Skip to content

OLMo 1B HF VRAM requirements

How much VRAM OLMo 1B HF needs
4-bit weights: 680 MBWith 2K context: 1.5 GB8-bit, 2K context: 2.0 GB
Parameters
1.2B
Longest context
2K
Released
Apr 2024
License
Apache 2.0
Strongest at
Not rated

How much VRAM OLMo 1B HF needs

Memory OLMo 1B HF needs at each precision and context length
PrecisionWeightsWith 2K
16-bit2.2 GB3.1 GB
8-bit1.2 GB2.0 GB
6-bit920 MB1.8 GB
5-bit798 MB1.6 GB
4-bit680 MB1.5 GB
3-bit548 MB1.4 GB
2-bit421 MB1.3 GB
1.6-bit231 MB1.1 GB

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

GPUs and Macs that run OLMo 1B HF

The best way OLMo 1B HF runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit2K142 tok/s0.2 sPDF: OLMo 1B HF on RTX 4060 report
RTX 50608-bit2K224 tok/s0.1 sPDF: OLMo 1B HF on RTX 5060 report
RTX 30708-bit2K224 tok/s0.1 sPDF: OLMo 1B HF on RTX 3070 report
RTX 3080 10 GB8-bit2K353 tok/s0.1 sPDF: OLMo 1B HF on RTX 3080 10 GB report
RTX 3060 12 GB8-bit2K184 tok/s0.2 sPDF: OLMo 1B HF on RTX 3060 12 GB report
Arc B5808-bit2K155 tok/s0.1 sPDF: OLMo 1B HF on Arc B580 report
RTX 40708-bit2K249 tok/s0.1 sPDF: OLMo 1B HF on RTX 4070 report
RTX 4070 Super8-bit2K249 tok/s0.1 sPDF: OLMo 1B HF on RTX 4070 Super report
RTX 50708-bit2K319 tok/s0.1 sPDF: OLMo 1B HF on RTX 5070 report
RTX 4060 Ti 16 GB8-bit2K150 tok/s0.1 sPDF: OLMo 1B HF on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit2K224 tok/s0.1 sPDF: OLMo 1B HF on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit2K184 tok/s0.1 sPDF: OLMo 1B HF on Arc A770 16 GB report
Radeon RX 9070 XT8-bit2K247 tok/s0.1 sPDF: OLMo 1B HF on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit2K319 tok/s0.1 sPDF: OLMo 1B HF on RTX 4070 Ti Super report
RTX 4080 Super8-bit2K344 tok/s0.1 sPDF: OLMo 1B HF on RTX 4080 Super report
RTX 5070 Ti8-bit2K404 tok/s0.1 sPDF: OLMo 1B HF on RTX 5070 Ti report
RTX 50808-bit2K427 tok/s0.1 sPDF: OLMo 1B HF on RTX 5080 report
Radeon RX 7900 XTX8-bit2K337 tok/s0.1 sPDF: OLMo 1B HF on Radeon RX 7900 XTX report
RTX 30908-bit2K419 tok/s0.1 sPDF: OLMo 1B HF on RTX 3090 report
RTX 40908-bit2K444 tok/s0.1 sPDF: OLMo 1B HF on RTX 4090 report
Radeon AI PRO R97008-bit2K247 tok/s0.1 sPDF: OLMo 1B HF on Radeon AI PRO R9700 report
RTX 50908-bit2K677 tok/s0.1 sPDF: OLMo 1B HF on RTX 5090 report
2× RTX 30908-bit2K419 tok/s0.1 sPDF: OLMo 1B HF on 2× RTX 3090 report
2× RTX 40908-bit2K444 tok/s0.1 sPDF: OLMo 1B HF on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit2K553 tok/s0.1 sPDF: OLMo 1B HF on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit2K427 tok/s0.1 sPDF: OLMo 1B HF on RTX 6000 Ada report
L40S8-bit2K393 tok/s0.1 sPDF: OLMo 1B HF on L40S report
2× RTX 50908-bit2K677 tok/s0.1 sPDF: OLMo 1B HF on 2× RTX 5090 report
4× RTX 30908-bit2K419 tok/s0.1 sPDF: OLMo 1B HF on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit2K677 tok/s0.1 sPDF: OLMo 1B HF on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit2K71 tok/s1.6 sPDF: OLMo 1B HF on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit2K71 tok/s1.6 sPDF: OLMo 1B HF on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit2K71 tok/s1.6 sPDF: OLMo 1B HF on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit2K146 tok/s0.8 sPDF: OLMo 1B HF on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit2K146 tok/s0.8 sPDF: OLMo 1B HF on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit2K146 tok/s0.8 sPDF: OLMo 1B HF on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit2K231 tok/s0.4 sPDF: OLMo 1B HF on M4 Max 128 GB report
M5 Max 128 GB8-bit2K250 tok/s0.1 sPDF: OLMo 1B HF on M5 Max 128 GB report
M3 Ultra 96 GB8-bit2K276 tok/s0.2 sPDF: OLMo 1B HF on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit2K276 tok/s0.2 sPDF: OLMo 1B HF on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit2K276 tok/s0.2 sPDF: OLMo 1B HF on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit2K347 tok/s0.1 sPDF: OLMo 1B HF on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit2K112 tok/s0.2 sPDF: OLMo 1B HF on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit2K112 tok/s0.2 sPDF: OLMo 1B HF on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit2K155 tok/s0.1 sPDF: OLMo 1B HF on RTX Spark report
DGX Spark8-bit2K142 tok/s0.1 sPDF: OLMo 1B HF on DGX Spark report
Dual DGX Spark8-bit2K142 tok/s0.1 sPDF: OLMo 1B HF on Dual DGX Spark report
DGX Station GB3008-bit2K1,367 tok/s0.1 sPDF: OLMo 1B HF on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit2K677 tok/s0.1 sPDF: OLMo 1B HF on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit2K677 tok/s0.1 sPDF: OLMo 1B HF on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit2K677 tok/s0.1 sPDF: OLMo 1B HF on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit2K737 tok/s0.1 sPDF: OLMo 1B HF on A100 80 GB SXM report
H100 SXM8-bit2K987 tok/s0.1 sPDF: OLMo 1B HF on H100 SXM report
4× H200 NVL8-bit2K1,174 tok/s0.1 sPDF: OLMo 1B HF on 4× H200 NVL report
8× H200 NVL8-bit2K1,174 tok/s0.1 sPDF: OLMo 1B HF 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/OLMo-1B-hf-GGUF
vLLM
vllm serve allenai/OLMo-1B-hf

Questions

How much VRAM does OLMo 1B HF need?

OLMo 1B HF needs 680 MB for its weights at 4-bit, 1.5 GB with an 2K context and 1.5 GB with 2K. At 8-bit, 2.0 GB.

What is the minimum VRAM for OLMo 1B HF?

About 1.4 GB at 3-bit with an 2K 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 OLMo 1B HF?

1.2B parameters: 2.2 GB of weights at 16-bit, 680 MB at 4-bit.

Can OLMo 1B HF run on an RTX 4090?

Yes. At 8-bit it holds 2K of context and writes about 444 tokens a second.

More models