So fundamental question time. If I plot with a stagger size of 8192, the plotter will consume 8192 memory to make the plots, but the miner will not use 8192mb of memory to mine? Or do both machines consume 8192MB of memory, both during the mining process and the plotting process?
Before optimizing the plots my miner is using up to 8192MB of memory while reading through the non-optimized plots, is that normal?
I think I missed something somewhere. What determines how much memory Blagos miner uses and the GPU Plotter? I assumed the amount of memory the GPU plotter uses is based on your globalworksize x number of cards and the amount of memory the miner used is your stagger size (when not optimized).
Also it seems as though I'm making less after optimizing then I was before optimizing (by a large margin).
I assume ur mining on windows? Windows caches everything it reads from large files and holds onto it in a 'standby list' of memory. For some reason windows doesnt know how to let go of this properly when reading large files and u have to run something like ranaurufu's memory cleaner which will purge the standby memory list once it hits a certain limit, but it does this for all memory mapped files on the standby list, therefore there is a systemwide decrease in IO performance, which is fine if ur windows box is dedicated solely to HDD mining. So, ur best bet is to set up mining on a linux machine that doesnt do this insanely aggressive caching. But, for ur first question, yes more RAM while plotting is better.
When using runaurufu's memory cleaner, u will want to make sure to set ur plot sizes to 200GB or less, that way, they can be flushed more often and the peak size of memory used is less. Because, it seems that windows wont allow a file to be flushed while its actively still being read, so a 4TB plot, will consume a shit ton of memory till the miner has completed reading it.
Very good answer!
I tried to use the program to set a maximum cache - unsuccessfully, no positive changes were seen...
http://www.microimages.com/downloads/SetFileCacheSize.htm (uses
https://msdn.microsoft.com/en-us/library/windows/desktop/aa965240(v=vs.85).aspx)
This program clears the cache
http://forum.sysinternals.com/rammap-empty-standby-list_topic27297.htmldetails can be read here
http://blogs.technet.com/b/askperf/archive/2012/12/19/where-has-all-my-physical-ram-gone.aspxYeah, the filecache is different for some reason then the standby list that these enormous plot reads go into, so it has no bearing...
Here is the memory cleaner I use, it pretty much calls the 'flush standby list memory' functions of windows but on a running threshold basis, which is good for mining, since it needs threshold rather than a timed schedule...
http://runaurufu.com/file/download/7Specify a number after the command & it will set that threshold, it defaults to 100MB. I find 1000MB works well.
On another note, there is a version of Uray's miner that was modified by someone, its the r2 build and it completely disables read caching the plot files alltogether, what would be even more awesome is if someone could do this w/ blago's miner, since thats my miner of choice. I wish I could get someone to do the same build but w/ Uray's latest release. =/
Here is a quick update:
Biggest part is finding a willing dev