Threadripper

The results were frankly staggering.

On a 64-thread Threadripper, he saw throughput jump from 13 million reads per second, all the way up to 304 million.

That is a massive win for reducing folio reference count contention.

But here is the catch, the code was not perfect.

So our editor nodes are typically more consumer side graphics cards, like something like a 4090 or a 5090, something like that.

Typically, one of the higher end, you know, Ryzens, not necessarily Threadrippers, usually like the Ryzen 9s, uh, like 16-core CPUs, and they vary, I wanna say, between 64 and 128 gigabytes of RAM.

And then our editor nodes are typically on the more workstation class cards by Nvidia, so typically something like an 6000 Ada or the new RTX Pro series, like the RTX 5000, RTX 6000, and typically between 128, 256 gigabytes of RAM.

Uh, and we typically stay on those same Ryzen 16-core CPUs.