I have a Dell R620 Harvesting my c07 farm. I have been told that I have compressed too much for CPU - but I don’t want to have to re-plot. Prior to Filter Halving I was getting lookup qualities of ~10 seconds.
Post filter halving it doesn’t seem to have changed all that much - screenshot below.
Can someone please explain the implications of leaving it like this?
All my plots are on a SAS controller/expander and connected to a CT with bindmounts.
Note: If I do have to replot - What would be the best compression level? It’s a 32 x Intel(R) Xeon(R) CPU E5-2665 0 @ 2.40GHz (2 Sockets) and I can dedicate almost all resources, but it would still be a CT with bindmounts. Farm is only 664 plots of size: 51.116 TiB on-disk, 65.752 TiBe (effective).
If you are getting a lot of stales, THEN you probably need to replot at least some of your plots to a lower compression level. Probably 1 level less compression is all you would need to do in that case, so C6.
If you are only getting a very few (say less than 1% or so) or NO stales, don’t worry about it.
First, the time to submit your partials (if you are pooling) is around 27 sec (3x 9.375s).
That 8 sec is basically a garbage value that was used for uncompressed plots but is irrelevant for compressed ones (as decompression needs to be added on the top of disk access). With uncompressed plots lookups were roughly about the same, so an average number was good enough; however, with compressed plots instead of depending on an avg value it is better to monitor lookup distribution (how far it goes and how often it happens). So, the fact that chia is spamming the logs is just annoyance, not really a warning or anything like that. I would start worrying if those lookups would start to push 19+ sec (two challenge intervals), as things may start going south if the third interval is hit (or two third intervals in the row - you will be getting GPU congestion from time to time in such case).
Maybe you can check what minimum GPU you can use to do decompression for the existing plots (I assume you have BB plots and I don’t know anything about that plotter). Maybe something like k2200 would do the job (maybe around $50 and will draw 20-40W max). This beats any replotting, if you want to stay with what you have.
Of course, if you want to replot then you will be looking at higher end GPUs, if you want to compress a lot. As for compressed plots the GPU is doing all the work (or rather most), if you want to keep the box you have right now, I would get some low-power chip instead of what you have right now.
CNI is going to switch to (initially) uncompressed plots in about a year or two (those will obsolete all the current plots), so you have plenty of time to decide which way to go.
It’s been a while, so things might have changed, but I think GH is much faster to compute BB plots, so your times will improve, as well as power usage.
It does appear though that you don’t have a real problem, since you’re not getting any stale shares?
In any case, lower power draw is probably worth the 1.5% fee still. Especially since you running v0 Xeons?
Could you fit in that box 90 deg PCIe adapter, so no holes in the case will be needed? Plenty of those on eBay for around $10 (I used one for a USB card and worked great). You could also use a PCIe flex cable, but that one runs over $20 or so (have my k2200 on such cable).
would a GPU not solve your problem? I use a tesla P4 in my dell R610 for all my plotting and farming, at the time they were $90US on ebay. Still good even after both halvings, and I have 6400 plots C7 compress
R620 is a 1U server - not much of a GPU is going to fit in it if any at all, and it might not have GPU power connections at all.
The slots it has are half-length, which is also VERY limiting on possible 1U GPUs.
For example, the A2000 does not need a power connector, IS short enough, but it’s too wide to fit in a single-slot card bay (it’s ballpark 1.5 card width).
FYI I only have a SAS expander PCIe card to connect to 5x12-bay enclosure daisy chained and GPU telsa p4 installed in the R610, and am running UNRAID on it for all my VM and dockers.
P4 is half height, with passive cooling that relies on the server fans for airflow (should not be an issue in a R620).
The LENGTH might be an issue though. NVIDIA Tesla P4 Specs | TechPowerUp GPU Database has several pictures.
@Spazhead - I SERIOUSLY doubt that removing the cover from the server helps cooling on it since it relies on airflow from the SERVER fans to stay cool.
Removing that shroud from P4 may not really lower the GPU temps. With the shroud, the air has only one intake at the end of the card, so all air that will exit through the bracket will go over the whole card.
On the other hand, when shroud is removed, the air will go “lazy” and will take the least resistant route, basically bunching up close to the bracket.
At least that would be my take.
Now, it may as well be that if you remove that shroud and add a 60-80mm fan toward the far end, that can drop temps more (if you have room for such fan), as the air from the fan will strip the heat from the card and disperse it toward the end of the box, letting it escape using different box exits.
One more thing that you could actually try is to use the shroud and add a fan at the far end forcing more air to go over the GPU radiator. The card will have some air resistance due to rather small channels below the shroud so potentially such fan can make a difference (I put such fan on the intake of the PSU, and that quiet down that PSU a lot; actually, after that I drilled 92mm hole in the server top cover right above the intake and added 92mm fan there; this way the PSU is getting ambient air, instead of already server heated air - it made a big difference).
If the server is mounted with other rack mount servers, the “add an external fan through the cover” isn’t an option. Not a bad idea if you DO have the space though.
In a 1u server, he won’t have the space to add a 60mm fan much less an 80 - 40 is the max that will fit in the case. That’s WHY most 1u servers tend to be screamers - 40mm fans turning high RPMs and often with 10+ blades per fan.
He MIGHT be able to mount 2 or 3 40mm fans to the end of the shroud.
