Skip to content

SmolLM3 3B VRAM requirements

How much VRAM SmolLM3 3B needs
4-bit weights: 1.7 GBWith 8K context: 3.1 GBWith 32K context: 4.7 GB8-bit, 8K context: 4.4 GB
Parameters
3.1B
Longest context
64K
Released
Jul 2025
License
Apache 2.0
Strongest at
Not rated

How much VRAM SmolLM3 3B needs

Memory SmolLM3 3B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 64K
16-bit5.7 GB7.0 GB8.7 GB11.0 GB
8-bit3.0 GB4.4 GB6.1 GB8.3 GB
6-bit2.3 GB3.7 GB5.4 GB7.6 GB
5-bit2.0 GB3.4 GB5.0 GB7.3 GB
4-bit1.7 GB3.1 GB4.7 GB7.0 GB
3-bit1.4 GB2.7 GB4.4 GB6.7 GB
2-bit1.1 GB2.4 GB4.1 GB6.3 GB
1.6-bit605 MB1.9 GB3.6 GB5.8 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run SmolLM3 3B

The best way SmolLM3 3B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit52K55 tok/s0.4 sPDF: SmolLM3 3B on RTX 4060 report
RTX 50608-bit52K87 tok/s0.3 sPDF: SmolLM3 3B on RTX 5060 report
RTX 30708-bit52K87 tok/s0.3 sPDF: SmolLM3 3B on RTX 3070 report
RTX 3080 10 GB8-bit64K138 tok/s0.2 sPDF: SmolLM3 3B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit64K71 tok/s0.5 sPDF: SmolLM3 3B on RTX 3060 12 GB report
Arc B5808-bit64K61 tok/s0.3 sPDF: SmolLM3 3B on Arc B580 report
RTX 40708-bit64K97 tok/s0.2 sPDF: SmolLM3 3B on RTX 4070 report
RTX 4070 Super8-bit64K97 tok/s0.2 sPDF: SmolLM3 3B on RTX 4070 Super report
RTX 50708-bit64K125 tok/s0.2 sPDF: SmolLM3 3B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit64K58 tok/s0.3 sPDF: SmolLM3 3B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit64K87 tok/s0.3 sPDF: SmolLM3 3B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit64K72 tok/s0.3 sPDF: SmolLM3 3B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit64K97 tok/s0.2 sPDF: SmolLM3 3B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit64K125 tok/s0.1 sPDF: SmolLM3 3B on RTX 4070 Ti Super report
RTX 4080 Super8-bit64K135 tok/s0.1 sPDF: SmolLM3 3B on RTX 4080 Super report
RTX 5070 Ti8-bit64K159 tok/s0.1 sPDF: SmolLM3 3B on RTX 5070 Ti report
RTX 50808-bit64K168 tok/s0.1 sPDF: SmolLM3 3B on RTX 5080 report
Radeon RX 7900 XTX8-bit64K134 tok/s0.3 sPDF: SmolLM3 3B on Radeon RX 7900 XTX report
RTX 30908-bit64K165 tok/s0.2 sPDF: SmolLM3 3B on RTX 3090 report
RTX 40908-bit64K175 tok/s0.1 sPDF: SmolLM3 3B on RTX 4090 report
Radeon AI PRO R97008-bit64K97 tok/s0.2 sPDF: SmolLM3 3B on Radeon AI PRO R9700 report
RTX 50908-bit64K271 tok/s0.1 sPDF: SmolLM3 3B on RTX 5090 report
2× RTX 30908-bit64K165 tok/s0.1 sPDF: SmolLM3 3B on 2× RTX 3090 report
2× RTX 40908-bit64K175 tok/s0.1 sPDF: SmolLM3 3B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit64K220 tok/s0.1 sPDF: SmolLM3 3B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit64K168 tok/s0.1 sPDF: SmolLM3 3B on RTX 6000 Ada report
L40S8-bit64K154 tok/s0.1 sPDF: SmolLM3 3B on L40S report
2× RTX 50908-bit64K271 tok/s0.1 sPDF: SmolLM3 3B on 2× RTX 5090 report
4× RTX 30908-bit64K165 tok/s0.1 sPDF: SmolLM3 3B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit64K271 tok/s0.1 sPDF: SmolLM3 3B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit64K28 tok/s4.1 sPDF: SmolLM3 3B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit64K28 tok/s4.1 sPDF: SmolLM3 3B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit64K28 tok/s4.1 sPDF: SmolLM3 3B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit64K57 tok/s2.0 sPDF: SmolLM3 3B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit64K57 tok/s2.0 sPDF: SmolLM3 3B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit64K57 tok/s2.0 sPDF: SmolLM3 3B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit64K92 tok/s1.0 sPDF: SmolLM3 3B on M4 Max 128 GB report
M5 Max 128 GB8-bit64K100 tok/s0.3 sPDF: SmolLM3 3B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit64K111 tok/s0.6 sPDF: SmolLM3 3B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit64K111 tok/s0.6 sPDF: SmolLM3 3B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit64K111 tok/s0.6 sPDF: SmolLM3 3B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit64K141 tok/s0.2 sPDF: SmolLM3 3B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit64K43 tok/s0.6 sPDF: SmolLM3 3B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit64K43 tok/s0.6 sPDF: SmolLM3 3B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit64K60 tok/s0.2 sPDF: SmolLM3 3B on RTX Spark report
DGX Spark8-bit64K55 tok/s0.2 sPDF: SmolLM3 3B on DGX Spark report
Dual DGX Spark8-bit64K55 tok/s0.1 sPDF: SmolLM3 3B on Dual DGX Spark report
DGX Station GB3008-bit64K576 tok/s0.1 sPDF: SmolLM3 3B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit64K271 tok/s0.1 sPDF: SmolLM3 3B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit64K271 tok/s0.1 sPDF: SmolLM3 3B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit64K271 tok/s0.1 sPDF: SmolLM3 3B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit64K297 tok/s0.1 sPDF: SmolLM3 3B on A100 80 GB SXM report
H100 SXM8-bit64K404 tok/s0.1 sPDF: SmolLM3 3B on H100 SXM report
4× H200 NVL8-bit64K487 tok/s0.1 sPDF: SmolLM3 3B on 4× H200 NVL report
8× H200 NVL8-bit64K487 tok/s0.1 sPDF: SmolLM3 3B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf bartowski/HuggingFaceTB_SmolLM3-3B-GGUF
vLLM
vllm serve HuggingFaceTB/SmolLM3-3B
MLX (Mac)
mlx_lm.chat --model mlx-community/SmolLM3-3B-4bit

Questions

How much VRAM does SmolLM3 3B need?

SmolLM3 3B needs 1.7 GB for its weights at 4-bit, 3.1 GB with an 8K context and 4.7 GB with 32K. At 8-bit, 4.4 GB.

What is the minimum VRAM for SmolLM3 3B?

About 2.7 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is SmolLM3 3B?

3.1B parameters: 5.7 GB of weights at 16-bit, 1.7 GB at 4-bit.

Can SmolLM3 3B run on an RTX 4090?

Yes. At 8-bit it holds 64K of context and writes about 175 tokens a second.

More models