RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

Can I run DeepSeek OCR locally?

DeepSeek OCR by deepseek-ai has 3.3B parameters (1.1B active per token, 64 experts, 6 used per token). At 4-bit with an 8K context it needs about 2.9 GB of GPU memory. It fits on 65 of 65 GPUs and Macs we track.

2.9 GBQ4_K_M · 8K context
RTX 4060smallest consumer GPU that fits
M1 16GBsmallest Mac that fits
8Kmax context
MHAattention type

Memory needed by quant and context length

GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use real file sizes from sabafallah/DeepSeek-OCR-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.

QuantQuality4K ctx8K ctx
2-bit (Q2_K)rough2.0 GB2.3 GB
3-bit (Q3_K_M)usable2.4 GB2.6 GB
4-bit (Q4_K_M) •good, most popular2.7 GB2.9 GB
5-bit (Q5_K_M)very good3.1 GB3.3 GB
6-bit (Q6_K)near lossless3.4 GB3.7 GB
8-bit (Q8_0) •effectively lossless3.8 GB4.0 GB
16-bit (F16/BF16) •original weights6.4 GB6.6 GB

Which GPUs and Macs run DeepSeek OCR?

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).

HardwareMemoryFits?Best quantNeedsMax tok/s (est.)
RTX 3060 12GB12 GBFitsF166.6 GB179details
RTX 4060 Ti 16GB16 GBFitsF166.6 GB143details
RTX 309024 GBFitsF166.6 GB466details
RTX 409024 GBFitsF166.6 GB502details
RTX 509032 GBFitsF166.6 GB892details
RTX 5070 Ti16 GBFitsF166.6 GB446details
RX 7900 XTX24 GBFitsF166.6 GB478details
Mac M4 Pro 48GB48 GBFitsF166.6 GB136details
Mac M4 Max 128GB128 GBFitsF166.6 GB272details
Mac M3 Ultra 512GB512 GBFitsF166.6 GB407details
DGX Spark128 GBFitsF166.6 GB136details
RTX 407012 GBFitsF166.6 GB251details
RTX 5060 Ti 16GB16 GBFitsF166.6 GB223details
RTX 40608 GBFitsF166.6 GB135details
Arc B58012 GBFitsF166.6 GB227details
RTX 3060 Ti8 GBFitsF166.6 GB223details
RTX 30708 GBFitsF166.6 GB223details
RTX 3080 10GB10 GBFitsF166.6 GB378details
RTX 3080 12GB12 GBFitsF166.6 GB454details
RTX 3090 Ti24 GBFitsF166.6 GB502details
RTX 4060 Ti 8GB8 GBFitsF166.6 GB143details
RTX 4070 Super12 GBFitsF166.6 GB251details
RTX 4070 Ti12 GBFitsF166.6 GB251details
RTX 4070 Ti Super16 GBFitsF166.6 GB334details
RTX 408016 GBFitsF166.6 GB357details
RTX 4080 Super16 GBFitsF166.6 GB366details
RTX 50608 GBFitsF166.6 GB223details
RTX 507012 GBFitsF166.6 GB334details
RTX 508016 GBFitsF166.6 GB478details
2x RTX 30902×24 GBFitsF166.6 GB466details
2x RTX 40902×24 GBFitsF166.6 GB502details
2x RTX 50902×32 GBFitsF166.6 GB892details
RTX A600048 GBFitsF166.6 GB382details
RTX 6000 Ada48 GBFitsF166.6 GB478details
RTX PRO 6000 Blackwell96 GBFitsF166.6 GB892details
L40S48 GBFitsF166.6 GB430details
A100 40GB40 GBFitsF166.6 GB774details
A100 80GB80 GBFitsF166.6 GB1015details
H100 PCIe80 GBFitsF166.6 GB995details
H100 SXM80 GBFitsF166.6 GB1667details
H200141 GBFitsF166.6 GB2388details
RX 76008 GBFitsF166.6 GB143details
RX 7700 XT12 GBFitsF166.6 GB215details
RX 7800 XT16 GBFitsF166.6 GB310details
RX 7900 GRE16 GBFitsF166.6 GB287details
RX 7900 XT20 GBFitsF166.6 GB398details
RX 907016 GBFitsF166.6 GB318details
RX 9070 XT16 GBFitsF166.6 GB318details
Ryzen AI Max+ 395 128GB128 GBFitsF166.6 GB127details
Arc A770 16GB16 GBFitsF166.6 GB279details
Mac M1 16GB16 GBFitsF166.6 GB34details
Mac M2 24GB24 GBFitsF166.6 GB50details
Mac M3 24GB24 GBFitsF166.6 GB50details
Mac M4 16GB16 GBFitsF166.6 GB60details
Mac M4 32GB32 GBFitsF166.6 GB60details
Mac M1 Pro 32GB32 GBFitsF166.6 GB100details
Mac M3 Pro 36GB36 GBFitsF166.6 GB75details
Mac M4 Pro 24GB24 GBFitsF166.6 GB136details
Mac M4 Pro 64GB64 GBFitsF166.6 GB136details
Mac M1 Max 64GB64 GBFitsF166.6 GB199details
Mac M2 Max 96GB96 GBFitsF166.6 GB199details
Mac M3 Max 128GB128 GBFitsF166.6 GB199details
Mac M4 Max 64GB64 GBFitsF166.6 GB272details
Mac M2 Ultra 192GB192 GBFitsF166.6 GB398details
Mac M3 Ultra 256GB256 GBFitsF166.6 GB407details

Questions

How much VRAM does DeepSeek OCR need?

About 2.9 GB at 4-bit (Q4_K_M) with an 8K context: 1.8 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 4.0 GB.

What is the smallest GPU that runs DeepSeek OCR?

At Q4_K_M and 8K context, the smallest single consumer GPU in our table that fits it is the RTX 4060 (8 GB).

Can I run DeepSeek OCR on a Mac?

Yes. At Q4_K_M with 8K context it fits on a Mac M1 16GB or larger, using the default macOS GPU memory limit.

Model facts

Hugging Face repodeepseek-ai/DeepSeek-OCR
Parameters3,336,106,240 (3.3B) · ~1.1B active (computed from config)
Layers / hidden size12 / 1280
AttentionMHA; 10 heads, 10 KV heads, head dim 128
KV cache per token (FP16)60.0 KiB
Max context8,192 tokens
Licensemit
Released on HFOct 17, 2025
Downloads (30d) / likes1,828,807 / 3,466

Sources: Hugging Face model API · config.json (deepseek-ai/DeepSeek-OCR) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.