Topic: Building Cheap Miners : My "Secret" - page 20. (Read 60237 times)

I am mining again on a Z400 with a single Vega 56 on Cast_XMR v0.9.2. The Vega 56 is averaging 1942 H/s.
http://www.gandalph3000.com

The CPU is a Xeon E5645 hex core processor with the AES instruction.
http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon%20E5645%20-%20AT80614003597AC%20(BX80614E5645).html

You need a processor from the Xeon 5600 family to mine with as those have the AES instruction.
http://www.cpu-world.com/CPUs/Xeon/TYPE-Xeon%205600.html

I am using XMR-Stak V2.3 and running it only on the CPU. I am running 1x threads on cores (2, 4, 6, 8 and 10). Core 0 is reserved for the Cast-XMR program. These settings use 10MB of the 12MB L3 cache. Again the last 2MB L3 is for Cast-XMR as I certainly don't want to throttle the Vega 56 just to get a few more hashes on the CPU.

https://github.com/fireice-uk/xmr-stak

With the above I am averaging 188 H/s on the Xeon E5645.

The top of the line Xeon for the Z400 is the X5687 quad core and it will produce 230 - 250 H/s. I am running my main system, a Dell T5500, with dual Xeon X5687 processors. I am running 1x, 2x, 2x threads on each processor keeping core 0 free on both so that I can use the system while the miner runs full out. I am getting 460 H/s on it.

I will be adding in a second Vega 56 to this system tomorrow and will update the above results.

-------------------------------

Edit: 04/08/2018

Well I did get the 2nd Vega 56 up and running about three days ago. With the Monero XMR hardfork the hashrates did drop slightly and now each Vega 56 gets 1900 H/s. Adding in the CPU hashrate of 188 give me a system hashrate of 3988 H/s.

My current update... still building with cards I got before crazy pricing. Finishing all my builds has taken a back-seat to my day-job & also changing pools & configurations on existing systems (*).

I have stopped buying new hardware completely, though, until things sort out...

But I am still making ~3x my power cost overall in mining so not shutting down any time soon.

(*) -- I am switching to Zergpool with Nemosminer on nVidia cards. On CPUs I am going to Zergpool / YescryptR16. On my AMD cards I am currently doing Neoscrypt on Zergpool but I am messing around with x16r with SGminer to see which I prefer.

I have no idea what the stock graphics cards do... but the stock CPUs are also generally very poor for mining as they don't have AES.

As for using these daily -- they are quite good for daily use / office use. I have bought several for use around my work office.

Any idea what hash rate these z400s in the newegg link in the OP get by themselves (I understand it'll be very low)? Also, would these z400s be solid for keeping wallets on as well as light internet use?

I've looked in the BIOS and did not see anywhere to disable the P410i controller.

When you say "in the firmware" do you mean in the actual array setup?


My box did come with the cable for the optical drive. However looking at the end the power portion will not fit the power connector on a 2.5" SSD. So I went to Fry's and bought the cable you mentioned earlier (Frys.com #6245160, Model #F03-168) but it does not fit at all. The power side of the Fry's cable is much too big. It measures 15mm whereas the HP Optical Cable PN 594765-001 power side measures 10mm.

Of course Fry's had only one in stock and it was a previous return so the cable inside may not even be the correct one.

I also checked the other two DL580's I have and they all have the same red optical connector.

EDIT: Online searches on the HP Optical Drive Cable PN 594765-001 turns up a lot of incorrect cables. This part number 6017B0240701 is also on the cable and does link to the CD Optical Cable. Too bad there are no specs to see what I need exactly.

https://www.techbuyer.com/6017b0240701-hp-proliant-dl980-g7-dvd-rom-cable-71996
http://www.l7inc.us/Search.php?start=105700&MFG=HP

EDIT2: This is the cable I need:

Slimline 13 pin SATA Male to 22 Pin SATA Female Cable Adapter-III
http://www.microsatacables.com/slimline-13-pin-sata-male-to-22-pin-sata-female-cable-adapteriii-adpc22pfwv13pm-iii


EDIT3 - This is even a better option as these cables fit the SATA connector on the DL580 G7. This allow you to directly hook up a 2.5" SSD to the DL580 G7.

(5-pack) 7+15 Pin SATA HDD Extension Cable Data+Power Male to Female, 19" / 50cm
https://www.ebay.com/itm/5-pack-7-15-Pin-SATA-HDD-Extension-Cable-Data-Power-Male-to-Female-19-50cm/232257234671

Note: There is a Internal solid state drive expansion bay inside the DL580 G7. It is located on the right side just behind the power supplies. The above cable connects the 2.5" SSD to the connector below the jumble of cables near the SPI board/VGA Power plugs. See pages 54 and 55 of the HP DL580 G7 Server Users Guide.

Actually I was able to buy 20x E7-8837's for $218 or $10.90 each. So a set of 4 cost me $43.60. I have bought five DL580's and will keep three for mining and sell the other two for profit. HiveOS has a limit of three free rigs.

You are correct in that more modern processors do use lower power and produce about the same hash rate in four core what these 8 cores do. The problem is that is a single processor. If you want two processor then you need a workstation class computer. Go to four processors puts you into server class systems. And modern versions of workstations and servers are magnitudes higher in price that of the HP DL580 servers that I and others here have purchased.

Why I purchased three DL580's for mining is that each has 11 PCIe slots, 4x 1200 watt power supplies (the server balances the load for efficiency and redundancy). The added hashing (1650 H/s on Cryptonight) using the 4x E7-8837 processors is a bonus.

So 4X E7-8837 at 130 watts each is around 520 watts, correct?  Maybe not too much to worry about if electricity is cheap or free but I just wonder if there are still better price-to-hashrate-to-power options.  I mean $20 for the 8837 is really awesome but then there are strong processors at 65 Watts instead of 130 Watts that are more expensive but might pay for themselves in power costs over the course of a year.

I've been trying to decide on a configuration for my next mining build that would include stronger CPU hash capability.

Thank you for your contributions here.  I really enjoy this thread.

I am also not going to be adding any more systems. I plan to keep three of the five DL580's and sell two of them to recover their initial cost and make some profit on those two. The two for sale will have 128gb ram, 4x E7-8837 processors and at least six 146GB SAS drives.

I plan to also sell the two R815's for the same reason. My original Dell T3610 systems will also be for sale along with the SuperMicro 24 Bay 2026TT-DLRF 2x X8DTT-HEF+ I picked up reacently.

The three DL580's and the Supermicro CSE-745 Dual Opteron 6348 will be kept and fully loaded with GPU's for mining.

The reason I am keeping the systems mentioned above is the high CPU hash rates and the boatloads of PCIe slots along with excellent power supplies.

All other systems will be sold to recoup initial cost and maybe even some profit.


Thanks. That is what I will do. The cheap 8gb usb stick that I am running HiveOS on just crapped out last night. I had to reload HiveOS on another stick to get the system up and running again. The cool thing about HiveOS is all I had to do to be up and running again was recopy the original HiveOS download to the usb stick, boot system from usb stick and then enter my rig number and password. I will be using HiveOS on my other two DL580's.

You can turn of the P410i controller in the firmware and it will decrease boot time.

You technically don't need the micro SATA adapter.  If your box came w/ the optical drive and cable, you can actually cut the connector in half to separate the data from the power.  That was my solution on box #1 due to THIS BOX MUST RUN NOW syndrome...

I've stopped spending money at this moment in time.  I'm pushing harder to learn trading.  More reward for less effort.  I have plenty of chassis' to add GPUs so I will continue to do that but no more CPU rigs for me at this time. 

Eventually all my rigs will turn to speculation only.

Well I'm going to be mining on Vega 56's again.

In early February I sold my two Vega 56's on eBay for $1575 (which I got bought new at MSRP $399.99 each). My total purchase price was $810 with shipping. After eBay and PayPal fees, shipping and insurance costs I cleared $1410 from the sale or a profit of $600.

With the price of XMR now at $166 a lot of my miners barely pay for the power they use. I believe that is why I see prices on eBay falling on Vega 56's.

So today I bought three Vega 56's for $1604 and got $100 eBay bucks in doing so (that I can spend next month).

Taking the $1604 and subtracting the $600 in profit results in a real cost of $1004 for three Vega 56's. If I count that $100 eBay bucks against the purchase then the cost is $904 or $301.33 each.

These Vega's were used for mining but the seller stated that he kept the fans running high and the AC low. These cards also still carry the full AMD warranty.

At least using these Vega's mining XMR will still be profitable on the systems they are in.

Hello,

I want to try the same system > Hp Proliant DL385 with two 6234 CPU + 4 or 6 GPUs

Do you need a usb splitter for install 4 GPUs? Because it DL385 have to 6 pcie ports.

Do you have any recomedation for start, config and run it Rig?

Thanks

Maybe you have one or more bad card. I have in my rig one RX 480 that I have underclocking because it freeze and reboot all rig.

Or maybe you need underclocking some card then you can run smoot.

 

That might save more power not having the SAS drives and backplane being powered.

With the SAS backplane power connector being disconnected did you run into any issues from the BIOS complaining when powering up?

I have some low end 2.5" SSD's that I have laying around so I think I will go the route you suggested earlier and run it off the DVD connector with the Fry's adapter.

You can also use regular 8 pin modular PSU cables (or even cheaper breakout board harnesses, I got 50 for $89 shipped) as well if you have them laying around, you just have to swap two pins.  They still lock, you just leave the two 3.3v pins unused.  I have one box on 2 internal PSU's that is running 3x1080 and a 1060 6gb all off the 4 internal power ports (repurposed SAS connector as well) and crunching 1650H/s on all 4 cpu's.  2 of the cards are installed inside and two of them are on risers.

One GPU power header on the 580 can drive TWO 1060s.

If you want to run a regular HDD but eliminate the SAS array, just use a 12v to 5v converter on one of the 12v lines from the SAS connector so you can power a regular drive.

For those who have HP DL580 G7 systems and need 6-pin GPU power cables the following seller has 10-pin to dual 6-pin GPU cables that fit the power distribution board inside the HP DL580 G7.

https://www.ebay.com/itm/10pin-to-6-6pin-Power-Adapter-Cable-for-HP-ProLiant-DL580-G7-and-GPU-50cm/141964325336

The seller is asking $10 + $4 for shipping one. I picked up 10 for a total of $83.50 by doing the Best Offer of $7.50 for quantity 10. Seller accepted. Shipping for 10 was $8.50.

EDIT: Additional DL580 G7 GPU Power Cables from this same supplier: Besides the 6+6 they have 8+8 and 8+6

https://www.ebay.com/sch/m.html?_odkw=&_ssn=computer-part-world&item=141964325336&_osacat=0&ssPageName=STRK%3AMEBIDX%3AIT&_from=R40&_trksid=p2046732.m570.l1311.R1.TR11.TRC1.A0.H0.XDL580.TRS0&_nkw=dl580+g7&_sacat=0

So what is the hash rate you are getting for the 4x E7-8837's?

Also on the DL580 G7 I removed four of the memory cartridges and now only run with one cartridge per processor. I also changed the two 4GB PC3-10600R memory to two 2GB PC3-8500R memory in each of those four cartridges.

The system now has 16GB total memory and generates the same hash rate. However the system now uses 120 watts less. So that means that each memory cartridge uses 30 watts.

I never had a problem w/ linux and XMR-Stak w/ the 8837's.  The default auto config for me worked right out the box.  I do think hwloc ends up putting the double memory ones up first though.

I've had a hell of a time with XMRig and that's because it doesn't support NUMA.

Thanks for the link. Your settings and mine were the same so after further investigation I found what the major problem that was causing the affinity errors and lower than expected hash rates.

Hive OS (and probably Linux in general) assigns 10 cores to each processor from the E7-4800 and E7-8800 family. So Processor #0 has cores 0 - 9, Processor #1 has Cores 10 - 19, Processor #2 has Cores 20 - 29, and Processor #3 has Cores 10 - 39. I was very confused in why mpstat -P ALL showed 40 cores even though the 4x E7-4830 (with HT disabled) only had 32 (4x 8 cores). It turns out that Linux disables two of the cores in each set of 10 for the 8-core processors. The two cores disabled are not at the end but core #4 & #5 from each group. If you have 6-core processors two more cores would be disabled.

So your configuration from the link above and what I was using would try to use non-existent cores (which caused the affinity errors) and would run more threads on some processors than what we thought and less on other processors. That is why I saw hashes of 11.5 instead of 33.5 that I was expecting.

This is the correct CPU threads table for 8-core E7-4800's and E7-8800's (including the E7-8837):

"cpu_threads_conf" :
   [
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 1  },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 2  },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 3  },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 6  },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 7  },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 8  },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 9  },

        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 10 },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 11 },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 12 },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 13 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 16 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 17 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 18 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 19 },

        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 20 },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 21 },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 22 },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 23 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 26 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 27 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 28 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 29 },

        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 30 },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 31 },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 32 },
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 33 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 36 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 37 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 38 },
        { "low_power_mode" : true,  "no_prefetch" : true, "affine_to_cpu" : 39 },
   ],

Using this the "affinity errors" went away and my overall CPU hash rate went from 1038 H/s to 1358 Hs for the processors. A 31% increase with no additional power used.

With 4x GTX 750's I am now hashing at 2250 H/s.

NOTE: the above configuration does not use Processor #0 Core #0 because I have that affined to from the Nvidia configuration file. If you have no GPU's then you can add this:
        { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 0  },
to the top of the config file for extra hashes.

--------------------------

EDIT: 04/08/2018

Well the above worked on Rig #1 but when I brought up Rig #2 I again had affinity errors on 4 of the cores.

Can I say Linux really really sucks because it really really does.

Even though both Rigs are EXACTLY the same Linux disables different 2 cores out of the 10 that it uses on each of the 8 physical cores of the E7-8837.

On Rig #1 of the 40 cores Linux defines it disables core # 4, 5 on Processor #0, cores #14, 15 on Processor #1, cores #24, 25 on Processor #2 and cores #34, 35 On Processor #3

But on Rig #2 it does this: disables core # 1, 8 on Processor #0, cores #11, 18 on Processor #1, cores #24, 25 on Processor #2 and cores #34, 35 On Processor #3

That means that the custom XMR CPU configuration that I set in HiveOS cannot be used because what worked on Rig #1 fails on Rig #2 and more that likely on my future Rig #3.

So those who are using HiveOS and XMR-Stak do not use any overrides in the XMR CPU configuration in HiveOS just let XMR-Stak auto-generate it.

EDIT: 04/26/2018

The above issues of dummy cores has been solved with HiveOS going to Linux Kernel 4.13. mpstat -P ALL now shows the correct 32 cores.

Thanks!



Reason i left hiveOS was that i never got it stable. Constant freeze and reboots, but it has a lot more features, on paper its far superior to SMOS.

I dont have the configs left i think, but please have a look at my post here on settings for E7-8837 with xmr-stak: https://bitcointalksearch.org/topic/m.28646698

The rack inside the server cabinet is custom built, bought the alu-profiles from my local hardware shop for ~75 euro, a bit expensive but very easy to build. I will look for a drawing, i only made one on paper

Any idea how to get 11 gpus working? I seriously wasted like all nights in a week to try to solve this. Changing hardware and waiting for multiple 15 min reboots is not something i like to do again.

Nice system.

I am trying Hive OS on my DL580 G7 with 4x E7-4830's and 4x GTX 750. It is up and running but something must be wrong with my configuration for the CPUs. I am getting some affinity errors on a few CPU cores and those cores only produce 11 H/s instead of the 33 H/s on cores where affinity works. I have disabled hyper-threading in the BIOS so the 32 cores shown are real CPU cores.

Could you post the CPU configuration that you used with Hive OS. Hopefully that will fix my problem.

Also is the rack that the DL580 and GPUs are mounted in a commercial rack or home made?

If home made can you provide assembly instructions.

Thanks