Can I run DeepSeek V3 locally?
DeepSeek V3 by deepseek-ai has 684B parameters (40B active per token, 256 experts, 8 used per token). At 4-bit with an 8K context it needs about 378 GB of GPU memory. It fits on 1 of 65 GPUs and Macs we track.
Memory needed by quant and context length
GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use real file sizes from unsloth/DeepSeek-V3-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.
| Quant | Quality | 4K ctx | 8K ctx | 32K ctx | 128K ctx |
|---|---|---|---|---|---|
| 2-bit (Q2_K) | rough | 240 GB | 240 GB | 242 GB | 248 GB |
| 3-bit (Q3_K_M) • | usable | 298 GB | 298 GB | 300 GB | 307 GB |
| 4-bit (Q4_K_M) • | good, most popular | 378 GB | 378 GB | 379 GB | 386 GB |
| 5-bit (Q5_K_M) • | very good | 444 GB | 444 GB | 446 GB | 452 GB |
| 6-bit (Q6_K) • | near lossless | 514 GB | 514 GB | 516 GB | 522 GB |
| 8-bit (Q8_0) • | effectively lossless | 665 GB | 665 GB | 667 GB | 674 GB |
| 16-bit (F16/BF16) | original weights | 1276 GB | 1276 GB | 1278 GB | 1284 GB |
DeepSeek V3 on Ollama: download size and VRAM
ollama run deepseek-v3:671b | Source | File | Download size | GPU memory at 8K ctx (est.) |
|---|---|---|---|
| Ollama download size | deepseek-v3:671b (Q4_K_M) | 404 GB | 377 GB |
| Our GGUF (Hugging Face) | unsloth/DeepSeek-V3-GGUF (Q4_K_M) | 404 GB | 378 GB |
Download sizes are decimal GB, as ollama.com lists them; memory figures on this site use binary GB (GiB). GPU memory for the Ollama tag is its file size plus this model's KV cache at 8K context and runtime overhead; Ollama's default context window may be smaller (set num_ctx to change it). Source: ollama.com/library/deepseek-v3:671b, checked Nov 17, 2026. Refreshed weekly; shown only when the tag's architecture and parameter count match this model. All Ollama models with sizes and VRAM.
Which GPUs and Macs run DeepSeek V3?
Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).
Prefer an API?
DeepSeek: DeepSeek V3 is on OpenRouter at $0.257 per 1M input tokens and $1.029 per 1M output tokens. Price details and history →
Questions
How much VRAM does DeepSeek V3 need?
About 378 GB at 4-bit (Q4_K_M) with an 8K context: 377 GB of weights (real GGUF file size), 0.5 GB of KV cache and 0.6 GB of runtime overhead. At 8-bit it needs about 665 GB.
What is the smallest GPU that runs DeepSeek V3?
No single consumer GPU in our table fits it at Q4_K_M with 8K context.
How big is the Ollama download for DeepSeek V3?
404GB for the deepseek-v3:671b tag (Q4_K_M), as listed on ollama.com. Run it with "ollama run deepseek-v3:671b". With an 8K context, plan on about 377 GB of GPU memory.
Can I run DeepSeek V3 on a Mac?
Not at Q4_K_M with 8K context on any Mac in our table, even a 512 GB Mac Studio, using the default GPU memory limit.
Similar models
Same family or a similar parameter count, with the VRAM each needs at Q4_K_M and 8K context.
DeepSeek R1
684B · 378 GB at 4-bitDeepSeek V3.1
685B · 382 GB at 4-bitDeepSeek V3 0324
685B · 378 GB at 4-bitGLM 5.3
753B · 430 GB at 4-bitGLM 5.2
753B · 435 GB at 4-bitGLM 5
754B · 426 GB at 4-bitEmbed this
Free badge and "Will it run on my GPU?" widget for model cards, READMEs and blogs. No cookies, no tracking. All options.
Copy a snippet
Markdown badge
[](https://runsonmy.com/model/deepseek-v3/) Widget (script)
<div class="runsonmy-widget" data-model="deepseek-v3"><a href="https://runsonmy.com/model/deepseek-v3/">Will DeepSeek V3 run on my GPU? Check on RunsOnMy</a></div>
<script async src="https://runsonmy.com/embed.js"></script> Widget (iframe)
<iframe src="https://runsonmy.com/embed/m/deepseek-v3/" width="100%" height="220" style="max-width:440px;border:0" loading="lazy" title="Will DeepSeek V3 run on my GPU? (RunsOnMy)"></iframe> Model facts
| Hugging Face repo | deepseek-ai/DeepSeek-V3 |
| Parameters | 684,489,845,504 (684B) · ~40B active (computed from config) |
| Layers / hidden size | 61 / 7168 |
| Attention | MLA; 128 heads, 128 KV heads, head dim 56 |
| KV cache per token (FP16) | 68.6 KiB |
| Max context | 163,840 tokens |
| License | — |
| Released on HF | Dec 25, 2024 |
| Downloads (30d) / likes | 1,418,977 / 4,463 |
Sources: Hugging Face model API · config.json (deepseek-ai/DeepSeek-V3) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.