SETI ASIC Chips


log in

Advanced search

Message boards : Number crunching : SETI ASIC Chips

1 · 2 · Next
Author Message
Brian
Send message
Joined: 20 Apr 13
Posts: 4
Credit: 644,840
RAC: 0
United States
Message 1386148 - Posted: 30 Jun 2013, 16:19:30 UTC

Could the basics of SAH be supercharged with ASIC chips? I mean that if we follow the BitCoin craze and have some chips specifically designed for this work? I would be interested in buying a bunch.

Brian

Alinator
Volunteer tester
Send message
Joined: 19 Apr 05
Posts: 4178
Credit: 4,647,982
RAC: 0
United States
Message 1386165 - Posted: 30 Jun 2013, 17:29:55 UTC
Last modified: 30 Jun 2013, 17:31:08 UTC

IIRC, there was some discussion here a few years back about using ASIC's and/or FPGA's as dedicated SAH crunchers.

However, acquiring the hardware and developing the software to run BOINC, as well as any projects you'd want to crunch wouldn't be cheap, and since there ain't no cake to be had for the effort and cost in the BOINC world one would need to have pretty deep pockets with no place better to put it. ;-)

OTOH, it would be a pretty cool and interesting project to sink ones' teeth into. :-D

Brian
Send message
Joined: 20 Apr 13
Posts: 4
Credit: 644,840
RAC: 0
United States
Message 1386202 - Posted: 30 Jun 2013, 20:22:00 UTC

I understand your argument. However, in respect to the "cake", getting larger numbers up on the board are all the value (sweetener) some people need. With all of the BOINC projects, what's in it for everyone? All I see is points. I mean there are projects like Rosetta that may give you some credit for crunching a solution, but no monetary prize. Yet, we still buy hardware and spend money on electricity to get higher up on the board!

That being said, if a Kickstarter project was started to fund these chips then I for one would invest. I would also buy extra chips outside of Kickstarter to run a rig.

My description would be something like using a Raspberry Pi to serve a box with the "SETI Chipsets" in it. The Pi or other cheep microcomputer could then manage the setup and report back the results. The chipsets could do most of the crunching, saving electricity and money for those who get enjoyment out of seeing their numbers rise.

Alinator
Volunteer tester
Send message
Joined: 19 Apr 05
Posts: 4178
Credit: 4,647,982
RAC: 0
United States
Message 1386208 - Posted: 30 Jun 2013, 20:40:07 UTC - in response to Message 1386202.
Last modified: 30 Jun 2013, 20:40:32 UTC

Mike Hewson posted this last November and an update this morning.

Looks like there's potential there, especially if they meet their pricing goals. ;-)

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 937
Credit: 22,552,830
RAC: 89,868
Australia
Message 1386233 - Posted: 30 Jun 2013, 22:30:12 UTC - in response to Message 1386208.

Mike Hewson posted this last November and an update this morning.

Looks like there's potential there, especially if they meet their pricing goals. ;-)


Apart from the Parallella there was an attempt to port SETI to FPGAs which is mentioned over at www.fpgaathome.org. I say attempt because there hasn't been an update since Jan 2013.
____________
BOINC blog

Profile James SotherdenProject donor
Avatar
Send message
Joined: 16 May 99
Posts: 8709
Credit: 33,211,665
RAC: 57,893
United States
Message 1386252 - Posted: 30 Jun 2013, 23:57:13 UTC

What are you asking for. That Seti@Home give bitcoins? Theres enough wailing and nashing of teeth now that V7 has arrived. You aint heard nothing yet if you start talking real cash.
____________

Old James

Brian
Send message
Joined: 20 Apr 13
Posts: 4
Credit: 644,840
RAC: 0
United States
Message 1386309 - Posted: 1 Jul 2013, 4:14:40 UTC - in response to Message 1386252.

No, no not giving BitCoins. I was saying that BitCoin mining now uses special chips designed for mining the coins. They are much faster than other methods. A mining rig can have unlimited BitCoin mining chips set up with one computer supplying data to them.

A rig of lets say 64 special chips designed for SETI could replace a bunch of computers, thus lowering computer costs. You would be able to add more boards with more chips for larger number crunching.

spitfire_mk_2
Avatar
Send message
Joined: 14 Apr 00
Posts: 443
Credit: 12,336,430
RAC: 6,693
United States
Message 1386311 - Posted: 1 Jul 2013, 4:24:02 UTC - in response to Message 1386309.


A rig of lets say 64 special chips designed for SETI could replace a bunch of computers, thus lowering computer costs. You would be able to add more boards with more chips for larger number crunching.

I think that would complete the circle. The original project started exactly because there was no money for hardware to crunching all the data. So the project team wrote the software and gave it away to the rest of us to crunch the data for them.
____________

Profile James SotherdenProject donor
Avatar
Send message
Joined: 16 May 99
Posts: 8709
Credit: 33,211,665
RAC: 57,893
United States
Message 1386314 - Posted: 1 Jul 2013, 5:02:10 UTC

Then I see the end where Seti@Home says we dont need you anymore.OR that so many do the mining bit that the pipes from the colocation gets just as jammed up as it was from the lab.
____________

Old James

bill
Send message
Joined: 16 Jun 99
Posts: 859
Credit: 23,006,139
RAC: 24,063
United States
Message 1386315 - Posted: 1 Jul 2013, 5:05:40 UTC - in response to Message 1386314.

That would only happen if the limits were removed.

Profile James SotherdenProject donor
Avatar
Send message
Joined: 16 May 99
Posts: 8709
Credit: 33,211,665
RAC: 57,893
United States
Message 1386325 - Posted: 1 Jul 2013, 5:56:06 UTC - in response to Message 1386315.
Last modified: 1 Jul 2013, 5:57:50 UTC

That would only happen if the limits were removed.


Explain why only if limits were removed. We had limits before the lab went to the relocation. The pipes were clogged up beyond repair. If every cruncher in here went to the magic mining chips and started to really pound out the work units, You dont think the pipes wouldnt get full?
Probally what would happen is we would outstrip the ability of the lab to make work to crunch. Then we'd hear all the crying about no work units and my rac is falling BS all over again.
____________

Old James

bill
Send message
Joined: 16 Jun 99
Posts: 859
Credit: 23,006,139
RAC: 24,063
United States
Message 1386329 - Posted: 1 Jul 2013, 6:09:36 UTC - in response to Message 1386325.

That would only happen if the limits were removed.


Explain why only if limits were removed. We had limits before the lab went to the relocation. The pipes were clogged up beyond repair. If every cruncher in here went to the magic mining chips and started to really pound out the work units, You dont think the pipes wouldnt get full?
Probally what would happen is we would outstrip the ability of the lab to make work to crunch. Then we'd hear all the crying about no work units and my rac is falling BS all over again.


I thought there were daily/max limits in place.

"Probally what would happen is we would outstrip the ability of the lab to make work to crunch."

Another kind of limit.

Profile James SotherdenProject donor
Avatar
Send message
Joined: 16 May 99
Posts: 8709
Credit: 33,211,665
RAC: 57,893
United States
Message 1386335 - Posted: 1 Jul 2013, 6:25:43 UTC - in response to Message 1386329.

That would only happen if the limits were removed.


Explain why only if limits were removed. We had limits before the lab went to the relocation. The pipes were clogged up beyond repair. If every cruncher in here went to the magic mining chips and started to really pound out the work units, You dont think the pipes wouldnt get full?
Probally what would happen is we would outstrip the ability of the lab to make work to crunch. Then we'd hear all the crying about no work units and my rac is falling BS all over again.


I thought there were daily/max limits in place.

"Probally what would happen is we would outstrip the ability of the lab to make work to crunch."

Another kind of limit.

There are still limits in place as far as I know. BUT if you crunch faster you use more work units faster. so my computer for the sake of argument is liminted to 100 for CPU and 100 for GPU and lets say I can crunch 100 a day for each. So i do 200 tasks per day. So if x computer now crunches 200 wu a day and the magic mining machine can turn that into lets say 500 a day I have more than doubled my crunching power. The limits are still in place for my cache size but ive have now manged to crunch that over twice a day. Now imagine that evertyone wants to get on the magic mining work unit gravy train.
I can see clogged pipes and waiting for work.

____________

Old James

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 937
Credit: 22,552,830
RAC: 89,868
Australia
Message 1386361 - Posted: 1 Jul 2013, 9:37:02 UTC
Last modified: 1 Jul 2013, 9:38:05 UTC

Really does it matter about limits? If they can't port the app to run on an FPGA then they certainly aren't going to be making a chip to run the app. All this talk about limits and bitcoin mining is hypothetical and irrelevant to the subject.
____________
BOINC blog

Profile ML1
Volunteer tester
Send message
Joined: 25 Nov 01
Posts: 8377
Credit: 4,106,547
RAC: 1,042
United Kingdom
Message 1386413 - Posted: 1 Jul 2013, 12:44:51 UTC - in response to Message 1386165.
Last modified: 1 Jul 2013, 12:47:36 UTC

IIRC, there was some discussion here a few years back about using ASIC's and/or FPGA's as dedicated SAH crunchers. ...

And then nVidia came along with their CUDA enabled GPUs...

And the rest is history!


Aside: Note that Berkerly already use FPGAs in some of their dedicated search hardware for such as SERENDIP. However, the s@h application performs a much deeper search and is perhaps better suited for more general purpose compute hardware to process. I'm sure the Lunatics have deservedly won their eponymous title over their years of dedicated knightly hacking to speed s@h towards the holy grail!!

I keep an eye on suitably large FPGA hardware and it always looks to be horribly expensive compared to just getting an off-the-shelf nVidia GPU.

Parallela + Epiphany is looking interesting for low-power/efficient computing. However, that is still a long way away from the top end nVidia GPUs...


Happy super-fast GPU crunchin',
Martin
____________
See new freedom: Mageia4
Linux Voice See & try out your OS Freedom!
The Future is what We make IT (GPLv3)

Alinator
Volunteer tester
Send message
Joined: 19 Apr 05
Posts: 4178
Credit: 4,647,982
RAC: 0
United States
Message 1386422 - Posted: 1 Jul 2013, 13:27:40 UTC - in response to Message 1386413.
Last modified: 1 Jul 2013, 13:37:30 UTC

IIRC, there was some discussion here a few years back about using ASIC's and/or FPGA's as dedicated SAH crunchers. ...

And then nVidia came along with their CUDA enabled GPUs...

And the rest is history!


Aside: Note that Berkerly already use FPGAs in some of their dedicated search hardware for such as SERENDIP. However, the s@h application performs a much deeper search and is perhaps better suited for more general purpose compute hardware to process. I'm sure the Lunatics have deservedly won their eponymous title over their years of dedicated knightly hacking to speed s@h towards the holy grail!!

I keep an eye on suitably large FPGA hardware and it always looks to be horribly expensive compared to just getting an off-the-shelf nVidia GPU.

Parallela + Epiphany is looking interesting for low-power/efficient computing. However, that is still a long way away from the top end nVidia GPUs...


Happy super-fast GPU crunchin',
Martin


That was it. Here's the link to it.

In our case, this baby is what does the preprocessing of the data coming off the Arecibo telescope (and others) before it gets sent to us to chew on.

Needless to say, this a very specialized and demanding application. ;-)

In any event, in the BOINC context I agree. It will be hard to displace the GPGPU's. Given they can put the pretty pictures on peoples' screen as well, it makes it fairly easy to get them out in the field in large numbers at a reasonable cost.

<edit> Here's another interesting link which came up in the google search. Looks like somebody has been playing around with idea.

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 937
Credit: 22,552,830
RAC: 89,868
Australia
Message 1386625 - Posted: 1 Jul 2013, 22:22:10 UTC - in response to Message 1386422.

<edit> Here's another interesting link which came up in the google search. Looks like somebody has been playing around with idea.


That's the one I mentioned before only I didn't link to it. Could have been interesting if they got it to work.

I am awaiting the Parallella to see how they go, although the small number of cores will presumably be one of its limitations (when compared with a GPGPU).

____________
BOINC blog

Profile Raistmer
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 16 Jun 01
Posts: 3397
Credit: 46,363,824
RAC: 9,877
Russia
Message 1386634 - Posted: 1 Jul 2013, 22:46:00 UTC - in response to Message 1386625.

64-core config is actually much more than most on current GPUs have, not less.
Each core corresponds 1 CU of GPU.
Their relative performance is another topic though.

____________

Brian
Send message
Joined: 20 Apr 13
Posts: 4
Credit: 644,840
RAC: 0
United States
Message 1387483 - Posted: 4 Jul 2013, 2:34:44 UTC

When I first brought up the possibility of ASIC chips, I was not so interested in speed as I was cost of ownership vs. number crunching. If we look at the price of a pc with all of its hardware to crunch and then add more and more expensive hardware to crunch then to some, it becomes cost ineffective to crunch. Not to mention the price of electricity a pc and it's associated hardware uses.

If we have chips that are low power and low cost, then we could build crunching rigs as we go. If I was not sending less on electricity, then I could add some more chips to a rig as I went. It would increase the crunching.

Now, I know about the Parallella, so do you think something like it could work?

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 937
Credit: 22,552,830
RAC: 89,868
Australia
Message 1387576 - Posted: 4 Jul 2013, 10:33:57 UTC - in response to Message 1387483.
Last modified: 4 Jul 2013, 10:35:33 UTC

When I first brought up the possibility of ASIC chips, I was not so interested in speed as I was cost of ownership vs. number crunching. If we look at the price of a pc with all of its hardware to crunch and then add more and more expensive hardware to crunch then to some, it becomes cost ineffective to crunch. Not to mention the price of electricity a pc and it's associated hardware uses.

If we have chips that are low power and low cost, then we could build crunching rigs as we go. If I was not sending less on electricity, then I could add some more chips to a rig as I went. It would increase the crunching.

Now, I know about the Parallella, so do you think something like it could work?


Sure it will work. We already can use the Raspberry Pi. The Parallella is a dual core ARM A9 so it's like two Pi's. The unknown is how well the Epiphany chip will run a SETI OpenCL app.
____________
BOINC blog

1 · 2 · Next

Message boards : Number crunching : SETI ASIC Chips

Copyright © 2014 University of California