[XPM]unofficial jhPrimeminer thread

[TheXev]

mumus, is there an IRC chatroom on freenode that people in this thread and use for discussion?  

I'm thinking of starting a mining hardware comparison page or something for primecoin.

With my A10-4600m, using the following settings for V6 AVX: -s 200000 -d 16 -t 4
I am managing between 25000-19000 PPS with 9.1 Val/h.. whatever that means.

The strange thing about the performance of my A10, is that i must disable Primorial Multiplier Auto Tunning, or it goes from its start up value of 67 (which is good for this cpu, as far as I can tell), down to 19.. which lowers PPS and Val/h.

Also, the startup tuning seems odd.  At parts of it, i have much higher PPS, but after awhile, it slows down... and its final value always seems to be slower.  Is there anyway we could get flags to disable the auto tuning, and manually program the L1cache value?  Also, another flag to disable Primorial Multiplier Auto Tunning on startup.  It simply doesn't work for my CPU.

I'm not sure what Val/h is, but my Core2 Duo downstairs is getting about the same, with much lower PPS.

If those two accounts on ypool do belong to you mumus, I am interested to hear what hardware you are mining with.

v6 AVX has been a champ for me on this machine.  I've gone from the 1900 PPS i could muster out of HP9, to this 20000 PPS setup, and I somehow managed to mine not 1, but two blocks lastnight!

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.

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).

[1713578246]

1713578246

[1713578246]

1713578246

[1713578246]

1713578246

[TheXev]

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.

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

The pool was getting DoSSed earlier or something.  Seems to have cleared up now.

[snaidervp]

how the -m values affect the found shares? i mean the lower value/higher value does what?

[mechs]

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.

I saw performance gains with AVX....please compile it with AVX included!

[mumus]

how the -m values affect the found shares? i mean the lower value/higher value does what?

This parameter used to be automatically adjusted in the miner. I can't say that a lower or higher is better because it depends on other parameters. I've added this parameter for users that already found the optimal value with one specific configuration and they want to preset and don't want the auto tuning to mess with it.

[dudeguy]

Just noticed this for all interested:

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

And later in the same thread:

to compare numbers:

v4 has 41 blocks found and 409 primecoins credited to the account.
--basically even with solo mining if i add the donation to the total

v6 has 51 blocks found and 370 credited to the account
--obviously this version is MUCH better at finding blocks and better for the pool over all. unfortunately, the share handling is killing this miner's profit potential.

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.

If these statistics are representative, then could we attribute them to any new functions added/omitted along with any new enhancements?

[Tommo_Aus]

Confirming there looks to be some sort of memory leak in v6, on one server without AVX I just restarted the process as it was using 11GB memory and only ~25% of the allocated threads. The other server with AVX was using 8GB RAM and same story, using only ~25% of allocated threads.

After restarting the processes they're both mining with full thread utilisation and using < 200MB memory.

[dudeguy]

Confirming there looks to be some sort of memory leak in v6, on one server without AVX I just restarted the process as it was using 11GB memory and only ~25% of the allocated threads. The other server with AVX was using 8GB RAM and same story, using only ~25% of allocated threads.

After restarting the processes they're both mining with full thread utilisation and using < 200MB memory.

Linux or Windows?

[Tommo_Aus]

Sorry I should've mentioned I'm running v6 through wine on Ubuntu. I'll give it a go through Windows once I get some more memory into my Windows machine. I don't remember v4 having this problem, but I see v7 is out now so I'll try that first and see if I have the same problem.

[geofflosophy]

Sorry I should've mentioned I'm running v6 through wine on Ubuntu. I'll give it a go through Windows once I get some more memory into my Windows machine. I don't remember v4 having this problem, but I see v7 is out now so I'll try that first and see if I have the same problem.

I think I saw earlier that you should stick with v4 if you're on Wine.

[psyc]

new version of jhPrimeMiner cabin?

[TheXev]

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.

Please post an AVX version.  I am seeing a serious reduction in performance in PPS (and in turn, chaining) on my A10-4600m.

[mumus]

Please post an AVX version.  I am seeing a serious reduction in performance in PPS (and in turn, chaining) on my A10-4600m.

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

[TheXev]

Awesome, i'll give this a good ol' knock around.

I had another great idea for a feature.  If you could add a button press, that'd just output all current settings (maybe even in command line format), that'd be killer.  We could tweak it, then output a useable command line to setup those settings in a .bat instantly.

[Roger3636]

Working a treat!

[Sterling]

Please post an AVX version.  I am seeing a serious reduction in performance in PPS (and in turn, chaining) on my A10-4600m.

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

Does this fix the prime multiplier 2x problem? I run into it daily, it's brutal waking up to 8 hours of errors and no XPM generated =\.

[mhps]

Please post an AVX version.  I am seeing a serious reduction in performance in PPS (and in turn, chaining) on my A10-4600m.

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

Would be nice to include msvcr110.dll.

[dudeguy]

Please post an AVX version.  I am seeing a serious reduction in performance in PPS (and in turn, chaining) on my A10-4600m.

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

Would be nice to include msvcr110.dll.

Seems to run fine without msvcr110.dll for me over 6 hours.  I'll keep my stats with it and see if anything changes, although I think it was intentionally left out for the AVX build.

[snakee]

Please post an AVX version.  I am seeing a serious reduction in performance in PPS (and in turn, chaining) on my A10-4600m.

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

Does this fix the prime multiplier 2x problem? I run into it daily, it's brutal waking up to 8 hours of errors and no XPM generated =\.

Would love to know that too, lost too much in the 2 nights i had v6 running with this error :/

[vingaard]

One little question...

I'm running v7 in a i5-2500 @3.3GHz and the question is...

could someone tell me which is the best configuration for the program adjustable parameters? (or at least... which configuration do you prefer or seems to get better results?)

Thank you in advance