Protoshares: Random FAQ from history


 *  Nota Bene: Once more, people are encouraged to follow me on Twitter @yvg1900 for futher announcements on upcoming miner update. 
 * Nota Bere: This are some questions from history of the chat. This is an working project ( still a lot to be written) and it is a very hard work going through all chat history. Of course, some mistake might be founded here.
 * Nota Bene: If you find this interesting please feel free to donate: PTS PfxFs91AoAw8PnDnzCDy6dXExpfMrAaWPc . LTC LfMdKBSNakfChhLTPQsrUjbzvZuMWQPVkU BTC 1JC4fd8wRSxK4YodqcopCDafbE421NP82m .XPM  AMzj1UHzmtm97fposVssSTQ1b1KTnq81ep
 * Read miner txt second.


 * ==== Invalid share Reason: Merkleroot is invalid or corrupted. Update your miner. There is no problem. This is very rare case of concurrent block info update. No need to solve it. Check for merkle validity removed due to performance reasons. This situation may happen, if you are using M7h and get other shares accepted no action required. ====
 * ==== Which one of your miners is coded for AMD Phenom II's? Windows, Linux or Debian. Penom II is barcelona, I think. This is an old chip series. ====
 * ==== Reference config: Dual Xeon E5-2697v2 512Gb RAM, OCed to 3.4 GHz, RAM at 1940. approx. 1200 CPM====
 * ==== which one would you suggest for AMD PHEOM II X6? find out which generation will work : start generic > barcelona > bulldozerv2 > bobcatv2 ====
 *  What do you recommend for my i5 3570K. win64generic or corei7avx? corei7avx if physical machine
 * ==== started mining using the core i7 jhprotominer on 2x Xeon L5639 with 2048mb/thread and I'm getting 80 collisions/min after 10 mins and dropping, do I want more or less collissions? Go 512M====
 * ==== Small guide for: single server 20 CPUs of E5-4650? (SGI UV2000) 640GB in total 1)        use corei7avx build 2) use 20 processes and numactl to pin them to individual sockets Check if ou have numactl installed numactl –help If yes, Then create a shell script in the same folder whre you have miner, Put launch line for jhprotominer with 16 threads   then add numactl -N o -m 0 – right before jprotominer (numactl -N 0 -m 0 -- ./jhprotominer -u username.worker -p x -t 16) Rest of the command line (after jhprotominer) keep as you have. Pick 512M version. this is perfect for this CPU. If it works and you see your worker connected on the site - you are almost done. Then stop it and duplicate .sh script 19 more times. Replacing 0 in -N 0 -m 0 with growing numbers fromm 1 to 19. Do you knwo how to use named screens? screen -R miner-cpu0 ====
 * ==== Have you got any optermized miners for AMD APU's? Bobcat2 and Bulldozerv1====
 * ==== I7 2500k is an i5 so should i stick with the plain version There are AVX and AVX-2 versions. AVX->Sandy/Iby bridge, AVX-2 -> Haswell ====
 * ====AMD Opteron(tm) Processor 4171 HE -        Barcelona====
 * ==== there are core i7 non-avx, avx and avx2 miner variations corei7sse4 for VPSes with i7/Xeon-E5 CPUs and buggy AVX support De facto bulldozerv1 uses sse4 core, and in new set there will be corei7sse4 Which miner should I use for my AMD Phenom II X6 1035t CPU? it supports SSE, SSE2, SSE3, SSE4a Barcelona====
 * ==== Always check CPU World for capabilities  So on some Xeon-based VPSes SSE4 will work while SSSE3 will fail ====
 *  which miner would you use for a 3570k corei7avx
 * ==== does your latest miner variation utilize memory bandwidth more effectively? Yes, it is====
 * ==== with i7-3770 CPU @ 3.40GHz should i use AVX or SSE version of miner? AVX should I use 4 threads or 8 threads, to utilize the hyper threading?   Normally this helps, because of HT is utilizing CPU while other HT accessing memory, and it is our case. But you shall experiment yourself. ====
 * ==== testing the new miner on an i5 760 here and coll/min seems down about 5-10% but seems to find more shares. Coincidence or intended? Coincidence. Which build? Try corei7, corei7sse4 or bulldozerv1. And run longer than 15 minutes needed to judge due to momentum nature of PoW. ====
 * ==== whats the best hardware to mine pts?   SGI UV 2000 http://www.sgi.com/products/servers/uv/====
 * ==== Dual Xeon E5-2697v2 does almost 1200 CPM ====
 * ==== i have a i5 4670k. wich optimized miner version i should run? coreavx2 ====
 * ==== I'm useing Win7 x64bit with i7@920 procesor. Choose among corei7, core2 and corei7sse4 ====
 * ==== I want to try it on a 32-bit operating systems...I do not optimize for 32-bit - not interested at all, sorry. Only special request and serious donation may turn me to 32-bit. Sorry.====
 * ==== I get Share value/h: 0.0000 all the time, but the CPU load is 100% You have no connection. port 8081 and it is down. Change port ====
 * ==== i5-3317U avx 77.7cpm, sse4 74.5 cpm. i5-3517U AVX 93.2cpm 78 shares, SSE4 91.2cpm 63 shares. i7-4500U avx version 88.5 cpm ====
 * ==== AVX2 only on Haswell CPUs. ====
 * ==== which of your new miners should i use for azure cloud comps? ATM it is barcelona, which has approx. +10% to the original miner. More opts for Azure need way more time ====
 * ==== What he did in his update is an extra check for block height every 256 shares. There are total of 62M shares checked for every nonce round, and adding similar check degrading performance to considerable amount. What I am calculating now is if it better to have a general performance gain or get rid of outdated shares. Say, if you'd have 2x performance, then you can afford losing 1 share of 4, so you are 1.5 times faster share-wise anyway. This is just matter of proper probabilistic calculations, this is what I am doing now - measuring actual probability of outdated shares comparing to perf degratation with the added check. ====
 * ====Why I receive so many "Invalid shares - Reaso: Share is outdated"? Where is the problem?This is the pool policy change. Formerly pool accepted all shares, which caused generation of way too many Orphaned Blocks, whoch caused a lot of anger on other pools. jh00 decided to change the policy, so now shares generated or received too late are getting rejected, do not count and not getting sent to the network ====
 * ==== core2 and corei7 have similar assembly code behind. Both SSSE3 instruction sets. If cat /proc/cpuinfo shows Xeon E5-XXXX then definitely go corei7avx if on phys machine, or test corei7sse4 and then corei7 if on VPS. Makes no sense to go core2 until you are on non-E5 xeon. ====
 * ==== can some1 tell me good miner for Intel L5410 http://ark.intel.com/products/33090/Intel-Xeon-Processor-L5410-12M-Cache-2_33-GHz-1333-MHz-FSB core2 or corei7sse4, try ====
 * ==== pre version ideas:There is a fix from jh00 (1.0e) that fixes most of outdated shares.This is done by periodically checking if new block arrived. So no work is made on the block already outdated. Doing so degrades performance if your machine is fast - you gain more from being faster than loose from doing outdated search at the end of the block, especially having in mind upcoming diff increase, so duration between blocks will increase, too. On the other hand, on slower machines (those well under 100 cpm) rejection ratio may be way too high because of nonce round just last longer for them. So here is my suggestion [actually, decision] I am doubling up the zoo and post 2 sets of builds One set will be faster, but will produce more outdates => targeted for faster machines with shorter nonce round time Second will have early nonce round termination => targeting slower machines This way people can decide for themselves Making command line option "is not an option" = we are speaking for single instruction that actually des we are speaking for single instruction that actually destroys perf by polluting CPU caches, which are precisely tuned for perf. So, as promised earlier, "I will make your life even more complicated" :) There are two more important improvements   1) Port randomization - miner will "auto-magically" try alternate ports if gets disconnected => may help load balancing => can be disabled via new -B cmdline switch. No need to change anything in the scripts, 100% compatible this time. 2) Improved display - now info on new blocks and their numbers, so you can see if miner is actually receiving blocks from the server; accepted counts; statistically correct Standard Deviation calculation, which will show you how stable your CPM metric is. Will show up as "warmup" until it collects reasonable amount of data. The point is that these builds are NOT just recompiles. They include COMPLETELY DIFFERENT code.M7f will contain nice -A option to control per-core affinity over NUMA.M7f will have an output for new blocks printed, so you will be able to see if miner receives jobs from the ====
 * ====celeron g1610: collisions/min: 58.7506, e=0.0230. a6-5400k: collisions/min: 40.3367, e=0.0157; from your stock of miners what's best for a Pentium Dual Core CPU E6300 @ 2.80 GHZ 32 bit? core2 ====
 * ==== M7f version will come alone with md5sums.txt file and separate signature file, so you can verify authencity of the miner archives right in your PTS wallet software using File/Verify Message command. ====
 * ==== somebody can try verifying message using the wallet? You'll need content of md5sums.txt and address and signature from md5sums-signature.txt This is how one can ensure that miners are not faked ProtoShares wallet menu File/Verify Message d5sums-signature.txt - address and signature info. md5sums.txt - content message to verify. This is what we are trying now - using wallet to make signatures :) ====
 * ==== e=standard deviation of coll/min value, to check how stable your perf is at the moment.e has nothing to do with the speed itself, it is just indication of how stable your speed is. Just for being able to compare comparable things. http://en.wikipedia.org/wiki/Standard_deviation ====
 * ====Intel Core i5-3570K CPU @ 3.40GHz corei7avx. intel(R) Xeon(R) CPU E5430 @ 2.66GHz core2 M7fs====
 * ==== Warmup is for stats calculation only. Means it is not enough data to be sure your CPM you see is stable enough for decidions. Stable values are when 1) warmup disappear; 2) you will see e=XXXXX, and that XXXXX is lower than 0,09. e is standard deviation of CPM over last 128 measures. So first 100-200 measurements most of the diffrerence in perf comes from pure luck, and after from pure performance ====
 * ==== why do you call it momentum? Because of it is momentum and can be simplest way described by differential equation CPM=d(Nonce)/d(T) ====
 * 