mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2019-09-04, 19:47   #1
aurashift
 
Jan 2015

11×23 Posts
Default How much of a problem is jitter?

https://support.hpe.com/hpsc/doc/pub...a00018313en_us


Still reading, and a lot of this is written for Xeons, but it sounds like every time there's a C or P state change to either boost performance or save power there's a 15 microsecond CPU delay, which to my layman's ears sounds like kind of a lot, especially in the CPU realm when other things like RAM have latencies of 70ns and infiniband has 300ns.


Unless I'm wrong about things I already know any CPU will never hit advertised turbo boost speeds at 100% phyCPU utilization.


I believe there's also things that save power at the cost of performance on the RAM and IO side of things, kind of wondering what the entire big picture is.


I could test out each of these workload profiles and use our 100M benchmark exponent 332220523 to see how it impacts things....what do you all think the number of iterations would need to be to get an accurate representation? test box is a single socket DL360, intel 6150 gold, 384GB of ddr4 2133
aurashift is offline   Reply With Quote
Old 2019-09-12, 01:56   #2
NookieN
 
NookieN's Avatar
 
Aug 2002

728 Posts
Default

It depends on your perspective maybe. At 2.4GHz, 15us is, what, 36,000 ticks? If you re-lock to 2.5GHz you should recover the lost time if you can stay at that frequency for more than 360us. Presumably the Pcode knows that and factors it in to decide if it will burn through its thermal headroom faster or slower than that.

As far as hitting advertised turbo, the specs list different max turbo frequencies for number of active cores. So they may advertise 4GHz as the headline number for 1 core and 2.8GHz for 16 cores, even if the nominal speed of the chip is 2.4GHz. It should sustain that turbo frequency for the number of active cores as long as the thermals allow. AVX and AVX512 also affect the max turbo the Pcode will allow, regardless of cooling.
NookieN is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
A new problem similar to the Collatz problem dabler Miscellaneous Math 1 2018-07-28 14:03
What is the problem here? didgogns Msieve 1 2016-11-15 03:31
problem I have science_man_88 Miscellaneous Math 2 2010-10-10 16:36
Problem with LMH derekg Lone Mersenne Hunters 2 2007-02-26 22:47
51 problem Neves Miscellaneous Math 5 2004-02-10 22:59

All times are UTC. The time now is 05:40.

Sun Mar 7 05:40:54 UTC 2021 up 94 days, 1:52, 0 users, load averages: 1.98, 2.08, 2.15

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.