Skip to content

Llama 4 Scout 17B 16E Instruct VRAM requirements

How much VRAM Llama 4 Scout 17B 16E Instruct needs
4-bit weights: 61.3 GBWith 8K context: 63.8 GBWith 32K context: 68.3 GB8-bit, 8K context: 110 GB
Parameters
109B
Active per token
18B
Longest context
10M
Released
Apr 2025
License
Llama4
Access
Gated on the Hub
Strongest at
Text and images

Hugging Face asks you to accept its terms before it serves the weights. Its licence (Llama4) is not a standard open-source one: check the terms before commercial use.

How much VRAM Llama 4 Scout 17B 16E Instruct needs

Memory Llama 4 Scout 17B 16E Instruct needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit202 GB205 GB209 GB227 GB
8-bit108 GB110 GB114 GB132 GB
6-bit83.0 GB85.5 GB90.0 GB108 GB
5-bit72.0 GB74.5 GB79.0 GB97.0 GB
4-bit61.3 GB63.8 GB68.3 GB86.3 GB
3-bit49.5 GB51.9 GB56.4 GB74.4 GB
2-bit37.9 GB40.4 GB44.9 GB62.9 GB
1.6-bit20.9 GB23.4 GB27.9 GB45.9 GB

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

GPUs and Macs that run Llama 4 Scout 17B 16E Instruct

The best way Llama 4 Scout 17B 16E Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 4060 report
RTX 5060Too bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 5060 report
RTX 3070Too bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 3060 12 GB report
Arc B580Too bigPDF: Llama 4 Scout 17B 16E Instruct on Arc B580 report
RTX 4070Too bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 4070 report
RTX 4070 SuperToo bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 4070 Super report
RTX 5070Too bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Llama 4 Scout 17B 16E Instruct on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 5070 Ti report
RTX 5080Too bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Llama 4 Scout 17B 16E Instruct on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 3090 report
RTX 4090Too bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Llama 4 Scout 17B 16E Instruct on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Llama 4 Scout 17B 16E Instruct on RTX 5090 report
2× RTX 30902-bit28K76 tok/s1.9 sPDF: Llama 4 Scout 17B 16E Instruct on 2× RTX 3090 report
2× RTX 40902-bit28K80 tok/s0.8 sPDF: Llama 4 Scout 17B 16E Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB2-bit32K99 tok/s1.7 sPDF: Llama 4 Scout 17B 16E Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada2-bit32K77 tok/s1.2 sPDF: Llama 4 Scout 17B 16E Instruct on RTX 6000 Ada report
L40S2-bit32K71 tok/s2.1 sPDF: Llama 4 Scout 17B 16E Instruct on L40S report
2× RTX 50903-bit48K101 tok/s0.6 sPDF: Llama 4 Scout 17B 16E Instruct on 2× RTX 5090 report
4× RTX 30906-bit24K39 tok/s1.0 sPDF: Llama 4 Scout 17B 16E Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell6-bit36K68 tok/s0.9 sPDF: Llama 4 Scout 17B 16E Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Llama 4 Scout 17B 16E Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB2-bit36K27 tok/s38 sPDF: Llama 4 Scout 17B 16E Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB6-bit36K23 tok/s19 sPDF: Llama 4 Scout 17B 16E Instruct on M4 Max 128 GB report
M5 Max 128 GB6-bit36K26 tok/s5.4 sPDF: Llama 4 Scout 17B 16E Instruct on M5 Max 128 GB report
M3 Ultra 96 GB4-bit32K37 tok/s11 sPDF: Llama 4 Scout 17B 16E Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit384K23 tok/s11 sPDF: Llama 4 Scout 17B 16E Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit1.3M23 tok/s11 sPDF: Llama 4 Scout 17B 16E Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit1.3M32 tok/s3.3 sPDF: Llama 4 Scout 17B 16E Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB2-bit36K21 tok/s11 sPDF: Llama 4 Scout 17B 16E Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB6-bit36K9.9 tok/s11 sPDF: Llama 4 Scout 17B 16E Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit28K11 tok/s3.8 sPDF: Llama 4 Scout 17B 16E Instruct on RTX Spark report
DGX Spark8-bit28K9.7 tok/s3.8 sPDF: Llama 4 Scout 17B 16E Instruct on DGX Spark report
Dual DGX Spark8-bit608K9.7 tok/s2.2 sPDF: Llama 4 Scout 17B 16E Instruct on Dual DGX Spark report
DGX Station GB3008-bit672K148 tok/s0.3 sPDF: Llama 4 Scout 17B 16E Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit384K55 tok/s0.5 sPDF: Llama 4 Scout 17B 16E Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit1.3M55 tok/s0.3 sPDF: Llama 4 Scout 17B 16E Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit3.2M55 tok/s0.1 sPDF: Llama 4 Scout 17B 16E Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM5-bit16K84 tok/s2.4 sPDF: Llama 4 Scout 17B 16E Instruct on A100 80 GB SXM report
H100 SXM5-bit16K120 tok/s0.8 sPDF: Llama 4 Scout 17B 16E Instruct on H100 SXM report
4× H200 NVL8-bit2.2M116 tok/s0.3 sPDF: Llama 4 Scout 17B 16E Instruct on 4× H200 NVL report
8× H200 NVL8-bit5.0M116 tok/s0.1 sPDF: Llama 4 Scout 17B 16E Instruct 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 unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF
vLLM
vllm serve meta-llama/Llama-4-Scout-17B-16E-Instruct

Questions

How much VRAM does Llama 4 Scout 17B 16E Instruct need?

Llama 4 Scout 17B 16E Instruct needs 61.3 GB for its weights at 4-bit, 63.8 GB with an 8K context and 68.3 GB with 32K. At 8-bit, 110 GB.

What is the minimum VRAM for Llama 4 Scout 17B 16E Instruct?

About 40.4 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 2× RTX 3090.

How big is Llama 4 Scout 17B 16E Instruct?

109B parameters (18B active per token): 202 GB of weights at 16-bit, 61.3 GB at 4-bit.

Can Llama 4 Scout 17B 16E Instruct run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the 2× RTX 3090.

More models