| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Microbiome Immunity Project
Thread: Low credits on windows |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 7
|
|
| Author |
|
|
fuzzydice555
Advanced Cruncher Joined: Mar 25, 2015 Post Count: 89 Status: Offline Project Badges: 
|
I did a little experiment with dual booting a Ryzen 3900X system and noticed that the points are much lower for windows than linux.
----------------------------------------I am limited to one MIP task per system, so CPU cache is not the issue here. Windows 3900X: Ubuntu 3900X: | Time | Credits | The runtimes are almost the same on ubuntu & windows, however the windows one received much lower credits for the same amount of work. I checked all my other linux machines and MIP tasks earn on average 40-60 points / hour, while these win tasks are around 8 points/hour. Was this just a bad batch of work units for windows? Can anyone running Ryzen on windows check their runtime/credits?  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
It's a function of the lowest common denominator in the credit pile, where integer does a much better computing job on Linux versus windows.
As knreed expansively wrote some week ago, NO NO NO, no revising of the credit system, zero chance. If you got something, present the case in full data analysis detail to the BOINC developers group with functional model how to remove the disparities. |
||
|
|
fuzzydice555
Advanced Cruncher Joined: Mar 25, 2015 Post Count: 89 Status: Offline Project Badges:
|
It's a function of the lowest common denominator in the credit pile, where integer does a much better computing job on Linux versus windows. As knreed expansively wrote some week ago, NO NO NO, no revising of the credit system, zero chance. If you got something, present the case in full data analysis detail to the BOINC developers group with functional model how to remove the disparities. 1. What is the credit pile? 2. I'm not looking to reform anything about the credit system. 3. On the same dual boot system OPN tasks show the exact same credits, which says to me that integer performance difference or the benchmark is not the cause. 4. I checked knreed's post: "In your case, since your computer was "overclaiming" the system will have lowered the efficiency rating" of your computer and will thus lower the amount of credit per cpu sec it estimates as what your computer should be awarded going forward." The tasks don't have a huge claimed runtime, so I'm not sure this applies to my case. |
||
|
|
TruboXL
Cruncher Malaysia Joined: Nov 22, 2015 Post Count: 29 Status: Offline Project Badges: 
|
I feel that MIP is missing a major performance benefit on Windows as they don't provide 64bit version of the science apps unlike on Linux and macOS
I think everyone knows the benefits of 64bit already so why not? Other active projects in WCG are providing 64bit version of science apps on Windows... |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
It's not just Windows:
----------------------------------------https://www.worldcommunitygrid.org/ms/device/...s.do?workunitId=287037503 https://www.worldcommunitygrid.org/ms/device/...s.do?workunitId=285029896 And it not just MIP: https://www.worldcommunitygrid.org/ms/device/...s.do?workunitId=289496946 [Edit 2 times, last edit by Former Member at Sep 16, 2020 1:21:23 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
It's I think all to do with maximum non-deviation of the results when all gets grouped and then analysed for the best common conclusion. IIRC there's no platform distinction with Rosetta either. Rosetta at Baker used to run in 32 bit only too. Don't know if still the case, but their raspberry arm version seemingly is 64 bit. See "https://www.google.com/search?q=rosetta at home in 64 bit" for the hits when search on just 64 bit at R@H.
----------------------------------------[Edit 1 times, last edit by Former Member at Sep 16, 2020 1:36:50 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
It's I think all to do with maximum non-deviation of the results when all gets grouped and then analysed for the best common conclusion... Not sure what you mean by that. Not that I am really concerned with points, but I know some people are so that's why I thought I would point it out as a warning to them. I order to trigger this behavior this is what I did: Launch an Ubuntu LXD container on a Ryzen 7 3700X. Configure it to use CPU's 0,1,2,3,8,9,10, and 11 (That way all 8 threads would be isolated on their own CCD and share the same L3 cache). Then install and configure BOINC to run 8 threads for a few hours to see what performance was like. After I had my performance data I dropped down to 4 threads and since then the points awarded for results from that container are much lower than they should be. |
||
|
|
|