Why some CPU's half as fast?

I have 30 identical desktops but two of them insist on having 8 S/s instead of 17. "top" looks the same as the others, hard drives seem OK, RAM swaps and checks were OK. I couldn't see anything in the log files, but I wouldn't know much about that. Any suggestions on how to diagnose?

BTW I can't just sell them and get GPUs because they haven't dropped in price for 6 months, and the electrical cost is only 3 W per S/s. R9 390's at 100 S/s doesn't bother me but 280x at 100 S/s I can only say ouch.

Is this drop on the pool or software side?

It's on my side.....

Thermal throttling maybe ?

Can't find it right now, and apologies your checks already covered this, but I'm almost sure I read a thread the other day where this problem was solved by getting the ram in dual channel mode. Not sure what you mean by 'top' looking the same, but if it means your ram is in the same slots as others, try the second dual channel configuration (often it's a1-b1 or a2-b2, consult your mobo documentation) and/or check bios settings to see if dual has to be enabled there.

RAM speed is good. I've swapped it out and everything. "top" is crucial for linux as it shows the processes.

CPU temps were 50 C, so it's not busy. I need to learn how to see the HDD accesses by proc number and see if those USB error messages I saw at boot are interrupting the processor a lot. I can turn them off in UEFI, but it seems too coincidental to see the same problem and speed in 2 of them.

Identical Desktops: two giving 8 s/s and 28 of them giving 17 S/s

All show:
* 100% CPU usage w/ miner
* same RAM usage
* Can swap RAM, no difference
* "top" shows same processes running
* lshw shows RAM exactly same after RAM is swapped (it's detected)

Differences:
* 50 C vs 70 C core temps.
* Read & write to virtual RAM is half as fast in slow machines.

Code to test speed of RAM:

#!/bin/bash
# Shows speed of CPU RAM.  Typically 3 and 6 GB/s on DDR3 1600 (17 GB/s) due to HDD/CPU
sudo mount tmpfs -t tmpfs RAM_test/
cd RAM_test
c=1
until [ $c -gt 20 ]; do
echo "writes"
dd if=/dev/zero of=data_tmp bs=1M count=512 2>&1 | cut -f4 -d, -s
dd if=/dev/zero of=data_tmp bs=1M count=512 2>&1 | cut -f4 -d, -s
dd if=/dev/zero of=data_tmp bs=1M count=512 2>&1 | cut -f4 -d, -s
echo "reads"
dd if=data_tmp of=/dev/null bs=1M count=512 2>&1 | cut -f4 -d, -s
dd if=data_tmp of=/dev/null bs=1M count=512 2>&1 | cut -f4 -d, -s
dd if=data_tmp of=/dev/null bs=1M count=512 2>&1 | cut -f4 -d, -s
let c=c+1
done
cd ..

Are the motherboards the same ?
Mobo VRM temps ?

Yes, they are identical. Is there a name for VRM in sensors data? I can only get 2 other temps out of it after running sensors-detect and after installing sensors-applet I still can't get any voltages. I don't have physical access to them while it is on.

For CPU mining, is around 2 S/s per core the best performance out there right now (as far as we know)?

No, 10 S/s per core on new, fast desktops. 30 S/s on new fast laptops. 20 S/s per $200 used desktop from ebay. All using nheqminer with xenoncat, sending it to a pool, who sends your payout to your exchange ZEC address. I have 43 desktops on a shelving rack giving 800 S/s. Probably $100 in electricity giving 2 ZEC, about the same as GPUs, but in the next few days they should pull ahead of me.

I've been following your posts since launch because I was intrigued by your setup. Are you going to keep mining when GPU's will pull ahead of you? I'm just a small hobby miner and I'm using 4 core I5 2500k desktops atm. We're living small and the girlfriend doesn't want open rigs with 6 GPU's each in the house and no high electricity bills so I'm limited in what I can do. Each processor gives me 15 sols/s, so very modest. 3 of them have pci-e 3.0 onboard though so I'm debating which GPU's to add to the little miners or none at all.

Thank you so much @zawy for all of your great posts. What are the technical reasons behind the new laptops being 3 times faster than the new desktops?

I'd max the desktops, which still is going to be far from silent. 280x for $100 or 470 for $200 at half the 280x electricity seem to be the choices. Half the electricity means it's going to be more quiet per S/s

1 Like

New desktops with 4 cards of DDR4 should be about 40 S/s on xenoncat. There are 2 laptops mentioned in the Intel/AMD CPU speed thread. One was 29 and the other was 32. 2012 CPUs are 18 S/s.

Having the same issue with some of my servers. Baffled.

Did you ever get this solved?

30 desktops you say?

How big do you want to go?

The good GPU options are now 1/3 the initial costs and the 2/3 the electricity. I'm 3 watt per Sol/s and $8 per S/s. I'm definitely not adding more and will probably sell them as soon as I have 2000 S/s of GPUs. I've got 44 working now that give me 800 S/s. I'm just curious as to these two giving half. It's certainly not worth the time to try to fix them, losing time on figuring out how to get GPUs in 10 or more of them.

1 Like

Looks like the same issue I've had: https://forum.zcashcommunity.com/t/solved-2-identical-cpus-i7-6700-gives-different-results/7568/17
you can check/compare # of RAM channels with following command:

dmidecode -t 17 | awk 'BEGIN { FS=":"; OFS="\t" } /Size|Channel/ { line = (line ? line OFS : "") $2 } /^$/ { print line; line="RAM" }' | grep -iv 'no'

In my case it was

RAM 16384 MB ChannelA-DIMM0
RAM 16384 MB ChannelA-DIMM1
RAM none ChannelB-DIMM0
RAM none ChannelB-DIMM1

Instead of

RAM 16384 MB ChannelA-DIMM0
RAM none ChannelA-DIMM1
RAM 16384 MB ChannelB-DIMM0
RAM none ChannelB-DIMM1

I have only two slots in these motherboards. The results of sudo dmidecode -t 17
were identical (excepting manufacturer's serial number of the RAM)