Someone posted this in our Chinese telegram comparing the two compressed plot standards. Max has stated that his plots are 2% smaller in the past, and it looks like he is exceeding that. Bladebit’s compressed plotter isn’t completed yet and likely Harold has more to do. However, it looks like currently Max’s fee is more than balanced by the gains.
(I assume the high C8 number for Bladebit is because Harold hasn’t finished C8 yet but am not sure)
Need ENG version… For the people to be able to compare info.
nice overview which summarizes all different available sources.
I myself have noticed a small anomaly between madmax cuda plotter C8 & madmax cpu plotter C8.
the cpu C8 plots are on average 50MB larger than the cuda C8 plots, this sounds insignificant but leads to the fact that a 18TB HDD can only filled with 234 CPU plots instead of 235 cuda plots.
and this is not a statistical inaccuracy due to a too small amount of plots, i am talking about the average of over 4000 CPU plots and over 15000 GPU plots. even with a quantity of 1000 plots this can be proven very well.
This one I just did up makes it easier to see.
Note that it is difficult to compare both precisely as the C levels for both do not match up. For instance, Bladebit C9 is closer to Gigahorse C7.
Yeah like this it is a bit of an apples to oranges comparison.
You need to move the table for gigahorse two rows down as Max starts at a higher compression level. So c1=c3 etc.
That way you will also see that the cutoff point for CPU is around the same point (c5 for gh, c7 for bb)
Then the real comparison will be the actual performance of the farmer. As in how much power does it take to farm 1PB of gigahorse C1 vs the power to farm 1PB of bladebit C3
From what has been said so far, the power draw will be a quadratic formula based on C levels, so most likely negligible for low C-levels (not sure, but maybe up to C3-5) but becoming potentially ~20-30% of HDs power consumption of the farm around c8 level. So, maybe rtx 4000 will be the way to go for higher compressions, if the power / performance is logarithmic for video card generations (I would assume so, but really don’t know).
What it kind of implies that it would be interesting to see whether the netspace gain for instance from C7 to C8 (~6%) would be worth the doubling of the electricity cost for the extra GPU power.
Something else to think about for the longer term is FPGA/ASIC which are a lot more efficient than GPUs ! 
i’m sorry how come hdd power consumption rely on it?
It doesn’t and I didn’t write it. Read it again.
ok my bad what is HDs ?
It doesn’t mean what you think it means. It’s just a ratio.
ohh alright then thankx
It’s gotten worse this past month.
NoSSD new plotter/farmer now does plots right around 54,000,000,000 bytes (ballpark 50.5 GiB) at max compression for a K32.
They DO need significantly more GPU to mine than than Gigahorse C9 plots though, but can be plotted almost as fast on the same hardware and CAN be plotted with quite a bit LESS RAM than the Gigahorse GPU plotter requires - and no SSD at all, abet at a speed penalty if you don’t have quite a bit of RAM.
Seems like they don’t overlap HD I/O and plotting work though, which slows them down some - but STILL quite fast.
sounds romantic on paper, but they will most likely shares the same fate as FlexPool…going down…then what noSSD is proprietary crap based on our network. Where do you farm afterwards?
With GH, you can at least solo farm…if GH goes down as well, you can also replot 
KISS in the official Chia network, and you will be happy if the project survives long enough.
If NoSSD goes down, you can always replot.
SAME as GH.
GH does have the advantage of not being a specific pool - but the DISADVANTAGE of needing to run a full node 100% of the time to farm. Otherwise seems like it will be AGAIN a tossup between GH and NoSSD (with CNI running WAY behind again!) once Max releases the new version he’s been working on, that’s supposed to be out in the next day or two.
GH wears HDDs faster than any other options I stick with official, while losing about 5% of space…living happily ever after…till some new improvement pops up. My 14TB SAS drive just died…only 3y old.
GH 2.0 may be different, presume you’ll test it?
i did test just to fill the void before Chia releases official plots. Chia C4 replot, 1/3rd already done.
GH, noSSD, they will all end up just like FlexPool or other pools/ideas. Imagine following, Chia gets finally official business - the real businessmen will ask “wtf are those parasites in your network. Have you ever seen AMG or M to sell their own versions of Mercedes or BMW?”
I prefer official, despite it uses more space. Not only because it was a great idea of great mind, but also to support devs. If the whole project goes south, so does any clones of Chia.
