On my laptop which I am using now, if I run BOINC natively under Linux, BOINC inevitably underestimates task length though not by a massive amount. However when I run the Windows version under WINE when I first run a new task type it overestimates by a factor of about 2.5.

Has anyone else noticed this?

I know it is a minor irritation and if it never got sorted it wouldn't affect my crunching but I would be interested to know a bit about why it is happening.

We would need to know which project, and which task types, you're asking about: there's probably somebody here with knowledge of most of them who could help explain it.

Do be aware that for most projects, the bulk of the runtime estimation work is done on the project servers, not on your computer. When you start running a new application type, the servers pay very little attention to the speed estimates supplied by your computer: instead, they start accumulating statistics on how well your computer performs on the new task. Once you have 'completed' 11 tasks of the new type - meaning that you have processed the computations successfully, returned the results, and had them accepted as valid by the server - the runtime estimates for newly issued work will be based of your reported experience so far: that would be a better point to make the comparison.

Thanks Richard,
When available I only crunch CPDN but as you will be aware that project has been down for a while but has displayed this difference between native linux and WINE in the past. I am currently seeing it on WCG with mainly the mapping cancer markers but also on other task types from them. The discrepancy between the two task types would be most simply explained to my mind by WINE doing something that misreports the CPU speed/type but I know there are a lot of variables so aren't assuming my first guess is correct.

In general, I think that CPDN is one of the projects that doesn't maintain speed statistics on the server, but instead relies on 'benchmarks' and a single DCF value.

The benchmark value for your CPU (there's no equivalent for GPUs) is calculated periodically by code embedded in the BOINC client. One interesting comparison would be the reported benchmark figures for the same machine, by the same (numeric) version of BOINC, once as a native Linux build, and again as a Windows build running under WINE.

The other variable to consider is the relative efficiency of the CPDN science applications - the same version of one particular program, as compiled separately for Linux and Windows. I'd suspect that in general scientific laboratory staff are better at configuring Linux compilers than the Windows equivalent, but there may be cases where labs have hired in Windows talent because they know it supports the greatest number of volunteers.

Pretty sure you are right about CPDN based on what various moderators have said over the years. I guess the other test would be to do a dual boot andsee what happens running natively under Windows, though as to do that legally means buying a copy and messing with my hard disks which are set up how I like them with SSD for the OS and larger HD for data. I assume that if WINE causes BOINC to read cpu speed/cpu type wrongly it would affect both those projects which keep the data on their own servers and others like CPDN alike.