Hacker News new | ask | show | jobs
by fabian2k 2565 days ago
If you want to compare to Xeons, the comparison aren't the consumer CPUs this is about, the AMD equivalent are the Epyc CPUs.

AMD doesn't disable ECC support entirely on consumer CPUs like Intel does, but as far as I know it's also not officially supported and guaranteed to work, it's up to the mainboard vendor how to handle this. In the Intel case you simply can't get ECC with non-Xeon CPUs.

3 comments

Well even on intel, its possible the firmware/os isn't doing the right thing. This was pretty common ~10 years ago, when the default linux behavior wasn't to report soft errors in the logs (due to missing drivers/whatever) so a lot of machines might just sit there and correct the errors and the only way to find out was to turn up some BMC/etc logging. I guess that is why you should buy machines from HP/Dell/Lenovo that are fully certified for your OS rather than random whitebox manufactures too, although given the problems I was having with HP equipment at the time its questionable.
> can't get ECC with non-Xeon CPUs

Not quite. There are some Core branded CPUs that support ECC, including funnily enough the i3's.

> There are some Core branded CPUs that support ECC, including funnily enough the i3's.

Hell, there are Celeron and Pentium chips that they have it enabled on. Not because they expect desktop users to buy them, but because it allows them to keep their Xeon brand premium while letting OEM's like Dell advertise the T140 "starting at $549" (in a configuration nobody would ever want to buy).

It depends on the use. For a home server or even a small office fileserver, you don't need massive threading capability, and in fact some of those low-core-count parts are fairly highly clocked, which makes them faster.

For example in the 7000 series, the i3 7100 has a 3.9 GHz base clock and you have to go almost to the top Xeon (the equivalent of an i7) to get anything equivalent. And even then it's a turbo, not a base clock, so in principle the motherboard should not let you turbo forever (PL2 time limit may actually be enforced on a server chipset).

Also depending on workload you may not even be able to exploit an increased threading capability anyway, without 10 GbE on the box, or link aggregation capability.

Oh, the i3's are fine for a general small business workload - compared to the socket-compatible Xeon's all they're really lacking is extra PCIe lanes if you need them. That's ultimately what Intel uses to segregate the Xeon and HEDT chips from their mainstream platform, after all.

The Celeron and Pentium chips that have infiltrated entry-level servers are absolute trash though.

There are even Atom chips with ECC: https://ark.intel.com/content/www/us/en/ark/products/97935/i... (for I assume the NAS products that use these?)

It seems like it's mostly any chip that would compete with the Xeon-W gets ECC removed.

There are no Epyc workstations chips with clock speeds comparable to Threadripper/Xeon-W. At least for the currently released products. And thus I consider Threadripper the closest competition to Xeon-W, not Epyc. AMD also lists ECC memory support as a feature for Threadripper.

That said, the new Xeon-W series has more memory channels (6) and supports more RAM (up to 2 TB) than any existing Threadripper product. I.e., AMD doesn't have an equivalent product for all use cases yet.

However, we don't know the Zen 2 Threadripper lineup and the frequencies for the different Zen 2 Epyc SKUs are also not public yet. AMD could release Threadripper with support for RDIMM/LRDIMM or Epyc chips with higher clock speeds to better compete against Xeon-W.