HP10 64 bit
My main rig (1.7 chainsperday) no blocks since 26-Aug
Another machine (1.1 chainsperday) no blocks since 10-Aug
I'm finding the variance quite unusual lately. Perhaps I need to sacrifice a goat to the Prime Gods.
Topic: [XPM] [ANN] Primecoin High Performance | HP14 released! - page 24. (Read 397588 times)
HP10 was the best, is there an ongoing work for an HP11 release soon?
Thanks
Thanks
Making the 64-bit builds on Windows has been.. traumatic. I can post the command-line only, windows 64 build if there is interest.. but I probably won't be able to do all the builds for all the platforms anytime soon. It is forked from hp 10.
I haven't used the command line miner before but if you post it I will give it a try.
It would be nice if the hp miner reports average rate of 8-chain and 9-chains found in the last N hours, or at least just how many 7,8,9,10 chains found since the start.
Yes I was thinking about this the other day. Could it be added to mininginfo or at least highest chain found ?
I made a fork that reports these things.. if Mikaelh or anyone else wants to grab the changes they are more than welcome to. 5-chains are skewed downwards a bit because I also no longer check the second bi-twin chain if it is too short to reach 9.
Thanks. Was it forked from hp10? Is there a win x64 binary?
Making the 64-bit builds on Windows has been.. traumatic. I can post the command-line only, windows 64 build if there is interest.. but I probably won't be able to do all the builds for all the platforms anytime soon. It is forked from hp 10.
It would be nice if the hp miner reports average rate of 8-chain and 9-chains found in the last N hours, or at least just how many 7,8,9,10 chains found since the start.
Yes I was thinking about this the other day. Could it be added to mininginfo or at least highest chain found ?
I made a fork that reports these things.. if Mikaelh or anyone else wants to grab the changes they are more than welcome to. 5-chains are skewed downwards a bit because I also no longer check the second bi-twin chain if it is too short to reach 9.
Thanks. Was it forked from hp10? Is there a win x64 binary?
Looks like this in the log:
-- 5-chains/h: 400.17133
-- 6-chains/h: 36.87732
-- 7-chains/h: 3.54956
-- 8-chains/h: 0.38116
-- Ratio 7/6: 0.0963 Ratio 8/7: 0.1074
2013-09-06 19:35:37 stats 599 prime/s 15892 test/s 0.721245 chain/d | run: 41.98, avg primorial: 31.8 block/d: 0.12639732
-- 5-chains/h: 400.17133
-- 6-chains/h: 36.87732
-- 7-chains/h: 3.54956
-- 8-chains/h: 0.38116
-- Ratio 7/6: 0.0963 Ratio 8/7: 0.1074
2013-09-06 19:35:37 stats 599 prime/s 15892 test/s 0.721245 chain/d | run: 41.98, avg primorial: 31.8 block/d: 0.12639732
The inverse ratios are more interesting, IMHO, since a value of 12 for each step has been discussed before, and it looks about right. The point was partly to illustrate the exponential nature of Primecoin difficulty, but it also enabled mining profitability estimates based on the 5-chain rate. Which seems like ancient history now, after all it's over 2 months since the launch :-j
It would be nice if the hp miner reports average rate of 8-chain and 9-chains found in the last N hours, or at least just how many 7,8,9,10 chains found since the start.
Yes I was thinking about this the other day. Could it be added to mininginfo or at least highest chain found ?
I made a fork that reports these things.. if Mikaelh or anyone else wants to grab the changes they are more than welcome to. 5-chains are skewed downwards a bit because I also no longer check the second bi-twin chain if it is too short to reach 9.
https://github.com/rsweny/primecoin
Looks like this in the log:
-- 5-chains/h: 400.17133
-- 6-chains/h: 36.87732
-- 7-chains/h: 3.54956
-- 8-chains/h: 0.38116
-- Ratio 7/6: 0.0963 Ratio 8/7: 0.1074
2013-09-06 19:35:37 stats 599 prime/s 15892 test/s 0.721245 chain/d | run: 41.98, avg primorial: 31.8 block/d: 0.12639732
I don't know the cause of the problem although I suspect memory leak or "garbage collection" problem of some sort.
I usually many times reboot miner... everyday.PS: Let's wait for the new version from Mikaelh)
It would be nice if the hp miner reports average rate of 8-chain and 9-chains found in the last N hours, or at least just how many 7,8,9,10 chains found since the start.
Yes I was thinking about this the other day. Could it be added to mininginfo or at least highest chain found ?
It would be nice if the hp miner reports average rate of 8-chain and 9-chains found in the last N hours, or at least just how many 7,8,9,10 chains found since the start.
+1
Since 02.09 I can not find a single block. My 10 computers working in the shuffle, what happened? Prior to this date I was stable find 1-3 block per day.
Does anyone else have such problems?
Does anyone else have such problems?
I have had similar problem on a smaller scale. After switching to ypool and observing the miner behavior, I think the cause is that the miners can enter a certain state that they mostly find lower difficulty shares with unproportionally few number of higher difficulty shares. You could spot the problem with high certainty after 6 hour (for 8-chains) and 24 hours (for 9-chains) with ypool, and restart the miner to hopefully "fix" the problem. But you can't spot this problem with solo miiners because they don't report long chain rates.
I don't know the cause of the problem although I suspect memory leak or "garbage collection" problem of some sort. It would be nice if the hp miner reports average rate of 8-chain and 9-chains found in the last N hours, or at least just how many 7,8,9,10 chains found since the start.
Comparing GFlops is meaningless here. FLOPS is FLoating-point Operations Per Second, and Primecoin uses integer arithmetic. So, even if the Phi is so much faster than X86 in floating point maths, it means nothing for Primecoin performance.
Yeah, you're right. This was the best i could come up with to compare integer maths between processors:
http://speedtraq.com/?s=1
Unfortunately the Phi isn't on here, but if we choose something that is relatively close to it:
Intel(R) Xeon(R) CPU L7555 @ 1.87GHz, the result is: 88.478x
Where as with the
AMD FX(tm)-8350 Eight-Core Processor, the result is: 29.681x
So even a Xeon that has more cores, and a higher clock speed than the Phi is still only 2.9x faster. Which means all it would take is three builds with those 8350s to outperform the Phi coprocessor. And now i think we are coming in cheaper than the Phi going with AMD, but it really depends on how much the rest of the builds cost (mobo, ram, hdd, psu, etc)
Since 02.09 I can not find a single block. My 10 computers working in the shuffle, what happened? Prior to this date I was stable find 1-3 block per day.
Does anyone else have such problems?
Does anyone else have such problems?
A more realistic situation is someone renting a server with Phi(s) installed. They'd still need to do a lot of coding. For all we know there may already be people doing this: not everyone has the incentive to disclose their works on public forums, especially if they can make a lot of money.
Although the Phi is an exceptional little package, there are many other co-processors on the market. In theory these too could be used for mining but again would require a good programming knowledge and understanding of the maths behind primecoin.
Although the Phi is an exceptional little package, there are many other co-processors on the market. In theory these too could be used for mining but again would require a good programming knowledge and understanding of the maths behind primecoin.
Comparing GFlops is meaningless here. FLOPS is FLoating-point Operations Per Second, and Primecoin uses integer arithmetic. So, even if the Phi is so much faster than X86 in floating point maths, it means nothing for Primecoin performance.
Serious question, what are your thoughts on mining XPM with a Xeon Phi?
Looks like those have a total of 57 cores @ 1.1 for a total of 62.7GHz of processing power. You can get 30GHz+ with an AMD Vishera for less than $200. I know the comparison isn't apples to apples, but I don't know why you'd spend that kind of money regardless.
Yeah, looking at that, i don't think you can really compare things like that just on GHz alone. Benchmarks put the AMD Vishera FX8350 at around 35 - 40 GFLops. That Xeon Phi 3120 is ~1000 GFLops. So about 25 times faster. Unless i am thinking about this wrong. That would mean i would need 25 computer builds with those $200 processors in them to achieve the same performance. That is in no way cheaper ($5000 for the processors alone).
Serious question, what are your thoughts on mining XPM with a Xeon Phi?
Looks like those have a total of 57 cores @ 1.1 for a total of 62.7GHz of processing power. You can get 30GHz+ with an AMD Vishera for less than $200. I know the comparison isn't apples to apples, but I don't know why you'd spend that kind of money regardless.
Serious question, what are your thoughts on mining XPM with a Xeon Phi?
Nobody knows. Nobody has the money to buy one of those. If you have the money, go ahead.
Although I think you might want to recompile the miner to support the new instruction sets that it provides, but you will never be able to get an ROI with each unit being so expensive, unfortunately.
Serious question, what are your thoughts on mining XPM with a Xeon Phi?
@needsmths
Don`t tuch anything ~2000pps is fine, you can`t compare to ypool miner
Don`t tuch anything ~2000pps is fine, you can`t compare to ypool miner
Don't change the parameters unless you have a computer science or mathematics degree 
w00t! went back to solo mining w/HP9 and hit a block within 12 hours.
{
"blocks" : 101723,
"chainspermin" : 5,
"chainsperday" : 0.65202060,
"currentblocksize" : 2807,
"currentblocktx" : 8,
"difficulty" : 9.47157931,
"errors" : "",
"generate" : true,
"genproclimit" : 7,
"roundsievepercentage" : 70,
"primespersec" : 1212,
"pooledtx" : 8,
"sievepercentage" : 18,
"sievesize" : 600000,
"testnet" : false
}
{
"blocks" : 101723,
"chainspermin" : 5,
"chainsperday" : 0.65202060,
"currentblocksize" : 2807,
"currentblocktx" : 8,
"difficulty" : 9.47157931,
"errors" : "",
"generate" : true,
"genproclimit" : 7,
"roundsievepercentage" : 70,
"primespersec" : 1212,
"pooledtx" : 8,
"sievepercentage" : 18,
"sievesize" : 600000,
"testnet" : false
}
Thanks.
Found something in a history. Seems like 2000 is really great )
I've only changed sievepercentage from default 10 to 5 and now see 2218 pps.
Face palm.
Exactly as Trillium said. Don't do that, you are increasing your PPS but you will get less blocks mined because your sieve is not producing enough candidates.
If you are playing with these settings always spend few hours on testnet trying one and the other version and comparing the results. These few hours are well worth it since you save many, many hours of false mining with inferior settings.
Jump to: