So, today, i was reading on the TrueNAS forum, an old hardware recommendation guide where the guide says “Hardware that has a FSB (front-side bus). The FSB will be a performance killer for ZFS”.
When i checked on Wiki, i got to know about it a bit and also that it was discontinued back in 2009 and it was used in Intel Atom, Celeron, Pentium, Core 2 and Xeon CPUs throughout 2008. The modern design is DMI 2.0, QPI and even discontinuance of the NB in favor of PCH.
While i was collecting some info about all flash NAS i’m trying to build, i’ve my eyes on Intel Ice Lake 8362 CPU. So, i was curious and checked the datasheet and it has:
I also have a workstation board that accepts Xeon W series CPU in the LGA4189 socket and when i checked the specs for W-3365 CPU on the official website, i see this:
Both the CPUs are 32 Cores and the 8362 has 11.2GT/s UPI Speed and 3 UPI Links but the W-3665 has 8 GT/s Bus Speed and 0 UPI Links.
Does that mean one should not use the Workstation processors as they do not have the QPI/UPI links?
Also, the links are on the CPU or on the both CPU and Motherboard as well?
With further searching, it seems like the UPI Links are meant for connecting multiple processors (such as on Dual socket systems). Is that true?
I also got to know that the Xeon W CPUs are optimized for tasks that require high single-threaded performance and fast memory access. Is that correct?
To be honest. For a NAS with max 10Gbit/s network connection, it doesnt matter. Intel XEON W / Scalable is way overpowered for a NAS in a home/ small business enviroment.
It will boil down to how much noise, heat and energy consumption you are willing to accept - and of course price and availability.
The bottelneck will not be in the CPU. On my System I can remotely play a AA Game over Parsec on a Win11 VM, write/read with full network speed to the NVMEs or HDs, have plex serving 2 TVs, and some apps in the background all at the same time at 30% CPU.
Things have changed a lot in the last 20 years. Memory controllers have moved into the CPU, parallel busses have been replaced by serial point-to-point connections, hard disk interfaces have changed etc etc.
Nowadays raw computing power is not that important any more. It’s more the quality of peripherials, data throughput, memory bandwidth etc.
Look here: https://www.supermicro.com/en/products/motherboard/a2sdi-h-tf
This board only has got a 25W CPU, but this somewhat special CPU also contains 12 SATA ports and two 10G Ethernet ports. With a small NVMe boot SSD and 32 or 64G of RAM this makes a very capable and energy efficient storage system. Synology and QNAP use the same chip for some of their systems.
Ok, running dozens of virtual machines or video transcoding will require different systems, but that’s why there are so many to choose from.
What does a dual port Intel 10G RJ45 NIC cost? That’s also included.
And also included is IPMI remote management capability. No professional will ever consider buying server boards without it. This has saved my a* more than once in the last 10 years. Consumer boards don’t have this.
Then: What’s the value of your data? Do you have a $/£/€ value for yours? Think about it.
Well, I said that it is a very nice board, didn’t I?
The 12 SATA ports might also save you the cost of a HBA.
A cursory search gave me prices of around 800 Euros, I expected around 400-500 Euros, but maybe those are only prices for used boards which I had in mind? Or maybe if you add the CPU and nic it more or less works out.
Even if you have server hardware, I still think having a sound backup strategy is the most important bit.
Yeah, if you’re willing to buy used, and get the HBA or the NIC separately, you can likely get similar functionality with a different board at 1/4 to 1/2 the price.
That said, the ready-to-go nature of the A2SDi-H-TF has definitely value as well.
