NeuroGenesysCan I run DeepSeek-V3 / R1?
Last updated • By Will Schulz
DeepSeek-V3 / R1 needs about 424 GB of memory at 4-bit (Q4_K_M) with an 8K-token context, so it needs 8 × 80 GB GPUs, since no single card in our list holds it. At 16-bit it needs 1384 GB, so it needs more than eight 80 GB GPUs.
How much memory does DeepSeek-V3 / R1 need?
| Precision | Weights | Total, 8K context | Total, 32K context | Total, 128K context |
|---|---|---|---|---|
| 16-bit (BF16 / FP16) | 1342 GB | 1384 GB | 1386 GB | 1392 GB |
| 8-bit (Q8_0) | 713 GB | 736 GB | 738 GB | 745 GB |
| 4-bit (Q4_K_M) | 410 GB | 424 GB | 426 GB | 433 GB |
| Native release (8 bits) | 671 GB | 693 GB | 694 GB | 701 GB |
Total = weights + FP16 KV cache + 3% of the weights + 1 GB of runtime overhead, as in the VRAM calculator. Multi-head latent attention caches 576 values per token per layer, which keeps the KV cache small. Longer contexts are only usable up to the model's own context window; check the model card. Released in FP8.
Which GPUs can run DeepSeek-V3 / R1?
| GPU memory | 16-bit (BF16 / FP16) | 8-bit (Q8_0) | 4-bit (Q4_K_M) |
|---|---|---|---|
| 8 GB | more than 8 | more than 8 | more than 8 |
| 12 GB | more than 8 | more than 8 | more than 8 |
| 16 GB | more than 8 | more than 8 | more than 8 |
| 24 GBRTX 3090, RTX 4090, L4 | more than 8 | more than 8 | more than 8 |
| 32 GBRTX 5090 | more than 8 | more than 8 | more than 8 |
| 40 GBA100 40GB | more than 8 | more than 8 | more than 8 |
| 48 GBRTX A6000, RTX 6000 Ada, L40S | more than 8 | more than 8 | more than 8 |
| 80 GBA100 80GB, H100 PCIe, H100 SXM | more than 8 | more than 8 | 8 cards |
| 141 GBH200 | more than 8 | 8 cards | 4 cards |
| 180 GBB200 | 8 cards | 8 cards | 4 cards |
| 192 GBAMD MI300X | 8 cards | 4 cards | 4 cards |
Assumes the runtime can allocate 95% of each card and every card pays the fixed overhead. Vendor memory sizes are binary (a 24 GB card holds 25.8 × 10⁹ bytes), and that is accounted for.
Can I run DeepSeek-V3 / R1 on a Mac?
| Unified memory | 16-bit (BF16 / FP16) | 8-bit (Q8_0) | 4-bit (Q4_K_M) |
|---|---|---|---|
| 24 GBM4 Pro | no | no | no |
| 32 GBM2 Max | no | no | no |
| 48 GBM3 Max (40-core GPU), M4 Pro, M4 Max (40-core GPU) | no | no | no |
| 64 GBM2 Max, M2 Ultra, M3 Max (40-core GPU), M4 Pro, M4 Max (40-core GPU) | no | no | no |
| 96 GBM2 Max, M3 Ultra | no | no | no |
| 128 GBM2 Ultra, M3 Max (40-core GPU), M4 Max (40-core GPU) | no | no | no |
| 192 GBM2 Ultra | no | no | no |
| 256 GBM3 Ultra | no | no | no |
| 512 GBM3 Ultra | no | no | no |
macOS lets the GPU use about 75% of unified memory by default; you can raise the limit with sudo sysctl iogpu.wired_limit_mb.
How fast does DeepSeek-V3 / R1 run at 4-bit?
At 4-bit, DeepSeek-V3 / R1 is too large for any single GPU or Mac we track, so it runs split across several cards. Use the inference speed estimator to model a multi-card setup.
Questions
How much VRAM does DeepSeek-V3 / R1 need?
About 424 GB at 4-bit (Q4_K_M), 736 GB at 8-bit (Q8_0) and 1384 GB at 16-bit (BF16), for one request with an 8K-token context. At 4-bit the weights are 410 GB; the rest is the KV cache and runtime overhead. With 128K tokens of context the 4-bit total rises to 433 GB.
Can I run DeepSeek-V3 / R1 on a 24 GB GPU like the RTX 4090?
Not on one card. At 4-bit it needs 424 GB, more than eight 24 GB cards.
Can I run DeepSeek-V3 / R1 on a Mac?
Not on any Mac we track, even at 4-bit: it needs 424 GB, more than the GPU can use on a 512 GB Mac Studio by default.
Why does DeepSeek-V3 / R1 need so much memory if only 37B parameters are active?
DeepSeek-V3 / R1 is a mixture-of-experts model: each token uses about 37B of its 671B parameters, but any token can be routed to any expert, so all 671B must be in memory. Memory follows the total; speed follows the active count, which is why it runs faster than a dense model of the same size.
Work it out for your setup
- Free toolVRAM calculator: DeepSeek-V3 / R1 at Q4_K_MChange the quantization, context, KV-cache precision and batch size.
- Free toolLLM inference speed estimatorTokens per second, time to first token and batching on any GPU or Mac.
- GuideCan I run it? All open-weight modelsMemory needed for every model we track, and what each GPU or Mac can run.
Sources
- DeepSeek-V3 / R1 model card and config.json — Hugging Face
