepSeek-R1-Distill-32B needs about 22.5 GB of VRAM at Q4_K_M — the quantization most people should use. Dropping to Q3_K_M brings that down to roughly 17.9 GB at some quality cost, while Q8_0 needs about 40.9 GB for the best quality this model can give.
What size graphics card does DeepSeek-R1-Distill-32B fit on?
DeepSeek-R1-Distill-32B needs about 22.5 GB at Q4_K_M, so a 24 GB card is the smallest common size that holds it entirely in VRAM. Anything smaller has to offload layers to system RAM, which typically costs you most of your generation speed.
Which quantization should I use for DeepSeek-R1-Distill-32B?
Use Q4_K_M unless you have VRAM to spare. It needs about 22.5 GB and loses very little quality against full precision. Q8_0 needs about 40.9 GB for a quality gain most people cannot detect in everyday coding and chat. Spend spare VRAM on a longer context instead.
How does context length affect the VRAM DeepSeek-R1-Distill-32B needs?
Model weights are fixed, but the KV cache grows linearly with context. For DeepSeek-R1-Distill-32B at a 32k context the cache is about 4.1 GB; doubling to 64k takes it to roughly 8.2 GB. Turning KV cache quantization off doubles those figures again.
Can the same GPU run other models similar to DeepSeek-R1-Distill-32B?
Yes. DeepSeek-R1-Distill-32B needs about 22.5 GB at Q4_K_M, and any model whose weights fit the same card runs on it — including OpenReasoning-Nemotron-32B (32B, based on Qwen2.5-32B), Qwen2.5-32B (32.5B), Qwen2.5-Coder-32B (32.5B), QwQ-32B (32.5B), Qwen3-32B (32.8B), Gemma 4 31B (31B). The weights fit the same GPU; generation speed varies (mixture-of-experts models are faster, dense models slower).
▸How token speeds are estimated
Two metrics are shown per GPU: Read tok/s (how fast the model ingests your prompt) and Decode tok/s (how fast it streams tokens back). They model fundamentally different bottlenecks.
Read (Prefill)
The prompt is processed in one parallel pass. This is compute-bound: it saturates the GPU's tensor cores.
Decode (Generation)
Each new token requires loading the entire model's active weights from VRAM. This is memory-bandwidth-bound: the GPU stalls waiting for data, not computing.
Weights = (activeParams × bits ÷ 8) × 1.15 overhead. KV cache per step = activeParams × multiplier × contextK.
Architecture utilization factors
Left: decode factor — Right: read factor
Data sources
TFLOPS and memory bandwidth are read from the GPU database. When missing, bandwidth falls back to a hardcoded dictionary.
Limitations
These are analytical estimates, not benchmark results. Use them as a relative comparison, not an absolute performance guarantee.