I used to love RunPod, but unfortunately, my experience with the platform has declined significantly. I have tried reaching out to a couple of people at RunPod to discuss these issues privately, but I have not received a response. Since I have been unable to connect with anyone, I have decided to share my experience publicly.
I genuinely used to enjoy using RunPod, which makes this all the more disappointing. However, the issues I have been experiencing have reached a point where I feel they need to be addressed.
1. Reliability
Reliability has become one of my biggest concerns with RunPod. I could write an entire article about this issue alone, but let me share what happened today.
I attempted to start a B200 instance several times, and each attempt failed. After spending nearly an hour waiting for the instance to start, I eventually gave up. Meanwhile, my project was delayed, and I had nothing to show for the time I had spent waiting.
I finally decided to try a B300 instance instead. Thankfully, that one started successfully.
Unfortunately, that was only the beginning of another problem.
It took approximately 20 minutes to download a 4 GB file. I still needed to download a 150 GB model, and at $7.89 per hour, I was not willing to continue paying for an instance while waiting for files to download at those speeds.
To put that into perspective, I was able to download the same 4 GB file on my local computer in just a couple of minutes.
For a platform that provides high performance GPU infrastructure, this level of network performance is simply unacceptable.
And before anyone assumes I am making a big deal out of one bad experience, I want to emphasize that this has been an ongoing issue for quite some time.
In fact, I previously preferred using RunPod’s community servers in Canada and Sweden because they seemed to offer better internet connectivity. Unfortunately, those servers now rarely appear to be available. More broadly, GPU availability across the platform has become another recurring frustration.
Over time, I estimate that I have lost hundreds of dollars simply waiting for pods to become operational or dealing with instances that fail to work as expected.
I also want to clarify that I am not using a complicated or unusual configuration. I use RunPod’s default template and have my own Python scripts that handle the rest of my environment setup.
My requirements are straightforward. I need to be able to start an instance, connect to it, download my files, and get to work.
Unfortunately, even getting to that point has become a frustrating and unreliable experience.
Why I’m Sharing This Publicly
I would have much preferred to work with RunPod directly to address these issues. I was more than willing to share my experiences, provide feedback, and help identify what might be going wrong.
I genuinely wanted to give them the opportunity to understand what I have been experiencing and hopefully improve the platform.
However, after attempting to reach out and receiving no response, I feel it is important to share these concerns publicly.
I am not writing this because I want RunPod to fail. Quite the opposite. I am writing this because I used to love the platform, and I would genuinely like to see it return to the level of reliability and performance that made me a customer in the first place.
When customers are paying premium prices for high performance computing, they should be able to expect reliable infrastructure, reasonable network speeds, and instances that actually work.
Right now, that has not been my experience with RunPod.