MagicTools
LLM VRAM Calculator (all models)

Can You Run DeepSeek V3 / R1 (671B MoE)? VRAM Requirements

DeepSeek V3 / R1 (671B MoE) has 671B total parameters (MoE, 37B active per token). Running it at Q4_K_M (GGUF) with 16K context takes roughly 413 GB of memory (381 GB weights + 1.1 GB KV cache + runtime overhead). The minimum viable hardware is about Mac M3 Ultra 512GB. All numbers are planning estimates, verified 2026-07-27.

VRAM by quantization (16K context)

QuantizationWeightsTotal neededMinimum hardware
FP16 / BF161.22 TB1.32 TB8× B200 192GB
FP8625 GB676 GB4× B200 192GB
Q8_0 (GGUF)662 GB716 GB5× B200 192GB
Q6_K (GGUF)512 GB554 GB8× H100 80GB
Q5_K_M (GGUF)444 GB480 GB7× H100 80GB
MXFP4331 GB359 GBMac M3 Ultra 512GB
Q4_K_M (GGUF)recommended381 GB413 GBMac M3 Ultra 512GB
Q3_K_M (GGUF)294 GB318 GBMac M3 Ultra 512GB
Q2_K (GGUF)219 GB237 GBMac M3 Ultra 512GB

How context length changes memory (Q4_K_M (GGUF))

ContextKV cacheTotal needed
4K0.3 GB412 GB
16K1.1 GB413 GB
64K4.3 GB416 GB
128K8.6 GB420 GB

KV cache grows linearly with context; FP8 KV cache halves it again (try it in the calculator below).

Customize the estimate

Back

LLM VRAM Calculator — Can I Run It?

Estimate how much GPU memory (VRAM) you need to run open-weight LLMs like Kimi K3, DeepSeek R1, Qwen3, or Llama locally. Pick a model, quantization, and context length — the calculator adds up weights, KV cache, and runtime overhead, then shows which GPUs or Macs can fit it. Runs entirely in your browser: no upload, no signup.

671B total params · 37B active (MoE) · max 128K context

Recommended balance · ~0.61 bytes/param

Estimated memory needed

420 GB

Model weights381 GB
KV cache (131,072 tokens)8.6 GB
Runtime overhead30.5 GB

Will it fit?

RTX 3060 12GB39× needed
RTX 4060 Ti 16GB29× needed
RTX 3090 / 4090 24GB20× needed
RTX 5090 32GB15× needed
A100 40GB12× needed
A100 / H100 80GB6× needed
H200 141GB4× needed
AMD MI300X 192GB3× needed
B200 192GB3× needed
Mac 36GB unified✗ too big
Mac 64GB unified✗ too big
Mac 128GB unified✗ too big
Mac M3 Ultra 512GB✓ fits

Assumes ~92% of device memory is usable. Multi-GPU counts are for tensor/pipeline parallel serving (vLLM, SGLang); Apple Silicon uses unified memory via llama.cpp or MLX.

FAQ

How much VRAM does DeepSeek V3 / R1 (671B MoE) need?

About 413 GB at the recommended Q4_K_M (GGUF) quantization with 16K context: 381 GB for weights, 1.1 GB for KV cache, plus runtime overhead. Longer context grows the KV cache.

What is the minimum hardware for DeepSeek V3 / R1 (671B MoE)?

Roughly Mac M3 Ultra 512GB, assuming 92% of device memory is usable. Lower-end setups can try more aggressive quantization (Q3/Q2) at a noticeable quality cost.

Can I run DeepSeek V3 / R1 (671B MoE) on a Mac?

Yes. At Q4_K_M (GGUF) with 16K context it needs about 413 GB, which fits the unified memory of a Mac M3 Ultra 512GB via llama.cpp or MLX.

Which quantization should I use for DeepSeek V3 / R1 (671B MoE)?

Q4_K_M is the recommended balance of size and quality; with headroom, Q6_K or Q8_0 reduce quality loss further.

VRAM requirements for other models