LLMLocal inference
DeepSeek V3
Weight estimates and planning VRAM for running DeepSeek V3 locally at each quantization level. Compare the lowest reference-cost devices that clear the plan.
Q4 plan
437 GB
Q8 plan
805 GB
FP16 plan
1495 GB
Context window
128 k tokens
Planning targets add 15% to weight estimates for runtime buffers and a modest KV cache. Long context or a different backend can require more. Official DeepSeek model card ↗
01 // GPUs that can run DeepSeek V3
Compatible hardware by quantization
Sorted by dated reference-cost estimates from Apr 30, 2026. These are not live offers.
Q4Q4_K_M (4-bit)
380GB weights · plan ≥437GBGPUVRAMPriceTier
Q8Q8_0 (8-bit)
700GB weights · plan ≥805GBNo single GPU tracked here clears the 805GB planning target for DeepSeek V3 at Q8. Multi-GPU, offload, or cloud inference is needed at this precision.
FP16FP16 (full precision)
1300GB weights · plan ≥1495GBNo single GPU tracked here clears the 1495GB planning target for DeepSeek V3 at FP16. Multi-GPU, offload, or cloud inference is needed at this precision.
02 // Frequently asked
DeepSeek V3 GPU questions
How much VRAM does DeepSeek V3 need?
DeepSeek V3 uses approximately 380GB for Q4 weights, 700GB at Q8, or 1300GB at FP16. GPU Hunter adds 15% planning headroom for runtime buffers and a modest KV cache, producing targets of 437GB, 805GB, and 1495GB respectively. Exact memory use varies by backend and context length.
What is the cheapest GPU to run DeepSeek V3?
Using GPU Hunter's 437GB Q4 planning target, the lowest reference-cost single device is the Apple M3 Ultra (512GB VRAM, dated estimate $9,499).
Can I run DeepSeek V3 at FP16?
DeepSeek V3 uses about 1300GB for FP16 weights and 1495GB under GPU Hunter's planning allowance—well beyond a single consumer GPU. Q4 or Q8 is more practical.
What quantization is best for DeepSeek V3?
Q4_K_M uses about 380GB for weights and is the most hardware-accessible option. Q8_0 uses about 700GB and trades more memory for fidelity. FP16 uses about 1300GB before runtime and context overhead. The right choice depends on the task, backend, and context window.