There are 4U server cases that hold 77 3.5” hard drives, and you could rig up empty cases with NAND of various forms and get even more density, if you could keep cooling under control.
Also, most datacenters will not use 4TB disks, you would use giant 20TB enterprise disks (for spinning rust) or 8TB enterprise SAS or nVME drives. You will lose some of that capacity to redundancy though, as RAID and region replication takes place.
So total capacity, minus disk first losses (so *0.85 as a rule of thumb) and then divide by the redundancy (probably / 3)
Realistically there’s going to be a mixture of cpu compute, gpu compute, storage, and general purpose. I feel like 10 drives per chassis is a decent average, and I intentionally used 4TiB to illustrate this truly absurd proposal.
There are 4U server cases that hold 77 3.5” hard drives, and you could rig up empty cases with NAND of various forms and get even more density, if you could keep cooling under control.
Also, most datacenters will not use 4TB disks, you would use giant 20TB enterprise disks (for spinning rust) or 8TB enterprise SAS or nVME drives. You will lose some of that capacity to redundancy though, as RAID and region replication takes place.
So total capacity, minus disk first losses (so *0.85 as a rule of thumb) and then divide by the redundancy (probably / 3)
Realistically there’s going to be a mixture of cpu compute, gpu compute, storage, and general purpose. I feel like 10 drives per chassis is a decent average, and I intentionally used 4TiB to illustrate this truly absurd proposal.