Is Available RAM taken into consideration for the overall Memory Mark score?

[grem75]

I'm using the latest Linux version of PerformanceTest and had a suspiciously low overall result on a kit of 1866 DDR3 even though my individual tests were right in the ballpark of others. It scored 1446 overall at the highest where others were in the 2400 range.

When I did the test the system had been running for a few days so cache was quite full and it showed ~1GB of "Available RAM". I purged my cache and it scored 2350 with no other changes and the individual results weren't that much different. Some were higher, some were lower, but nothing that should account for a 900 point difference.

Surely "Available RAM" has nothing to do with the performance, but I can't see any other reasons for that low score. Is there any documentation for how that calculation is done?

https://i.imgur.com/L7Ykd0X.png

https://i.imgur.com/uAbaER5.png

[Simon (PassMark)]

Available RAM is part of the calculations, see formula here:
https://forums.passmark.com/performa...4964#post54964

[David (PassMark)]

Surely "Available RAM" has nothing to do with the performance

Trust me, a system with 16GB of free RAM will, on average, be significantly more responsive than a system with 1GB of free RAM.
With 1GB of RAM it is very likely you are going to end up swapping memory out of disk, which incurs a huge performance penalty.

[grem75]

That was I/O cache, which can be freed immediately if something actually needs it. I was really only using about 5GB, I didn't close any applications between those tests.

Using that unused RAM as cache is a performance benefit with no penalty, especially with my SATA disk. Also, I don't swap to disk.

That makes the score pointless if I can increase it by using less RAM. That isn't benchmarking the hardware. Just because I boot a system and run the benchmark in 100MB doesn't mean the hardware is better.

[David (PassMark)]

But if you have to dump the cache to load an app, then that will have a performance impact.
Having lots of RAM does increase performance. Up to a point of diminishing returns anyway, but with just 1GB free you are some way from the point of diminishing returns.

To be fair we made it this way as it was consumer expectation that having, for example, 32GB of RAM is faster than 8GB of RAM. Of course the RAM isn't faster. But often the overall user experience will be better because of the additional RAM.

You can just disregard the test if you like and only look at the other test results.

[grem75]

It is instantly available as application memory. It is like a deleted file on your hard drive, there is data there but it is marked as available for writing to the memory management of the kernel. It doesn't have to be written as 0s before you can write something else to it. Even if it did have to write it to 0s first my crusty old DDR3 can do 9GB/s, how long do you really think it would take? I can write 1GB to RAM before I can read 100MB from SSD.

Windows has the exact same caching behavior, it just presents it differently because people will panic. There is a reason https://www.linuxatemyram.com/ exists and many system monitoring utilities have changed their behavior on Linux to make the difference more clear.

I'll bet Passmark is reading from a value that excludes cache on Windows, just from looking at other people's results. Since the algorithm can't be changed to ditch the useless variable without invalidating previous results, at least fix that bug so it excludes caches on Linux too.