Skip to content

Intern S1 VRAM requirements

How much VRAM Intern S1 needs
4-bit weights: 136 GBWith 8K context: 138 GBWith 32K context: 143 GB8-bit, 8K context: 241 GB
Parameters
241B
Active per token
27.8B
Longest context
64K
Released
Jul 2025
License
Apache 2.0
Strongest at
Text and images

How much VRAM Intern S1 needs

Memory Intern S1 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 64K
16-bit448 GB451 GB455 GB461 GB
8-bit238 GB241 GB245 GB251 GB
6-bit184 GB186 GB191 GB196 GB
5-bit159 GB162 GB166 GB172 GB
4-bit136 GB138 GB143 GB149 GB
3-bit110 GB112 GB116 GB122 GB
2-bit84.1 GB86.4 GB90.8 GB96.7 GB
1.6-bit46.2 GB48.6 GB53.0 GB58.9 GB

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

GPUs and Macs that run Intern S1

The best way Intern S1 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Intern S1 on RTX 4060 report
RTX 5060Too bigPDF: Intern S1 on RTX 5060 report
RTX 3070Too bigPDF: Intern S1 on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Intern S1 on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Intern S1 on RTX 3060 12 GB report
Arc B580Too bigPDF: Intern S1 on Arc B580 report
RTX 4070Too bigPDF: Intern S1 on RTX 4070 report
RTX 4070 SuperToo bigPDF: Intern S1 on RTX 4070 Super report
RTX 5070Too bigPDF: Intern S1 on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Intern S1 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Intern S1 on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Intern S1 on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Intern S1 on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Intern S1 on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Intern S1 on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Intern S1 on RTX 5070 Ti report
RTX 5080Too bigPDF: Intern S1 on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Intern S1 on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Intern S1 on RTX 3090 report
RTX 4090Too bigPDF: Intern S1 on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Intern S1 on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Intern S1 on RTX 5090 report
2× RTX 3090Too bigPDF: Intern S1 on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Intern S1 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Intern S1 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Intern S1 on RTX 6000 Ada report
L40SToo bigPDF: Intern S1 on L40S report
2× RTX 5090Too bigPDF: Intern S1 on 2× RTX 5090 report
4× RTX 30902-bit20K46 tok/s1.7 sPDF: Intern S1 on 4× RTX 3090 report
RTX PRO 6000 Blackwell2-bit32K72 tok/s1.5 sPDF: Intern S1 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Intern S1 on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Intern S1 on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Intern S1 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Intern S1 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Intern S1 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Intern S1 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB2-bit32K26 tok/s31 sPDF: Intern S1 on M4 Max 128 GB report
M5 Max 128 GB2-bit32K28 tok/s9.0 sPDF: Intern S1 on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Intern S1 on M3 Ultra 96 GB report
M3 Ultra 256 GB5-bit64K20 tok/s18 sPDF: Intern S1 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit64K15 tok/s18 sPDF: Intern S1 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit64K20 tok/s5.5 sPDF: Intern S1 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Intern S1 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB2-bit32K13 tok/s18 sPDF: Intern S1 on Ryzen AI Max+ 395 128 GB report
RTX Spark3-bit20K14 tok/s6.3 sPDF: Intern S1 on RTX Spark report
DGX Spark3-bit20K13 tok/s6.3 sPDF: Intern S1 on DGX Spark report
Dual DGX Spark6-bit64K8.0 tok/s3.6 sPDF: Intern S1 on Dual DGX Spark report
DGX Station GB3008-bit64K86 tok/s0.5 sPDF: Intern S1 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell5-bit64K47 tok/s0.9 sPDF: Intern S1 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit64K34 tok/s0.5 sPDF: Intern S1 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit64K34 tok/s0.2 sPDF: Intern S1 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Intern S1 on A100 80 GB SXM report
H100 SXMToo bigPDF: Intern S1 on H100 SXM report
4× H200 NVL8-bit64K69 tok/s0.5 sPDF: Intern S1 on 4× H200 NVL report
8× H200 NVL8-bit64K69 tok/s0.2 sPDF: Intern S1 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/internlm_Intern-S1-GGUF
vLLM
vllm serve internlm/Intern-S1

Questions

How much VRAM does Intern S1 need?

Intern S1 needs 136 GB for its weights at 4-bit, 138 GB with an 8K context and 143 GB with 32K. At 8-bit, 241 GB.

What is the minimum VRAM for Intern S1?

About 86.4 GB at 2-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 4× RTX 3090.

How big is Intern S1?

241B parameters (27.8B active per token): 448 GB of weights at 16-bit, 136 GB at 4-bit.

Can Intern S1 run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the 4× RTX 3090.

More models