Times (Elapsed / CPU) for BRP5/6/6-Beta on various CPU/GPU combos - RESULTS ONLY

UPDATE: 09-Mar-2015 UTC There is a new (further optimised - v1.52) beta and an extra line just for it in the template below. If you are submitting results, please segregate and please use the appropriate line in the table.

PLEASE NOTE: Please read these guidelines.

This is intended to be a Results ONLY thread. Its main purpose is to give hard data on how the new, optimised, BRP6-beta app is performing in relation to either BRP5 or the standard non-optimised BRP6 app. If you don't have a decent sample size of BRP6-beta results, please don't post here.

If you do have such results (say a minimum of around 30 data points but preferably more) here are detailed instructions and a template to make things as easy as possible for you. It's NOT mandatory to use this template and you don't have to fill in every field but there are minimum requirements - which will be enforced. The intended audience includes the Devs, who have already expressed interest in having reliable information like this. I'm particularly interested in creating information on which volunteers can base their decisions about how to make their contribution more efficient and/or effective. I think there may be quite a few people wondering about adding/upgrading a GPU in an existing (perhaps older) rig. Hopefully this stuff may help them.

INSTRUCTIONS - for preparing and submitting results using the template.

1. Use the 'quote' button to reply to this post.

2. From the very top of the text in the text box that appears, delete everything up to and INCLUDING the first line that reads **** Make FIRST CUT here ****. You can do this later if you need to leave these instructions in place whilst you are editing the data in the template below. If you read these notes whilst looking at the template below, hopefully you will understand them better.

3. What immediately follows this FIRST CUT line is what I'm calling the 'template'. Just fill in your data to replace the placeholders that are there. Remove placeholders if you don't have particular items to report.

4. Be very careful with 'data table' type fields between the 'pre' and '/pre' tags. Make sure you replace the placeholder with the same number of characters, with leading or trailing blanks as appropriate. If you don't, the formatting will be stuffed up.

5. If you are happy for others to see your results list and general computer details (NOTE: NO PERSONAL INFORMATION IS VISIBLE) then

• * Open a separate browser tab or window and go to your computer list on the website.
  * For the 'details' link to your particular host that you see there, use right-click -> Copy link location
  * In the template, replace the *** LINK to host *** with 'url='...'/url' tags and paste your copied link right between the '=' and the ']' that follows the '=' in the opening 'url=]' tag.
  * Replace '...' in the previous bullet point with whatever name or description you want the link to show as.
  * If in doubt, look at one of my completed examples (before you delete it from your reply to this post :-) ).
  * Don't forget to replace 'NN' with a number appropriate for your host.
  

6. For all the lines that follow until you get to the actual data table just replace *** .... *** with the information you wish to submit about your setup.

7. For the actual data table itself, just replace the '#####' strings with exactly the same number of characters (digits plus leading or trailing blanks as appropriate) to make things line up nicely. I also insert commas to make large numbers more readable - eg 1,234,567 instead of 1234567. You can leave blank any fields for which you don't have data. BRP6-beta data PLUS sample size are mandatory.

8. The COMMENTS section is optional but is useful to document any particular points you noticed and to draw attention easily to them.

9. When finished data editing, immediately below you should find the line that reads **** Make SECOND CUT here ****. Delete that line and everything that follows - likely to be quite a lot :-). To do this, I would click the start of the line to mark it, drag the scrollbar of the text-box to the very end of its travel and shift-click after the very last text you see there. Everything in between the two points gets selected and can be deleted with a single 'backspace'.

If anything in the above list of instructions (or what follows) is not clear, or perhaps even blatantly wrong, please complain in the DISCUSSION thread. I'd like to fix it so it's clear to all who wish to contribute. Suggestions for improvement are very welcome.

**** Make FIRST CUT here ****

HOST NN - *** LINK to host ***

[pre]
CPU: *** Specify at least the model -- eg 'Intel Core i3 2120' plus any more you wish - eg '3.30GHz Sandy Bridge' - give speed if overclocked ***
Cores/Threads: *** Specify both numbers if you have a HT capable CPU -- eg 2 / 4 - this makes it clear if HT is on ***
Motherboard: *** Specify at least Manufacturer and Model -- eg 'ASUS P5-KPL AM/PS' or 'Dell xxxx' if proprietary ***
PCIe slot *** Specify PCIe version (generation) and electrical width of the bus -- eg 'V1.x x16' - add more lines if additional slots are in use ***
1st GPU: *** Specify GPU model and RAM size -- eg 'AMD HD7850 2GB' plus speeds if not standard clocks (or if you know them) ***
2nd GPU: -
3rd GPU: -
RAM: *** Specify number, size, speed of each module -- eg '2 x 2GB 1333MHz' as a minimum ***
Concurrency: *** Specify number of GPU tasks in parallel and resource shares (from project or from app_config.xml perhaps -- eg '4 @ 0.5 CPUs + 0.25 GPUs' ***
CPU Tasks: *** Specify number and type -- eg '2 x FGRP4' or '4 x Seti' or 'none' etc, as appropriate ***
Free CPU cores: *** Specify number -- add (virtual) or something else if Hyper-threading is on to flag this ***
OS: *** Specify the details ***
Driver: *** Specify whatever details you know ***
BOINC Version: *** Specify details -- eg '7.2.42 64bit'

Elapsed Time Statistics CPU time Statistics
----------------------------------- ------------------------------ Sample
Search Min Mean Max S.D. Min Mean Max S.D. Size Notes / Comments
====== ======= ======= ======= ===== ====== ====== ====== ==== ====== ================
BRP5 ####### ####### ####### ##### ###### ###### ###### #### #### *** Add something here if you wish ***
BRP6 (non-beta) ####### ####### ####### ##### ###### ###### ###### #### #### *** Add something here if you wish ***
BRP6-beta (Mean +- 1 x SD): 10
Longest CPU time in sample: 1,571
Shortest CPU time in sample: 953
[/pre]

COMMENTS

• * Nice improvement but not as spectacular as for the older CPU in HOST 01. This host has a version 2 PCIe bus which obviously was coping better.
  * The optimised app is just under 10% faster than the standard app.
  * With the low CPU component in the crunch time, I'll be interested in seeing if 1 free core is enough to support 4 GPU tasks.
  

====####====

HOST 03 - fx_6300-01

[pre]
CPU: AMD FX 6300 Hex core
Threads: 6
PCIe slot x16 Version 1.x (I suspect) (single slot)
1st GPU: AMD HD7850 2GB
2nd GPU: -
3rd GPU: -
RAM: 2 x 4GB
Concurrency: 4 @ 0.67 CPUs + 0.25 GPUs
CPU Tasks: 3 x FGRP4
Free CPU cores: 3

Mean Times - secs Elapsed time Stats CPU time Stats
----------------- --------------------- ------------------- Sample
Search Elapsed CPU part Std Devn Variance Std Devn Variance Size Notes / Comments
====== ======= ======== ======== ======== ======== ======== ====== ================
BRP5 ~19,100 ~2,020 - - - - - Long term averages - little variation
BRP6 25,622 2,641 343 118,068 21 432 33 Stats results from online calculator
BRP6-beta 18,554 1,426 1,415 2,002,159 232 53,816 56 Stats results from online calculator

Extra information for BRP6-Beta results above
---------------------------------------------
Longest elapsed time in sample: 23,770
Shortest elapsed time in sample: 17,694
Longest CPU time in sample: 2,214
Shortest CPU time in sample: 1,302
[/pre]

COMMENTS

• * As for previous hosts, BRP6 times line up very well with BRP5 long term averages, adjusted by 33% for the bigger tasks.
  * This host shows a significant improvement for BRP6-beta tasks. They even out-perform the previous BRP5 times, but not by quite as much as for HOST 01 :-).
  * On average, around 27% performance improvement over the non-optimised app on this host.
  * I'm not a fan of this CPU for three main reasons. (1) - It uses more power than a low end Intel i3. (2) - It doesn't drive the HD7850 GPU as well and needs more free cores. (3) - The three CPU threads in use take significantly longer to crunch FGRP4 tasks than the virtual cores of an i3 (around 9hrs per task compared to 6hrs). With the new beta app the gap has certainly been narrowed but I'd still rather be using a Pentium dual core or an i3. Either are likely to get more out of the GPU.
  

====####====

HOST 04 - g3258-01

[pre]
CPU: Intel Pentium Dual Core G3258 (Haswell Refresh)
Threads: 2
PCIe slot x16 Version 2 (single slot)
1st GPU: AMD HD7850 2GB
2nd GPU: -
3rd GPU: -
RAM: 2 x 4GB
Concurrency: 4 @ 0.45 CPUs + 0.25 GPUs
CPU Tasks: 1 x FGRP4
Free CPU cores: 1

Mean Times - secs Elapsed time Stats CPU time Stats
----------------- --------------------- ------------------- Sample
Search Elapsed CPU part Std Devn Variance Std Devn Variance Size Notes / Comments
====== ======= ======== ======== ======== ======== ======== ====== ================
BRP5 ~15,600 ~1,160 - - - - - Long term averages - little variation
BRP6 ~20,200 ~1,540 - - - - ~20 ~33% larger than BRP5 as expected
BRP6-beta 18,092 685 930 854,500 108 11,622 90 Stats results from online calculator

Extra information for BRP6-Beta results above
---------------------------------------------
Longest elapsed time in sample: 20,776
Shortest elapsed time in sample: 16,347
Longest CPU time in sample: 1,188
Shortest CPU time in sample: 604
[/pre]

COMMENTS

• * An excellent barebones budget GPU cruncher. Only 1 core was needed to support BRP5 4x so that will certainly be true for BRP6-beta.
  * The G3258 has unlocked multiplier and is overclockable in a budget H81 board (I use Asrock) with the appropriate BIOS version.
  * Overclocking the CPU improves crunch time of CPU tasks but had little observable effect on BRP5 GPU crunch times. I assume that is likely to continue for BRP6-beta.
  * On average, around 10% performance improvement over the non-optimised app on this host.
  

====####====

Please Note: At this point I changed the template to show the stats values of mean, std dev, and variance for each time (CPU and Elapsed) grouped together for easier reference rather than having the mean separated, as previously. I think this will be easier to decipher. Sorry for any inconvenience caused.

HOST 05 - 13_2120-01 - Updated Stats: 11-Mar-2015

[pre]
CPU: Intel Core i3 2120 (Sandy Bridge)
Cores/Threads: 2 / 4
PCIe slot x16 Version 2 (single slot)
1st GPU: AMD HD7850 2GB
2nd GPU: -
3rd GPU: -
RAM: 2 x 4GB
Concurrency: 4 @ 0.5 CPUs + 0.25 GPUs
CPU Tasks: 2 x FGRP4
Free CPU cores: 2 (virtual)

Elapsed Time Statistics CPU time Statistics
----------------------------------- ------------------------------ Sample
Search Min Mean Max S.D. Min Mean Max S.D. Size Notes / Comments
====== ======= ======= ======= ===== ====== ====== ====== ==== ====== ================
BRP5 - ~16,300 - - - ~2,260 - - - Long term averages prior to arrival of BRP6.
BRP6 (non-beta) - *21,400* - - - *2,900* - - - *Inferred value* - from BRP5. All BRP6 crunched with beta app.
BRP6-beta 1.52
BRP6-beta 1.52 18,053 18,214 18,298 60 1,337 1,371 1,403 22 24 First tasks all done with beta 1.52

Note: All beta tasks stats were done in LibreOffice calc after importing directly from the website tasks list.
There are two lines for beta 1.47. I left the previous values when I added the extra points so you could easily see the precise change.
[/pre]

COMMENTS

• * Sandy Bridge i3's have a much larger CPU component when compared later architectures like Haswell.
  * Despite this, the elapsed times for GPU tasks are surprisingly close to those for the same GPU on Haswell.
  * By inferring a standard BRP6 time of around 21,400 from the BRP5 results, the approx improvement for BRP6-beta is around 14% on this host.
  * By increasing the sample size from 107 to 162 there was little change in the mean and just a small decrease in standard deviation.
  * Still too early for beta 1.52 but there does seem to be great reduction in variability without any adverse change in the mean. What a ripper!!!
  

====####====

Please note: At this point I added a set of instructions and an actual template full of editable placeholders so that it should be very clear as to the details we would like to see in future submissions.

I would like to thank Jeroen and Gavin for their excellent contributions of results. I would also like to thank Gord for pointing out the 'Import from URL' feature in modern spreadsheets which should help people to collect data and calculate the stats.

====####====

HOST 06 - hebe

[pre]
CPU: Intel core2 quad Q8400 (Yorkfield) 2.66GHz clocked to 3.04GHz
Cores/Threads: 4 / 4
Motherboard: ASUS P5QPL-AM
PCIe slot PCIe 1.x x16 (single slot)
1st GPU: AMD HD7850 2GB - default clocks: core=900MHz mem=1200MHz temp=73C GPU load=95%
2nd GPU: -
3rd GPU: -
RAM: 2 x 2GB DDR2 800MHz
Concurrency: 4 @ 0.5 CPUs + 0.25 GPUs
CPU Tasks: 2 x FGRP4
Free CPU cores: 2
OS: PCLinuxOS 2014.04 - kernel 3.12.16-pclos3
Driver: fglrx driver and OpenCL libs from Catalyst 13.12
BOINC Version: 7.2.42 64bit

Elapsed Time Statistics CPU time Statistics
-------------------------- --------------------------- Sample
Search Mean Std Dev Variance Mean Std Devn Variance Size Notes / Comments
====== ====== ======= ======== ====== ======== ======== ====== ================
BRP5 ~23,200 - - ~4,800 - - - Estimated long term average
BRP6 ~30,200 - - ~6,500 - - ~12 Small sample size - averages estimated but agree with 1.33 x BRP5
BRP6-beta 20,045 1,675 2,807,166 2,276 680 461,983 97 Stats calculated from website data imported into LibreOffice

Extra information for BRP6-Beta results above
---------------------------------------------
Longest elapsed time in sample: 24,550
Shortest elapsed time in sample: 18,737
Longest CPU time in sample: 4,164
Shortest CPU time in sample: 1,773
[/pre]

COMMENTS

• * I chose to do this host because it is pretty much identical with HOST 01 with the exception of the motherboard. It has always performed better on BRP5 by a significant margin than all my other Q8400 hosts (all with that different motherboard) - mean crunch time = 23,200 compared to 28,500 for HOST 01 (which is typical of the others). I was keen to see how this one ended up on BRP6-beta in comparison to the others. I have my answer - mean crunch time = 20,045 compared to 20,884 for HOST 01. Still better but not by that much any more.
  * The improvement of the beta app over the standard app seems to be better than 30% for this host.
  

====####====

Please note: At this point I added an extra BRP6-beta 1.52 line to the template to allow for separate reporting of the new optimisations now added to the app.