[XPM] Finding the best sievesize for mining

[Tamis]

Hi guys,

one of the interesting question regarding primecoin mining is what is the best sievesize to mine.

Some say it's 1M, others claim to find a whole lot of more blocks with 4M and so on.

My experience so far is that I find a block a day with default settings (1M) on a i7 3770k @ 4400.
After reading some users claiming that 4M gave them a lot more blocks I did a 24 hours test run with sievesize=4M

I found a block in 4 hours with this setting and nothing more for the next 20 hours.
Now the randomness of the mining process makes it hard to draw conclusions on such short time frames but if we all try to keep notes and play with the sievesize we might get some intersting data to share.

So please come and share your tests with us if you happen to be tweaking this parameter.

I'm going to run at 2M for 48 hours starting now and will share my results once done.

[ampere]

Testing it on testnet would probably give you better statistics to work worth.

[Tamis]

Well this has been done by several people and they do not get the same result on testnet and real mining.

No idea why.

[Bitmong]

Well, testnet has much lower difficulty, so that could play a part there.

[Tamis]

I've been playing around with testnet and different sievesize. I launched the client for 5 minutes each time and took note of the amount of block found on each run.

1M = 12b, 19b
2M = 16b
3M = 12b, 15b
4M = 19b, 16b, 13b, 17b
5M = 14b

This does not give much and the variance is a bitch so i would need to do the same on longer time periods and a lot more run.
Still it would look like 4M is the best candidate for the moment !

I found thousands of XPM while doing this, it's a shame its on testnet :)

Please feel free to give your results so we can progress on that question.

[achillez]

sievesize depends on cache size

[Lauda]

You have to test it out on your CPU, it varies and there is no right universal answer in this case.

[Tamis]

The difficulty passing 9 is causing a crash on hp5 and hp6 for several people when sievesize is too high.
I was running fine at 4.6M and the client crashed while I was taking a nap. Since then I cannot go over 3.6M without crashing.

achillez and LaudaM what sievesize did you end up using and on what cpu ?

[eCoinomist]

I already tested this in real live client at 8.* difficulty, 1M give better result in my exp. I had 2 block/hr for 3 hours, so the statistics should be decent enough.

[Vigil]

Been running for 96 hrs at various sievesizes, mainly 1000000. I'm getting about 4000 pps. I haven't found a block.