Optimal Xeon Performance Question

I have a couple of idle Cisco UCS B200 M3 Blades I was considering using for mining - they are in a datacenter so the power bill isn't a concern for me.

The specs are Dual Xeon E5-2697 V2 (AVX 12 core / 24 thead) so 48 threads per blade.
256GB 1866 DDR3 ECC

Trying to decide of just a huge bare metal OS would be better, vs a bunch of VM's - or if its even worth it at all.

My initial testing only yielded about 50 sol/s on bare metal windows.
VMware ESX running Linux VM's using 8 vCPU's (free esx limitation) and varying amounts of memory all settle in at about 12 sols. So if I ran 6 of these VM's it would yield about 60-70 sols.

Seems really low.

Any optimization tips, or are these older CPU's lacking AVX2 simply not worth it?

1 Like

sudo apt-get upgrade
sudo apt-get update
sudo apt install qt5-default
sudo apt-get install cmake build-essential libboost-all-dev
git clone https://github.com/sarath-hotspot/nheqminer.git
cd nheqminer/nheqminer
mkdir build
cd build
cmake ..
make

./nheqminer -l us1-zcash.flypool.org:3333 -u taddress.worker -t addnumberofthreads

This is the fastest CPU miner out there. Use screen so they don't stop when you exit putty etc.

i7-4790k with AVX2 doing 28 sol/s
i7-4790k running AVX code (-e 1) doing 25 sol/s

so AVX2 is about 12% faster

cpu is 4core/8thread running with 6 threads (-t 6) cause 75% is fastest for ht
compare to your 2xE5-2697, you have 6x more cores, but speed clock is 1,7 slower, and no avx2, you should have about 86 sol/s... but you have to remeber that zcash using heavy memory bandwitch consumption algorithm, with that many cores using the same bus it is no way to get the same GB/s transfer per core
used https://github.com/nicehash/nheqminer for windows, it runs xenoncat solver for both codes, it is written in assembler, so should be the same fast on linux