Topic: Xeon Phi Coprocessors (x100 & x200) CryptoNight V7 CPU - Need CPU Miner Support (Read 799 times)

bump

Currently setting up 5110p on a supermicro board, kernel version need to be compatible with the phi / intel's MPSS, so currently installing centos 7, possibly opensuse 42.3, have yet trialled neither but will endaevour to report findings here. Surprised hash rate on the x100 reported so low considering they are not far off on core count in comparison with the x200 cards, there are 4 threads per core on around 60 cores, so approx 240threads per card. Theory being once configured the cards should work right off the bat provided the mining software picks up the threads (cmd lscpu should count the cores on the phi... i think).

Any software used on the phi needs to be either cross compiled or natively compiled so that it is compatible with mic architecture (linux based, x86-64), from what I have gathered applications with a high input/output are not suitable for native compilation. Intel provides opensource dependencies by way of it's MPSS (many process software stack) that can be added to the stack in order to compile software through GCC, Autotools etc. Results during compilation are variable, source code is generally developed with a debian based build in mind, some dependencies are therefore not available for building with the MPSS stack which may be causing complications. Generally encountering C++ compilation errors (perhaps due to inexperience  in software development), intel does however offer base API and HPC toolkit to community, not necessarily for commercial use.

From what I have read XEON Phi was phased out mainly due to security concerns, so for server applications the hardware is now purported to be relatively useless; however, for HPC it would seem they are well suited for application.

Update:

Prior to compilation miners require the following dependencies to build from source. In order to cross compile these dependencies would therefore probably need to be compiled and then installed in the environment prior to cross compilation of the actual miner, going to attempt at working on each.

libjansson-dev
libcurl4-openssl-dev
libssl-dev
zlib1g-dev
pthreads
zlib

In order to achieve native compilation, mpss toolchain needs installation which is dependant on the following dependencies (in order):

rpm -ihv libperl5-5.14.2-r7.k1om.rpm
rpm -ihv perl-5.14.2-r7.k1om.rpm
rpm -ihv m4-1.4.16-r2.k1om.rpm
rpm -ihv libgdbm4-1.10-r3.k1om.rpm
perlmodules - all

rpm -ihv autoconf-2.69-r7.k1om.rpm
rpm -ihv gnu-config-20111111-r1.k1om.rpm

install autoconf and gnu-config w/ rpm -ivh *.rpm  in different directory

rpm -ihv autoconf-doc-2.69-r7.k1om.rpm
rpm -ihv automake-1.11.2-r3.k1om.rpm
rpm -ihv automake-doc-1.11.2-r3.k1om.rpm
rpm -ihv binutils-2.22+mpss3.8.6-1.k1om.rpm
rpm -ihv binutils-symlinks-2.22+mpss3.8.6-1.k1om.rpm
rpm -ihv bison-2.5-r1.k1om.rpm
rpm -ihv byacc-20120115-r0.k1om.rpm

rpm -ihv libarchive13-3.2.1-r0.k1om.rpm
rpm -ihv cmake-2.8.7-r2.3.k1om.rpm

rpm -ihv cpp-5.1.1+mpss3.8.6-1.k1om.rpm
rpm -ihv cpp-symlinks-5.1.1+mpss3.8.6-1.k1om.rpm
rpm -ihv flex-2.5.35-r3.k1om.rpm
rpm -ihv g++-5.1.1+mpss3.8.6-1.k1om.rpm
rpm -ihv g++-symlinks-5.1.1+mpss3.8.6-1.k1om.rpm

rpm -ihv libmpfr4-3.1.0-r1.k1om.rpm

rpm -ihv libmpc2-0.8.2-r0.k1om.rpm
rpm -ihv gcc-5.1.1+mpss3.8.6-1.k1om.rpm
rpm -ihv gccmakedep-1.0.2-r3.k1om.rpm
rpm -ihv gcc-symlinks-5.1.1+mpss3.8.6-1.k1om.rpm

rpm -ihv gcov-5.1.1+mpss3.8.6-1.k1om.rpm
rpm -ihv gcov-symlinks-5.1.1+mpss3.8.6-1.k1om.rpm
rpm -ihv gperf-3.0.4-r0.k1om.rpm
rpm -ihv libtool-2.4.2-r2.0.k1om.rpm
rpm -ihv libtool-dbg-2.4.2-r2.0.k1om.rpm
rpm -ihv libtool-dev-2.4.2-r2.0.k1om.rpm
rpm -ihv libtool-doc-2.4.2-r2.0.k1om.rpm
rpm -ihv makedepend-1.0.3-r0.k1om.rpm
rpm -ihv patch-2.6.1-r0.k1om.rpm
rpm -ihv pkgconfig-0.25-r3.k1om.rpm

rpm -ihv make-3.82-r2.k1om.rpm

I added the following for attempts at native compilation (with cryptozeny's opt and sugarmaker)

rpm -ihv libcidn1-2.14.1+mpss3.8.6-1.k1om.rpm
rpm -ihv linux-libc-headers-dev-2.6.38+mpss3.8.6-1.k1om.rpm
rpm -ihv glibc-extra-nss-2.14.1+mpss3.8.6-1.k1om.rpm
rpm -ihv libc6-dev-2.14.1+mpss3.8.6-1.k1om.rpm

rpm -ihv pkgconfig-dbg-0.25-r3.k1om.rpm
rpm -ihv pkgconfig-doc-0.25-r3.k1om.rpm
rpm -ihv glibc-doc-2.14.1+mpss3.8.6-1.k1om.rpm
rpm -ihv glibc-utils-2.14.1+mpss3.8.6-1.k1om.rpm
rpm -ihv libc6-dbg-2.14.1+mpss3.8.6-1.k1om.rpm

rpm -ihv automake-doc-1.11.2-r3.k1om.rpm
rpm -ihv libgmp10-5.0.4-r0.k1om.rpm   -
rpm -ihv libgmp-dbg-5.0.4-r0.k1om.rpm
rpm -ihv libgmpxx4-5.0.4-r0.k1om.rpm
rpm -ihv libgmp-dev-5.0.4-r0.k1om.rpm
rpm -ihv libgmp-doc-5.0.4-r0.k1om.rpm
rpm -ihv libgmp-staticdev-5.0.4-r0.k1om.rpm

rpm -ihv libcurl4-7.58.0-r0.k1om.rpm  -
rpm -ihv libcurl-dev-7.58.0-r0.k1om.rpm
rpm -ihv libcurl-doc-7.58.0-r0.k1om.rpm

rpm -ihv libssl1.0.0-1.0.2j-r0.k1om.rpm   -

Generally receiving c++ compilation errors when building the miners from source, I have log files if anyone interested. I think I may have missed a step on cross compilation so focusing my efforts there.

how do you do it cheers

Aw, shucks. I was hopeful. Thank you.

Price per hash per watt its getting tight with difficulties increasing.

If I'm correct, its performance is similar to Radeon Vega GPUs, so probably not worth it

708360-001 HP Intel Xeon PHI 5110P Hexaconta 60-Core 1.053GHz 30MB L2 Cache 22NM 225W Coprocessor
under $300 - is this a good deal?

Asrock made a 4x server specifically for crypto mining with 4 phi chips.  Spec was around 12kh on CN Heavy for the unit, but for $6000, gpus made more sense.
https://www.asrockrack.com/general/productdetail.asp?Model=2U4N-F/X202#Specifications

What do you mean by " Asrock 4x system"?

Hi

I'm using Xeon phi Co-processor.

I'm currently mining various currencies using cpuminer-opt on Xeon phi. At this time it seems that only coins of argon2d correspond to AVX512.

It supports the following algorithm.

allium      SSE2 AES SSE4.2 AVX2.
anime      SSE2 AES AVX2.
argon2      SSE2 AVX2.
argon2d250   SSE2 AVX2 AVX512.
argon2d-crds   SSE2 AVX2 AVX512.
argon2d500   SSE2 AVX2 AVX512.
argon2d-dyn   SSE2 AVX2 AVX512.
argon2d4096   SSE2 AVX2 AVX512.
axiom      None.
bastion      SSE2 AES.
blake      AVX2.
blakecoin      SSE4.2 AVX2.
blake2s      SSE4.2 AVX2.
bmw         error
c11         SSE2 AES AVX2.
cryptolight      SSE2 AES.
cryptonight   SSE2 AES.
cryptonightv7   SSE2 AES.
decred      AVX2.
deep         SSE2 AES AVX2.
dmd-gr      SSE2 AES.
drop         None.
fresh         None.
groestl      SSE2 AES.
heavy      None.
hmq1725      SSE2 AES AVX2 SHA.
hodl         AES SSE4.2 AVX2.
jha         SSE2 AES AVX2.
keccak      AVX2.
keccakc      AVX2.
lbry         AVX2 SHA.
luffa         None.
lyra2h      SSE4.2 AVX2.
lyra2re      SSE2 AES SSE4.2 AVX2.
lyra2rev2      SSE2 AES SSE4.2 AVX2.
lyra2z      SSE4.2 AVX2.
lyra2z330      SSE4.2 AVX2.
m7m         SHA.
myr-gr      AES AVX2.
neoscrypt      SSE2.
nist5         SSE2 AES AVX2.
pentablake   AVX2.
phi1612      SSE2 AES AVX2.
pluck      None.
polytimos      SSE2 AES AVX2.
quark      SSE2 AES AVX2.
qubit         SSE2 AES AVX2.
scrypt      SSE2 AVX2.
scrypt:N      unknown
scryptjane:nf   unknown
sha256d      None.
sha256t      SSE4.2 AVX2 SHA.
shavite3      error
skein         AVX2 SHA.
skein2      AVX2.
skunk      SSE2 AVX2.
timetravel      SSE2 AES AVX2.
timetravel10   SSE2 AES AVX2.
tribus      SSE2 AES AVX2.
vanilla      SSE4.2 AVX2.
veltor      SSE2 AES AVX2.
whirlpool      None.
whirlpoolx      None.
x11         SSE2 AES AVX2.
x11evo      SSE2 AES AVX2.
x11gost      SSE2 AES AVX2.
x12         SSE2 AES AVX2.
x13         SSE2 AES AVX2.
x13sm3      SSE2 AES AVX2.
x14         SSE2 AES AVX2.
x15         SSE2 AES AVX2.
x16r         SSE2 AES AVX2.
x16s         SSE2 AES AVX2.
x17         SSE2 AES AVX2.
xevan      SSE2 AES AVX2.
yescrypt      SSE2 SHA.
yescryptr8      SSE2 SHA.
yescryptr16   SSE2 SHA.
yescryptr24   SSE2 SHA.
yescryptr32   SSE2 SHA.
yespower      SSE2 SHA.
yespowerr8   SSE2 SHA.
yespowerr16   SSE2 SHA.
yespowerr24   SSE2 SHA.
yespowerr32   SSE2 SHA.
zr5         SSE2 AES.

I have several Xeon Phi cards and being forced to use Lukminer is my biggest nuisance.

More compatible miners would be wonderful.

Have you ran lukminer in native mode?  I can't find intel c++ compiler anywhere. I think you need version 18.0 right?

Yea that would be great also but I don't know how plausible it would be since they are essentially a bunch of CPU cores which have been known to perform best in CryptoNight.

I had read about people trying to mine some Ethereum with some x100s a while back with not great results.

I agree there are definitively many less Co-Processors floating around than GPUs but even so devs take the time to support even ancient architectures like the Radeon 5000 series (from 2009) that yield absurdly low dev fees (it's almost not even worth it) yet nobody seems to take interest in cards that have been seen hitting 3.5Kh/s+ consistently... it just seems odd to me. The co-processors are x86 which (at least I would think) wouldn't be a huge hurdle to add compatibility too.

XMR-STAK would make a world of a difference.

As a software dev, since you get a % of hashes for your work, would you rather write for a device that has 10’s of thousands in use, or a device that has maybe 1000 in use (just guessing on the number).

Another consideration is that the x100 cards haven’t been made for several years now, so there’s no warranty whatsoever.  Even the x200 cards were discontinued a year or two ago.  Complete crapshoot on buying used ones.

On the other hand, if Phi prices came down and you were able to get STAK or another miner to access them, it would make an interesting option.

what would be more interesting is to see some people release miners that support these phi cards on something other then crytonight ..such as lyra2z neoscript etc

Those xilinx cards have been talked about for almost a year now. I still haven't even seen one verified running. Those guys are barely supposedly buying them, they don't even have the software for the miners yet lol. The numbers they show aren't even that impressive either (if they're even true) considering you have to tear half the card apart to get those numbers (voiding the warranty on a $4k card). Also remember that 4k price is just the card, not including the other hardware needed to get the thing working..

Yea 4% is still tolerable (although I still think it's quite high) but that isn't the only issue. If say the miner could run on XMR-STAK you could mix several Phi PCI-E cards and several GPUs together which would maximize the performance of the boards.

Imagine running a Thread ripper 1950X system (hashes ~5200 on CN-LiteV7) + some Xeon Phi7220As + some GUPs all in the same system. That would yield some very impressive results.

Even better: Imagine a Threadripper 2 system (would not be surprised if it hashes 10kh/s+ on CN-liteV7).

X100 5110s hash around 600-650H/s. Not exactly great but it's expected since they're quite a bit old.

I probably have access to some  5110 at very reasonable price of 150$
Any idea of the hashrate I can expect by one of those?

The problem is that you need a specific miner. 4% is still tolerable. The exact same problem and the FPGA Xilinx - also its miner. Now, if Wolf0 or Claymore would turn their attention to them. Although there may already be " private " versions of these miners...

Yea I agree, but I have seen a few 7220A and 7240s being sold under 800 USD which makes them a great value. I have my 7220A on a riser with a cheap Z270 board. The overall cost would be quite good if I could run atleast 4 cards per board. My main problem is the lack of support otherwise I would buy more.

Although lukminer works fine, it is quite crippling at the same time. The forced dev fee is really high + high cost of equipment + lack of compatibility with other hardware (GPUs) turns a lot of people off.

I think full offload support from XMR-STAK or XMRig would bring alot more people in. I have also heard many people say they get better performance running the cards natively with xmr stak (I haven't seen any proof though). If that's true they would be extremely compeling.

The main problem is lack of support from other miners.

I’ve been following the Phi developments, but for me, it comes down to costs/mhs.  Phi’s are expensive and for the cost of a chip or board, I get better performance from a 480/8 GPU.  The only one that comes close is the Asrock 4x system, but they come from Asia and a 2-3 month lead time as they are custom built.

Hey all!

It has recently come to my attention that more and more Xeon Phi Co-Processors are starting to appear on Ebay and such sites. I have seen more and more people selling KNL (Knights landing) 7220A/P and 7240P quite often.
There has also been a lot of "pre built" servers racks using Xeon Phis being sold lately. Like this "Neon Miner" --> (I.e: https://bit.ly/2l75w4K)

Currently there is only one single miner that officially supports the Xeon phi lineup (Lukminer) but he also is a supplier of the cards. He's holding a mini-monopoly (lol) while forcing a quite high 4% fee on the closed source miner.

I have heard others say you can run xeon phis on XMR-Stak by compiling it "Natively" on the card opposed to offloading it (like LukeMiner does). I have yet to replicate those results.

Seeing how KNL Xeon Phi cards are one of the most powerful option (3.2Kh/s on CN-Heavy) when it comes to mining CryptoNight V7 and it's varients I find it rather odd there is very little support specially since it's x86.

I'll post a picture of my PHI7220A running LukMiner incase anyone wants to see the performance.