Skip to content

Internlm2 Chat 20B VRAM requirements

How much VRAM Internlm2 Chat 20B needs
4-bit weights: 11.2 GBWith 8K context: 13.5 GBWith 32K context: 18.0 GB8-bit, 8K context: 22.0 GB
Parameters
19.9B
Longest context
32K
Released
Jan 2024
License
Custom
Strongest at
Not rated

Its licence (Custom) is not a standard open-source one: check the terms before commercial use.

How much VRAM Internlm2 Chat 20B needs

Memory Internlm2 Chat 20B needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit37.0 GB39.3 GB43.8 GB
8-bit19.7 GB22.0 GB26.5 GB
6-bit15.2 GB17.5 GB22.0 GB
5-bit13.2 GB15.5 GB20.0 GB
4-bit11.2 GB13.5 GB18.0 GB
3-bit9.0 GB11.4 GB15.9 GB
2-bit6.9 GB9.3 GB13.8 GB
1.6-bit3.8 GB6.1 GB10.6 GB

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

GPUs and Macs that run Internlm2 Chat 20B

The best way Internlm2 Chat 20B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Internlm2 Chat 20B on RTX 4060 report
RTX 5060Too bigPDF: Internlm2 Chat 20B on RTX 5060 report
RTX 3070Too bigPDF: Internlm2 Chat 20B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Internlm2 Chat 20B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Internlm2 Chat 20B on RTX 3060 12 GB report
Arc B580Too bigPDF: Internlm2 Chat 20B on Arc B580 report
RTX 4070Too bigPDF: Internlm2 Chat 20B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Internlm2 Chat 20B on RTX 4070 Super report
RTX 5070Too bigPDF: Internlm2 Chat 20B on RTX 5070 report
RTX 4060 Ti 16 GB4-bit16K17 tok/s1.9 sPDF: Internlm2 Chat 20B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB4-bit16K26 tok/s1.8 sPDF: Internlm2 Chat 20B on RTX 5060 Ti 16 GB report
Arc A770 16 GB4-bit16K22 tok/s2.0 sPDF: Internlm2 Chat 20B on Arc A770 16 GB report
Radeon RX 9070 XT4-bit16K30 tok/s1.2 sPDF: Internlm2 Chat 20B on Radeon RX 9070 XT report
RTX 4070 Ti Super4-bit16K38 tok/s0.9 sPDF: Internlm2 Chat 20B on RTX 4070 Ti Super report
RTX 4080 Super4-bit16K41 tok/s0.8 sPDF: Internlm2 Chat 20B on RTX 4080 Super report
RTX 5070 Ti4-bit16K50 tok/s0.9 sPDF: Internlm2 Chat 20B on RTX 5070 Ti report
RTX 50804-bit16K53 tok/s0.7 sPDF: Internlm2 Chat 20B on RTX 5080 report
Radeon RX 7900 XTX6-bit32K33 tok/s1.8 sPDF: Internlm2 Chat 20B on Radeon RX 7900 XTX report
RTX 30906-bit32K39 tok/s1.2 sPDF: Internlm2 Chat 20B on RTX 3090 report
RTX 40906-bit32K42 tok/s0.5 sPDF: Internlm2 Chat 20B on RTX 4090 report
Radeon AI PRO R97008-bit32K18 tok/s1.2 sPDF: Internlm2 Chat 20B on Radeon AI PRO R9700 report
RTX 50908-bit32K56 tok/s0.4 sPDF: Internlm2 Chat 20B on RTX 5090 report
2× RTX 30908-bit32K31 tok/s0.7 sPDF: Internlm2 Chat 20B on 2× RTX 3090 report
2× RTX 40908-bit32K33 tok/s0.3 sPDF: Internlm2 Chat 20B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K43 tok/s0.6 sPDF: Internlm2 Chat 20B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K31 tok/s0.4 sPDF: Internlm2 Chat 20B on RTX 6000 Ada report
L40S8-bit32K28 tok/s0.7 sPDF: Internlm2 Chat 20B on L40S report
2× RTX 50908-bit32K56 tok/s0.2 sPDF: Internlm2 Chat 20B on 2× RTX 5090 report
4× RTX 30908-bit32K31 tok/s0.4 sPDF: Internlm2 Chat 20B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K56 tok/s0.3 sPDF: Internlm2 Chat 20B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Internlm2 Chat 20B on Mac mini M4 16 GB report
Mac mini M4 24 GB4-bit16K8.1 tok/s26 sPDF: Internlm2 Chat 20B on Mac mini M4 24 GB report
Mac mini M4 32 GB6-bit20K6.1 tok/s26 sPDF: Internlm2 Chat 20B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB4-bit16K18 tok/s13 sPDF: Internlm2 Chat 20B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K10 tok/s13 sPDF: Internlm2 Chat 20B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K10 tok/s13 sPDF: Internlm2 Chat 20B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K19 tok/s6.6 sPDF: Internlm2 Chat 20B on M4 Max 128 GB report
M5 Max 128 GB8-bit32K21 tok/s1.9 sPDF: Internlm2 Chat 20B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K24 tok/s3.7 sPDF: Internlm2 Chat 20B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K24 tok/s3.7 sPDF: Internlm2 Chat 20B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K24 tok/s3.7 sPDF: Internlm2 Chat 20B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K33 tok/s1.2 sPDF: Internlm2 Chat 20B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K7.4 tok/s3.8 sPDF: Internlm2 Chat 20B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K7.4 tok/s3.8 sPDF: Internlm2 Chat 20B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K10 tok/s1.3 sPDF: Internlm2 Chat 20B on RTX Spark report
DGX Spark8-bit32K9.2 tok/s1.3 sPDF: Internlm2 Chat 20B on DGX Spark report
Dual DGX Spark8-bit32K9.2 tok/s0.8 sPDF: Internlm2 Chat 20B on Dual DGX Spark report
DGX Station GB3008-bit32K180 tok/s0.1 sPDF: Internlm2 Chat 20B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K56 tok/s0.2 sPDF: Internlm2 Chat 20B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K56 tok/s0.1 sPDF: Internlm2 Chat 20B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K56 tok/s0.1 sPDF: Internlm2 Chat 20B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K63 tok/s0.8 sPDF: Internlm2 Chat 20B on A100 80 GB SXM report
H100 SXM8-bit32K98 tok/s0.3 sPDF: Internlm2 Chat 20B on H100 SXM report
4× H200 NVL8-bit32K133 tok/s0.1 sPDF: Internlm2 Chat 20B on 4× H200 NVL report
8× H200 NVL8-bit32K133 tok/s0.1 sPDF: Internlm2 Chat 20B 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 mradermacher/internlm2-chat-20b-i1-GGUF
vLLM
vllm serve internlm/internlm2-chat-20b

Questions

How much VRAM does Internlm2 Chat 20B need?

Internlm2 Chat 20B needs 11.2 GB for its weights at 4-bit, 13.5 GB with an 8K context and 18.0 GB with 32K. At 8-bit, 22.0 GB.

What is the minimum VRAM for Internlm2 Chat 20B?

About 11.4 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 Ti 16 GB.

How big is Internlm2 Chat 20B?

19.9B parameters: 37.0 GB of weights at 16-bit, 11.2 GB at 4-bit.

Can Internlm2 Chat 20B run on an RTX 4090?

Yes. At 6-bit it holds 32K of context and writes about 42 tokens a second.

More models