Can you run DeepSeek V4 & V4-Flash locally?
DeepSeek V4 family · July 2026 (preview) · 284.00B parameters (13B activated per token) · Hugging Face model card
DeepSeek V4-Flash has 284 billion total parameters with 13 billion activated per token, and it needs about 187 GB of VRAM at Q4_K_M, so no consumer GPU can run it — the practical floor is a single 192 GB NVIDIA B200 or four 80 GB data center GPUs. DeepSeek V4-Pro has 1.6 trillion total parameters (49B activated) and needs about 1056 GB at Q4_K_M, which is data-center-only hardware. On consumer hardware, distilled or smaller open models are the realistic local options.
Minimum: 1× NVIDIA B200 192GB (Q4_K_M, ~187 GB) or 4× 80GB (H100/A100) for V4-Flash · Recommended: 2× NVIDIA B200 192GB or 8× H100 80GB with FP8 weights for V4-Flash; multi-node for V4-Pro
How much VRAM does DeepSeek V4 & V4-Flash need at each quantization?
DeepSeek V4 & V4-Flash needs 187.4 GB of VRAM at Q4_K_M. The table below lists weights-only size and total VRAM including overhead for each common quantization level with 4k and 32k token contexts.
| Quantization | Bits / weight | Weights only | Total + KV @4k ctx | Total + KV @32k ctx |
|---|---|---|---|---|
| Q4_K_M | 4.8 | 170.4 GB | 187.4 GB | 187.4 GB |
| Q5_K_M | 5.7 | 202.4 GB | 222.6 GB | 222.6 GB |
| Q6_K | 6.6 | 234.3 GB | 257.7 GB | 257.7 GB |
| Q8_0 | 8.5 | 301.8 GB | 331.9 GB | 331.9 GB |
| FP16 | 16 | 568.0 GB | 624.8 GB | 624.8 GB |
DeepSeek V4 uses compressed sparse attention (CSA) with a compressed KV cache; the exact per-token cache size is not published as a simple formula, so the table shows weights + 10% overhead and you should budget extra GB for long contexts.
Method: weights = parameters × bits-per-weight (Q4_K_M ≈ 4.8, Q5_K_M ≈ 5.7, Q6_K ≈ 6.6, Q8_0 ≈ 8.5, FP16 = 16), plus 10% loading overhead, plus KV cache from the verified architecture config (43 layers, KV heads, head dim). Source: V4-Flash: 284B total / 13B activated, 43 layers, 256 routed experts, 1M context — HF model card + config.json (deepseek-ai/DeepSeek-V4-Flash). V4-Pro: 1.6T total / 49B activated — HF model card (deepseek-ai/DeepSeek-V4-Pro).
Which GPUs can run DeepSeek V4 & V4-Flash locally?
At Q4_K_M with a 4k context, DeepSeek V4 & V4-Flash needs 187.4 GB of VRAM. The lists below are computed live from our GPU database and grouped by how much headroom the card has. Disclosure: CompareAIHardware.com participates in the Amazon Associates program and earns from qualifying purchases through links on this page. Affiliate relationships do not influence our recommendations.
Runs DeepSeek V4 & V4-Flash comfortably (25%+ VRAM headroom)
| GPU | VRAM | Bandwidth | Type | |
|---|---|---|---|---|
| AMD Instinct MI325X | 256 GB | 6000 GB/s | Data Center GPU | |
| AMD Instinct MI355X | 288 GB | 8000 GB/s | Data Center GPU | |
| NVIDIA B300 (Blackwell Ultra) | 288 GB | 8000 GB/s | Data Center GPU |
Minimum GPUs that fit DeepSeek V4 & V4-Flash at Q4_K_M
These GPUs hold the model but leave little headroom — keep contexts short.
| GPU | VRAM | Bandwidth | Type | |
|---|---|---|---|---|
| AMD Instinct MI300X | 192 GB | 5300 GB/s | Data Center GPU | |
| NVIDIA B200 | 192 GB | 8000 GB/s | Data Center GPU |
Needs 2+ GPUs to run DeepSeek V4 & V4-Flash
One of these cards is too small on its own, but a pair (tensor or pipeline parallel, ~90% efficiency) covers the 187.4 GB requirement. See our multi-GPU guide for setup.
| GPU | VRAM | Bandwidth | Type | |
|---|---|---|---|---|
| H200 SXM | 141 GB ×2 | 4800 GB/s | Data Center GPU |
Can a Mac run DeepSeek V4 & V4-Flash?
Yes — these Apple Silicon machines fit DeepSeek V4 & V4-Flash at Q4_K_M, since macOS lets the GPU use about 75% of unified memory. Generation speed is bound by memory bandwidth, so the GB/s column matters as much as capacity.
| Apple Silicon | Unified memory | Usable by GPU (~75%) | Bandwidth |
|---|---|---|---|
| M3 Ultra (Mac Studio) | 512 GB | 384 GB | 819 GB/s |
Which pre-built systems can run DeepSeek V4 & V4-Flash?
These mini PCs and workstations from our database fit DeepSeek V4 & V4-Flash at Q4_K_M. Usable-memory figures are conservative: Windows shares about half of system RAM with the GPU by default (Linux can expose more), macOS lets Apple Silicon GPUs use about 75% of unified memory, and Linux unified-memory systems such as GB10 expose roughly 90%. Generation speed is bound by memory bandwidth, so compare the GB/s column before buying.
| System | Type | Memory | Usable for AI | Bandwidth | GPUs |
|---|---|---|---|---|---|
| HP Z8 Fury G5 (4x RTX 6000 Ada) | Workstation | 192 GB | 192 GB | 960 GB/s | 4 |
Frequently asked questions
Can you run DeepSeek V4-Flash locally?
Only on data center hardware. DeepSeek V4-Flash needs about 187 GB of VRAM for its Q4_K_M weights, so it requires a single 192 GB NVIDIA B200, or at least four 80 GB GPUs. No consumer graphics card can run it without heavy offloading.
Why is DeepSeek V4-Flash so large if only 13B parameters activate per token?
DeepSeek V4-Flash is a Mixture-of-Experts model: all 284 billion parameters must sit in memory even though only 13 billion are active per token, according to the Hugging Face model card. VRAM use follows total parameters, while generation speed follows activated parameters.
Can a Mac Studio run DeepSeek V4-Flash?
A Mac Studio with M3 Ultra and 512 GB of unified memory can hold the Q4_K_M weights of DeepSeek V4-Flash, which occupy about 187 GB. Generation speed is limited by the M3 Ultra’s 819 GB/s memory bandwidth, so expect several tokens per second rather than data center throughput.