In this case, I think an out-of-order CPU for a normal ISA has much better information about what and how to optimize than a compiler could discern for a VLIW CPU except for a few special cases, like old fashioned number crunching.

https://x.com/JavierCasares/status/1843963074904227945

4 Sierra Forest cores form a cluster that shares 4MB of L2 cache, each Granite Rapids core has its own 2MB L2 cache.

Sierra Forest has 3MB L3 cache per cluster, for a total of 108MB for the 144 core parts which is a smaller total than the L2 caches. No firm word on L3 shared caches for Granite Rapids, but its predecessor Emerald Rapids has 5MB per core. At that level, the 6980P would have a total of 640MB, but rumor at the beginning of the year is saying "up to 480MB."

So much more space is being dedicated to Granite Rapids caches, and the end of Dennard scaling a couple of decades ago means they'll draw plenty of power at rest. As theandrewbailey says, you really have to benchmark what you plan to run on these CPUs, but many loads will benefit from the bigger cache hierarchy even if some of these bigger ones are slower as they tend to be.

As I'm finding with a Raptor Lake CPU and conservative stock Debian 12 bookworm and it's 6.1 kernel, largest format video playing has some tearing and so on due to driver support based on the error messages I'm seeing, 720p is much better. At the bottom of the stack root ZFS is rock solid: https://openzfs.github.io/openzfs-docs/Getting%20Started/Deb...

Today availability should be different and skimming their rackmount SKUs the vast majority claim "Ships Within 3-5 Business Days" which was my experience with a tower workstation.

I went to the above few minutes of effort because the company has done very well for me and mine in the last couple of decades, and people who are buying small quantities which certainly includes small or startup businesses as well as "hobbyists" shouldn't think the old shortages still hold in a period where I hear general demand is down outside of AI platforms. And prices are certainly pretty good.

See also its Nvidia equivalent https://www.supermicro.com/en/products/system/gpu/8u/as-8125... which is implicitly on allocation because TSMC has been contracted to make only so many H100 chips per unit of time, I think I heard 550,000 for 2023. I remember seeing that the Nvidia cards list at around $30K each....

I think the store is more for stuff that's in more plentiful supply and less ITAR restricted, with a limited number of SKUs and options for each that they're prepared to support for a long time.

Shipped a SYS-531A-IL workstation in less than a week including a weekend as my latest workstation configured with my chosen, albeit from a limited selection of memory, good CPUs and NVME and SATA drives (enterprise options for all of those), all properly kept cool per lm-sensors and smartmontools. I used their live chat to quickly get a couple of answers, it was packed well, and in theory for a reasonable up front cost they'll cross ship a replacement for the next 5 years.

When Mr. Rsync.net was composing his message today I was adjusting my scripts to do their nightly backup to his service that uses their servers.

For a couple of decades have bought a bunch of their motherboards through resellers for myself and family, and built some rack mount systems for a friend bought I forget from where with excellent results. The current swapping of hardware due to the above purchase took down a system that had been in service for a dozen years.

You asked a very good question, I've been hearing very good things about PC Engines for years (except of course the now common supply chain problems), and I'd have bought my own from them except I'd already invested the Mini-ITX ecosystem for these sorts of machines long ago.