Bitcoin Forum

Bitcoin => Hardware => Topic started by: willphase on October 02, 2012, 09:00:42 PM



Title: Parallella
Post by: willphase on October 02, 2012, 09:00:42 PM
http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone

Interesting - could this be used for mining?

also - http://news.ycombinator.com/item?id=4588741

Will


Title: Re: Parallella
Post by: pieppiep on October 02, 2012, 09:07:39 PM
Probably
My guess it's power efficienty in Mhash/s will be somewhere between the cpu's and gpu's we already have.


Title: Re: Parallella
Post by: YokoToriyama on October 02, 2012, 10:35:00 PM
lol there asking for 750k..... aiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
i don't know about you but that sounds like way to much money.


Title: Re: Parallella
Post by: davidspitzer on October 02, 2012, 10:53:49 PM
lol there asking for 750k..... aiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
i don't know about you but that sounds like way to much money.

$750,000 is for the whole project it looks like it will be about $500 for a board


Title: Re: Parallella
Post by: Bogart on October 02, 2012, 10:56:48 PM
I saw that the other day.  It sounded kind of odd.  Like a product in search of a market.

Their talk of equating clock speed with performance seemed somewhat odd.

I didn't see any mention of the instruction set used on the "workhorse" processor, only mention of a "supervisory" processor that runs ARM.

It will be neat if it turns into a whole new class of processor.


Title: Re: Parallella
Post by: YokoToriyama on October 02, 2012, 11:39:50 PM
http://www.kickstarter.com/projects/1523379957/oculus-rift-step-into-the-game (http://www.kickstarter.com/projects/1523379957/oculus-rift-step-into-the-game)

this right here is worth around 2 million its an amazing 3D technology.

if there chip is so power where are the comparisons with the current cpu market?

http://s3.amazonaws.com/ksr/assets/000/176/636/f655b21138f173cf2c6dc2d3cff7c8f6_large.png?1348542744

they can't even spell reduced...

I'm not buying it here.


Title: Re: Parallella
Post by: Stephen Gornick on October 27, 2012, 07:42:45 PM
The $750K goal on Kickstarter was reached (and now over $850K):
 - http://arstechnica.com/information-technology/2012/10/99-raspberry-pi-sized-supercomputer-hits-kickstarter-goal



Title: Re: Parallella
Post by: tacotime on October 27, 2012, 07:59:10 PM
probably similar to fpga consumption for sha256 mining.  with 64 processing cores at ~700 MHz, that's maybe 12 MH/s
probably irrelevant to litecoin mining due to the shitty ~2GB/s memory bus

https://s3.amazonaws.com/ksr/assets/000/204/098/e08e9ae5194d6b7fb57216d8fac119ba_large.jpg
This is interesting, as the current litecoin protocol uses only 128kb of memory per thread.  if they have 128kb on onboard memory per process, these will be crazy litecoin miners


Title: Re: Parallella
Post by: ElectricMucus on October 27, 2012, 08:07:28 PM
probably similar to fpga consumption for sha256 mining.  with 64 processing cores at ~700 MHz, that's maybe 12 MH/s
probably irrelevant to litecoin mining due to the shitty ~2GB/s memory bus

Once these things start packing sufficient memory in the cores themselfes it will get pretty efficient for scrypt.
If litecoin lasts so long as their current roadmap they might even be the status quo by then.


Title: Re: Parallella
Post by: tacotime on October 27, 2012, 08:38:33 PM
yeah, i didn't see that image until after.  regardless, we can just make a new coin or ltc could modify its protocol so that it becomes more memory hard over time.


Title: Re: Parallella
Post by: ElectricMucus on October 27, 2012, 08:44:19 PM
yeah, i didn't see that image until after.  regardless, we can just make a new coin or ltc could modify its protocol so that it becomes more memory hard over time.

I'm not sure we'd want that. I think down the road this kind of architecture will drive regular PCs as well as embedded DSPs.
Because once you start using a sufficient number of cores that architecture becomes an necessity not a feature.

They might even start removing external memory altogether, since once memory becomes sufficiently large the bandwidth required to write to all of it becomes a serious bottleneck. The 2022 version would have 16 GB core ram, I'd say that is sufficient for a desktop even by then. This isn't really a problem at this scale yet, but once go even a magnitude higher it becomes one.
Hard-disks are already at this state where reliable data storage requires double or triple redundancy because the probability of another drive failure starts to matter during the mirroring process.