[erlang-questions] AMD Ryzen anyone?

[Alex Alvarez]

The only thing I'd add for reference is that in high performance
computing, the default is to disable virtual threads (e.g.,
hyper-threads) as they actually slow down things otherwise.  Under
normal execution, having more CPU threads available is a good thing,
but not if the application needs a lot more out of the CPU.  The
biggest time issue is actually not the CPU, but memory access.  In
anything that really pushes the system, like when you perform parallel
calculations (e.g., openmp), you simply don't want virtual threads
anywhere.  As far as I know, all super computer disable virtual
threads by default.  Each case could vary, but this is something to
test for.  And in the case Erlang, I'd try with/out to see what
results you get. If possible, I'd disable the virtual threads and try
CouchDB again.

Cheers,
Alex

On 1/4/2018 at 12:58 PM, "Eric des Courtis"  wrote:+1 for Threadripper
or Ryzen Pro
On Wed, Jan 3, 2018 at 7:48 PM, Grzegorz Junka  wrote:
 On 03/01/2018 08:21, Iblis Lin wrote:
  forgot to CC the list.
 On 01/03/2018 01:21 PM, Iblis Lin wrote:
  well, purchasing decision is up to you, but I want to share my
personal
 experience:
 I will vote -100 for the Ryzen build. For me, the term "Ryzen" means
 malicious trolling.
 It wasted my time and mental efforts.
 I will show you my journal:
    - I set up the Ryzen box in 2017/5. Mine is 1700 (8 cores 16
threads).
    - I install Arch Linux, ran CouchDB on it... and found the machine
 crash and reboot regularly in an interval of 2-3 days.
    - I also tried to build mainline kernel, (at that moment, it's
4.12
 IIRC), not work , still crashing or freezing.
    - Another test: disabling all deamon, the box can be alive over a
week.
    - Changing memory module in 2017/6, not work.
    - 2017/9 I RMA-ed it.
    - After RMA, it still keeps crashing, hard freezing, soft
freezing...
 within 3-6 hr if I run something on it.
      the kill-ryzen script DOES kill my box, even it's the new CPU
back
 from RMA.
      The *improvement* is the crashing interval shrunk. Thank you
AMD.
    - 2017/10, I changed my OS from Arch to Ubuntu. not work.
 On 01/03/2018 06:46 AM, Adam Rutkowski wrote:
  Hey all,
 I'm thinking of building a Ryzen machine, mainly for Erlang
development. There's this rather obscure bug with some Ryzen
processors on Linux/BSD [1] and I'm worried I might be getting a unit
from the faulty batch; I have no means of verifying this before
purchasing from my local suppliers. I'm wondering if I should I expect
any issues with Erlang VM doing the multi-core heavy lifting.
 Cheers
 [1]:
https://www.phoronix.com/scan.php?page=news_item&px=Ryzen-Segv-Response
 Interestingly, it's not that not choosing Ryzen guarantees you a
worry-free reality:
 http://www.dailymail.co.uk/sciencetech/article-5232037/Security-flaw-Intel-chips-past-decade.html
 Arguably 50% slowdown or a security flaw is better than a random
crash. But maybe the real lesson is to choose a processor that doesn't
have any known vulnerability so far, e.g. Threadripper or a dedicated
server processor?
 GrzegorzJ
 _______________________________________________
 erlang-questions mailing list
 erlang-questions@REDACTED
 http://erlang.org/mailman/listinfo/erlang-questions
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20180105/5fc9d8d0/attachment.htm>