You mentioned Bladebit is a great solution for middle range consumer plotters but what about low range consumer plotters like me, who use HDD (in raid0) for plotting, can Bladebit improve plotting speed for them? :$ Thanks for the great video as always my brother and keep on being awesome
Using madmax, my 5950x system produces a plot every 18.5 minutes on average, without any overclocking, and is fairly consistent. The alpha2 build of bladebit seems to only be able to get down to 26.5 minutes on the same system with similar settings. How I invoke madmax: -r 16 -K 32 -u 512 -v 1024 I've found that, with madmax at least, it is faster to use fewer threads on the first and second phases. Madmax doesn't max out my CPU cores at any point, but bladebit does immediately and throughout. Madmax has no problem using 110G of RAM for temp space via tmpfs, where as bladebit chokes on anything over 105G on my system. I'm assuming there are a lot of things just stuck there in the alpha which will be remedied as it gets closer to release. Command arguments are a little funny. Bladebit only seems to work when the arguments are in a certain order, which is a little odd, and seems unnecessary. I'm sure that'll be ironed out before release. I'm very much looking forward to bladebit surpassing madmax in performance, if possible.
Your vision of a budget system is very different from my own... Can you do some benchmark in a PC with "only" 16 GB or RAM? Plus 4 cores 8 threads and 8 cores and 16 threads?
It's always going to be fun to tune for performance: especially on some of those larger systems! But with netspace still slowly sliding... is there any need? Maybe this work is to assist plot compression next year?
just putting it out there, why isn't an arm specific plotter planned? there's potential for an arm only gpu plotter, realistically raspberry pi, jetson nano, etc are never going to match a threadripper speed wise. Almost all present GPU cards are x86/64 based not arm. But could achieve use case for smaller old hard drives, plot with them until they die, While the rpi or whatever also farms. 280w + 20w vs 20w which would be used anyway. remember you're competing with proof of stake on carbon footprint in the near future.
So tricky question. On skylake and cascade lake 192GB is full memory bandwidth for dual socket (12 channels of DDR4). But as far as actual memory, it won’t help much beyond 128 until you get to completely in memory with 416
