Skip to content

FastVLM 0.5B VRAM requirements

How much VRAM FastVLM 0.5B needs
4-bit weights: 439 MBWith 8K context: 1.3 GBWith 32K context: 1.6 GB8-bit, 8K context: 1.6 GB
Parameters
759M
Longest context
32K
Released
Aug 2025
License
apple-amlr
Strongest at
Not rated

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

How much VRAM FastVLM 0.5B needs

Memory FastVLM 0.5B needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit1.4 GB2.3 GB2.6 GB
8-bit769 MB1.6 GB1.9 GB
6-bit593 MB1.4 GB1.7 GB
5-bit515 MB1.4 GB1.6 GB
4-bit439 MB1.3 GB1.6 GB
3-bit354 MB1.2 GB1.5 GB
2-bit271 MB1.1 GB1.4 GB
1.6-bit149 MB1.0 GB1.3 GB

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

GPUs and Macs that run FastVLM 0.5B

The best way FastVLM 0.5B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit32K202 tok/s0.1 sPDF: FastVLM 0.5B on RTX 4060 report
RTX 50608-bit32K304 tok/s0.1 sPDF: FastVLM 0.5B on RTX 5060 report
RTX 30708-bit32K304 tok/s0.1 sPDF: FastVLM 0.5B on RTX 3070 report
RTX 3080 10 GB8-bit32K447 tok/s0.1 sPDF: FastVLM 0.5B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit32K255 tok/s0.1 sPDF: FastVLM 0.5B on RTX 3060 12 GB report
Arc B5808-bit32K201 tok/s0.1 sPDF: FastVLM 0.5B on Arc B580 report
RTX 40708-bit32K333 tok/s0.1 sPDF: FastVLM 0.5B on RTX 4070 report
RTX 4070 Super8-bit32K333 tok/s0.1 sPDF: FastVLM 0.5B on RTX 4070 Super report
RTX 50708-bit32K411 tok/s0.1 sPDF: FastVLM 0.5B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K212 tok/s0.1 sPDF: FastVLM 0.5B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K304 tok/s0.1 sPDF: FastVLM 0.5B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K232 tok/s0.1 sPDF: FastVLM 0.5B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K303 tok/s0.1 sPDF: FastVLM 0.5B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K411 tok/s0.1 sPDF: FastVLM 0.5B on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K438 tok/s0.1 sPDF: FastVLM 0.5B on RTX 4080 Super report
RTX 5070 Ti8-bit32K499 tok/s0.1 sPDF: FastVLM 0.5B on RTX 5070 Ti report
RTX 50808-bit32K521 tok/s0.1 sPDF: FastVLM 0.5B on RTX 5080 report
Radeon RX 7900 XTX8-bit32K385 tok/s0.1 sPDF: FastVLM 0.5B on Radeon RX 7900 XTX report
RTX 30908-bit32K513 tok/s0.1 sPDF: FastVLM 0.5B on RTX 3090 report
RTX 40908-bit32K537 tok/s0.1 sPDF: FastVLM 0.5B on RTX 4090 report
Radeon AI PRO R97008-bit32K303 tok/s0.1 sPDF: FastVLM 0.5B on Radeon AI PRO R9700 report
RTX 50908-bit32K734 tok/s0.1 sPDF: FastVLM 0.5B on RTX 5090 report
2× RTX 30908-bit32K513 tok/s0.1 sPDF: FastVLM 0.5B on 2× RTX 3090 report
2× RTX 40908-bit32K537 tok/s0.1 sPDF: FastVLM 0.5B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K634 tok/s0.1 sPDF: FastVLM 0.5B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K521 tok/s0.1 sPDF: FastVLM 0.5B on RTX 6000 Ada report
L40S8-bit32K487 tok/s0.1 sPDF: FastVLM 0.5B on L40S report
2× RTX 50908-bit32K734 tok/s0.1 sPDF: FastVLM 0.5B on 2× RTX 5090 report
4× RTX 30908-bit32K513 tok/s0.1 sPDF: FastVLM 0.5B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K734 tok/s0.1 sPDF: FastVLM 0.5B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit32K99 tok/s0.9 sPDF: FastVLM 0.5B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K99 tok/s0.9 sPDF: FastVLM 0.5B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K99 tok/s0.9 sPDF: FastVLM 0.5B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K181 tok/s0.5 sPDF: FastVLM 0.5B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K181 tok/s0.5 sPDF: FastVLM 0.5B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K181 tok/s0.5 sPDF: FastVLM 0.5B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K257 tok/s0.2 sPDF: FastVLM 0.5B on M4 Max 128 GB report
M5 Max 128 GB8-bit32K272 tok/s0.1 sPDF: FastVLM 0.5B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K291 tok/s0.1 sPDF: FastVLM 0.5B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K291 tok/s0.1 sPDF: FastVLM 0.5B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K291 tok/s0.1 sPDF: FastVLM 0.5B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K339 tok/s0.1 sPDF: FastVLM 0.5B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K155 tok/s0.1 sPDF: FastVLM 0.5B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K155 tok/s0.1 sPDF: FastVLM 0.5B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K219 tok/s0.1 sPDF: FastVLM 0.5B on RTX Spark report
DGX Spark8-bit32K202 tok/s0.1 sPDF: FastVLM 0.5B on DGX Spark report
Dual DGX Spark8-bit32K202 tok/s0.1 sPDF: FastVLM 0.5B on Dual DGX Spark report
DGX Station GB3008-bit32K1,133 tok/s0.1 sPDF: FastVLM 0.5B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K734 tok/s0.1 sPDF: FastVLM 0.5B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K734 tok/s0.1 sPDF: FastVLM 0.5B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K734 tok/s0.1 sPDF: FastVLM 0.5B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K778 tok/s0.1 sPDF: FastVLM 0.5B on A100 80 GB SXM report
H100 SXM8-bit32K940 tok/s0.1 sPDF: FastVLM 0.5B on H100 SXM report
4× H200 NVL8-bit32K1,042 tok/s0.1 sPDF: FastVLM 0.5B on 4× H200 NVL report
8× H200 NVL8-bit32K1,042 tok/s0.1 sPDF: FastVLM 0.5B 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 gguf-org/fastvlm-gguf
vLLM
vllm serve apple/FastVLM-0.5B

Questions

How much VRAM does FastVLM 0.5B need?

FastVLM 0.5B needs 439 MB for its weights at 4-bit, 1.3 GB with an 8K context and 1.6 GB with 32K. At 8-bit, 1.6 GB.

What is the minimum VRAM for FastVLM 0.5B?

About 1.2 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 FastVLM 0.5B?

759M parameters: 1.4 GB of weights at 16-bit, 439 MB at 4-bit.

Can FastVLM 0.5B run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 537 tokens a second.

More models