GPU to 72 status... - Page 218

NBtarheel_33 · 2013-09-18, 08:46

Quote:

Originally Posted by James Heinrich

1 GHz-day = 5.075 P-90 years (original GIMPS work measurement). Assuming a single stock GTX 580 processes 433.2 GHz-days/day (running mfaktc), that single GPU will do what took George's original Pentium 90 an entire year of work... every 17 minutes.

Other way around: 1 P-90 year = 5.075 GHz-days.

So 433.2 GHz-days/day = 85.36 P-90 years/day = 1 P-90 year of work completed every 1/85.36 * 1440 = 16.87 minutes.

flashjh · 2013-09-19, 01:19

So assume the same level of progress in computers, 20 years from now, how long would it take to do all the work G72 has done?

TheMawn · 2013-09-19, 01:48

GPU72 has a yearly rate of 2,500,000 GHz-Days as of now. Let's forget the 1,000,000 GHz-Days of saved work. We could factor it in later quite easily.

The thing to keep in mind is we're looking at GPU vs CPU which isn't exactly fair as the GPU architecture is extremely well suited to some of the computing we're doing. GPU's aren't exactly an improvement on a CPU.

Considering the improvements CPU-wise, where a 4.6 GHz Core i5-3570k is worth 12 GHz of Core 2, which in turn is 2.4 P90-years, or 876 P90-days. A single core of mine is 876 times faster than the P90. 3504 times faster if we count the increase in core count, but let us not.

If the same level of increase can be attributed to a GTX 580, then a 876-fold increase over 430 GHz-Days per day gives 376,780 GHz-Days per day. Divide the yearly GPU72 production by this value and you get 6.635.

In other words, a high end (compute-wise the GTX 580 is probably close to a GTX 780) GPU in twenty years will likely have the capacity to beat the entire GPU72 team's yearly output in one week.

kracker · 2013-09-19, 02:06

Quote:

Originally Posted by TheMawn

GPU72 has a yearly rate of 2,500,000 GHz-Days as of now. Let's forget the 1,000,000 GHz-Days of saved work. We could factor it in later quite easily.

The thing to keep in mind is we're looking at GPU vs CPU which isn't exactly fair as the GPU architecture is extremely well suited to some of the computing we're doing. GPU's aren't exactly an improvement on a CPU.

Considering the improvements CPU-wise, where a 4.6 GHz Core i5-3570k is worth 12 GHz of Core 2, which in turn is 2.4 P90-years, or 876 P90-days. A single core of mine is 876 times faster than the P90. 3504 times faster if we count the increase in core count, but let us not.

If the same level of increase can be attributed to a GTX 580, then a 876-fold increase over 430 GHz-Days per day gives 376,780 GHz-Days per day. Divide the yearly GPU72 production by this value and you get 6.635.

In other words, a high end (compute-wise the GTX 580 is probably close to a GTX 780) GPU in twenty years will likely have the capacity to beat the entire GPU72 team's yearly output in one week.

?

I didn't know there was a yearly done and saved on gpu72... I just see lifetime.

Calculations for the gpu72 team on Primenet seem right, but many aren't in the team.

ixfd64 · 2013-09-19, 02:19

I wonder if Curtis Cooper knows about this sub-project. Surely there must be GPU-enabled workstations at UCM?

blahpy · 2013-09-19, 05:18

Quote:

Originally Posted by ixfd64

I wonder if Curtis Cooper knows about this sub-project. Surely there must be GPU-enabled workstations at UCM?

Running mfaktx at higher bit levels while computer is in use means lag, though...

VictordeHolland · 2013-09-19, 20:36

Quote:

Originally Posted by TheMawn

GPU72 has a yearly rate of 2,500,000 GHz-Days as of now. Let's forget the 1,000,000 GHz-Days of saved work. We could factor it in later quite easily.

The thing to keep in mind is we're looking at GPU vs CPU which isn't exactly fair as the GPU architecture is extremely well suited to some of the computing we're doing. GPU's aren't exactly an improvement on a CPU.

Considering the improvements CPU-wise, where a 4.6 GHz Core i5-3570k is worth 12 GHz of Core 2, which in turn is 2.4 P90-years, or 876 P90-days. A single core of mine is 876 times faster than the P90. 3504 times faster if we count the increase in core count, but let us not.

If the same level of increase can be attributed to a GTX 580, then a 876-fold increase over 430 GHz-Days per day gives 376,780 GHz-Days per day. Divide the yearly GPU72 production by this value and you get 6.635.

In other words, a high end (compute-wise the GTX 580 is probably close to a GTX 780) GPU in twenty years will likely have the capacity to beat the entire GPU72 team's yearly output in one week.

That is PROBABLY not going to happen, transistors can't get that small. We are now at 28nm, which is about 100-150 atoms. For GPUs to get 876 times faster, you need either 876x higher clock speeds (no way that is going to happen). Or 876x amount of transistors, and in order to pack those transistors on a GPU, they need to be very small, and we will have huge difficulties past 10nm (due to quantum tunnelling).

chalsall · 2013-09-19, 20:58

Quote:

Originally Posted by VictordeHolland

That is not going to happen....

How can you say that with such certainty?

VictordeHolland · 2013-09-19, 21:34

Quote:

Originally Posted by chalsall

How can you say that with such certainty?

Bought a timemachine from Ebay, travelled to the year 2033, asked the robots (who wiped out mankind): how many transistors are there in your body, asked them for the next Mersenne Primes (which they wouldn't give to me), spoke to a half a dozen aliens, travelled back into time, woke up with a headache.

chalsall · 2013-09-19, 22:36

Quote:

Originally Posted by VictordeHolland

Bought a timemachine from Ebay, travelled to the year 2033, asked the robots (who wiped out mankind): how many transistors are there in your body, asked them for the next Mersenne Primes (which they wouldn't give to me), spoke to a half a dozen aliens, travelled back into time, woke up with a headache.

Actually, compared to some of the answers I've been receiving recently, this is actually passable... 8-)

kracker · 2013-09-19, 23:09

Quote:

Originally Posted by chalsall

Actually, compared to some of the answers I've been receiving recently, this is actually passable... 8-)

+1

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Status	Primeinator	Operation Billion Digits	5	2011-12-06 02:35
62 bit status	1997rj7	Lone Mersenne Hunters	27	2008-09-29 13:52
OBD Status	Uncwilly	Operation Billion Digits	22	2005-10-25 14:05
1-2M LLR status	paulunderwood	3*2^n-1 Search	2	2005-03-13 17:03
Status of 26.0M - 26.5M	1997rj7	Lone Mersenne Hunters	25	2004-06-18 16:46

2013-09-19, 01:19	#2389
flashjh "Jerry" Nov 2011 Vancouver, WA 1123₁₀ Posts	So assume the same level of progress in computers, 20 years from now, how long would it take to do all the work G72 has done?

2013-09-19, 01:48	#2390
TheMawn May 2013 East. Always East. 11×157 Posts	GPU72 has a yearly rate of 2,500,000 GHz-Days as of now. Let's forget the 1,000,000 GHz-Days of saved work. We could factor it in later quite easily. The thing to keep in mind is we're looking at GPU vs CPU which isn't exactly fair as the GPU architecture is extremely well suited to some of the computing we're doing. GPU's aren't exactly an improvement on a CPU. Considering the improvements CPU-wise, where a 4.6 GHz Core i5-3570k is worth 12 GHz of Core 2, which in turn is 2.4 P90-years, or 876 P90-days. A single core of mine is 876 times faster than the P90. 3504 times faster if we count the increase in core count, but let us not. If the same level of increase can be attributed to a GTX 580, then a 876-fold increase over 430 GHz-Days per day gives 376,780 GHz-Days per day. Divide the yearly GPU72 production by this value and you get 6.635. In other words, a high end (compute-wise the GTX 580 is probably close to a GTX 780) GPU in twenty years will likely have the capacity to beat the entire GPU72 team's yearly output in one week.

2013-09-19, 02:19	#2392
ixfd64 Bemusing Prompter "Danny" Dec 2002 California 95B₁₆ Posts	I wonder if Curtis Cooper knows about this sub-project. Surely there must be GPU-enabled workstations at UCM?