[XMG] M7M CPU mining discussion thread

[dga]

i7-4770 CPU @ 3.40GHz
Linux
Miner:  wolf-m7m-cpuminer-V2+dga   (grin)
Threads: 8

Code:

[2015-07-03 22:17:17] thread 6: 211920 hashes, 14.46 khash/s
[2015-07-03 22:17:17] accepted: 67/68 (98.53%), 115.19 khash/s (yay!!!)
[2015-07-03 22:17:19] thread 3: 225705 hashes, 14.14 khash/s
[2015-07-03 22:17:19] accepted: 68/69 (98.55%), 114.85 khash/s (yay!!!)
[2015-07-03 22:17:20] thread 5: 245311 hashes, 14.47 khash/s
[2015-07-03 22:17:20] accepted: 69/70 (98.57%), 114.74 khash/s (yay!!!)

(original wolf miner gets:

Code:

[2015-07-03 22:39:11] accepted: 37/37 (100.00%), 95.20 khash/s (yay!!!)

)

20% boost.

[dga]

i7-4770 CPU @ 3.40GHz
Linux
Miner:  wolf-m7m-cpuminer-V2+dga   (grin)
Threads: 8

Code:

[2015-07-03 22:17:17] thread 6: 211920 hashes, 14.46 khash/s
[2015-07-03 22:17:17] accepted: 67/68 (98.53%), 115.19 khash/s (yay!!!)
[2015-07-03 22:17:19] thread 3: 225705 hashes, 14.14 khash/s
[2015-07-03 22:17:19] accepted: 68/69 (98.55%), 114.85 khash/s (yay!!!)
[2015-07-03 22:17:20] thread 5: 245311 hashes, 14.47 khash/s
[2015-07-03 22:17:20] accepted: 69/70 (98.57%), 114.74 khash/s (yay!!!)

(original wolf miner gets:

Code:

[2015-07-03 22:39:11] accepted: 37/37 (100.00%), 95.20 khash/s (yay!!!)

)

20% boost.

Well done!

Thanks.  I'll be a little mercenary with this one and not try to mine with it myself:  It's for sale.  "Sale" can be open sourcing via a bounty, or exclusive private sale to an individual or coalition.  2 BTC.  "exclusive" means nobody else gets it, I send you an update for any small-percentage improvements I make, and you get right of first refusal if I find another more than 10% speedup.

(Pricing is to get paid for the time I put into it, not make a massive profit.)

[dga]

i7-4770 CPU @ 3.40GHz
Linux
Miner:  wolf-m7m-cpuminer-V2+dga   (grin)
Threads: 8

Code:

[2015-07-03 22:17:17] thread 6: 211920 hashes, 14.46 khash/s
[2015-07-03 22:17:17] accepted: 67/68 (98.53%), 115.19 khash/s (yay!!!)
[2015-07-03 22:17:19] thread 3: 225705 hashes, 14.14 khash/s
[2015-07-03 22:17:19] accepted: 68/69 (98.55%), 114.85 khash/s (yay!!!)
[2015-07-03 22:17:20] thread 5: 245311 hashes, 14.47 khash/s
[2015-07-03 22:17:20] accepted: 69/70 (98.57%), 114.74 khash/s (yay!!!)

(original wolf miner gets:

Code:

[2015-07-03 22:39:11] accepted: 37/37 (100.00%), 95.20 khash/s (yay!!!)

)

20% boost.

Well done!

Thanks.  I'll be a little mercenary with this one and not try to mine with it myself:  It's for sale.  "Sale" can be open sourcing via a bounty, or exclusive private sale to an individual or coalition.  2 BTC.  "exclusive" means nobody else gets it, I send you an update for any small-percentage improvements I make, and you get right of first refusal if I find another more than 10% speedup.

(Pricing is to get paid for the time I put into it, not make a massive profit.)

I should have done that with mine. Then again, it was pretty easy...

Actually, you raise a good point - I should feed up the chain a little, since I built on your code:

2 BTC gets it for the first {whatever} to commit.  Of that, 10% goes to Wolf0.

[djm34]

may-be I should buy it 

[dga]

Code:

[2015-07-04 10:17:04] accepted: 71/72 (98.61%), 122.79 khash/s (yay!!!)
[2015-07-04 10:17:06] thread 5: 949595 hashes, 15.83 khash/s
[2015-07-04 10:17:10] thread 5: 58609 hashes, 15.84 khash/s
[2015-07-04 10:17:10] accepted: 72/73 (98.63%), 122.80 khash/s (yay!!!)
[2015-07-04 10:17:13] thread 0: 797582 hashes, 14.72 khash/s

30% speed boost.  Still part of that same original offer, but if I get it fast enough to be profitable on AWS before anyone takes me up on it, the offer disappears.

[djm34]

Code:

[2015-07-04 10:17:04] accepted: 71/72 (98.61%), 122.79 khash/s (yay!!!)
[2015-07-04 10:17:06] thread 5: 949595 hashes, 15.83 khash/s
[2015-07-04 10:17:10] thread 5: 58609 hashes, 15.84 khash/s
[2015-07-04 10:17:10] accepted: 72/73 (98.63%), 122.80 khash/s (yay!!!)
[2015-07-04 10:17:13] thread 0: 797582 hashes, 14.72 khash/s

30% speed boost.  Still part of that same original offer, but if I get it fast enough to be profitable on AWS before anyone takes me up on it, the offer disappears.

the problem with that coin is that if the hashrate goes beyond certain values the block reward decrease to very low value

[go6ooo1212]

Code:

[2015-07-04 10:17:04] accepted: 71/72 (98.61%), 122.79 khash/s (yay!!!)
[2015-07-04 10:17:06] thread 5: 949595 hashes, 15.83 khash/s
[2015-07-04 10:17:10] thread 5: 58609 hashes, 15.84 khash/s
[2015-07-04 10:17:10] accepted: 72/73 (98.63%), 122.80 khash/s (yay!!!)
[2015-07-04 10:17:13] thread 0: 797582 hashes, 14.72 khash/s

30% speed boost.  Still part of that same original offer, but if I get it fast enough to be profitable on AWS before anyone takes me up on it, the offer disappears.

How is the cpu temp on such speeds?

[dga]

Code:

[2015-07-04 10:17:04] accepted: 71/72 (98.61%), 122.79 khash/s (yay!!!)
[2015-07-04 10:17:06] thread 5: 949595 hashes, 15.83 khash/s
[2015-07-04 10:17:10] thread 5: 58609 hashes, 15.84 khash/s
[2015-07-04 10:17:10] accepted: 72/73 (98.63%), 122.80 khash/s (yay!!!)
[2015-07-04 10:17:13] thread 0: 797582 hashes, 14.72 khash/s

30% speed boost.  Still part of that same original offer, but if I get it fast enough to be profitable on AWS before anyone takes me up on it, the offer disappears.

the problem with that coin is that if the hashrate goes beyond certain values the block reward decrease to very low value

I know.  There's a great prisoners dilemma in coins with this type of inverse-hashrate block reward (Primecoin, too).  It's quite funny.

[dga]

Code:

[2015-07-04 10:17:04] accepted: 71/72 (98.61%), 122.79 khash/s (yay!!!)
[2015-07-04 10:17:06] thread 5: 949595 hashes, 15.83 khash/s
[2015-07-04 10:17:10] thread 5: 58609 hashes, 15.84 khash/s
[2015-07-04 10:17:10] accepted: 72/73 (98.63%), 122.80 khash/s (yay!!!)
[2015-07-04 10:17:13] thread 0: 797582 hashes, 14.72 khash/s

30% speed boost.  Still part of that same original offer, but if I get it fast enough to be profitable on AWS before anyone takes me up on it, the offer disappears.

How is the cpu temp on such speeds?

Code:

cor CPU    %c0  GHz  TSC SMI    %c1    %c3    %c6    %c7 CTMP PTMP   %pc2   %pc3   %pc6   %pc7  Pkg_W  Cor_W GFX_W
        100.00 3.70 3.40   0   0.00   0.00   0.00   0.00   80   80   0.00   0.00   0.00   0.00  42.99  37.99  0.00

Physical id 0:  +80.0°C  (high = +80.0°C, crit = +100.0°C)

43W, running at full turbo (3.7ghz is the max for the 4770).  80 degC.  Stock cooler.  Nothing very interesting.

[dga]

Code:

[2015-07-04 10:17:04] accepted: 71/72 (98.61%), 122.79 khash/s (yay!!!)
[2015-07-04 10:17:06] thread 5: 949595 hashes, 15.83 khash/s
[2015-07-04 10:17:10] thread 5: 58609 hashes, 15.84 khash/s
[2015-07-04 10:17:10] accepted: 72/73 (98.63%), 122.80 khash/s (yay!!!)
[2015-07-04 10:17:13] thread 0: 797582 hashes, 14.72 khash/s

30% speed boost.  Still part of that same original offer, but if I get it fast enough to be profitable on AWS before anyone takes me up on it, the offer disappears.

I should stop.  But it's too much fun.

[2015-07-04 15:59:36] accepted: 112/116 (96.55%), 137.84 khash/s (yay!!!)
45% speed boost

Code:

[2015-07-04 16:07:11] accepted: 59/59 (100.00%), 141.65 khash/s (yay!!!)

Finally hit 50% speed boost.  Cool beans, though it's now AVX2 (haswell and later)-specific.  Any takers yet? ;-)

This is at 45W package on the i7-4770 (up from 43, but a large gain in perf/watt).  Still not thermally throttled - CPU is at 82 degC.

Wonder if it's more profitable than Monero at this point.  Hmmm...

[111magic]

I don't know the answer on that one!
Maybe time will tell!

[smolen]

the problem with that coin is that if the hashrate goes beyond certain values the block reward decrease to very low value

I know.  There's a great prisoners dilemma in coins with this type of inverse-hashrate block reward (Primecoin, too).  It's quite funny.

Yes, the question of how and when attack the blockchain is game-theoretic

may-be I should buy it 

... just to pull underpriced offer from the market

[dga]

the problem with that coin is that if the hashrate goes beyond certain values the block reward decrease to very low value

I know.  There's a great prisoners dilemma in coins with this type of inverse-hashrate block reward (Primecoin, too).  It's quite funny.

Yes, the question of how and when attack the blockchain is game-theoretic

may-be I should buy it 

... just to pull underpriced offer from the market

I'm happy to raise the price if that makes it more interesting.

I've found two blocks thus far on my little i7-4770.  Not too bad - it's actually power-profitable, though not hugely so.

[111magic]

the problem with that coin is that if the hashrate goes beyond certain values the block reward decrease to very low value

I know.  There's a great prisoners dilemma in coins with this type of inverse-hashrate block reward (Primecoin, too).  It's quite funny.

Yes, the question of how and when attack the blockchain is game-theoretic

may-be I should buy it 

... just to pull underpriced offer from the market

I'm happy to raise the price if that makes it more interesting.

I've found two blocks thus far on my little i7-4770.  Not too bad - it's actually power-profitable, though not hugely so.

Oh thats nice with the i7
Nice to hear that Magi is power-profitable.
Magi likes to grow step by step.
Value will raise in time.
 

[dga]

the problem with that coin is that if the hashrate goes beyond certain values the block reward decrease to very low value

I know.  There's a great prisoners dilemma in coins with this type of inverse-hashrate block reward (Primecoin, too).  It's quite funny.

Yes, the question of how and when attack the blockchain is game-theoretic

may-be I should buy it 

... just to pull underpriced offer from the market

Indeed.  I don't think I've seen quite as poorly-thought-out a design choice in an altcoin in a while.

Attacking this blockchain would be trivial.  This is *one* machine using my code:

Code:

[2015-07-06 11:33:26] accepted: 407/408 (99.75%), 793.46 khash/s (yay!!!)

A single AWS c4.8xlarge is a little faster than this.  One could double-spend Magi for about $30USD/hour.

[djm34]

the problem with that coin is that if the hashrate goes beyond certain values the block reward decrease to very low value

I know.  There's a great prisoners dilemma in coins with this type of inverse-hashrate block reward (Primecoin, too).  It's quite funny.

Yes, the question of how and when attack the blockchain is game-theoretic

may-be I should buy it 

... just to pull underpriced offer from the market

Indeed.  I don't think I've seen quite as poorly-thought-out a design choice in an altcoin in a while.

Attacking this blockchain would be trivial.  This is *one* machine using my code:

Code:

[2015-07-06 11:33:26] accepted: 407/408 (99.75%), 793.46 khash/s (yay!!!)

A single AWS c4.8xlarge is a little faster than this.  One could double-spend Magi for about $30USD/hour.

which is rather expensive considering you won't make that much

[dga]

the problem with that coin is that if the hashrate goes beyond certain values the block reward decrease to very low value

I know.  There's a great prisoners dilemma in coins with this type of inverse-hashrate block reward (Primecoin, too).  It's quite funny.

Yes, the question of how and when attack the blockchain is game-theoretic

may-be I should buy it 

... just to pull underpriced offer from the market

Indeed.  I don't think I've seen quite as poorly-thought-out a design choice in an altcoin in a while.

Attacking this blockchain would be trivial.  This is *one* machine using my code:

Code:

[2015-07-06 11:33:26] accepted: 407/408 (99.75%), 793.46 khash/s (yay!!!)

A single AWS c4.8xlarge is a little faster than this.  One could double-spend Magi for about $30USD/hour.

which is rather expensive considering you won't make that much

It's true.  But the design discourages the hash rate from going up as the value of the coin increases, which is exactly the source of security in bitcoin -- one might suspect that this would just as likely cap the value of the coin as anything else.

At least it helps answer the question of whether it's worth implementing it on a GPU.

But next month will be funny for Magi when the first Skylake CPUs come out with Intel's SHA extensions.  Easy ~2-3x speed boost to the miner once again.  Pretty soon there will only be enough hash rate headroom for a single machine to do all the mining.  And then we can rename it "central bank coin".  Has a nice ring to it. :-)

[smolen]

But next month will be funny for Magi when the first Skylake CPUs come out

Skylake is going to be quiet game-changer for X11, Quark and other non memory hungry algorithms. Just imagine AVX512 rented on demand! Well, the market already priced that in with Litecoin

[dga]

Also, can I just say:  What the hell is with that weird Gauss-Legendre numerical integration that doesn't involve any variables from the block?

I personally think it looks much nicer this way - not that it reduces the runtime much, but getting rid of unnecessary complexity in a hash function always brings a bit of sunshine to my day:

Code:

double GaussianQuad_N2(const double x1, const double x2)
(blah blah blah)

---->
uint32_t sw2_(int nnounce)
{
    double xm = (sqrt((double)(nnounce)))/900 + 50;
    double x[4], w[4];
    double z;

    double s =  0x1.23456789abcdfp-1 * xm * swit2_(xm);

    z = 0x1.cff6ce0533a69p-1;
    x[0] = xm - xm*z;
    x[1] = xm+xm*z;
    w[1] = w[0] = 0x1.e539ec36e028ap-3 * xm;

    z = 0x1.13b23fd99b705p-1;
    x[2]=xm-xm*z;
    x[3]=xm+xm*z;
    w[3] = w[2] = 0x1.ea1da25aaa962p-2 * xm;
    
    for(int j=0; j<=3; j++) s += w[j]*swit2_(x[j]);

    return ((uint32_t)(s*2277.4239655685828));
}

There you go.  Much nicer to not have any of that silly for loop garbage in there when you're just multiplying through by some constants.

[djm34]

Also, can I just say:  What the hell is with that weird Gauss-Legendre numerical integration that doesn't involve any variables from the block?

I personally think it looks much nicer this way - not that it reduces the runtime much, but getting rid of unnecessary complexity in a hash function always brings a bit of sunshine to my day:

Code:

double GaussianQuad_N2(const double x1, const double x2)
(blah blah blah)

---->
uint32_t sw2_(int nnounce)
{
    double xm = (sqrt((double)(nnounce)))/900 + 50;
    double x[4], w[4];
    double z;

    double s =  0x1.23456789abcdfp-1 * xm * swit2_(xm);

    z = 0x1.cff6ce0533a69p-1;
    x[0] = xm - xm*z;
    x[1] = xm+xm*z;
    w[1] = w[0] = 0x1.e539ec36e028ap-3 * xm;

    z = 0x1.13b23fd99b705p-1;
    x[2]=xm-xm*z;
    x[3]=xm+xm*z;
    w[3] = w[2] = 0x1.ea1da25aaa962p-2 * xm;
    
    for(int j=0; j<=3; j++) s += w[j]*swit2_(x[j]);

    return ((uint32_t)(s*2277.4239655685828));
}

There you go.  Much nicer to not have any of that silly for loop garbage in there when you're just multiplying through by some constants.

sshhh  
(the idea was that double precision is bad for gpu... )