Hello Bipben and miner friends,
Sorry for taking You time on this topic, but I dare to ask You for advice.
I am in mining for long time, but with GPUs, now, with HDD PoC mining, I am missing something.
I just got 8 pcs of 4Tb HDD WD40PURX sata drives.
And have motherboard for it, A88X chipset.
All drives are going to same mobo. I also got 16Gb of ram (2x8Gb).
And some low end CPU (A4-3700).
It is dedicated PoC miner machine.
I also have 280X graphic cards ans I 3770k CPU on other comp (where I work).
What I am missing, or do not understand is this:
1) Plotting HDDs. Should I use wplotgenerator to plot hdd? Or Gpuplotgenerator? Is result same?
Can You please tell me parametars for gpu plot for configuration on my machine with R9 graphic. Cpu ram size is actually CPU cache size (on 3770k cpu it is 8Mb)? And I have 8Gb RAM DDR3. Radeon is with 3072Mb ram. Platform 0 device 0. I know rule of plotting and done it with earlier version of Gpuplotminer on 1tb spare drive. But it is different.
2) Is 16Gb ok for mining machine, or is also 8Gb ok. On that machine will be 8x4Tb drives? And what miner to use for optimal results?
Thank You.
@Yanakitu Tenatako
1) As you seem to have a great amount of GPU power, it could be a good idea to use the GPU plotter. They both produce the same output files. The only thing that varies is the device used for plot generation.
When I speak about CPU RAM size, I mean the available amount of main memory, not CPU cache size.
2) About mining, you most likely should use one of the C ones, as they are faster and more stable in memory. With higher staggerSize values at plot generation, you will reduce the disk stress when mining.
@Alex Coventry I've looked at your code. I think that the N CPU buffers are really bottleneck as it will require a lot of RAM to plot N disks at the same time, or will force the user to reduce its value, thus increasing disk-stress when mining. Maybe a limted amount of rotating buffers would be enough and more RAM efficient. Maybe a stagger-less version (staggerSize = fileSize / PLOT_SIZE) of the plotter could solve this RAM issue (need some tests as it will increase IO operations).
Anyhow I will work on that part with the already implemented multi-GPU support, begining with the ideas behind your version.
What do you mean by "much faster"? Do you speak about a performance difference between the 3.0.0 and the 2.1.1 or between the 2.1.1 and your modded one?
Yes, there is a lot of room for improvement. I only took it as far as I needed for my purposes.
You only need as many rotating buffers as it takes to max out transfer bandwidth.
By much faster, I mean faster than the same code without asynchronous writes.
Thanks for publishing your revision history.
@fivebells I totally agree with you
The only hard part is to guess at runtime which number of buffers would max out bandwidth. I think that I will make it an input parameter to be as polyvalent as possible.
