Topic: [XPM]unofficial jhPrimeminer thread - page 32. (Read 180246 times)

Thanks for getting back to me.  I benchmarked my A10-4600m at various points, but the fastest i have managed 4ch was at the above settings.

A few examples:

-s 200000 -d 16 -t 4:
Val/h 3.283 - PPS: 23282 - 4ch/h: 1786.01 - 5ch/h: 140.29 - 6ch/h: 12.93 - Last 4ch/h: 1346.97 - Last 5ch/h: 0.00

That was an extended length of time, almost a day, after a few hours (restarted), i can manage more along the lines of 2100 4ch/h.. so, I am not sure if its variance, or a leak causing a slow down.  v6 AVX offers me a rather large performance gain over v5 with the same settings (unsure why).

If I enable Primorial Multiplier Auto Tunning with that serv and -d setting, then my performance drops of drasticly:
Sieve/Test ratio: 85.7 / 14.3 %  - New PrimorialMultiplier: 37 (this is about where it settles to tune in at)

Val/h 0.000 - PPS: 11012 - 4ch/h: 1725.09 - 5ch/h: 175.43 - 6ch/h: 0.00 - Last 4 ch/h: 897.87 - Last 5ch/h: 0.00

Sometimes my PPS will drop as low as 800 or worse, and my shares per round dip to almost nothing.  My CoreDuo is beating at my A10 in those instances!?


-s 1200000 -d 16 -t 4:

Primorial Multiplier Auto Tunning Disabled:
Val/h 0.861 - PPS: 10702 - 4ch/h: 1308.35 - 5ch/h: 94.68 - 6ch/h: 8.61 - Last 4ch/h: 0.00 - Last 5ch/h: 448.99

That rate wasn't a quick benchmark either.  I let it run for almost 45mins. T_T

Primorial Multiplier Auto Tunning enabled:
Sieve/Test ratio: 85.4 / 14.6 %  - New PrimorialMultiplier: 67
Val/h 0.589 - PPS: 7439 - 4ch/h: 1355.81 - 5ch/h: 106.11 - 6ch/h: 5.89 - Last 4ch/h: 448.99 - Last 5ch/h: 0.00

Once again, things get worse with this enabled.  Bulldozer+ cores must be a bitch to tune with these settings.  Can AVX be compiled and optimized for AMD in your binaries, or is Intel simply getting better optimizations at the compiler level?  I know this is a laptop cpu, but something seems off (I did check, I'm not running any power saving features.  I'll do some checking against v5 and v6 non-AVX tomorrow (I haven't tired v6 non-AVX).

08/11/13 - 11:56:25 - SHARE FOUND !!! (Th#: 0) ---  DIFF: 7.195531 - VAL: 1.000
   >7
Invalid share
Reason: Share data time overflow

Edit:

Fixed

[18:18] MarcoPolo: i have seen Share data time overflow before. i have a machine mining with a bad cmos battery. so sometimes it looses its time settings. update the time to the correct date and time and it should be gone after a miner restart

This same thing was happening to me..IDk whats the deal.. I dont think it has to do with your internet connection.

"500 - Internal server error.
There is a problem with the resource you are looking for, and it cannot be display"

http://ypool.net seems to be down. DDOS or updates/server restarts?

Edit: Lasted a brief moment. Everything back to normal.

And later in the same thread:

Strictly speaking, the highlighted can be put under doubt. Standard deviation for 46 found blocks is about 0.5*sqrt(46) = 3.39. For 99.7% confidence interval it's 3*SD = 10.17, so the two values are within error. Need some more testing.

just downloaded and used, on the top when it starts it says v6 and autotune is still automatic

Just noticed this for all interested:

http://community.ypool.net/index.php?page=Thread&threadID=58

I'm getting the same problem. My other miners are still working however, which his weird. I could not re-connect after disconnecting my miner and trying to connect again immediately. Don't know why. Help?

I saw some disconnected messages a few minuets ago. But seems to not be doing that.

Thanks for all your work mumus

Weird stuff... I'm using v6 and the last 3 shares as soon as I find one I get disconnected.

08/11/13 - 16:08:09 - SHARE FOUND !!! (Th#: 2) ---  DIFF: 6.717313 - VAL: 0.100    >6
xpt: Disconnected, auto reconnect in 30 seconds
Sieve/Test ratio: 74.7 / 25.3 %  - New PrimorialMultiplier: 59
Sieve/Test ratio: 100.0 / 0.0 %  - New PrimorialMultiplier: 61
Sieve/Test ratio: 0.0 / 100.0 %  - New PrimorialMultiplier: 59
Sieve/Test ratio: 0.0 / 100.0 %  - New PrimorialMultiplier: 53
Sieve/Test ratio: 0.0 / 100.0 %  - New PrimorialMultiplier: 47
Sieve/Test ratio: 0.0 / 100.0 %  - New PrimorialMultiplier: 43

Anybody saw something similar?


Edit: It keeps happening now all the time, already like 5 times now at least and maybe some more times I didn't saw it.

Hi,
A minor release to fix some of the problems we encountered lately and some features mostly for advanced users. As usual this version is not well tested, (time issue). The source is on github.
1. Implemented some kind of heartbeat monitoring, if a thread get stuck it's automatically restarted. Hopefully this will resolve the problem some of you reported that after a while the mining speed is dropped.
2. By default the initial cache auto tuning is disabled. It can be enabled using the "-tune true" command line parameter or by pressing the 'c' button in the miner.
3. The size of L1CacheElements that the above mentioned auto tuning used to set, is by default 256000 and can be set using the '-c' parameter (must be between 64000 - 2000000 and multiply of 32)
4. The auto tuning for the round sieve percentage can be also disabled by specifying a fixed value for the primorialMultiplier with the '-m' parameter (must be between 5 - 1009 and should be a prime number)

I intentionally left out these new parameters from the help because these are intended for more advanced users.
https://dl.dropboxusercontent.com/u/1189851/jhPrimeminer-GMP-v7.zip
I won't post the AVX version because I think it has no performance gain. In case somebody saw any improvement with AVX let me know.
32bit - https://dl.dropboxusercontent.com/u/1189851/jhPrimeminer-GMP-v7-32bit.zip

Edit: the 32 bit version is broken, don't use it until I'll find out what's the problem with it.

AVX - https://dl.dropboxusercontent.com/u/1189851/jhPrimeminer-GMP-v7-AVX.zip

i'm on an amd phenom x6.. I was solo mining for a few weeks at about 2-3000pps.  I tried out ypool right when it began, got like 300pps, so I was like f this.  today, I tried v4, got val/h of 1.7 & 5000pps.  now trying v6: val/h 8 and pps ~22000!  way better.

Mumus you are the man bro! V6 works like a champ! It's the fastest so far. Keep up the good work!

Thanks for the miner men great job will be sending you a tip soon

I confirm this. On wine, v4 does better than any other miner, including hp9. Also, more than 6 threads per process cause decrease in performance. Probably I could get more rate with hp9 but running 8 instances of it on my 48-core node would be pain in the ass (8 virtual machines, 8 wallets, 8 blockchain copies etc.)

P.S. Unfortunately, jhPrimeminer crashes when I try to point it to my local node with getwork.

Personally I rarely on any chat room, with a full time job and 3 kids I just simply can't.
The performance you have is indeed not so good. First the -s 200000 is too low I think. I got poor performance every time I set lower than 500000. I'm usually using 2000000.
Regarding to PPS I rarely using as a reference. First I'm checking the 4ch/h and I'm trying to tune to max. Then based on longer test period I try to go for maxing out 5ch/h or 6ch/h, depending how fast is the computer. on a slow machine 6ch/h  must be measured for too long time to overcome the random factor.
The startup tuning works well with higher sieve size and higher SievePercentage. No wonder that it's odd for -s 200000 -d 16.
Those two accounts you mentioned (unfortunately) doesn't belongs to me as I also write in one of my previous post. Regarding to hardware I use i7 3770 for development and I've just started some testing on AMD Opteron 4180 which meantime doesn't performs so well. I expected much more per core.
Right now I'm working on a new release with some of the features that you requested and I'll publish the code but I don't want to build the binaries yet. Some more testing needs to be done.

Mumus,

just fyi, your v6 is much faster then the current .4 version. I dont know where the differences are, but my shares in the pool more then halved since the change earlier today. And its not luck as it are more then 50 computers running which so far had a pretty steady rate even though it's off course slowly declining.

Yes, only works through wine on Linux. Performance is poor through wine so I solo mine now on Linux.

For Windows v6 and Linux v4. These are just my test results and others will probably be different.

I suggest you do your own tests to find out what works best on your kit.