CPU time difference

Message boards : Number crunching : CPU time difference
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · Next

AuthorMessage
Matthew@SETI@home

Send message
Joined: 14 Feb 16
Posts: 40
Credit: 9,278,146
RAC: 6
United States
Message 1793718 - Posted: 5 Jun 2016, 18:39:55 UTC - in response to Message 1793714.  
Last modified: 5 Jun 2016, 18:50:54 UTC

Ok, I made that change, although it's not clear whether BOINC is now using one thread of each core or both threads of one core. Given that total CPU usage per Task Manager dropped from 99% to around 55-65%, I don't see how that could be good for total throughput.
ID: 1793718 · Report as offensive
Rasputin42
Volunteer tester

Send message
Joined: 25 Jul 08
Posts: 412
Credit: 5,834,661
RAC: 0
United States
Message 1793722 - Posted: 5 Jun 2016, 18:54:48 UTC - in response to Message 1793718.  

One important thing is the memory bandwidth.
The more tasks are running, the worse the speed gets(per core), as all cores have to negotiate the same memory.
This is why running fewer cores can, over all, improve performance as there are fewer memory request collisions.
ID: 1793722 · Report as offensive
Matthew@SETI@home

Send message
Joined: 14 Feb 16
Posts: 40
Credit: 9,278,146
RAC: 6
United States
Message 1793724 - Posted: 5 Jun 2016, 19:04:14 UTC - in response to Message 1793722.  
Last modified: 5 Jun 2016, 19:37:16 UTC

In which case a large fraction of that 99% might have been wasted time, I gather. I'll watch for a dramatic change in host RAC, beyond its gradual increase and its "normal" erratic behavior. If there is a better metric, even involving freeware, let me know.
ID: 1793724 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1793737 - Posted: 5 Jun 2016, 19:46:06 UTC - in response to Message 1793724.  
Last modified: 5 Jun 2016, 19:47:12 UTC

In which case a large fraction of that 99% might have been wasted time, I gather. I'll watch for a dramatic change in host RAC, beyond its gradual increase and its "normal" erratic behavior. If there is a better metric, even involving freeware, let me know.

RAC is slow tool. After few tasks completion you could notice difference.
To make comparison aways look at "true angle range" in stderr of result. Only close values can be compared directly. Also, exclude overflows (error -9) results from comparison.
With 2 CPUs "disabled" throughput can little decrease (or may even increase - it depends) but at least you get the answer on initial question - why so big difference.
ID: 1793737 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1793740 - Posted: 5 Jun 2016, 19:50:55 UTC - in response to Message 1793722.  

One important thing is the memory bandwidth.
The more tasks are running, the worse the speed gets(per core), as all cores have to negotiate the same memory.
This is why running fewer cores can, over all, improve performance as there are fewer memory request collisions.

Quick question, I was looking at my new machine I just got fired up this morning, and it seemed to run a little bit off, kind of stuttery (best word I can think of) even before BOINC was installed, not all the time, but off and on. I brought up resource monitor, and noticed at the bottom on the right there was a graph named Memory, which appears to track Hard Faults. Is this similiar to what we're talking about when we say memory collisions? My guess is it's not, but do you have any idea what they are? I am on my LotzaCores writing this, and pulled it up here just to check, and noticed that this one seems to be having none at this time, and BOINC is running all 48 cores at once. That is worrisome, as it appears that there may possibly be an issue with the memory on that new machine?

ID: 1793740 · Report as offensive
Rasputin42
Volunteer tester

Send message
Joined: 25 Jul 08
Posts: 412
Credit: 5,834,661
RAC: 0
United States
Message 1793751 - Posted: 5 Jun 2016, 20:36:54 UTC
Last modified: 5 Jun 2016, 20:42:56 UTC

Do you have an SSD or conventional HDD?

Win 7 is not normally expended to run 48 cores.
There is a way of setting it up, to be better suited for that.
ID: 1793751 · Report as offensive
Matthew@SETI@home

Send message
Joined: 14 Feb 16
Posts: 40
Credit: 9,278,146
RAC: 6
United States
Message 1793753 - Posted: 5 Jun 2016, 20:43:03 UTC - in response to Message 1793751.  
Last modified: 5 Jun 2016, 20:56:45 UTC

Per http://www.brighthub.com/computing/windows-platform/articles/52249.aspx, "a hard fault happens when the address in memory of part of a program is no longer in main memory, but has been instead swapped out to the paging file". These are not memory collisions, rather, higher hard faults would indicate a higher RAM demand-supply ratio. The new computer should also show a much higher total utilization % at the top of the Memory column in Task Manager, Processes tab (or an equivalent if not Win10). Solution is a memory upgrade.

My guess is that memory collisions are invisible to the OS, but that's just a guess.
ID: 1793753 · Report as offensive
Rasputin42
Volunteer tester

Send message
Joined: 25 Jul 08
Posts: 412
Credit: 5,834,661
RAC: 0
United States
Message 1793759 - Posted: 5 Jun 2016, 21:02:37 UTC

I could not find your processor on the internet.
What i found is this:

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Xeon+E5-2692+v2+%40+2.20GHz&id=2761

But it does not have 48 processors (only 12/24)
ID: 1793759 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1793764 - Posted: 5 Jun 2016, 21:16:28 UTC - in response to Message 1793751.  

SSD drive on that machine.

ID: 1793764 · Report as offensive
Rasputin42
Volunteer tester

Send message
Joined: 25 Jul 08
Posts: 412
Credit: 5,834,661
RAC: 0
United States
Message 1793765 - Posted: 5 Jun 2016, 21:18:21 UTC - in response to Message 1793764.  

Thanks, what about your processor?
ID: 1793765 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1793766 - Posted: 5 Jun 2016, 21:19:00 UTC - in response to Message 1793753.  

Memory upgrade? It's got 32 gig of ram in it, how much more could it need? lol I know Windows is a pig, but that's ridiculous.

ID: 1793766 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1793767 - Posted: 5 Jun 2016, 21:20:33 UTC - in response to Message 1793765.  
Last modified: 5 Jun 2016, 21:21:48 UTC

This is the system I am referring to.

GenuineIntel Intel(R) Core(TM) i7-3770K CPU

It actually has 2 950's in it and 2 670's that were pulled from the machine that I am putting the water cooled cards into.

ID: 1793767 · Report as offensive
Rasputin42
Volunteer tester

Send message
Joined: 25 Jul 08
Posts: 412
Credit: 5,834,661
RAC: 0
United States
Message 1793769 - Posted: 5 Jun 2016, 21:25:23 UTC - in response to Message 1793767.  

Did you do an "sfc/scannow" on it under an administrative prompt?
Maybe, there is some system corruption?
ID: 1793769 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1793775 - Posted: 5 Jun 2016, 21:58:47 UTC - in response to Message 1793769.  

I'll do that once I get it back up and booting again. Stupid, lazy me, I had an ASUS P8Z77 WS motherboard that was going to go into another project that I decided not to do, so thought I'd get it fired up and see how it worked. So, back to stupid, lazy me, instead of O/Cing it the normal way thru the BIOS, I thought I would give Asus's AI Suite II V2.00.01 a shot. One of the utilities in it automatically overclocks your board in small steps, reboots and tries again till it fails, and then reverts to the last good, stable overclock. As it was written by ASUS, it Had to be stable, and effective, Right? That's what it said it would do. And I believed them. I knew better than that, but like I said, stupid, and lazy.. lol

Well, it rebooted, tried, and got thru the first overclock, rebooted again, and that was it. Machine is on, but it's hung. That's where I gave up on it this morning, I had too much stuff to do today to screw with it, but I can hit reset, and go into safe mode no prob. Tried uninstalling it in safe mode, no do, says the program crashed trying to uninstall and aborted, so then tried a system restore from about an hour before I ran it, it restored fine, but after the reboot, same hang.

I'll have to do some more research on it tonite, hit some forums and such, but it was my own stupid mistake. Worst case, I wasted the 2-3 hours I spent loading and updating it, but I can always copy the BOINC data directory to a USB drive, and Fdisk it, then reload it. Since I am loading it off of a USB3 drive, that part doesn't take _too_ long, it's the Windows updates that are a killer. Even with the using the SP rollup that they just came out with last month, it still takes forever to do the 1st windows update after that. Def going to grab a frosty one before I start all that foolishness.

Oh well, the world isn't coming to an end, but I sure would like to know what the heck is causing those errors. What exactly does sfc/scannow do? Some sort of a chkdsk or something?

ID: 1793775 · Report as offensive
Rasputin42
Volunteer tester

Send message
Joined: 25 Jul 08
Posts: 412
Credit: 5,834,661
RAC: 0
United States
Message 1793781 - Posted: 5 Jun 2016, 22:10:55 UTC

What exactly does sfc/scannow do? Some sort of a chkdsk or something?


It checks the integrity of the system.Very useful to stop a system from going "bananas".

Out of curiosity:Your System

https://setiathome.berkeley.edu/show_host_detail.php?hostid=8012837

Lists 48 cores, but the processor only has 12/24 cores.
Do you have two procs in that machine?
ID: 1793781 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1793782 - Posted: 5 Jun 2016, 22:20:18 UTC - in response to Message 1793781.  

What exactly does sfc/scannow do? Some sort of a chkdsk or something?


It checks the integrity of the system.Very useful to stop a system from going "bananas".

Out of curiosity:Your System

https://setiathome.berkeley.edu/show_host_detail.php?hostid=8012837

Lists 48 cores, but the processor only has 12/24 cores.
Do you have two procs in that machine?

That's because it's a dual CPU system which the details don't show. ;-)

Cheers.
ID: 1793782 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1793786 - Posted: 5 Jun 2016, 22:46:37 UTC - in response to Message 1793781.  

What exactly does sfc/scannow do? Some sort of a chkdsk or something?


It checks the integrity of the system.Very useful to stop a system from going "bananas".

Out of curiosity:Your System

https://setiathome.berkeley.edu/show_host_detail.php?hostid=8012837

Lists 48 cores, but the processor only has 12/24 cores.
Do you have two procs in that machine?

Yep, dual socket Supermicro server MB -> Pics of it.

ID: 1793786 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1793787 - Posted: 5 Jun 2016, 22:50:37 UTC

Oh, and after messing around with it, I got it back up and running. Yay, must have been the beer that coaxed it back to working. That's my story and I'm sticking to it. Well, I will try running sfc/scannow and see what happens. Just run it in a command prompt windows started as run as admin, correct?

ID: 1793787 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1793788 - Posted: 5 Jun 2016, 22:54:08 UTC - in response to Message 1793782.  
Last modified: 5 Jun 2016, 23:00:19 UTC

Yeah, I've found that BOINC leaves a bit to be desired when it comes to displaying all the hardware in a given system. It seems to have issues with both CPU's and GPU's, as the one mentioned above has dual 24 core procs, and this one has 4 video cards, and it displays only the first one. I am going to go out on a limb and guess that it really doesn't matter what is displayed, as long as the program in the background knows what's what. Otherwise I'd assume that they would have fixed it long ago.

*edit* Display it on the server system summary page that is, it displays it locally just fine, every time for me.

ID: 1793788 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1793876 - Posted: 6 Jun 2016, 6:20:12 UTC - in response to Message 1793788.  

*edit* Display it on the server system summary page that is, it displays it locally just fine, every time for me.

BOINC not only "displays it locally just fine", it also reports all this info to the server.
Check your sched_request_setiathome.berkeley.edu.xml
Find in it strings <host_info> and <coproc_

(I have only one <coproc_ati> but I think you'll find in your file several GPUs listed)
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1793876 · Report as offensive
Previous · 1 · 2 · 3 · 4 · Next

Message boards : Number crunching : CPU time difference


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.