Japanese Super Computer - 8.2 petaflops per second

Gabi

legendary

Activity: 1148

Merit: 1008

If you want to walk on water, get out of the boat

Quote from: Jack of Diamonds on June 20, 2011, 01:21:19 PM

Quote from: Gabi on June 20, 2011, 10:47:17 AM

Not my fault they bough nvidia cards Cheesy

Nvidia cards are superior for distributed computing in general. Take F@h protein folding, or space research such as SETI for example;
Nvidia cards fetch 4-10x more performance than a comparable ATI/AMD card.

Also most supercomputers in the world use Nvidia cards or ASICs.

http://www.top500.org/

You named the 2 projects where the ATI client is a fail: Folding and SETI. Folding ATI client is so much fail, and the staff state this too, it's a known fact it's not optimized. As for SETI, well, same apply, it use mostly the CPU for the computations.

Let's speak about Milkyway, Collatz Conjecture, DNETC (and Moo! Wrapper), where both the nvidia and ati client are fully optimized. ATI simply kickass nvidia in all these projects, especially in Milkyway where double precision is used.

minerX

newbie

Activity: 56

Merit: 0

Quote from: Jack of Diamonds on June 20, 2011, 01:21:19 PM

Quote from: Gabi on June 20, 2011, 10:47:17 AM

Not my fault they bough nvidia cards Cheesy

Nvidia cards are superior for distributed computing in general. Take F@h protein folding, or space research such as SETI for example;
Nvidia cards fetch 4-10x more performance than a comparable ATI/AMD card.

Also most supercomputers in the world use Nvidia cards or ASICs.

http://www.top500.org/

They aren't actually fast then AMD. It's just no-one got around to writing a decent software program for the AMD cards. The v7 client has tried to fix this and they are performing 10x better then they were before. This just shows its not a hardware issue, its a software issue.

Whereas with bitcoin we can assume its a hardware issue. Guarantee if there was profit to be made someone would have figured something out to fix the nvidia GPUS. Whereas in the nonprofit sector there is a much less drive.

Jack of Diamonds

sr. member

Activity: 252

Merit: 251

Quote from: Gabi on June 20, 2011, 10:47:17 AM

Not my fault they bough nvidia cards Cheesy

Nvidia cards are superior for distributed computing in general. Take F@h protein folding, or space research such as SETI for example;
Nvidia cards fetch 4-10x more performance than a comparable ATI/AMD card.

Also most supercomputers in the world use Nvidia cards or ASICs.

http://www.top500.org/

Mousepotato

hero member

Activity: 896

Merit: 1000

Seal Cub Clubbing Club

Quote from: ||bit on June 20, 2011, 12:58:26 AM

I think that's about 10^12 times faster than an AMD 6990 Cool

Non OC'd maybe Tongue

Gabi

legendary

Activity: 1148

Merit: 1008

If you want to walk on water, get out of the boat

Not my fault they bough nvidia cards Cheesy

Jack of Diamonds

sr. member

Activity: 252

Merit: 251

Tianhe-1 has 7168 Nvidia S1070 tesla gpu's.

Their hashing power comes to around 1.1thash/s which is about 12.5% of the total bitcoin network capacity, or less depending on the timeframe.

It's ranked #1 on top500.org as well.

Gabi

legendary

Activity: 1148

Merit: 1008

If you want to walk on water, get out of the boat

You are right, my data was a bit old, the hash rate of bitcoin grew a lot in these days Cheesy

But well for 190€ you can buy a 6950 that can easily do 350mhash/s. You would need like 23285 of them to double the current network hashrate, and that's less than 5 millions of €. Sure, $ is worth less than €, but with 10 millions of $ you can get that.

So saying "omg bitcoin hashrate is higher than the whole top500 supercomputers together" is a bit nonsense, when the bitcoin hashrate is worth like only 10millions $ (i think those supercomputers are more expensive...)

nebiki

sr. member

Activity: 742

Merit: 250

Quote from: Gabi on June 20, 2011, 05:22:22 AM

Do you realize that the computing power of bitcoin exceeds that of the supercomputers because they are mostly cpu based???

With a simple 5 millions $ investment you can buy enough GPUs to have more than 50% of the total mining power.

i thought with 5 mil $ you could get a few batches of fully customized asics. your statement sounds a bit exaggerated, because at 2$/Mhash/s you need like 16 mil $, which does not include boards/psus/location etc. i'm not too sure if a customized machine which supports 2000 gpus can be build for 5 mil $. probably not. so i'd say you need at least 10-15 mil $. but that would not be as fun as a machine pulling 10^12 the amount of the current network speed. that would still (even if it won't come true) be the funniest thing for bitcoin miners. imagine 10^12 * difficulty after only a couple minutes/hours. we'd need years and years to get the difficulty back to what we can handle with our rigs.

Gabi

legendary

Activity: 1148

Merit: 1008

If you want to walk on water, get out of the boat

Do you realize that the computing power of bitcoin exceeds that of the supercomputers because they are mostly cpu based???

With a simple 5 millions $ investment you can buy enough GPUs to have more than 50% of the total mining power.

SgtSpike

legendary

Activity: 1400

Merit: 1005

Quote from: nebiki on June 20, 2011, 02:06:28 AM

this is bullshit. even if there was a super-fast-at-hashing computer, it could only get a maximum of 2016 blocks, which is about 2 mil $, before making the difficulty skyrocket. and you know what happens if you want to cash out on mt gox with a lot of money at once.

edit: actually this is quite interesting. what if some super computer decides to join the network for a few cycles? every investment made in mining would be dead. oh dear, i so want this to happen. oh, and actually, i was wrong. the difficulty would only increase by 400% max at each cycle, right? but then again, the network atm is already that much faster than any super computer available, i guess.

This.

The total hashing power of the network far exceeds any of the supercomputers out there, so it's really irrelevant.

nebiki

sr. member

Activity: 742

Merit: 250

this is bullshit. even if there was a super-fast-at-hashing computer, it could only get a maximum of 2016 blocks, which is about 2 mil $, before making the difficulty skyrocket. and you know what happens if you want to cash out on mt gox with a lot of money at once.

edit: actually this is quite interesting. what if some super computer decides to join the network for a few cycles? every investment made in mining would be dead. oh dear, i so want this to happen. oh, and actually, i was wrong. the difficulty would only increase by 400% max at each cycle, right? but then again, the network atm is already that much faster than any super computer available, i guess.

Auspician

full member

Activity: 126

Merit: 103

Interestingly, based upon the power usage given, the K computer does eat up about 32 cents worth of electricity *per second*. Still, I'd be more than happy to go in to rent a few minutes worth of processing power.

Naturally, after a day or so of this computer mining bitcoins, all other miners would likely be priced out of the market.

||bit

hero member

Activity: 924

Merit: 506

Quote from: hugolp on June 20, 2011, 01:06:41 AM

Quote from: Synaesthesia on June 20, 2011, 01:01:53 AM

Bitcoin mining is all about integer operations per second (IOPS) not floating point operations per second (FLOPS). But yeah they could probably still make a couple BTC with that thing!

I bet that computer is not very profitable in regard to consumption MHas/J though.

If the guesstimate I posted was even close to correct, I'd think it would have to be profitable. Unless, of course, they spend more than $150 million per second on power. Tongue

hugolp

legendary

Activity: 1148

Merit: 1001

Radix-The Decentralized Finance Protocol

Quote from: Synaesthesia on June 20, 2011, 01:01:53 AM

Bitcoin mining is all about integer operations per second (IOPS) not floating point operations per second (FLOPS). But yeah they could probably still make a couple BTC with that thing!

I bet that computer is not very profitable in regard to consumption MHas/J though.

||bit

hero member

Activity: 924

Merit: 506

Quote from: SgtSpike on June 20, 2011, 01:00:25 AM

I thought the entire bitcoin network, if it actually used petaflops, was around 26 PF's. So I can't see $150M profit in the period of a second.

Though I suppose it is unrealistic somehow, I'm just basing this on my understanding of the speed of an AMD 6990 [per AMD's specs below]. No other factors were considered. It seems to be between about 1 to 5 terraflops/second:

■Default (BIOS1)
◦Up to 830MHz Engine Clock
◦5.10 TFLOPs Single Precision compute power
◦1.27 TFLOPs Double Precision Compute Power
■Overclocked (BIOS2)
◦Up to 880MHz Engine Clock
◦5.40 TFLOPs Single Precision compute power o 1.37 TFLOPs Double Precision Compute Power

http://www.amd.com/us/products/desktop/graphics/amd-radeon-hd-6000/hd-6990/Pages/amd-radeon-hd-6990-overview.aspx#3

Synaesthesia

sr. member

Activity: 546

Merit: 253

Bitcoin mining is all about integer operations per second (IOPS) not floating point operations per second (FLOPS). But yeah they could probably still make a couple BTC with that thing!

SgtSpike

legendary

Activity: 1400

Merit: 1005

I thought the entire bitcoin network, if it actually used petaflops, was around 26 PF's. So I can't see $150M profit in the period of a second.

||bit

hero member

Activity: 924

Merit: 506

"Japanese 'K' Computer Is Ranked Most Powerful"

http://www.post-gazette.com/pg/11171/1155010-96-0.stm

I think that's about 10^12 times faster than an AMD 6990 Cool

Disregarding power costs, Profitability would be somewhere around:

Revenue per day: 13,759,214,349,570.81 USD
Revenue per time frame[1 month]: 418,796,086,765,061.50 USD

Anyone want to pitch in to buy a mere one second of usage?... Revenue would be about $150,000,000

My math might need double checking. Tongue

Topic: Japanese Super Computer - 8.2 petaflops per second (Read 5000 times)