Bitcoin Forum
December 10, 2016, 10:57:37 AM *
News: Latest stable version of Bitcoin Core: 0.13.1  [Torrent].
 
   Home   Help Search Donate Login Register  
Pages: [1]
  Print  
Author Topic: How many PetaFLOPS are we at?  (Read 2231 times)
epi 1:10,000
Full Member
***
Offline Offline

Activity: 154


View Profile
June 16, 2011, 01:12:11 PM
 #1

I know bitcoin watch has a TeraFLOP readout but it is supposedly inaccurate based on standard mining hardware, so what is it based on ATI hardware specs?  What is it if you take ASIC computing into account?   The weekly increase in hashing power never ceases to tickle my technolust.
Advertised sites are not endorsed by the Bitcoin Forum. They may be unsafe, untrustworthy, or illegal in your jurisdiction. Advertise here.
1481367457
Hero Member
*
Offline Offline

Posts: 1481367457

View Profile Personal Message (Offline)

Ignore
1481367457
Reply with quote  #2

1481367457
Report to moderator
1481367457
Hero Member
*
Offline Offline

Posts: 1481367457

View Profile Personal Message (Offline)

Ignore
1481367457
Reply with quote  #2

1481367457
Report to moderator
1481367457
Hero Member
*
Offline Offline

Posts: 1481367457

View Profile Personal Message (Offline)

Ignore
1481367457
Reply with quote  #2

1481367457
Report to moderator
Sukrim
Legendary
*
Offline Offline

Activity: 1848


View Profile
June 16, 2011, 05:46:56 PM
 #2

You can by definition give no FLOPS rating for the "bitcoin cluster":
http://en.wikipedia.org/wiki/FLOPS

There are 0 floating point operations required for Bitcoin hashing, so the correct answer is: 0 FLOPS (also 0 PetaFLOPS).

https://bitfinex.com <-- leveraged trading of BTCUSD, LTCUSD and LTCBTC (long and short) - 10% discount on fees for the first 30 days with this refcode: x5K9YtL3Zb
Mail me at Bitmessage: BM-BbiHiVv5qh858ULsyRDtpRrG9WjXN3xf
epi 1:10,000
Full Member
***
Offline Offline

Activity: 154


View Profile
June 16, 2011, 05:52:06 PM
 #3

I understand that the hashing function doesn't need FLOPs, my question is what is the potential combined x86 FLOP capacity of the mining network?  Say, lets define it in terms of distributed deconvolution capacity,  weather modeling, folding, or whatever you like.   I want to create a new sentient life form dam it!
iya
Member
**
Offline Offline

Activity: 81


View Profile
June 16, 2011, 06:18:12 PM
 #4

The network currently does ~ 9 Thash/s.
Assuming it's entirely composed of Radeon HD 5970, which each do ~700 Mhash/s and have a theoretical peak performance of 4.64 TeraFLOPS, the cumulative capacity is ~60 PetaFLOPS (12857 Radeon HD 5970).
Sukrim
Legendary
*
Offline Offline

Activity: 1848


View Profile
June 16, 2011, 06:21:27 PM
 #5

I understand that the hashing function doesn't need FLOPs, my question is what is the potential combined x86 FLOP capacity of the mining network?  Say, lets define it in terms of distributed deconvolution capacity,  weather modeling, folding, or whatever you like.   I want to create a new sentient life form dam it!

Noone can know:
ASICs might not be able to do a single FLOP while nVidia GPUs might crank out a LOT of FLOPS, potentially even more than what's useful in mining.

It's like watching the tour de france and then estimating how fast these bikers would swim.
Some might be faster than others, some might drown, some might be used to a triathlon and be excellent swimmers! You can NOT however judge from thier biking abilities how good their swimming abilities will be.


The network currently does ~ 9 Thash/s.
Assuming it's entirely composed of Radeon HD 5970, which each do ~700 Mhash/s and have a theoretical peak performance of 4.64 TeraFLOPS, the cumulative capacity is ~60 PetaFLOPS (12857 Radeon HD 5970).
You forgot Giga, which is between Tera and Mega... - no, the system is NOT as tiny as ~13000 high-end GPUs!  Roll Eyes

https://bitfinex.com <-- leveraged trading of BTCUSD, LTCUSD and LTCBTC (long and short) - 10% discount on fees for the first 30 days with this refcode: x5K9YtL3Zb
Mail me at Bitmessage: BM-BbiHiVv5qh858ULsyRDtpRrG9WjXN3xf
iya
Member
**
Offline Offline

Activity: 81


View Profile
June 16, 2011, 06:26:06 PM
 #6

You forgot Giga, which is between Tera and Mega... - no, the system is NOT as tiny as ~13000 high-end GPUs!  Roll Eyes
I did?
bitcoinwatch.com estimates Network Hashrate at 111537 TeraFLOP/s, or about twice my value...
Sukrim
Legendary
*
Offline Offline

Activity: 1848


View Profile
June 16, 2011, 06:33:32 PM
 #7

Argh, too many zeroes...
Yes, you are of course right, currently ~13000 5970 GPUs would make up the Bitcoin network.

The stat "FLOPS" is still completely useless in this regard and actually misleading. Currently the network might be able to produce quite a few FLOPS too - once people switch to ASICS however, these days are over.

https://bitfinex.com <-- leveraged trading of BTCUSD, LTCUSD and LTCBTC (long and short) - 10% discount on fees for the first 30 days with this refcode: x5K9YtL3Zb
Mail me at Bitmessage: BM-BbiHiVv5qh858ULsyRDtpRrG9WjXN3xf
epi 1:10,000
Full Member
***
Offline Offline

Activity: 154


View Profile
June 16, 2011, 06:51:02 PM
 #8

If the estimated computing capacity of the human brain is 1 to 10 PetaFLOPS.  If a bunch of neurologist can make an estimated guess why not do it just for the fun of it...... You know for bragging rights.
Sukrim
Legendary
*
Offline Offline

Activity: 1848


View Profile
June 16, 2011, 08:33:54 PM
 #9

If the estimated computing capacity of the human brain is 1 to 10 PetaFLOPS.

Why do you not solve a few billion hashes yourself then with your brain but let your GPU do this 1000 times slower?  Roll Eyes

https://bitfinex.com <-- leveraged trading of BTCUSD, LTCUSD and LTCBTC (long and short) - 10% discount on fees for the first 30 days with this refcode: x5K9YtL3Zb
Mail me at Bitmessage: BM-BbiHiVv5qh858ULsyRDtpRrG9WjXN3xf
epi 1:10,000
Full Member
***
Offline Offline

Activity: 154


View Profile
June 16, 2011, 09:19:18 PM
 #10

I really don't care about hashing or bitcoins in the context of this thread only the potential scientific applications of a distributed GPU network, such as real time 2xHD deconvolution with FRET analysis and 3D protien reconstruction of a membrane bound protein with a binding partner protein, using billions of freeze fracture tunneling EM images.  Using Bitcoin and Folding as proof of distributed computing concept, I am sure such a distributed network could be used for other research projects too.
Gabi
Legendary
*
Offline Offline

Activity: 1050


View Profile
June 17, 2011, 04:56:35 PM
 #11

If the estimated computing capacity of the human brain is 1 to 10 PetaFLOPS.

Why do you not solve a few billion hashes yourself then with your brain but let your GPU do this 1000 times slower?  Roll Eyes

Remember that the brain has excellent image-analizying properties, how many PFLOPs are required to make a computer with the same image-analize power as a brain?  Cheesy
Sukrim
Legendary
*
Offline Offline

Activity: 1848


View Profile
June 17, 2011, 05:07:07 PM
 #12

I really don't care about hashing or bitcoins in the context of this thread only the potential scientific applications of a distributed GPU network, such as real time 2xHD deconvolution with FRET analysis and 3D protien reconstruction of a membrane bound protein with a binding partner protein, using billions of freeze fracture tunneling EM images.  Using Bitcoin and Folding as proof of distributed computing concept, I am sure such a distributed network could be used for other research projects too.

Again: If you need 20 PFLOPS just get 2-20 humans to do these calculations then... Roll Eyes

Not possible? That's exactly my point! Some things are easier to calculate for humans, some are easier to calculate for GPUs, some are easier to calculate by the chips used in simple pocket calculators!

You can NOT measure an integer algorithm in FLOPS just like you can't measure/estimate how many meters are in a glass of water! These are different units of measurement and they are NOT convertible. At all.

If you want to use the Bitcoin network power, just create an easy to use open source client and pay bitcoins for the proof of work that you receive with your client. BOINC already offers a readily available infrastructure for this. You can however NOT deduct from the x TH/s the Bitcoin cluster is doing atm. how fast your program would run then.

https://bitfinex.com <-- leveraged trading of BTCUSD, LTCUSD and LTCBTC (long and short) - 10% discount on fees for the first 30 days with this refcode: x5K9YtL3Zb
Mail me at Bitmessage: BM-BbiHiVv5qh858ULsyRDtpRrG9WjXN3xf
xalex
Jr. Member
*
Offline Offline

Activity: 43



View Profile
June 17, 2011, 05:30:20 PM
 #13

Argh, too many zeroes...
Yes, you are of course right, currently ~13000 5970 GPUs would make up the Bitcoin network.

The stat "FLOPS" is still completely useless in this regard and actually misleading. Currently the network might be able to produce quite a few FLOPS too - once people switch to ASICS however, these days are over.

"currently ~13000 5970 GPUs would make up the Bitcoin network"

That doesn't sound like much at all... :S
I'm running a 5970 and a 2x5870 (just over the 5970 speed).. this would mean I'm now providing about 1/6500 of the entire mining capabilities right now.
Simply said, there are 6499 other computer geeks on this planet who's friends think they are mad scientists with 3 rusty machines with overpowered GPU's in there basement keeping it at a nice 24 degrees Celcius? Tongue

Hey! I'm not one in a million! I'm one in 6500! Cheesy
epi 1:10,000
Full Member
***
Offline Offline

Activity: 154


View Profile
June 17, 2011, 11:08:44 PM
 #14

Learning neural computing units using mouse neurons are already in development.  I don't know much about the matlab program that runs our hardware but our computational needs are estimated in FLOPS.  It would be nice to know if we could expand our computing capacity w/ GPUs as Matlab is developing acceleration for it.  It would be nice if there was an OpenSource project but I don't know enought to be able contribute currently.  HPC time is expensive for a small academic lab at the moment.
Sukrim
Legendary
*
Offline Offline

Activity: 1848


View Profile
June 17, 2011, 11:54:09 PM
 #15

Switch to SAGE (which is open source - matlab isn't!) and code an OpenCL interface to it or pay someone else to do that for you. If you're academic, there are a lot of academic clusters worldwide that you could use. Bitcoin is NOT made for arbitrary computations and if your's are using floats, then sorry pal - in a few years from now there might not even be a lot of devices in the Bitcoin network that even know what a float is.

https://bitfinex.com <-- leveraged trading of BTCUSD, LTCUSD and LTCBTC (long and short) - 10% discount on fees for the first 30 days with this refcode: x5K9YtL3Zb
Mail me at Bitmessage: BM-BbiHiVv5qh858ULsyRDtpRrG9WjXN3xf
epi 1:10,000
Full Member
***
Offline Offline

Activity: 154


View Profile
June 18, 2011, 02:49:25 AM
 #16

Thx for the SAGE suggestion, I'll look into it.  Maybe we can get allot of cheap ATI cards when Miners stop using GPU's and move to ASIC.  The Academic clusters out there require a collaboration agreement (except for our Universities HPC center, which is a pain in the ass to use).
Pages: [1]
  Print  
 
Jump to:  

Sponsored by , a Bitcoin-accepting VPN.
Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!