Allow Bitblade CUDA to create more than 1 plot per run (-n)

[j4s0n]

Describe the request
Add a configuration option to allow the COUNT flags to be greater than 1 via a configuration setting. This would more than double the speed of Bladebit CUDA plotting. Every time Bladebit starts, it performs an allocating buffers phase, which takes a significant amount of time (58% of the total plot time, on my system). This is only done once, regardless of how many plots are being created (ie, -n 100).

Additional comments
On my testing, creating 1 c7 plot takes 4m9s. Creating 2 plots takes 6m34.8s (1m44.7s savings). Creating 3 plots takes 8m36.3s.

I understand that this is architecturally a divergence from how plotman traditionally schedules plotting (single plot tasks in parallel). This could be accomplished either by using something like a minimum_plotting_count option which would passed to bladebit -n. This would cause the plotter to fail once the disk is full, so some disk space checks to override may be desired.

Alternately, using a boolean-style setting could be used by detecting disk space combined with utilizing the suspend or kill commands once the disk is full.

[altendky]

Sorry to leave you hanging here. I'm not going to promise anything, but there is some discussion about putting some work into plotman again. It is becoming of personal interest, so there's some hope it actually happens.

Thanks for the report. As you point out, the needs around plotting have certainly changed drastically from the initial design of plotman. However much time, and whenever, we get around to applying to this certainly starts with getting familiar with the present needs around plotting and figuring out how to support those best.

[j4s0n]

Thanks! No rush -- The Machinaris project (which uses a modified plotman) has implemented this. It's a solid work-around for now, if anyone else has the same concern.