Question 1

How much VRAM does Falcon3-7B-Instruct-1.58bit need?

Accepted Answer

Falcon3-7B-Instruct-1.58bit (2.5B) requires approximately 5.516517314314842 GB of VRAM at full precision (FP16), 2.8 GB at 8-bit quantization (Q8), and 1.4 GB at 4-bit quantization (Q4). The actual VRAM usage may be slightly higher due to KV cache and runtime overhead.

Question 2

What GPUs can run Falcon3-7B-Instruct-1.58bit?

Accepted Answer

We have 10 compatible GPUs indexed for Falcon3-7B-Instruct-1.58bit, including L4, M60, RTX A4000, T4, A40 and 5 more. Compatibility is determined by whether the GPU's VRAM can fit the model's memory footprint, potentially using quantization to reduce memory usage.

Question 3

What is the cheapest GPU to run Falcon3-7B-Instruct-1.58bit?

Accepted Answer

The cheapest GPU to run Falcon3-7B-Instruct-1.58bit in the cloud is the L4 at $0.20/hr. Other affordable options include M60 ($0.23/hr) and RTX A4000 ($0.25/hr). Prices reflect the lowest on-demand rate across cloud providers.

Question 4

Can Falcon3-7B-Instruct-1.58bit run on consumer hardware?

Accepted Answer

Yes, Falcon3-7B-Instruct-1.58bit can run on consumer GPUs like the RTX A4000 (16 GB VRAM) at full precision (FP16). Other compatible consumer GPUs include RTX 3090 and RTX A4500.

Question 5

What quantization options are available for Falcon3-7B-Instruct-1.58bit?

Accepted Answer

Falcon3-7B-Instruct-1.58bit supports the following quantization levels: FP16 (full precision): ~5.516517314314842 GB, Q8 (8-bit): ~2.8 GB, Q6 (6-bit): ~2.1 GB, Q4 (4-bit): ~1.4 GB. Lower quantization reduces VRAM usage at the cost of some quality loss. Q8 offers a good balance between quality and memory savings, while Q4 provides maximum compression.

Falcon3-7B-Instruct-1.58bit

VRAM Requirements

Compatible GPUs

Falcon3-7B-Instruct-1.58bit FAQ

GPU	VRAM	Best Quant	From	Rent
L4	24 GB	FP16	$0.20/hr
M60	8 GB	FP16	$0.23/hr
RTX A4000	16 GB	FP16	$0.25/hr	Rent
T4	16 GB	FP16	$0.35/hr
A40	48 GB	FP16	$0.44/hr	Rent
RTX 3090	24 GB	FP16	$0.46/hr	Rent
RTX A4500	20 GB	FP16	$0.50/hr	Rent
P4	8 GB	FP16	$0.60/hr
L40S	48 GB	FP16	$0.86/hr	Rent
RTX 5090	32 GB	FP16	$0.99/hr	Rent

VRAM requirements for Falcon3-7B-Instruct-1.58bit at different quantization levels
Quantization	VRAM Required
FP16	5.5 GB
Q8	2.8 GB
Q6	2.1 GB
Q4	1.4 GB