Mayayana wrote:
-
"Shadow" wrote

| Have you tried measuring speed? CrystalDiskMark shows quite
| impressive numbers for RAMDisks. If it's a faux RAMDisk, numbers will
| be much lower.

See my post to Paul. I got it set up. Then I tried
Crystal, which I'd never seen before. I don't know
how to interpret it. The ramdisk had only 94 MB free
so I tested a run of 64 MB. The result was 10 times
higher than read/write to a normal partition. I don't
know what that means. Maybe I'd need to set up a
folder on the ramdisk that's not swap in order to test
it.

You can use HDTune free version to test the
"whole surface" of your RAMDisk in read mode.
You don't need to test write mode particularly,
as there's no reason for read versus write to
vary. There's no "physical process" involved here.

My HDTune read is a flat line at around 4GB/sec, which
isn't really all that good of a result. The best
results I can get in the room, are around 7GB/sec.
And this is nowhere near what a Streams benchmark
should be reporting, especially on the other machine.

When you work out the bandwidth claimed based on
the memory numbers, the results on the RAMDisk won't
even be close. You get what you can from it.

I discovered one other anomaly, which is that performance
is more consistent on OSes like Windows 10, if you set
the Power schema to "High Performance" and jam the CPU
to the nominal clock. That'll help if your graph line isn't
flat, when HDTune testing. You'll get stairsteps
in the graph, if the machine is running the "Balanced"
schema. I reported a previous result for Windows 10,
where HDTune didn't seem to be returning the right
numbers on Windows 10, and perhaps this latest
discovery of the schema setting, might help with
that too.

Paul