Post
I was checking a faculty candidate what the #HPC is offering on its newest iteration of the cluster (that is fee per CPU second use). The high mem instance? 64 GB of RAM. Which is what my <checks notes> decade+ ancient server.
Within the last two weeks I was running a relatively simple random forest that needed 80+ GB on version 1 of the cluster. Well I guess I'm not doing that after it's decomissioned. Which sent me down the rabbit hole. What would it take for me to build a 256 GB RAM server?
@thatdnaguy
High mem is 64G? The 1990s called and wants their supercomputer back.
Our regular nodes are all 512G each. The large men nodes have more. And we are due to upgrade
@jannem
I'm watching our dept trying to dismantle and divest from our very functional cluster with 256GB nodes. They haven't seen a single cloud bill yet because the vendor is giving them a free trial period.
I absolutely cannot wait.
The major impetus seems to be that they think this is a clever way to abandon 6PB of old research data
@thatdnaguy
Considering the current market, the 90s might indeed be calling. Unless a department is willing to pay for high mem nodes or suffers from the illusion that cloud providers are a charity.
512G standard nodes, however, do not seem like a necessity. Not even for genomics works. There are use cases for 1T+ nodes, but bread and butter tasks, surely for I/O intensive tasks too, does not require that much.
Install bonfire.cafe
Get the full app experience
