Author

Topic: Nexus - Pure SHA3 + CPU/GPU + nPoS + 15 Active Innovations + More to Come - page 343. (Read 785531 times)

hero member
Activity: 799
Merit: 1000
i7 now mining fine Smiley and stats nice and smooth on both rigs:
i5 ~ 75PPS
i7 ~ 85PPS

Thank you very nice work! Smiley

Edit:
Thats i5 + 40 PPS i7 + 45 PPS.
wow!
legendary
Activity: 868
Merit: 1058
Creator of Nexus http://nexus.io
yes did git pull then make clean and rebuild.

Edit: just found a block on the i5,
[METERS] 82686.090909 SPS | 97.636364 PPS | Height = 1774
[MASTER] Prime Cluster of Difficulty 4.106306 Found on Thread 3

911f331bd073984e095a7cc8a35fc084ea5750fb3d4bbb6b1a8bd8d0de1e289bd011dea2bb3219c 05f6814e990a10c4ca44e5bc4405b15af275093b8bbfcc72ce22eda11a9b46c453e5169f4025fd9 61b08c34b6e05e18162c0111c946b190c482680a4abeab02341f2a622c8a3f3ee733198a57eaa5a 1097068162697eb9c68

[MASTER] Block Accepted By Coinshield Network.


I just removed the Mutexes from Meter. It appears this worked.  Smiley

edit: What is your PPS increase?

Viz.
hero member
Activity: 799
Merit: 1000
yes did git pull then make clean and rebuild.

Edit: just found a block on the i5,
[METERS] 82686.090909 SPS | 97.636364 PPS | Height = 1774
[MASTER] Prime Cluster of Difficulty 4.106306 Found on Thread 3

911f331bd073984e095a7cc8a35fc084ea5750fb3d4bbb6b1a8bd8d0de1e289bd011dea2bb3219c 05f6814e990a10c4ca44e5bc4405b15af275093b8bbfcc72ce22eda11a9b46c453e5169f4025fd9 61b08c34b6e05e18162c0111c946b190c482680a4abeab02341f2a622c8a3f3ee733198a57eaa5a 1097068162697eb9c68

[MASTER] Block Accepted By Coinshield Network.
legendary
Activity: 868
Merit: 1058
Creator of Nexus http://nexus.io
Of course:
i7 4770k @ 4.2ghz ubuntu 13.10
hangs on New block directly after connecting to wallet.
i5 3570k @ 4.2ghz ubuntu 13.10 runs ok but the output is like so
[MASTER] Coinshield Network: New Block 1767
[METERS] 90721.636364 SPS | 145.000000 PPS | Height = 1767
[METERS] 92843.181818 SPS | 98.090909 PPS | Height = 1767
[METERS] 63305.363636 SPS | 85.727273 PPS | Height = 1767
[METERS] 61715.363636 SPS | 104.545455 PPS | Height = 1767
[METERS] 83212.181818 SPS | 120.909091 PPS | Height = 1767
[METERS] 74373.909091 SPS | 91.000000 PPS | Height = 1767
[METERS] 69206.000000 SPS | 92.636364 PPS | Height = 1767
Stats Output is much slower than earlier builds and very inconsistent (ie values jump around alot)

Ah deadlock, thank you. I'll post a fix. You compiling from Github?

edit: removed mutex from meter. Let me know if this brings different results.

Viz.
hero member
Activity: 799
Merit: 1000
Of course:
i7 4770k @ 4.2ghz ubuntu 13.10
hangs on New block directly after connecting to wallet.
i5 3570k @ 4.2ghz ubuntu 13.10 runs ok but the output is like so
[MASTER] Coinshield Network: New Block 1767
[METERS] 90721.636364 SPS | 145.000000 PPS | Height = 1767
[METERS] 92843.181818 SPS | 98.090909 PPS | Height = 1767
[METERS] 63305.363636 SPS | 85.727273 PPS | Height = 1767
[METERS] 61715.363636 SPS | 104.545455 PPS | Height = 1767
[METERS] 83212.181818 SPS | 120.909091 PPS | Height = 1767
[METERS] 74373.909091 SPS | 91.000000 PPS | Height = 1767
[METERS] 69206.000000 SPS | 92.636364 PPS | Height = 1767
Stats Output is much slower than earlier builds and very inconsistent (ie values jump around alot)
legendary
Activity: 868
Merit: 1058
Creator of Nexus http://nexus.io
windows7 64 Bit Speed ​​does not improve?

I'm seeing my Windows 7 64 bit go from 4 PPS to 10 PPS.

Viz.
sr. member
Activity: 490
Merit: 250
Mining will automatically stop!
sr. member
Activity: 490
Merit: 250
Optimized Miner: I am getting 2 - 3x PPS now [about 40-50 PPS on 4 cores from 18-24 PPS]. We'll see about the Segmentation Faults.

Github Source
Windows x86 [No x64 yet, I still haven't slept]


Had some fun all night with the Sieve of Eratosthenes and Mutexes: some things are more important than sleep! Thank someone for waking me up at 3:00 AM, this and insomnia led me to this update  Cheesy

Viz.



windows7 64 Bit Speed ​​does not improve?
legendary
Activity: 868
Merit: 1058
Creator of Nexus http://nexus.io
Are you using libgmp if so what version?

No sir, check prime.cpp. I wrote my own sieve of Eratosthenes for divisor tests before Miller-Rabin and Fermat. You can play around with the nSieveSize to get optimal results. I set these binaries at 200, and this seems to be the most optimal thus far.

edit: could you give me more data? I have two miners running with no issues.

Viz.

hero member
Activity: 799
Merit: 1000
Are you using libgmp if so what version?

[MASTER] Coinshield Network: New Block 1758
[METERS] 57883.727273 SPS | 85.272727 PPS | Height = 1758
[METERS] 68272.272727 SPS | 103.636364 PPS | Height = 1758
[METERS] 65839.000000 SPS | 88.181818 PPS | Height = 1758
[METERS] 61276.909091 SPS | 87.000000 PPS | Height = 1758
[METERS] 63026.818182 SPS | 88.636364 PPS | Height = 1758
[METERS] 68276.000000 SPS | 110.090909 PPS | Height = 1758
[METERS] 104962.000000 SPS | 179.545455 PPS | Height = 1758
[METERS] 116121.818182 SPS | 185.636364 PPS | Height = 1758
[METERS] 129446.636364 SPS | 262.636364 PPS | Height = 1758
[METERS] 205827.090909 SPS | 186.090909 PPS | Height = 1758
[METERS] 95890.727273 SPS | 127.636364 PPS | Height = 1758
[METERS] 61557.000000 SPS | 102.636364 PPS | Height = 1758
[METERS] 112538.545455 SPS | 132.818182 PPS | Height = 1758
[METERS] 93726.363636 SPS | 140.454545 PPS | Height = 1758
[MASTER] Coinshield Network: New Block 1759
[METERS] 78634.818182 SPS | 123.363636 PPS | Height = 1759
[METERS] 73791.363636 SPS | 86.909091 PPS | Height = 1759
[METERS] 62523.454545 SPS | 77.727273 PPS | Height = 1759

something is just not right with this build... this is my i5 my i7 just hangs on Newblock.
legendary
Activity: 868
Merit: 1058
Creator of Nexus http://nexus.io
Optimized Miner: I am getting 2 - 3x PPS now [about 40-50 PPS on 4 cores from 18-24 PPS]. We'll see about the Segmentation Faults.

Github Source
Windows x86 [No x64 yet, I still haven't slept]


Had some fun all night with the Sieve of Eratosthenes and Mutexes: some things are more important than sleep! Thank someone for waking me up at 3:00 AM, this and insomnia led me to this update  Cheesy

Viz.

hero member
Activity: 666
Merit: 500
any compiled .exe available?
legendary
Activity: 868
Merit: 1058
Creator of Nexus http://nexus.io
Thanks for the data guys, I'll post another update tomorrow; end this bug once and for all. [And get some x64 windows binaries too]

If you are having issues with the Failed to Update Height, try a larger timeout, the default is 10 seconds. This is the 4th commandline argument. 127.0.0.1 9325 4 20. Does it get stuck in a loop of Failed to Update Height, Failed to Connect, etc, or does it eventually reconnect?

I did notice higher PPS on my machine as well. The real tell here though will be SPS, the PPS is prime numbers found per second, SPS is Searches per Second. Prime numbers can be in random clusters, so this is why you see variations with higher PPS and lower PPS each meter update. Searches per Second will tell you how quickly your miner is searching nNonce values, similar to hash rate.

This at least is a decent improvement. We'll see what others can be made.  Smiley

Goodnight Everybody,
Viz.
hero member
Activity: 820
Merit: 1000
Added Mutex Locks to Primeminer, new source on my Github. Still encountered Segmentation Fault though  Undecided
Let me know your results as a degree of improvement.

Thank You,
Viz.

Viz, I think you can simplify your solution by moving the mutex lock in the ServerThread loop INSIDE the if(BLOCK) statement around line 313 of miner.cpp. 

Code:
if(BLOCK)
{
LOCK(THREADS[nIndex]->MUTEX);
 
THREADS[nIndex]->BLOCK = BLOCK;

THREADS[nIndex]->fBlockFound = false;
THREADS[nIndex]->fNewBlock   = false;
}


You can then remove all other places that you are locking accept for the lock in PrimeMiner, which is fine.  As I understand it, you don't need to protect against asynchronous access to integral data types as they are inherently thread safe as they are on the stack, so only protect access to the BLOCK pointer on the miner thread instance.  By doing this, you remove the need for the server thread to have synchronous access to the state of each thread when updating the METERS, so they update regularly rather than having to wait for each miner thread to finish as it is doing now - it mean each METERS read out happens regularly and with consistent results which it is not doing right now.

Unfortunately I'm still seeing the segmentation fault (even with your code) so I think there is somewhere else that needs making threadsafe.

Cheers
Paul
hero member
Activity: 820
Merit: 1000
previous build was ~ 33PPS
Why say "much much slower" if you have same results?)
Quote
[METERS] 22679.023529 SPS | 31.741176 PPS | Height = 1592
Sorry for the confusion but the actual display of the output happens less often. I hope that makes sense.

Yeah that makes sense and I'm seeing the same result.  I'm also seeing very frequent segmentation faults still which is not good.  Maybe hold off on this miner until it is resolved.
hero member
Activity: 799
Merit: 1000
previous build was ~ 33PPS
Why say "much much slower" if you have same results?)
Quote
[METERS] 22679.023529 SPS | 31.741176 PPS | Height = 1592
Sorry for the confusion but the actual display of the output happens less often. I hope that makes sense.
Hix
legendary
Activity: 1971
Merit: 1036
previous build was ~ 33PPS
Why say "much much slower" if you have same results?)
Quote
[METERS] 22679.023529 SPS | 31.741176 PPS | Height = 1592
hero member
Activity: 799
Merit: 1000
sr. member
Activity: 433
Merit: 250
I compiled miner with cygwin and the PPS rate went up greatly.
hero member
Activity: 820
Merit: 1000
Bit rusty on linux still; did "git pull" to update but ubuntu@ubuntu:~/primeminer$ make clean
make: *** No rule to make target `clean'.  Stop.
how to clean build?

Edit: Might be worth mentioning only getting segfault from previous build on i7 4770k, i5 3570k has run since launch without segfault.

make -f makefile.unix

or make -j8 -f makefile.unix (the 8 being number of threads to build with)


I know the cmd to compile  Smiley  I was wondering how to remove the already built files, from memory with other miners i've run "make clean"

Edit: Also the output of the new build seems "strange" the Meters output seems much much slower than previous build.

[METERS] 22679.023529 SPS | 31.741176 PPS | Height = 1592
[METERS] 25558.700000 SPS | 46.600000 PPS | Height = 1592
[METERS] 70294.142857 SPS | 110.142857 PPS | Height = 1592
[METERS] 51540.625000 SPS | 59.750000 PPS | Height = 1592
[METERS] 38107.833333 SPS | 50.500000 PPS | Height = 1592
[METERS] 44799.000000 SPS | 84.500000 PPS | Height = 1592
[METERS] 58842.166667 SPS | 75.666667 PPS | Height = 1592
[METERS] 64534.800000 SPS | 123.900000 PPS | Height = 1592
[MASTER] Coinshield Network: New Block 1593
[METERS] 42817.125000 SPS | 32.562500 PPS | Height = 1593
[METERS] 27897.500000 SPS | 51.125000 PPS | Height = 1593
[METERS] 39008.250000 SPS | 65.625000 PPS | Height = 1593
[MASTER] Coinshield Network: New Block 1594
[METERS] 29257.444444 SPS | 35.444444 PPS | Height = 1594
[METERS] 54324.333333 SPS | 120.000000 PPS | Height = 1594

Edit2: Then some connection issue to the wallet, previous build is chugging away fine on another machine.
[MASTER] Coinshield Network: New Block 1595
[METERS] 32235.000000 SPS | 39.800000 PPS | Height = 1595
[METERS] 89994.000000 SPS | 203.000000 PPS | Height = 1595
Failed to Update Height...



You still have to provide the make file was my point.
So make clean -f makefile.unix
Jump to: