It Looks Like AMD's Linux Developers Have Begun Work On Zen 2 / EPYC 2 "Rome" Support

[phoronix]

Phoronix: It Looks Like AMD's Linux Developers Have Begun Work On Zen 2 / EPYC 2 "Rome" Support

Ahead of the Zen 2 processors expected in 2019, it appears AMD developers have begun working on their Linux kernel support patches for these next-generation CPUs. In particular, it appears the flow of Linux kernel code for supporting EPYC 2 "Rome" processors has begun...

http://www.phoronix.com/scan.php?page=news_item&px=AMD-EDAC-Kicks-Off-Zen-2

[ermo]

If you're getting an 8-core/16-thread (or even a 16-core/32-thread) workstation with 16-64 GB of RAM in order to do your job, why *wouldn't* you want ECC?!

[boxie]

If you're getting an 8-core/16-thread (or even a 16-core/32-thread) workstation with 16-64 GB of RAM in order to do your job, why *wouldn't* you want ECC?!

I have an 8 core/16 thread pc, no need for ecc - although it would be nice. ecc usually means slower throughput though, which is not all that ideal for gaming

[MagicMyth]

From what I've seen in the past and tested the performance impact from ECC is at most 2% so as consumers push more and more data through their system it might be a worth while trade off. Something Michael could test perhaps?

[L_A_G]

Didn't they start getting things merged to mainline for the original Zen CPUs about two years before they came out? If they are on track for a Q1 2019 launch as previously advertised, this is a bit late seeing how 4.20/5.0 won't be getting it's first stable release until December or January (if the period from the merge window closing to the first stable release ends up unusually hectic).

[torsionbar28]

ecc usually means slower throughput though, which is not all that ideal for gaming

Source? Plenty claim a small performance hit, but every benchmark I've seen comparing the two shows no performance difference at all, zero.

[Michael]

Source? Plenty claim a small performance hit, but every benchmark I've seen comparing the two shows no difference at all, zero.

I do have some extra DDR4 ECC RDIMMs arriving today, before putting them in POWER9 server, will see if any of them work in any of my Threadripper boxes (haven't checked on compatibility from that angle in a while; though I guess it probably only works for ECC UDIMMs).

[chris200x9]

Didn't they start getting things merged to mainline for the original Zen CPUs about two years before they came out? If they are on track for a Q1 2019 launch as previously advertised, this is a bit late seeing how 4.20/5.0 won't be getting it's first stable release until December or January (if the period from the merge window closing to the first stable release ends up unusually hectic).

I don't think it's comparable. Zen was an ENTIRELY new arch like nothing ever seen before, zen 2 is just an iteration on zen. How long before launch was zen+ support merged?

[Michael]

I don't think it's comparable. Zen was an ENTIRELY new arch like nothing ever seen before, zen 2 is just an iteration on zen. How long before launch was zen+ support merged?

Zen+ I don't think really needed any new kernel patches, at least none come to mind, besides the simple k10temp CPU temperature reporting patch I ended up doing after receiving review samples.

[schmidtbag]

I'm also interested to see the performance difference of ECC vs non-ECC. I would think that there would be little to no performance loss at all (assuming all other specs are equal), but, if the actual ECC starts to kick in and fixes errors, then I could see there being some performance loss. But, that performance loss would be practically unnoticeable, especially compared to the actual result of the error, which most definitely would be noticeable.

From what I've seen in the past and tested the performance impact from ECC is at most 2% so as consumers push more and more data through their system it might be a worth while trade off. Something Michael could test perhaps?

Depends on your workload. If you're doing something that needs 100% accuracy the first time around, then yes, the performance loss and extra expense are not a priority. However, I would say the vast, VAST majority of end-user computers don't warrant the need of ECC, especially something like a gaming PC, HTPC, or a family computer.