Its not only the vector extensions that are added, its switching between march=native to everything up to core-avx2. And O2 to O3. I dont know what exactly did the trick, has to be something thats added here.

I see a difficulty below 17. Thats <100k pps total.

Let me show you how to optimize the miner:

My I7-4770k with 8 threads, stock-miner:
pps: 31821 / 31160  10g/h 4473 / 3429  15g/h 0 / 0

Now go on and build your own libgmp from source. When done, run the tune utility for optimum results. Copy your final libgmp.so.10 to /usr/local/lib/.

Now we're at:
pps: 34039 / 34041  10g/h 6077 / 4479  15g/h 0 / 0

Optimizing sieve size:
open src/main.cpp

line 252:
  1048576);

to
   8 * 1048576);


and line 256
  500000);

to
  15000000);

make sure to also change the shift, otherwise you wont be able to submit most blocks! You could specify it on command line, but this is safer:

line 128:
  uint16_t shift = (opts->has_shift() ?  atoi(opts->get_shift().c_str()) : 20);

change to:
  uint16_t shift = (opts->has_shift() ?  atoi(opts->get_shift().c_str()) : 23);


Now its mining at
pps: 38276 / 37068  10g/h 5156 / 5265  15g/h 0 / 0


Lets also add the tuning for Haswell AVX2:
open Makefile and set

OTFLAGS   = -march=core-avx2 -mtune=core-avx2 -O3
(if you have a haswell, of cause)

Results:
pps: 39691 / 39553  10g/h 6059 / 5635  15g/h 0 / 0

I've just added CACH to my blockchain-explorer:

https://bchain.info/CACH/

Comes with a richlist:

https://bchain.info/CACH/rich/


edit: removed it again. Did not know about what was going on here.

To sum it up: Its a lot more than just "cause we can".

Vanitygen assumes the private key magic bytes are (public key) + 128, which is not true for lots of coins. For Gap its +59.

go to vanitygen.c

Line 399
   privtype = 128 + addrtype;

change to
   privtype = 59 + addrtype;

then hit make again

Now it generates valid private keys

Let me explain it to you:

Difficulty increases just like with every other coin. But as the effort needed to find gaps increases exponentially with higher merit, the log of actual difficulty is shown.

For everyone who is tuning the miner:
Max. sieve size is 2^shift

How many instances do you use?

Verification is quite fast. Even a 7k gap is verified within seconds.
With O(log(p)) the max. merit is already above 256, with a larger shift difficulty can raise further. I don't see any limit here.


Its a bit of a tradeoff here. Larger shifts allow for larger, more efficient sieves, while lower ones speed up the fermat tests. I figured out the optimum is somewhere between 20 (the default value) and 26. Its only a minor improvement anyway.

./gapcoind signmessage "GMhhQ2Evh5P6mjoQRHokkz17gakvLjAizC" "Yes thats my address - dcct"
H654Z+DUgc78Q3L/gcg7lHCxiARH3unyEC08tZ6/d868ZZvtHz1P3j/G0k8cijEUbHpa5sRL3pv3xIoL1XonhkM=


I think we will find a lot more high merits soon. 35.4245 is the largest known.

The i7-4790 has hyperthreading. Try 8 threads.

I was using 4.8.2. 4.9 was about the same speed.

you did make clean first?
While tune is running you need to stop the miner of cause.

Try us-east-1e.

Also there is a beautiful api you can use. If it gets terminated just let it automatically spawn a new one in a different data center. Copy the miner and start it using a automated small script.


Let me guess: Not optimized for avx, using stock libgmp.

-O3 -march=corei7-avx -mtune=corei7-avx

+tune for libgmp.

And you are at ~144K pps

With spot instances its a lot cheaper than that. One c3-8xlarge is ~25 cents per hour für ~135,000 pps.

Thats ~15$ per hour with on average about 5 Blocks. Actually profitable!

At the moment its a useful metric as everyone is using the same implementation. I agree this needs to be changed soon.

At the moment you could still buy cheaper @polo, so why not sell for 30k.

But with rising difficulty you wont be able to buy this cheap soon.

Just started the node on a fresh server (Core 2 dual core), and it finished syncing in < 10 minutes.

Dosn't change the fact that its a CPU miner  

It took only minutes here. Whats slowing it down? Network? HDD?