Enterprise-grade bare metal aimed at teams that would otherwise colocate. Obsidian Rack ships current-generation EPYC and Xeon Scalable builds with ECC RAM as standard, dual 25G uplinks, and BGP session support on request. Custom builds are made to order with signed hardware manifests, and every machine includes redundant PSUs on separate feeds.
Share your experience to help others choose the right provider.
Out-of-band access is genuinely first class here. Remote ISO mount works, serial-over-LAN works, rescue mode boots in about forty seconds, and the power controls are not buried behind three menus. I have used a lot of vendors where the KVM is a decade-old Java applet held together with hope. Not this one.
No complaints about the hardware or the network at all. The HTML5 KVM is fast enough that I have done full OS installs over it from a hotel wifi. My only real gripe is that almost nothing I want is in stock inventory - our last two custom builds took six and nine working days respectively. They were upfront about it, but it does mean you cannot treat them as elastic capacity.
We ran our analytics cluster on shared virtual instances for three years and spent most of that time arguing with a vendor about steal time. Moved the whole thing to Obsidian Rack over a weekend. The machines are exactly what the spec sheet said - I pulled the manifest, checked the DIMM part numbers against what dmidecode reported, and they matched. p99 query latency dropped 40% and it has stayed there for eight months. Their network team set up a BGP session for us in under a day.
This is the review I wanted to write since day one and finally got the chance. NVMe in slot 3 threw SMART errors overnight, their monitoring caught it before I did, ticket was already open when I logged in. Hot-swapped, RAID rebuilt, zero downtime, and someone actually wrote me a short post-mortem explaining which batch the drive came from. That is the whole reason you pay a premium for dedicated.
Everything works. That is not in dispute. But we are a 9-person company running a fairly boring Rails app and we are paying enterprise money for redundant PSUs on separate feeds that we will probably never notice. The build we ordered also slipped past the quoted date by three days with no proactive update - I had to chase. If your workload genuinely justifies it, great. Ours maybe doesnt.
Out-of-band access is genuinely first class here. Remote ISO mount works, serial-over-LAN works, rescue mode boots in about forty seconds, and the power controls are not buried behind three menus. I have used a lot of vendors where the KVM is a decade-old Java applet held together with hope. Not this one.
No complaints about the hardware or the network at all. The HTML5 KVM is fast enough that I have done full OS installs over it from a hotel wifi. My only real gripe is that almost nothing I want is in stock inventory - our last two custom builds took six and nine working days respectively. They were upfront about it, but it does mean you cannot treat them as elastic capacity.
We ran our analytics cluster on shared virtual instances for three years and spent most of that time arguing with a vendor about steal time. Moved the whole thing to Obsidian Rack over a weekend. The machines are exactly what the spec sheet said - I pulled the manifest, checked the DIMM part numbers against what dmidecode reported, and they matched. p99 query latency dropped 40% and it has stayed there for eight months. Their network team set up a BGP session for us in under a day.
This is the review I wanted to write since day one and finally got the chance. NVMe in slot 3 threw SMART errors overnight, their monitoring caught it before I did, ticket was already open when I logged in. Hot-swapped, RAID rebuilt, zero downtime, and someone actually wrote me a short post-mortem explaining which batch the drive came from. That is the whole reason you pay a premium for dedicated.
Everything works. That is not in dispute. But we are a 9-person company running a fairly boring Rails app and we are paying enterprise money for redundant PSUs on separate feeds that we will probably never notice. The build we ordered also slipped past the quoted date by three days with no proactive update - I had to chase. If your workload genuinely justifies it, great. Ours maybe doesnt.