Well, additional file-download is 65 KB or something, so is easily done in 1 second. Startup of task can take some time, starting 7 + resuming 1 (on i7-HT) at the same time took roughly 4 minutes before cpu-usage hit 100%. Disk-usage quickly increased to 300-something MB, during initial startup, for so increasing to 500 MB a few minutes afterwards.

Not sure on later disk-usage, since didn't bother baby-sitting them. But, atleast the 1st. task hit 684 MB before doing a checkpoing and dropping-down to roughly 500 MB again...

More interesting in this run is the following:
7 of 8 used 4 minutes to startup.
Minimum run-time 10.19 hours, max run-time 12.07 hours, average 11.297 hours.
Minimum cpu-time 10.08 hours, max 11.97, average 11.19 hours.
Wasted time: minimum 353 second, max 422 seconds, average 390 seconds or 6.50 minutes.
CPU-efficiency, minimum 98.9%, maximum 99.2%, average 99.04%.

So, majority of the wasted time was on the initial startup, and while no current way to test this, atleast my guess is this initial startup will take less time "normally", when not all cores starts at the same time, but instead on average is 1+ hours between each start.

Per 24 hours a 99% efficiency will give less than 2 hours wasted cpu-time on an 8-way HT-system, so a quad should be somewhere between 1 and 2 hours wasted cpu-time per day.

Since a 3.5 GB ram-disk on average will lead to 2 hours idle time per day on a quad-core, even in the unlikely instance you'll manage 100% cpu-efficiency, you're still worse off performance-wise using the ram-disk than using an old HD...

Well, this thread started as a ramdisk for Windows, so until CEP2 is released for windows, knowing much of actual performance will mostly be guess-work...

Now, maybe my assumtion is wrong, but atleast my assumption is that an i7 running 8 CEP2 will put larger load on the disk-subsystem than a quad-core running 4 CEP2 will do...

So, while cpu-efficiency maybe isn't a good indication, my test atleast gave conclusive answers to some of my initial expectations:
1: Downloading next task takes only 1 second.
2: Disk-usage is atleast 510 MB continuously (except the very short startup-period).
3: Disk-usage can increase to atleast 700 MB (*), if this was the peak or not I don't know.
4: Application-files is 204 MB.

This means, a quad-core should max-out at:
700 MB * 4 + 204 MB = 2.75 GB.
To get new work, if assumes not at peak usage, you have:
3 running tasks = 510 MB * 3 + 204 MB = 1.50 GB, too large. (**)
2 running tasks = 510 MB * 2 + 204 MB = 1 GB, ok.

So, as expected, on a quad-core, you'll only re-fill with work if max 2 CEP2 is running...

Meaning, only on a dual-core can you max-out CEP2-production.

As for finding a dual-core with enough memory to run a 3.5 GB ramdisk...

Even for a quad-core, having more than 4 GB memory installed is not common.

If you've got more than 4 cores (real or HT), you can't let CEP2 cun even in a mix of projects, since if 6 CEP2 start at once, you'll use minimum 3.2 GB just for CEP2, and if a couple of them starts using a little more, chances are all will error-out...

So, summing-up, my conclusions are:
a: Single-core can't fulfill the memory-requirements of a ramdisk regardless of what you're doing.
b: Dual-core can fulfill the memory-requirements, if upgrades the memory. This commonly would mean changing all memory in the computer.
c: Quad-core can't max-out CEP2-production, it will either sit partially idle, or you'll need to run with a mix of other projects. Also, chances are you'll need to upgrade memory.
d: Hex-cores and higher must be limited to max 1 CEP2 at a time, if you're not going to constantly baby-sit the crunching that is...



(*) The 2 GB disk-requirement wasn't taken-out of thin air, someone did violate a limit of 800 MB or there-around, if someone also violated a larger 1.5 GB limit I don't know. Also, seeing the long beta-phase and upload-server isn't ready before now, it's not unrealisic if they've started with the "small", "easy" wu's initially, and is planning on releasing the "big" wu's needing 1 GB+ later-on...
(**) You can't set free disk-space to zero in BOINC.

Works for me running combined with HFCC and C4CW. Start with running a mix and if you can set a RAMdisk as big as 6-7GB, you may have a big winner. Key is that at work fetch time CEP2 tests if 2GB is free. After, it can't care less so a work around is for instance to do a multi day fetch and then run it near dry and then repeat the cycle. Of course, some members report that the gap time (elapsed-cpu time differential) for the some I7 such as 980x is minimal, 1-2%, so then it's not worth the effort. The reported CPU time itself is no indication. Run times are all over the place for CEP2. You have to look with tools such as BOINCtasks that logs the efficiency. Without this utility it's pencil and paper.

I might have misunderstood what you meant here, but a RAM disk is a virtual hard disk created in memory, hence RAM disk.

http://en.wikipedia.org/wiki/RAM_disk

Unless the conversation is about some other kind of RAM disk?

Hello Speedy51,

That is my opinion, too. The problem is that CEP2 runs reasonably well on Linux but loses efficiency on Windows when running multiple CEP2 tasks due to excessive disk accesses and the way Windows handles Disk I/O. RamDisk should certainly solve the problem on Windows, but that much RAM costs a great deal. The cheap solution is to establish a Linux partition and run CEP2 in Linux64 on a fast disk or run Windows with just 1 CEP2 task at a time (an option in your profile / projects) or choose to run multiple CEP2 tasks on Windows while ignoring the disk thrashing and loss of efficiency.

Lawrence