I’ll share plotter logs and all the system info once it starts happening again. I have restarted the system and it seems to be working fine… For now.
1 Like
What is known to cause gradual slowdowns over time:
- RAM overheating
- SSDs running out of trimmed space
- RAM fragmentation (too little free RAM)
3 Likes
GUYS WHO USE 64 GB RAM - TELL ME YOUR EXPERIENCE PLEASE! I WAS TRYING FEW MONTHS AGO AND IT WAS SLOW - A PLOT LIKE IN 30 MINS. please tell how are you doing.
I noticed the swap space was getting full.
I’ve removed swap and seems to be running okay now… More time will tell 
2 Likes
I’m getting 30 min as well on my 64G desktop. It’s because the SSD cannot read/write any faster. There’s a reason why we say RAM is king.
2 Likes
Gigahorse C8 - 64GB RAM quad channel DDR4 2133MHz, GTX 1060 6GB, 2x NVMe Transcend 220S 1TB - RAID0 > 15 minutes; with one NVMe 26 minutes. With 192GB RAM, RAID0 > 6 minutes.
2 Likes
Hey madMax43v3r, you on the official discord now as well?
Any further developments in the pipeline?
Yeah, keeping up with Chia updates / soft forks. (I have to review all the code changes, to make sure there is no backdoor being added, it’s quite some work.)
Gonna focus on MMX again.
2 Likes
Please keep working on keeping your plotting/decompression tools competitive.
Not much left to optimize now, unless there is new ideas.
Which single GPU can I use for 5 PB C8 k32 farming until plot filter 256? One single 3090 is enough?
And if I need to replot to K33 in the future which C level should I use for Same 5 PB ?
told ya … k33/c8 gonna be a king
and 128 gb - how faster it should be?
but why K33? less GPu needed for checks?
980 pros 1 TB - total 2 Tb?
Gonna need two 3080 for 5 PB C8 at filter 256. Same for K33.
Should be around 3-4x faster than 64G mode.
Thanks for answer.
So, one 3090 will be not enough?
Another question: Let me make some description of my system first. Now my system is working with standard uncompressed plots around 500 disks mixed size ,9 harvester and two farmers for now. (One is the old farmer that’s working with old plots for now)
I make one new farmer separated with one 3090 (for now) and I am plotting with one plotter with 2 x 3090, 256 Ram and 2 nvme plotting time around 84 sec /c8 for one plot! My network is 1Gb but for copy plots from plotter i use 10GB network Pcie cards and for example plots on 20 disk on additional harvester(each harvester around 55-60 disk) and over 10GB network i am sending and receiving via plot sink to those 20 disks, the rest of 35 disks are harvesting over 1GB network but also the new harvester with Giga horse farmer is farming with brand new plots after it was saved to deleted disks(only thing i should do is in new farmer config must have the new destination folders and automatically start to farm after it is saved on disk) !
Before I run my old system in farmer harvester mode after i change my network for 100MBit to 1Gbit i try other way i share all the disks and in farmer config i put the shred destination folders and it was working well (have some little problems sometimes but not to often I have 1 stale in 3-4 days)
My question is how my compressed plots number will rise does 1Gb network can handle this system this way?
Sorry for long description but i want to clarify only for clear picture!
The point I do this way is I want to minimalize the lost until replotting my disks with compressed plots.
There was someone on discord saying they got 6min c8 plot with 64GB and a 3090 + 980pro
which would be quite impressive (if true).
My results with one single 3090 and 128 ram is 3,9 min and 256 ram with two 3090 is around 81-84 sec
Is 1 Gbit Ethernet enough for 5 PB C7 farming? I want to set up my farming machine as a separate one.
And does plotting and farming simultaneously impair plotting performance? E.g. should I plot and farm on different machines so that plotting is achieved at a maximum possible speed?