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?

It is real hpm, or cpm*2.

Quote from: reorder on November 24, 2013, 06:11:51 pm

Quote from: crendore on November 24, 2013, 06:08:53 pm

Python is significantly slower than C.  Buyer beware.  Personally i would wait for a C implementation than buy a python one.

Sorry, but so what? You do not need ultra performance to send 10-20 packets of data/sec over network Most pools are running python, for that matter.

Anyway, sold.

EDIT: sold exclusively.

Right, and i should take whatever a salesman says at face value too.  There's a reason that all decent miners are written in C, and it certainly can't be hand-waved away by some silly argument about network performance.

Not selling anymore, but still. There are multiple reasons for that indeed, speed being not on the list. Think poclbm and how it is still alive and well.

Call it nitpicking, but it is written in Python using pyopencl My plan is to try to get some reward for the work first. Mining is apparently not an option anymore (not with my 2 noisy 7870s), so I am selling the miner. The buyer, if there will be one, can opensource it of course at his discretion.

EDIT: sold exclusively.

Quote from: reorder on November 24, 2013, 08:15:34 am

Quote from: bytemaster on November 24, 2013, 08:05:28 am

Quote from: reorder on November 24, 2013, 07:56:09 am

Quote from: bytemaster on November 23, 2013, 10:09:27 pm

Quote from: m0rph on November 23, 2013, 10:04:40 pm

So how about some actually useful information regrading the GPU miner then?

I don't have any information other than the algorithm I would implement given the time:

Generate 512 MB of birthdays (512 Hashes) in parallel
Use the GPU to sort them
Use the GPU to find collisions

Someone who wants an open source GPU miner could easily pay to have one developed for them and it would cost about $10,000 or perhaps less.   Considering what people are paying Amazon for cloud computing, it only makes sense that someone would develop this.

I have implemented a GPU miner, and this algo (with bitonic parallel sorting which is considered best for the task) yields just 160hpm on 7870 at best. There are better ways, currently I can squeeze >500hpm out of the card at stock clocks.

Thank you for sharing.  So with an optimized algorithm you can get about a 3x gain from the GPU vs a CPU as many CPUs are getting 160 HPM.   Will you share your algorithm?

Just a hashtable, fast and dirty in a sense that it may skip valid collisions.

So if you accept losses due to race conditions then you gain on average.  Sounds like how I optimized my CPU miner.

Curiously, the more you are willing to lose, the more you gain. HPM was increasing as I went from full sort to cuckoo hashtable to plain synchronized hashtable to unsynchronised one. And then the same with worsening the hash function and reducing the table size.

Quote from: m0rph on November 23, 2013, 10:04:40 pm

So how about some actually useful information regrading the GPU miner then?

I don't have any information other than the algorithm I would implement given the time:

Generate 512 MB of birthdays (512 Hashes) in parallel
Use the GPU to sort them
Use the GPU to find collisions

Someone who wants an open source GPU miner could easily pay to have one developed for them and it would cost about $10,000 or perhaps less.   Considering what people are paying Amazon for cloud computing, it only makes sense that someone would develop this.

I have implemented a GPU miner, and this algo (with bitonic parallel sorting which is considered best for the task) yields just 160hpm on 7870 at best. There are better ways, currently I can squeeze >500hpm out of the card at stock clocks.