Topic: Sixth alt coin thread I forgot to mod last thread. - page 45. (Read 80193 times)

my generic k17  seems good   does 7  1070ti's and 1 1070   all  evga hybrid  water cooled smos about 1000 watts  using a delta 2400 watt server

 

Very cool - definitely interested to find out as well!

Office space or warehouse? Also curious how are you ventilating the space?

It will be in the farm in a few hours - the farm will be the real test if this guy survives.

The farm runs 34-35c on normal days and can get higher during hotter days.

The miner in question is the black box behind the 8 x R7-370 on a "k17" China mobo.

Wow, that's a really nice deal. How are the temperatures with those 8 GPUs enclosed like that? Is it super loud in order to move enough air through it to not bake? Is the PSU any good (both for effieciency and reliability)?  Having a 8 GPU plug and play system for that price is very interesting.

Thanks Mashy81 -- thats encouraging news. Hows the power consumption like - I want to try 8 x Vega64 if the temps are manageable since these are reference blower style GPUs.

I am running the onda 1800 with 4 vega 56 atm. 64g ssd (cheap $25 one from ali) win 10 4g ram and it been running sweet for a week. I plan to try an fill the extra 2 slots. I am running 36g virtual memory.
I have put them all next to each other in the slots and maximum temp has been 65. With no extra fans and my ambient temp is 30 C.
7600hs stable.

I like the idea of a NAS box for swap.    Assuming your network is fast enough, I am wondering on how many miners you could run on this type of scheme.....    I would assume a fiber link to this share would be ideal;  maybe even a separate fiber network just for the swap shares.....


I have these massive old 4U SCSI raids I wanna do something with eventually... maybe ill pop them on a miner and turn it into solely swap space just to hear what it's 'song' sounds like.......   The geek in me really coming out in this one....    They are massive and heavy... and I have two of them.    They need two full SCSI channels per unit to interface.   I don't have any PCI interfaces with 2x connectors...... yet.......

My group got the first batch A4s and had many sleepless nights and nightmares - took months to fix before it ran smoothly.

Hope the A4+ has better firmware this time round.

 Writing to a SSD eventually kills the memory cells - they have a limit on how many times they CAN be written to, which varies depending on the drive and variations on each individual cell and most drive "manage" writes to even them out among all the cells in the drive, which tends to cause lots of the drive to fail at once when they DO start dying, as opposed to concentrating the writes on certain cells and having a gradual deterioration of drive space (there are good arguments on both sides of that debate though).

 Mining generally doesn't do a lot of writing to the drive (ETH in the days of DAG file creation on the drive was an exception) so it should be much less of an issue in a mining rig - though Windows usage would tend to increase the issue a little.

I think the cast miner works within Win10 virtual memory parameters and would be doing DAG type read-writes I bet.

If you are into fast video depository -- using SSDs will be expensive approach. You should consider a NAS with RAID-5 or RAID 1/0 type configuration. The HDDs like WD-Black on RAID is just as fast and cheaper to setup in a NAS box.

 If they didn't have that "minimum 3 units" order requirement, I'd likely have some A4 units by now.

My concern is the large file writes.

See;  I have killed 4 SSD's last year by continually shuffling large files from my gopro to my laptop (256Gb SSD), then when home; laptop to server.   Sometimes its downloading a 1-3Gb torrent, but I always preallocate space as to relieve strain on the file tables....   But with SSD's over time of handling very large files, you will notice the machine having pauses with a seemingly solid HDD light, but it will always recover... then one day it will happen and never come back...  Just dead to system drives now.  When it pauses, its like a processing lag glitch.... you don't pick up on it until its already very serious and it takes 15+seconds or more to continue doing what it is doing.

This behavior has happened on business class drives (Apacer), user class drives such as sanforce chipped units.... and in expensive samsung 840EVO black series drives.    I have a bunch of them I need to try and hack into the serial port on them and see if I can recover the data.....   I lost a lot of important files more than once....     I dont want to have to remove the flash chips, read them in a flasher off the board, then reassemble the image on a pc and recover....     It's a LOT of work, and the NAND readers aren't exactly cheap themselves......

... the rig is all powered by the Delta 2400w PSU -- so I am utilizing everything onboard the motherboard. If I use a normal SSD, I can - but... I would need to a 6pin to sata for the SSD. With the M2 -- no other power needed.

The Vega64 XMR 2000h/s per GPU guide that I am studying at the moment indicates that up to 40-50GB of virtual memory is needed by that cast-amd-xmr miner to run smoothly. The registry hack method is also needed to lower the power consumption and bump up the hash speed in excess of 2000h/s per GPU -- which is amazing -- well we will see.

@VyperBTC
some more rig-porn for you... these two will go into production tomorrow - they ran 24 hours non stop with smOS EWBF and sgminer-gm flawlessly.

K17 Generic China motherboard 8-slot board, Celeron and 4gb DDR4 RAM with 8 x R7-370s running Cryptonite/sgminer-gm-v5.5.5.8-NH

Ready made rig like Pandaminer (8 x P106-100 on B250 Colorful motherboard) -- one of my group members bought 1 x unit at 2264$USD which is hard to believe for a fully functional system (Celeron, 4GB, 64GB M2-Sata, 1600w ATX PSU). If I DIY the whole system -- I think parts alone and 8 x P106s will exceed 2264$ !

When it arrived, we tested it and it booted up with Windows 10 in Chinese ! Anyways - we took out the M2 and run smOS via USB3 stick - ran EWBF Equihash at NH for 24 hours without any issues.

https://www.aliexpress.com/item/Cheap-Mining-Rig-Machine-Antminer-8-NP106-Cards-1600W-ETH-185MH-s-ZEC-2500-Blackchain-Win10/32836106822.html?spm=a2g0s.9042311.0.0.m0cV3A

At what point is the SSD going to be detrimentally effected by the swap space?   If its a static DAG file... I could see it being ok.... but, that's a lot of read/writes on a SSD if its written a bunch.....


I know a 128Gb SSD is probably cheaper than a 64Gb SODimm would ever be...... but the thought did cross my mind.

Loving all your test rigs, beautiful beautiful things my man!

2 x vega64 + 2 vega56 Cryptonite rig testing in progress using the Colorful B250 8-slot motherboard with WIN10, LGA1511 Celeron, 8GB DDR4, 64GB M2-Sata SSD. This board has 8 x 6pin input to power the mobo and 8 x 6pin output to GPUs.

Notes:
1. May need to change that M2 to 120GB because cast-amd-xmr miner seems to need a lot of virtual memory to  run properly.
2. I have 5 more Vega64s still unboxed, final config maybe 2 rigs with 4 x Vega64s and 1 x Vegas56 = 5 x GPU rig - the D1800 maybe able to do the job.
3. If the overall power can be contained around 1400w-1600w and good temps with 8 x GPU - I will do a full blown 8 x Vega64 rig with the Colorful board. Spacing will be tight!

I would like to test the 9 gpus board server style, but it's not in stock at the moment..

@Storx

I had the Code 12 error on my one ASROCK motherboard.  I had to do one of these things, I don't recall exactly which.  I can check MB settings when I get home from work if you would like me to.

1) Go into the bios try to find, TOLUD "Top Of Lower Usable Dram", set it to 3.5G
2) Bios > try set pci-e in your bios to GEN 1
3) Bios > "above 4g decoding" to enable

I am 90% sure the setting that fixed my problem was the TOLUD setting it to 3.5G.
I am 85% sure the setting you need tweaked is in bios.

Hope this helps.

Baikal is one of the better Asics out there - had a wonderful and profitable experience with them since their 150MH miner.

Unfortunately, Baikal has sold out the X10s - its a mean, low power, great quality and powerful ASIC miner supporting X11/Quark/Qubit/Myriad-Groestl/Skein ... a few new algos in future.

Buying from resellers will cost you almost 3x up to US$4k per miner -- too expensive - don't waste your money.

I missed the pre-sale on this one, although I was in the Baikal early buyer list earlier this year.

I think Storx managed to get a few units.

very true 

has anyone tried this vendor?

https://www.alibaba.com/product-detail/New-Baikal-Giant-X10-ATX-Power_60696776433.html?spm=a2700.7724838.2017115.1.71174b8ZKmvKq