s R.
> I rented a Verified RTX 4090 instance…
> I rented a Verified RTX 4090 instance that advertised over 600 Mbps network speeds. GPU training and disk access worked, but outbound transfers were extremely slow and unstable. Throughput sometimes dropped to 10–20 KB/s, and connections repeatedly > terminated while I was trying to retrieve my trained checkpoint. > > I tested direct SSH/SCP, Vast proxy SSH, Jupyter HTTPS, and plain HTTP, with similar results. This wasted more than ten hours of my time and caused additional rental charges. > > Support only stated that bandwidth is shared and not guaranteed, refused compensation, and suggested renting another instance to copy the data—even though that copy would still depend on the affected host’s network. > > I understand that shared bandwidth can fluctuate, but a Verified host delivering effectively unusable throughput with repeated disconnections should be investigated. Customers should not bear all additional costs when they cannot retrieve their > completed work. Based on this experience and the refusal to provide a refund or credit, I will move important training workloads to another platform.
Response from Vast.aiSep 2, 2026
We’re sorry to hear about the difficulties you encountered. We’d appreciate the opportunity to review your case again and escalate it to one of our Support managers. Please reach out to us through our support web chat and we'll be happy to take another look.




