Looking for Opinions on Reviews

[Patrick]

With Linux-Bench's 11 main tests generating ~55 data points, I think we have hit the point where the script is generating too many data points. Especially hard is that the scale is completely different between the tests.

Here is the big question:
What would make sense in terms of presenting data? Eventually they will all be in a sortable format on Linux-Bench.com. Until then, what makes the most sense for presenting results?

E.g. I am working on a comparison of two lower-end Intel Xeon E5 processors. My first instinct was to present all of the data. Now I am thinking that may not make sense.

One idea I had was to use a baseline, then show a performance delta in percentage for all of the tests. Then have the raw data available in the application to sort/ make graphs as folks want with the raw data.

Another idea is to present 5-10 different graphs but focus on specific bits. Like Dhrystone/ Whetstone for UnixBench as an example.

Thoughts/ suggestions would be appreciated.

Regards,
Patrick

 

[Stanza]

I personally have not looked at or run the Linux bench as yet.

Some things I would be interested in tho are

Performance per watt.

Eg

Xyz CPU took abc amount of time to run the bench which used 123watt hours of juice to do it

Bbb cpu took longer to run the bench, but only used 86watt hours of juice to do it.

Would give me meaningful results to make a purchase decision .... Depending if I was looking for raw speed vs lowest power vs decent performance with reasonable power use etc.

?

.

 

[Patrick]

I personally have not looked at or run the Linux bench as yet.

Some things I would be interested in tho are

Performance per watt.

Eg

Xyz CPU took abc amount of time to run the bench which used 123watt hours of juice to do it

Bbb cpu took longer to run the bench, but only used 86watt hours of juice to do it.

Would give me meaningful results to make a purchase decision .... Depending if I was looking for raw speed vs lowest power vs decent performance with reasonable power use etc.

?

.

Tom's used to do something like that. Chris was using the same Extech power meter I use (I aligned to ensure we could be compatible) so I am guessing I could do it.

There are a lot of variables with that though such as fans, power supplies, onboard motherboard components. What about having 24x 32GB DIMMs v. having 4x8GB? The 4x 8GB will use less power but does that make them less power but is it meaningful?

Great thoughts, just not sure if people use servers to measure workloads/ power except maybe in cryptocurrency mining scenarios for example. In many cases, faster page loads mean more sales or faster compute times mean more iterations/ less idle time for folks. In either example, the speedup is generally worth the power differential (these days.)

Will certainly look into it.

BTW - you should try Linux-Bench. Ultra easy to run. I wrap the entire bash command in time() just to see the differences.