Question 1

How much VRAM does gpt-oss-safeguard-20b need?

Accepted Answer

gpt-oss-safeguard-20b (21.5B) requires approximately 46.73285937309265 GB of VRAM at full precision (FP16), 23.4 GB at 8-bit quantization (Q8), and 11.7 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 gpt-oss-safeguard-20b?

Accepted Answer

We have 10 compatible GPUs indexed for gpt-oss-safeguard-20b, including A40, L40S, L40, RTX A6000, A100 80GB PCIe 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 gpt-oss-safeguard-20b?

Accepted Answer

The cheapest GPU to run gpt-oss-safeguard-20b in the cloud is the A40 at $0.44/hr. Other affordable options include L40S ($0.60/hr) and L40 ($0.82/hr). Prices reflect the lowest on-demand rate across cloud providers.

Question 4

Can gpt-oss-safeguard-20b run on consumer hardware?

Accepted Answer

Yes, gpt-oss-safeguard-20b can run on consumer GPUs like the RTX A6000 (48 GB VRAM) at full precision (FP16).

Question 5

What quantization options are available for gpt-oss-safeguard-20b?

Accepted Answer

gpt-oss-safeguard-20b supports the following quantization levels: FP16 (full precision): ~46.73285937309265 GB, Q8 (8-bit): ~23.4 GB, Q6 (6-bit): ~17.5 GB, Q4 (4-bit): ~11.7 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. As a Mixture of Experts (MoE) model with 32 experts (4 active), gpt-oss-safeguard-20b has a larger total weight size but only activates a subset of parameters per token, which can affect effective memory usage during inference.

GPU	VRAM	Best Quant	From	Rent
A40	48 GB	Q8	$0.44/hr	Rent
L40S	48 GB	Q8	$0.60/hr	Rent
L40	48 GB	Q8	$0.82/hr	Rent
RTX A6000	48 GB	Q8	$1.09/hr
A100 80GB PCIe	80 GB	FP16	$1.39/hr	Rent
A100 80GB	80 GB	FP16	$1.39/hr	Rent
H100 MEGA	80 GB	FP16	$1.80/hr	Rent
H100 SXM	80 GB	FP16	$1.80/hr	Rent
H100 NVL	94 GB	FP16	$1.80/hr	Rent
H100 PCIe	80 GB	FP16	$1.80/hr	Rent

gpt-oss-safeguard-20b

VRAM Requirements

Compatible GPUs

gpt-oss-safeguard-20b FAQ

VRAM requirements for gpt-oss-safeguard-20b at different quantization levels
Quantization	VRAM Required
FP16	46.7 GB
Q8	23.4 GB
Q6	17.5 GB
Q4	11.7 GB