GPU Miners are here... if rumors are to be believed!

[barwizi]

meh, i only believe when i see

[iruu]

@reorder
http://www.intel.com/content/www/us/en/intelligent-systems/intel-technology/fast-sha512-implementations-ia-processors-paper.html

8.59cycles/bytes for sha512 on large data buffer. Thanks to AVX. It's ridiculous.

Taking into account cost of capital, expected roi and other things, gpus obviously win, but that's quite another point.
(nobody sane is going to buy any hardware for mining this coin)

[reorder]

Theres a guy on irc freenode #protoshares testing a gpu miner for pts mining
He says 1 GFX card with 2 gig DDR is like the largest amazon ec2 instance,two xeons

If that is all the faster a GPU is then I would have to say that we are VERY close to having a CPU only algorithm.   What is the relative power consumption and cost of the systems?

It's because of sha512, not birthdays.
New intel cpus are extra fast in generating sha512s due to many specialized vector instructions. 32 ht cores 2.1Ghz xeon (2x16) can generate one nonce range in 37ms, radeon 5870 in ~70ms. That's it. Additionally intel cpu can be oced even 2x.
All that birthday searching does nothing, in fact, it probably helps gpu. 

Its 40nm gpu vs 22nm cpu though, a three generation gap! Maybe Rx200 Radeons are again much faster than cpu. When the two gpu version comes out I'm going to buy it and write optimized version for it out of curiosity.

Still, few have such high-end cpus.

Also, there's absolutely no resistance to fpga/asics. It can be drastically faster on such things, I mean several orders of magnitude. Not that it's something important anywany - mining is too short lived.

I'd say it is not as much SIMD as it is 64-bit registers that gives x86 such an advantage in sha512. If only radeons had anything like SHLD, they would blow intel away.

[Sy]

My 7950 at 1025/1250 ran at 515 cpm, dual gpu 510 cpm each - power draw about 200-225W per GPU, 50W for the System -> 450-500W for 1000 cpm -> around 2 cpm per W

My i5-3570k at home draws 90W under load (whole system) for 120 cpm, thats 1.33 cpm/w

e5 Xeons are better cpm/w (i think they even beat gpus) but are not realistic to compare since one e5 costs around 1000-2000€ and a 7950 200€ max.

Your i5 is better cpm/w than a GPU which means that those who own i5's can still mine profitably and cover electric costs.   Those who are investing in capital equipment just for mining care about  CPM/w and CPM/capital in which case they will probably opt for the GPU approach.  But for the average casual miner capital costs are '0' which means all that matters is CPM/w.    I believe that an i7 is probably around 1 cpm/w. 

As an individual miner, it is cheaper to mine with your CPU than your GPU if you care about return on electricity.  That is a win in my book!   

1.2 CPM per Watt is worse than 2 CPM per Watt...but yeah, close enough to make cpu mining still a viable option.

[bytemaster]

My 7950 at 1025/1250 ran at 515 cpm, dual gpu 510 cpm each - power draw about 200-225W per GPU, 50W for the System -> 450-500W for 1000 cpm -> around 2 cpm per W

My i5-3570k at home draws 90W under load (whole system) for 120 cpm, thats 1.33 cpm/w

e5 Xeons are better cpm/w (i think they even beat gpus) but are not realistic to compare since one e5 costs around 1000-2000€ and a 7950 200€ max.

Your i5 is better cpm/w than a GPU which means that those who own i5's can still mine profitably and cover electric costs.   Those who are investing in capital equipment just for mining care about  CPM/w and CPM/capital in which case they will probably opt for the GPU approach.  But for the average casual miner capital costs are '0' which means all that matters is CPM/w.    I believe that an i7 is probably around 1 cpm/w. 

As an individual miner, it is cheaper to mine with your CPU than your GPU if you care about return on electricity.  That is a win in my book!   

[iruu]

Theres a guy on irc freenode #protoshares testing a gpu miner for pts mining
He says 1 GFX card with 2 gig DDR is like the largest amazon ec2 instance,two xeons

If that is all the faster a GPU is then I would have to say that we are VERY close to having a CPU only algorithm.   What is the relative power consumption and cost of the systems?

It's because of sha512, not birthdays.
New intel cpus are extra fast in generating sha512s due to many specialized vector instructions. 32 ht cores 2.1Ghz xeon (2x16) can generate one nonce range in 37ms, radeon 5870 in ~70ms. That's it. Additionally intel cpu can be oced even 2x.
All that birthday searching does nothing, in fact, it probably helps gpu. 

Its 40nm gpu vs 22nm cpu though, a three generation gap! Maybe Rx200 Radeons are again much faster than cpu. When the two gpu version comes out I'm going to buy it and write optimized version for it out of curiosity.

Still, few have such high-end cpus.

Also, there's absolutely no resistance to fpga/asics. It can be drastically faster on such things, I mean several orders of magnitude. It's infinitely parallelizable - with 2^23 sha512 cores the collisions for one nonce range could be generated in just a few dozen cycles. Not that it's something important anyway - mining is too short lived for fpga/asics.

[Sy]

My 7950 at 1025/1250 ran at 515 cpm, dual gpu 510 cpm each - power draw about 200-225W per GPU, 50W for the System -> 450-500W for 1000 cpm -> around 2 cpm per W

My i5-3570k at home draws 90W under load (whole system) for 120 cpm, thats 1.33 cpm/w

e5 Xeons are better cpm/w (i think they even beat gpus) but are not realistic to compare since one e5 costs around 1000-2000€ and a 7950 200€ max.

[bytemaster]

Theres a guy on irc freenode #protoshares testing a gpu miner for pts mining
He says 1 GFX card with 2 gig DDR is like the largest amazon ec2 instance,two xeons

If that is all the faster a GPU is then I would have to say that we are VERY close to having a CPU only algorithm.   What is the relative power consumption and cost of the systems?

[smokim11]

Theres a guy on irc freenode #protoshares testing a gpu miner for pts mining
He says 1 GFX card with 2 gig DDR is like the largest amazon ec2 instance,two xeons

Using a high end video card should get some big hashrates, I wouldn't expect it to be released for free yet.

[Anddos]

Theres a guy on irc freenode #protoshares testing a gpu miner for pts mining
He says 1 GFX card with 2 gig DDR is like the largest amazon ec2 instance,two xeons

[smokim11]

Is that actually HPM, as in hashes per minute, or collisions per minute (which coyote miner prints as hpm for some reason)?

Coyote: collisions per minute reported as hpm (at least on github which is 12 days old! - the binary probably multiplies this by 2 also, otherwise coyote miner would be two times slower)
Beeer and ypool: collisions per minute * 2 reported as col/m

Which one is it?

edit:
http://bitsharestalk.org/index.php?topic=965.0
Ok, so 520HPM is impossible on amazon instance. So it's actually col/m*2 for all public miners.

So unless you specifically modified a miner to report true hpm you're getting 500 col/m. Did you?

Sorry must of been a mistake, on amazon aws instance: 32 core server i was getting 500+ col/min using jhprotominer for ypool.

[iruu]

Original client worked this way:

Code: [Select]

pblock->nNonce=pblock->nNonce+1;
testHash=pblock->CalculateBestBirthdayHash(pindexPrev, outdated);
nHashesDone++;
nHashesDone/time = original meter

So cpm is inflation, because as you said, there's less than one collision pair per work.

Not that it really matters now

[reorder]

That's not hpm, hpm is hashes per minute, or are you certain there's on average exactly one collision per try? I don't blame you for the stats inflation though - the pools started it.

Well, that depends on which hashes you meter. What you mean must probably be called 'works per minute', it is related to cpm linearly on average and the coefficient must be >1 (average number of collisions in a work is just a tad less than 1). So, it is a deflation instead

[3dtr]

If there exist GPU miner. What's about to change the hashing algorithm to something else?

[iruu]

That's not hpm, hpm is hashes per minute, or are you certain there's on average exactly one collision per try? I don't blame you for the stats inflation though - the pools started it.