Question 1

What AI models can run on P4?

Accepted Answer

The P4 (8 GB VRAM) can run models including paligemma-3b-ft-refcoco-seg-896-jax, Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8, Llama-3.2-3B-Instruct-QLORA_INT4_EO8, Llama-3.2-3B-Instruct-SpinQuant_INT4_EO8, paligemma-3b-ft-vizwizvqa-448-jax and more — 10 compatible models indexed so far. Compatible models are determined by whether the model's memory footprint fits within the GPU's 8 GB VRAM, potentially using quantization (Q8, Q6, or Q4) to reduce memory usage.

Question 2

What is the cheapest cloud option for P4?

Accepted Answer

The cheapest on-demand option for the P4 is Google Cloud at $0.60/hr (). Spot/preemptible pricing is available at $0.19/hr. We track pricing across 7 providers to help you find the best deal.

Question 3

Is P4 good for fine-tuning?

Accepted Answer

The P4 has limited capacity for fine-tuning with only 8 GB of VRAM. It can handle QLoRA fine-tuning of small models (up to ~3B parameters) but is not recommended for larger training workloads. Consider a GPU with more VRAM for serious fine-tuning.

Question 4

How much VRAM does P4 have?

Accepted Answer

The P4 has 8 GB of VRAM (GDDR5). It also offers 192 GB/s memory bandwidth, 5.5 TFLOPS of FP16 compute.

Question 5

What quantization formats work on P4?

Accepted Answer

The P4 supports the following quantization formats: Q8 (8-bit), Q6 (6-bit), Q4 (4-bit). Q4 quantization can fit models requiring up to ~32 GB at FP16 into 8 GB.

P4

Specifications

Cloud pricing

Compatible models

Compare this GPU

Frequently asked questions

VRAM	8 GB
Memory type	GDDR5
Bandwidth	192 GB/s
FP16 TFLOPS	5.5
FP8 support	No

P4