RunsOnMy ModelsGPUs & MacsBy VRAMAPI pricesVRAM calculatorNew

Can I run deepseek moe 16b base locally?

deepseek moe 16b base by deepseek-ai has 16B parameters (2.8B active per token, 64 experts, 6 used per token). At 4-bit with an 8K context it needs about 12.5 GB of GPU memory. It fits on 59 of 65 GPUs and Macs we track. See all models that fit in 16 GB of VRAM.

12.5 GBQ4_K_M · 8K context
RTX 4060 Ti 16GBsmallest consumer GPU that fits
M2 24GBsmallest Mac that fits
4Kmax 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 mradermacher/deepseek-moe-16b-base-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.

QuantQuality4K ctx
2-bit (Q2_K) •rough7.7 GB
3-bit (Q3_K_M) •usable9.4 GB
4-bit (Q4_K_M) •good, most popular11.6 GB
5-bit (Q5_K_M) •very good13.0 GB
6-bit (Q6_K) •near lossless15.1 GB
8-bit (Q8_0) •effectively lossless17.7 GB
16-bit (F16/BF16)original weights32.0 GB

Which GPUs and Macs run deepseek moe 16b base?

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 GBFitsQ3_K_M10.2 GB246details
RTX 4060 Ti 16GB16 GBFitsQ4_K_M12.5 GB154details
RTX 309024 GBFitsQ8_018.6 GB311details
RTX 409024 GBFitsQ8_018.6 GB335details
RTX 509032 GBFitsQ8_018.6 GB596details
RTX 5070 Ti16 GBFitsQ4_K_M12.5 GB478details
RX 7900 XTX24 GBFitsQ8_018.6 GB319details
Mac M4 Pro 48GB48 GBFitsQ8_018.6 GB91details
Mac M4 Max 128GB128 GBFitsF1632.8 GB97details
Mac M3 Ultra 512GB512 GBFitsF1632.8 GB145details
DGX Spark128 GBFitsF1632.8 GB48details
RTX 407012 GBFitsQ3_K_M10.2 GB344details
RTX 5060 Ti 16GB16 GBFitsQ4_K_M12.5 GB239details
RTX 40608 GBNo———details
Arc B58012 GBFitsQ3_K_M10.2 GB311details
RTX 3060 Ti8 GBNo———details
RTX 30708 GBNo———details
RTX 3080 10GB10 GBTightQ2_K8.6 GB656details
RTX 3080 12GB12 GBFitsQ3_K_M10.2 GB623details
RTX 3090 Ti24 GBFitsQ8_018.6 GB335details
RTX 4060 Ti 8GB8 GBNo———details
RTX 4070 Super12 GBFitsQ3_K_M10.2 GB344details
RTX 4070 Ti12 GBFitsQ3_K_M10.2 GB344details
RTX 4070 Ti Super16 GBFitsQ4_K_M12.5 GB359details
RTX 408016 GBFitsQ4_K_M12.5 GB383details
RTX 4080 Super16 GBFitsQ4_K_M12.5 GB393details
RTX 50608 GBNo———details
RTX 507012 GBFitsQ3_K_M10.2 GB459details
RTX 508016 GBFitsQ4_K_M12.5 GB512details
2x RTX 30902×24 GBFitsF1632.8 GB165details
2x RTX 40902×24 GBFitsF1632.8 GB178details
2x RTX 50902×32 GBFitsF1632.8 GB317details
RTX A600048 GBFitsF1632.8 GB136details
RTX 6000 Ada48 GBFitsF1632.8 GB170details
RTX PRO 6000 Blackwell96 GBFitsF1632.8 GB317details
L40S48 GBFitsF1632.8 GB153details
A100 40GB40 GBFitsF1632.8 GB275details
A100 80GB80 GBFitsF1632.8 GB360details
H100 PCIe80 GBFitsF1632.8 GB354details
H100 SXM80 GBFitsF1632.8 GB592details
H200141 GBFitsF1632.8 GB848details
RX 76008 GBNo———details
RX 7700 XT12 GBFitsQ3_K_M10.2 GB295details
RX 7800 XT16 GBFitsQ4_K_M12.5 GB333details
RX 7900 GRE16 GBFitsQ4_K_M12.5 GB307details
RX 7900 XT20 GBFitsQ6_K16.0 GB316details
RX 907016 GBFitsQ4_K_M12.5 GB341details
RX 9070 XT16 GBFitsQ4_K_M12.5 GB341details
Ryzen AI Max+ 395 128GB128 GBFitsF1632.8 GB45details
Arc A770 16GB16 GBFitsQ4_K_M12.5 GB299details
Mac M1 16GB16 GBFitsQ2_K8.6 GB59details
Mac M2 24GB24 GBFitsQ5_K_M13.9 GB47details
Mac M3 24GB24 GBFitsQ5_K_M13.9 GB47details
Mac M4 16GB16 GBFitsQ2_K8.6 GB104details
Mac M4 32GB32 GBFitsQ8_018.6 GB40details
Mac M1 Pro 32GB32 GBFitsQ8_018.6 GB66details
Mac M3 Pro 36GB36 GBFitsQ8_018.6 GB50details
Mac M4 Pro 24GB24 GBFitsQ5_K_M13.9 GB128details
Mac M4 Pro 64GB64 GBFitsF1632.8 GB48details
Mac M1 Max 64GB64 GBFitsF1632.8 GB71details
Mac M2 Max 96GB96 GBFitsF1632.8 GB71details
Mac M3 Max 128GB128 GBFitsF1632.8 GB71details
Mac M4 Max 64GB64 GBFitsF1632.8 GB97details
Mac M2 Ultra 192GB192 GBFitsF1632.8 GB141details
Mac M3 Ultra 256GB256 GBFitsF1632.8 GB145details

Questions

How much VRAM does deepseek moe 16b base need?

About 12.5 GB at 4-bit (Q4_K_M) with an 8K context: 10.1 GB of weights (real GGUF file size), 1.8 GB of KV cache and 0.6 GB of runtime overhead. At 8-bit it needs about 18.6 GB.

What is the smallest GPU that runs deepseek moe 16b base?

At Q4_K_M and 8K context, the smallest single consumer GPU in our table that fits it is the RTX 4060 Ti 16GB (16 GB).

Can I run deepseek moe 16b base on a Mac?

Yes. At Q4_K_M with 8K context it fits on a Mac M2 24GB or larger, using the default macOS GPU memory limit.

Similar models

Same family or a similar parameter count, with the VRAM each needs at Q4_K_M and 8K context.

DeepSeek Coder V2 Lite Instruct

16B · 10.5 GB at 4-bit

DeepSeek V2 Lite

16B · 9.8 GB at 4-bit

DeepSeek V2 Lite Chat

16B · 10.5 GB at 4-bit

Kimi VL A3B Thinking 2506

16B · 10.8 GB at 4-bit

Kimi VL A3B Instruct

16B · 10.8 GB at 4-bit

Kimi VL A3B Thinking

16B · 10.3 GB at 4-bit

Model facts

Hugging Face repodeepseek-ai/deepseek-moe-16b-base
Parameters16,375,728,128 (16B) · ~2.8B active (computed from config)
Layers / hidden size28 / 2048
AttentionMHA; 16 heads, 16 KV heads, head dim 128
KV cache per token (FP16)224.0 KiB
Max context4,096 tokens
Licenseother
Released on HFJan 8, 2024
Downloads (30d) / likes25,584 / 170

Sources: Hugging Face model API · config.json (deepseek-ai/deepseek-moe-16b-base) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.