Starting a new thread for the discussion of Benchmarks as it has rather taken over another thread.

Same physical computer.
Linux, BOINC8.1.0 Measured floating point speed 3.9 billion ops/sec
Measured integer speed 10.32 billion ops/sec

WINE BOINC8.0.4 Measured floating point speed 6.97 billion ops/sec
Measured integer speed 30.1 billion ops/sec

Win10 (in VM) BOINC8.0.3 Measured floating point speed 6.3 billion ops/sec
Measured integer speed 22.08 billion ops/sec

My understanding is that it is the floating point speed that is critical for CPDN. When I finish the current lot of work using WINE, I will start up the Linux client and check that benchmarks have been run on it. I should for the sake of rigour also get the two windows clients to the same BOINC version. I can't see why it should affect anything but I could also tell the WINE client to report itself as Win10 to match the VM one.

Anything else I should do to ensure my initial data actually means something?

You don't have to wait. Go to your computers in the account and at the top left select to view all computers, not just the ones active in last 30 days. From there find the ones you want to check and view details. If values are 1 and 1 then benchmarks haven't been run (or reported to project yet).

Dave, It's hard to believe those numbers: the linux speed is half that of both the VM & Wine? Nah. If anything it should be the other way around but much closer.

I wondered why the client even bothers to do its own benchmarks as it could easily tabulate them using the SPEC CPU database once the CPU has been identified. I presume it's to take overclocking into account for credit purposes.

CPDN models need as fast single core speed as you can get, followed by memory bandwidth.

---

---
CPDN Visiting Scientist

We ran a similar thread at SETI some years ago. A similar question had arisen: running on the identical hardware (a dual-boot machine), the Linux benchmark was substantially lower than the Windows benchmark. To save you wading through a long and sometimes techy thread, I'll drop you in on what turned turned out to be the final answer:

Windows vs. Linux

By default, BOINC runs science apps at the lowest possible thread priority, equivalent to 'idle'. And at the time (over 15 years ago), Ubuntu would run idle processors at a reduced clock rate ...

So the problem turned out to be real, and could be remedied by changing an OS configuration setting.

My understanding is that the Ubuntu default setting was subsequently changed, so that the problem won't recur on newer installations. But I remember it as a useful warning to look at a question like this from all angles.

Dave, It's hard to believe those numbers: the linux speed is half that of both the VM & Wine? Nah. If anything it should be the other way around but much closer.

This was probably benchmarked while running OIFS tasks which would have meant running the benchmarks with a very low number of cores set. I am wondering if that has an effect?

Mine is the reverse of Dave's but follows the same pattern. On both the newer Ryzen and older Intel, both of which are Windows10, the VMs (Linux) have floating point values at least 1 unit higher than host. WSL2 Ubuntu being the highest and Hyper-V Ubuntu 2nd. Host Windows is lowest. So Linux is higher than WIndows but in my case Linux is VMs and Windows is host. What's interesting is the integer speed on LInux is ~4 to 9 fold higher than Windows.

BOINC suspends all work while running benchmarks. Dave's values do seem kind of low, for it being a 7950X cpu.

My understanding is the benchmarks are the old Whetstone (floating point) and Dhrystone (integer) benchmarks.

I'm not really sure that benchmarks do much anymore besides possibly not being able to get initial work if one's cache is set really, really low, for most projects. For large CPDN tasks, it makes a bigger difference, otherwise cache would need to be really high. After initial task(s) complete, I think BOINC learns the capability of one's system and uses that to get more work.

I am wondering if that has an effect?

Yes but not the one I expected! I cut Wine installation of BOINC down to one core, ran the benchmarks, and FP performance has gone up to Measured floating point speed 7.55 billion ops/sec from 6.97 billion ops/sec. Having demonstrated that the number of cores in use affects it I now need to check all installations with the same goalposts. (Could it have been thermal throttling with more cores running?) Yet another variable!