The value for money is probably the biggest thing for me. Being able to spin up an incredibly powerful machine within a few minutes, use it for exactly as long as I need, and then shut it down is much cheaper than buying the hardware myself.

It's especially useful for me personally because I travel a lot. Carrying around a machine powerful enough for the kind of AI workloads I run just isn't practical, so being able to access the same level of hardware from anywhere in the world makes my work considerably easier.

I also really like the amount of choice. I can test different GPUs depending on what I am doing and compare performance without having to commit to buying anything. That's been really useful for AI video generation in particular, where the difference between GPUs can be huge.

Another big positive is the API. They have made it simple enough that I can basically pass the documentation to an AI coding assistant and have something working within minutes. That's been genuinely useful for GPUs that are often busy. Instead of manually refreshing and waiting for the GPU I want to become available, I can have my own system check for GPUs that fit my parameters, launch the pod, run what I need, and terminate it again when I'm finished.

The dashboard is also pretty straightforward. Even when I first started using Runpod, it didn't take long to understand how to launch a pod, attach storage, choose a GPU and get everything running. The documentation is generally clear as well, especially for the API.

I mainly use Runpod for AI workloads, particularly running ComfyUI and video generation models, and it works really well for that. Being able to choose the exact GPU and VRAM I need for a particular model is much more useful to me than using a more restricted cloud service where the hardware is hidden from you.

The network volume system is useful as well because I can keep my main setup and models stored there rather than having to start completely from scratch every time I launch a new machine.

Overall, the main thing I like is that Runpod gives me access to hardware that would otherwise be very expensive and inconvenient for me to own, while still giving me enough control to automate most of the process. Review collected by and hosted on G2.com.

The main thing I dislike is that their AI support can be a bit hit or miss. They have what should be a really useful feature where, if a pod fails to start properly or is having an issue, you can ask the AI to take a look at it for you. In practice though, it often doesn't give me enough useful information to actually understand what the problem is. I've also had cases where it recommends restarting the pod, I agree, it tells me it has done it, but nothing seems to happen.

Another issue I sometimes have is inconsistent performance between machines. I can select the exact same GPU in different regions and one can be considerably slower than another. I've even had this happen within the same region. That makes it difficult to know how a machine is going to perform before actually starting it.

The most frustrating part of this is startup time. Occasionally a pod can spend a very long time downloading the Docker image or getting everything ready, and you're still being charged while you wait. If it's been over 20 minutes and it still hasn't finished, I normally terminate it and try another machine. That means I've paid for the failed attempt, and I then have to wait for another suitable GPU to become available and start the whole process again.

I think more transparency around the individual machines before launching them would help a lot. If there was some indication of recent startup speed, network performance or reliability for a particular host, it would make it easier to avoid slower machines in the first place. Review collected by and hosted on G2.com.