Yes, that is what I was thinking. But if the difference is too small to matter, then this would be a dead end.
Cheers

Regarding "efficiency" and Ralf's remark from the previous post, quoted below...

I don't have any systems that run both BOINC and "real-world" production work (and they're all running Linux), so my efficiency experience is likely to be different to Ralf's... However, I do want to note that I start wondering what's going on if task efficiency drops below 97 or 98% as it usually indicates a system problem, and efficiency well over 90% is very common if my wingmen are anything to go by :-)

To provide some data points I just did some ad hoc queries on my Wingmen database (SQLite3) when I saw Ralf's post. With no access to medians, standard deviations and such like to do a proper analysis in a hurry. I'll just summarize :-) Note that I don't let BOINC have more than 75% of "CPUs" - experiments using more resulted in various throughput and efficiency issues on my systems.

Of the WCG [non-GPU] projects I find ARP1 the least "efficient" (no surprise with the amount of output it does, which is not going to SSDs on any of my main systems at present...) and I get over 98% on both the systems well enough resourced to run ARP1. (And wingmen's total CPU divided by total elapsed yields over 95% on ARP1!)

My laptop (which I use quite a lot for browsing and such like, only manages 95..96% for MCM1 and OPN1, but my other systems manage over 99% efficiency for MCM1 and OPN1. The wingman "grand average" is about 95% for MCM1 and about 93% for OPN1. (Interestingly, the laptop manages 99% at TN-Grid!)

My Raspberry Pi (which uses a USB3 SSD) gets over 99% efficiency on OPN1 (the only WCG project it can run) and almost 100% efficiency at TN-Grid (less I/O!) -- I think that highlights where the efficiency drop is :-)

Now, of course, it can be pointed out that my sample of about 16,500 tasks (and about 13,200 wingmen's tasks - lower because of adaptive replication on OPN1) gathered since WCG resumption is an insignificant sample, but if others can speculate so can I :-) And I find I can't agree with that sub-90% assessment if we're talking about BOINC running on machines that are not "real work" platforms as well...

Cheers - Al.

P.S. The OP has only clocked up just over 15 years total according to the member statistics comparison, and as there are 10 years or more of OPN1 in there (suggesting MCM1 is possibly only a recent fall-back given it's only at 1 year?) I think the earlier comments about said user's distance from the next badges was on target :-)

Thanks Alan for providing some additional numbers here. And I must admit, I am a bit surprised.

To be clear about my post, as "efficiency" I am counting the "CPU Time" used for a specific WU against the "Elapsed Time" of that WU. This data is what I got by downloading daily (with a 2 day exception due to work out of town) the CSV file from the Results page of the WCG web site from August 23rd 2022 until today (well, yesterday). and process that data with my own stats program, because at that time (end of August) the WCG internal stats had been running for the 2 months since WU started being send out "for testing". One of the things I was always curious about the last few years was how much time the BOINC client is actually using on my hosts.

As I mentioned before, ALL my hosts are actually computers that are being used, in one form or another, some more, some less, during regular work hours and some beyond. Not a single one of my hosts is operating purely to crunch for WCG (or other BOINC projects). This is currently roughly a total of two dozen machines (haven't bothered to do an exact count), with about half of those being machines I have either physical access or at least remote access (like some servers). The rest are machines that I got permission to use their "spare CPU cycles but which I do not have access to for months because they are at friends or customers without remote access. A few of those are machines I haven't touched for at least a couple of years since I have set them up.
This is also the reason why I do not modify the default BOINC client settings after install. Not only do I not have the time and desire to babysit each and every one of my hosts, I just don't have access to them. For me, once BOINC is installed and projects (99% WCG, a couple have at least one other, non-WCG, project added) activated, it is pretty much hands-off, fire and forget.That's why I am so frustrated when the system doesn't work and we have to deal with download errors or other issues, as those are events that at some point "stuff up" the hosts and result in messages, more or less annoying for specially those for me not accessible machines and their users.

One additional observation that I have made in recent days is that the number of WUs in PV jail (Pending Validation and Pending Verification) in my local database derived from the Results page download is deviating from the number of such results shown with the filters set accordingly on the Results page, by at least a couple hundred WUS (more precisely, Results). This is currently about a 100% difference (747 in my database vs 382 shown on the web site). Those PVa/PVe WUs don't count in my efficiency calculation, I do count only valid resuls. But that should not account for the difference between my general efficiency numbers and those from Alan.
Interesting for me is also while ARP1 shows for me the highest percentage of all active projects, while for Alan, this is apparently the project with the lowest efficiency. But with a very small sample size of only 71 valid results in 4 months counted, I am not sure how valid that statistic would be on my side.

Ralf

I am showing the following efficiency ratings for my current 4 machines.
General use Windows 7 I7-3770 96.95%
Dedicated Linux I7-2600 99.08%
Dedicated Linux Xeon E5-2665 99.67%
Dedicated Linux Xeon E5645 99.92%

This is just for the last 683 results which are all MCM.
For a while I had kept pretty detailed statistics with over 100,000 results and these numbers are consistent across projects.

Noted these have nothing to do with efficiency per kilowatt hour. This is cpu time divided by elapsed time.

Cheers

Now that everybody is running MCM, and lots of big resources are chomping those tasks, why isn't there a 500 year badge? I mean, even IBM doesn't have one, so I would conclude it doesn't exist.

Disclaimer: I hope to get there in about 100 days, and I want one!