SF memory hard faults issue

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
yorkman
Posts: 85
Joined: Thu Jul 27, 2017 8:59 pm

SF memory hard faults issue

Post by yorkman » Sun May 27, 2018 6:10 pm

Is it just me on does SF and its derivatives have memory hard fault issues? In windows 2012 for example on a dual e5-2699v3, looking at the Resource Monitor I have full bars all the time (around 400 hard faults) just like the cpu usage monitor. That's crazy! I have 64 GB of ram and I'm using 32 GB of hash which has always been fine before. Unfortunately I don't know when this problem started as I haven't looked at this for months.

User avatar
Eelco de Groot
Posts: 4168
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: SF memory hard faults issue

Post by Eelco de Groot » Mon May 28, 2018 6:08 am

Sorry don't know much about it. Had not even heard of the term 'memory hard fault' before. Usually the term 'Page fault' is used I believe. Windows 10 does not have this in its Task Manager even, at least could not find it. No report of 'memory hard faults' anywhere so no way to measure. The Task Manager might be wrong, is not very reliable in its output. The only sensible way to measure is if you get a speed up with a Stockfish programmed for Large Pages or an OS that does that automatically like Linux, most flavors.
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

Modern Times
Posts: 2417
Joined: Thu Jun 07, 2012 9:02 pm

Re: SF memory hard faults issue

Post by Modern Times » Mon May 28, 2018 7:16 am

Eelco de Groot wrote:
Mon May 28, 2018 6:08 am
Windows 10 does not have this in its Task Manager
Yes it does, search for Resource Monitor on the task bar.
.

Opinions expressed here are my own, and not necessarily those of the CCRL Group.

User avatar
Eelco de Groot
Posts: 4168
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: SF memory hard faults issue

Post by Eelco de Groot » Mon May 28, 2018 7:53 am

Thanks Ray! Windows Start -> Windows Systeembeheer -> Broncontrole in my Dutch version Windows 10. (Edit: now I see I can access that from Task manager as well, like you said) I see very little pagefaults with the Stockfish that I compiled on this system at first glance (using small hash tables, 512 megabytes). Or with Kaissa, that is practically the same. I see a blue, almost flat line at 60 pagefaults per second but that does not seem the measurement itself? It is not completely flat but I don't think that is the measured level of page faults. Not sure Green spikes maybe once a minute below ten spikes per second. But as I said not sure what the blue line is
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

Milos
Posts: 3387
Joined: Wed Nov 25, 2009 12:47 am

Re: SF memory hard faults issue

Post by Milos » Mon May 28, 2018 11:53 am

Eelco de Groot wrote:
Mon May 28, 2018 6:08 am
Sorry don't know much about it. Had not even heard of the term 'memory hard fault' before. Usually the term 'Page fault' is used I believe. Windows 10 does not have this in its Task Manager even, at least could not find it. No report of 'memory hard faults' anywhere so no way to measure. The Task Manager might be wrong, is not very reliable in its output. The only sensible way to measure is if you get a speed up with a Stockfish programmed for Large Pages or an OS that does that automatically like Linux, most flavors.
Ah just ignore it, it's another meaningless question. Ofc it is page "fault", and ofc it is not fault it is just stupid windows name when page is not found in cache and needs to be retrieved from the disk/swap.
He is seeing a high number of those because he has enabled Large pages and is using half of memory for hash and his memory is extremely fragmented. Easy solution, just restart the machine or use less hash. :)

yorkman
Posts: 85
Joined: Thu Jul 27, 2017 8:59 pm

Re: SF memory hard faults issue

Post by yorkman » Mon Jun 04, 2018 9:33 pm

Milos wrote:
Mon May 28, 2018 11:53 am
Eelco de Groot wrote:
Mon May 28, 2018 6:08 am
Sorry don't know much about it. Had not even heard of the term 'memory hard fault' before. Usually the term 'Page fault' is used I believe. Windows 10 does not have this in its Task Manager even, at least could not find it. No report of 'memory hard faults' anywhere so no way to measure. The Task Manager might be wrong, is not very reliable in its output. The only sensible way to measure is if you get a speed up with a Stockfish programmed for Large Pages or an OS that does that automatically like Linux, most flavors.
Ah just ignore it, it's another meaningless question. Ofc it is page "fault", and ofc it is not fault it is just stupid windows name when page is not found in cache and needs to be retrieved from the disk/swap.
He is seeing a high number of those because he has enabled Large pages and is using half of memory for hash and his memory is extremely fragmented. Easy solution, just restart the machine or use less hash. :)
The system was of course rebooted. I know very well that pages can get fragmented with LP. My pages are not fragmented however. And even without using LP I see this issue.

Post Reply