mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > CADO-NFS

Reply
 
Thread Tools
Old 2021-11-11, 21:05   #1
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

11·461 Posts
Default How do various CPU architectures compare for CADO speed?

My homebuilt haswell-generation desktop seems to have gone kablooey. Time for a new (likely used xeon) machine!

Before I pull the trigger, I thought I'd ask if any of y'all have relative speed comparisons of Haswell vs Skylake or Ryzen for CADO work.

I am considering a dual Xeon v3 (Haswell), a v4 (Broadwell), a first-gen Silver (Skylake), or AMD (would be new machine and new experience to me, but I think Threadripper is out of price range). Total machine cost should be $1200-1400.

Skylake has the most memory bandwidth, Ryzen the least. But I don't have any clue how the various generations compare in work-per-cycle for CADO. My software interests are, in order, CADO, LLR, ECM.

If anyone has data, what I am most interested in is e.g. "skylake runs 24% faster than Haswell at the same ghz on [cado, ecm, llr]".

Last fiddled with by VBCurtis on 2021-11-11 at 21:06
VBCurtis is offline   Reply With Quote
Old 2021-11-11, 21:20   #2
paulunderwood
 
paulunderwood's Avatar
 
Sep 2002
Database er0rr

5·787 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
My homebuilt haswell-generation desktop seems to have gone kablooey.
What are the symptoms? Sometimes all it takes is a new CR 2032 battery for the mainboard.

Last fiddled with by paulunderwood on 2021-11-11 at 21:26
paulunderwood is offline   Reply With Quote
Old 2021-11-11, 22:02   #3
charybdis
 
charybdis's Avatar
 
Apr 2020

22316 Posts
Default

I find that Haswell and Skylake have very similar speed/GHz for CADO. Skylake runs about 10% faster than Haswell for ECM.

Some of the newer Intel architectures are substantially faster per GHz for CADO. Coffee Lake beats Skylake/Haswell by ~40% even when accounting for the faster turbo boost.
charybdis is offline   Reply With Quote
Old 2021-11-11, 23:45   #4
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

11·461 Posts
Default

Quote:
Originally Posted by paulunderwood View Post
What are the symptoms? Sometimes all it takes is a new CR 2032 battery for the mainboard.
Good idea, and I have one of those lying around since it's the same battery for my car keyfob. I'll give that a try.

Symptoms are irregularly-spaced crashes that are not temperature-related; I've checked all fans, CPU has an all in one water/radiator system that is functioning, and crashes happen when when CPU is under 45C according to linux "sensors". Some crashes freeze the screen but computation continues, others cause a reboot without OS doing anything to signify instability. System is not dusty, and has quite a lot of airflow.

The last two days the machine POSTs and goes through the boot process, but usually produces a solid single-color display to screen (the color changes, but is usually from the family of colors of the ubuntu login screen). Once or twice after letting machine sit powered off for an hour or more I get to login, and then within a few seconds the single-color to the whole screen happens.

I've tried ctrl-alt-F2 or F3 to try to get to a terminal window, doesn't work.

I just tried swapping to another CPU I have lying around, as the used Xeon I upgraded to never seemed perfectly stable after the upgrade; dunno if I seated something incorrectly or disturbed a plug, but now I don't POST- I get a long series of fast beeps followed after ~15 sec by a slower series of beeps. It's an X99 board socket 2011-3 from when 5820 CPU came out, late 2013 or early 2014.
VBCurtis is offline   Reply With Quote
Old 2021-11-12, 07:18   #5
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

2·3·5·29 Posts
Default

If you can get it to post again you could try starting in recovery mode to directly boot to a root shell, or one-time edit the grub command to boot your normal kernel in text mode (which should start all non-graphical services instead of barebones so it may crash where recovery mode might not).
  • Open grub, possibly by holding shift (bios) or spamming escape (uefi)
  • Advanced
  • For recovery mode pick second kernel then pick root from menu, to boot normal kernel in text mode press e on first kernel then add text or possibly 3 to the boot option




A memtest would be a good idea from the grub menu too, a chunk of bad memory sounds plausible.
M344587487 is offline   Reply With Quote
Old 2021-11-12, 07:36   #6
henryzz
Just call me Henry
 
henryzz's Avatar
 
"David"
Sep 2007
Cambridge (GMT/BST)

2·2,969 Posts
Default

Do you have a spare power supply?
henryzz is offline   Reply With Quote
Old 2021-11-12, 08:45   #7
kruoli
 
kruoli's Avatar
 
"Oliver"
Sep 2017
Porta Westfalica, DE

28·3 Posts
Default

For the current 3,748+ team sieving, on my 5950X, I have four clients with --override t 8 running. The average WU takes 6m22s, so the computer returns 1 WU every 95s. This is with slightly suboptimal 3,200MT memory. This is the computer the server is hosted on, so this might also decrease the WU throughput slightly. The CPU is not overclocked. The full system sucks around 200W from the wall, so maybe a BIOS settings lets it keep turboing to some extend.

Edit: Sorry for not having a relative comparison. Maybe you could post a sample command for las, then I could execute it once (while having the system idle) or alternatively fill all cores of the system up with that command.

Last fiddled with by kruoli on 2021-11-12 at 08:52 Reason: Additions.
kruoli is offline   Reply With Quote
Old 2021-11-12, 17:29   #8
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

11·461 Posts
Default

Quote:
Originally Posted by henryzz View Post
Do you have a spare power supply?
I wish, as this seems second-most-likely after a motherboard failure.

When the crashes first started a few months ago, they were quite rare (like a batch of 2-3 with 20-40 days of stability in between). My RAM config was 4x4 + 4x8; I removed the 4x8 and the crashes got quite a bit less frequent. Then I tried with the 4x8 and without the 4x4, and crashes were more common. So, I've been running on 16GB for a few months.

The machine does not POST with the i7 back in place; I'm not currently motivated to re-seat the CPU again to try to get it to POST, since that's merely a new issue on top of all the old issues.

Thanks to everyone for the suggestions- a couple of these I had not thought of.

Also thanks to Kruoli for the timing data- I'll determine my average WU time and discuss in a future post.
VBCurtis is offline   Reply With Quote
Old 2021-11-12, 22:10   #9
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

11·461 Posts
Default

If I were to create a CADO benchmark thread and include in the first post a standard-test file to run (input and parameters), is RSA120 an acceptable size to benchmark? On my older multi-core boxes that's an hour or so.
VBCurtis is offline   Reply With Quote
Old 2021-11-12, 22:30   #10
kruoli
 
kruoli's Avatar
 
"Oliver"
Sep 2017
Porta Westfalica, DE

30016 Posts
Default

I would be willing to run this test on my 5950X when sieving is done and also on a c5.metal instance.
kruoli is offline   Reply With Quote
Old 2021-11-12, 23:25   #11
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

24·257 Posts
Default

@VBCurtis: I have had this type failure many times with various machines and quite often for me it was the Power Supply. If I didn't take care of it in a timely manner, on more than one occasion the motherboard also went. I think the latter was due to not being timely in my PS replacement, but I can't be sure. I would at least do a visual inspection of the motherboard and if you're adventurous, pull the cover off the power supply and do a visual as much as you can with it as well. Not so much in the more recent machines, but I have found many electrolytic capacitors that have failed and bulged the top of the can visibly showing distortion. If they are replaced in time, it can save the motherboard from total failure, or at least let it perform for a little longer.
EdH is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
CADO NFS Shaopu Lin CADO-NFS 522 2021-05-04 18:28
CADO help henryzz CADO-NFS 4 2017-11-20 15:14
How should we compare GPUs to CPUs? chalsall GPU to 72 56 2012-07-15 19:24
Compare interim files with different start shifts? zanmato Software 12 2012-04-18 14:56
CADO R.D. Silverman Factoring 4 2008-11-06 12:35

All times are UTC. The time now is 05:06.


Sun Dec 5 05:06:09 UTC 2021 up 134 days, 23:35, 0 users, load averages: 2.74, 1.72, 1.50

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.