Processing Power Comparision for 50th anniversary of Moore's Law

[Napper198]

Original Article by experts-exchange.com

Found on Gizmodo

 

It's basically just a diagram which tries to visualize and compare old and modern supercomputers and other devices like phones, consoles and even smart watches. Nothing special but I find it interesting anyway.

 

Some direct comparisions 

A single Apple iPhone 5 has 2.7 times the processing power than the 1985 Cray-2 supercomputer.

 

 

The question remains, how far has computing power come over the last 60 years? Current supercomputers exceed commercially available products; however, it's still fun to compare what it would take to rival the titans (and Tianhes) of the industry. See the comparison of a Tianhe-2 Supercomputer and a Playstation 4 below.

 

 

I personally like this one because according to cnet.com Sony has sold 20.2 million PS4s by March 2015 so all units combined would have more than 1000 times the processing power of the fastest supercomputer right now. Also the Tianhe-2 costs about 390 million USD whereas 18400 PS4s would only burn a 9.2 million USD hole into your pocket. Also 24MW vs. 2.76MW (source Wikipedia 1, 2)

 

Full diagram ahead, warning long:

 

 

 

Edit: I did a bit of googling to determine if Moore was right:

 

So an Intel 8086 (father of the x86 architecture) introduced in 1978 has a transistor count of 29,000, a fabrication process of 3000nm (3µ), a die size of 33mm² and a whopping 4.77MHz clock speed.

Comparing it to Linus new favorite CPU, the E5-2699v3 introduced in 2014 has a transistor count of 5,560,000,000, a fabrication process of 22nm (0.022µ), a die size of 661mm² and runs up to 3.6GHz.

Separated by 36 years equals 2^18 or a multiplier of 262144 so 29,000x262144=7,602,176,000. Not quite Intel

 

(source: Wikipedia again)

[Suffokation]

This was an awesome read. Thank you.

[george357]

Nice post!

[azurezeed]

but, can the Tianhe-2 run minesweeper?

[NumLock21]

but, can the Tianhe-2 run minesweeper?

Probably, it can play "minesweeper".

[Guest Kloaked]

"I personally like this one because according to cnet.com Sony has sold 20.2 million PS4s by March 2015 so all units combined would have more than 1000 times the processing power of the fastest supercomputer right now. Also the Tianhe-2 costs about 390 million USD whereas 18400 PS4s would only burn a 9.2 million USD hole into your pocket. Also 24MW vs. 2.76MW"

 

I don't think that counts the higher quality components (such as power delivery possibly?) and the man hours put into making the Tianhe-2 a possibility (pre-development time and actually putting it together).

[kpreg]

That's Moore of an increase than I expected

[MrRavens]

Wow, and it just keeps on increasing as we speak. 

[MrRavens]

That's Moore of an increase than I expected

 

Really? REALLY!?

[Napper198]

I don't think that counts the higher quality components (such as power delivery possibly?) and the man hours put into making the Tianhe-2 a possibility (pre-development time and actually putting it together).

 

of course there is no ECC RAM, power delivery fail over or high speed inter connectivity and therefore it's impossible to compute tasks which can't be broken up into pieces. It's just a number game after all. 

[RexinOridle]

Thanks a lot. Nice post.

[Doughnutnator]

"I personally like this one because according to cnet.com Sony has sold 20.2 million PS4s by March 2015 so all units combined would have more than 1000 times the processing power of the fastest supercomputer right now. Also the Tianhe-2 costs about 390 million USD whereas 18400 PS4s would only burn a 9.2 million USD hole into your pocket. Also 24MW vs. 2.76MW"

I don't think that counts the higher quality components (such as power delivery possibly?) and the man hours put into making the Tianhe-2 a possibility (pre-development time and actually putting it together).

Maybe it has something to do with paralellism? Maybe PS4's can't be used for that kind of work.

I read somewhere about a supercomputer made of PS3's... Anyways I think that CPU's like the Xeon and Xeon Phi and GPU's like Nvidias Tesla are really designed and optimized to crunch numbers and make "supercomputer stuff", "potatostations" and "peasantboxes" may not have the potential to be raised at a superultrapeasant level

(Just joking, I love you console gamers)

[Guest Kloaked]

 

 

 

 

I should have worded what I said a different way. What I mean is, you can't just tally up the cost of all of those PS4's and say "all of these PS4's cost <x> less and potentially have <x> more processing power compared to the cost and performance of the Tianhe-2"

 

What isn't talked about are the engineer's time and the amount of work put into making the Tianhe-2, thus the cost of it.

[Trik'Stari]

So where does a 4690k and a 970 sit on that list? Well above the PS4 I assume. Just using that as a benchmark/reference, I'd honestly like to know how many times more powerful my PC is compared to the cray-2

[Opcode]

So where does a 4690k and a 970 sit on that list? Well above the PS4 I assume. Just using that as a benchmark/reference, I'd honestly like to know how many times more powerful my PC is compared to the cray-2

Not that hard with the Cray-2 at 1.9 GFLOPS. My HD 5870 is rated for 2.72 TFLOPS (2720 GFLOPS). That's 1431x faster than the Cray-2 supercomputer.

[Trik'Stari]

Not that hard with the Cray-2 at 1.9 GFLOPS. My HD 5870 is rated for 2.72 TFLOPS (2720 GFLOPS). That's 1431x faster than the Cray-2 supercomputer.

Dayum.....

[Opcode]

Dayum.....

GTX 970 is rated for a peak of 3.9 TFLOPS (3900 GFLOPS) that's 2052x faster than the Cray-2.

[BuckGup]

Just wait until Quantum computers destroy that chart and theory

[mackncheesiest]

Because of how crazy exponents are, it kind of depends on what year you use as your starting point. If, instead of 1978, you used 1975 (the year Moore supposedly revised his prediction from "doubling every year" to "doubling every two years"), things can turn out a bit differently.

 

The one processor Wikipedia lists as being introduced in 1975 is the MOS Technology 6502 with a whopping 3510 transistors.

Applying the same analysis you had there yields a predicted value of 3510*2^((2014 - 1975)/2) = 2,602,507,753. So from that point of view, REKT.

 

Since Moore was with Intel, though, and everyone else around that time was a bunch of chumps, let's stick with Intel processors. The two closest processors for Intel are the 8080 in 1974 and the 8085 in 1976, with 4500 and 6500 transistors, respectively. A simple average would estimate that a hypothetical 1975 processor would have ~5500 transistors. Using, again, that same analysis, you end up with 5500*2^(19.5) = 4,078,003,601 -- so, from that point of view, Moore's Law still basically REKT.

 

If you take into account that it wouldn't be exactly 5500 in 1975 because of the fact that it was actually growing exponentially and not linearly, it ends up a bit closer, but not quite. Assuming it was actually doubling every year from 1974 to 1975 (the original statement of the law), let's say there was a processor with 9000 transistors in 1975. Then, 9000*2^(19.5) = 6,673,097,802 -- so finally, we overshot by a bit.

 

The thing is, though, if you use 2015 instead of 2014 in some of these numbers (like the 5500 number up there), you end up with 5500*2^(20) = 5,767,168,000 transistors, which is pretty astoundingly close.

 

Going off of these results, though, I think we've successfully lower bounded it and upper bounded it, so I'd say that Moore's Law is, indeed BigTheta(2^(n/2)). I'm just sure Moore's glad he changed it from "double once a year" to "double once every two years" because, if he hadn't, we'd need to be on the order of 5*10^15 = 5,000,000,000,000,000 = 5 quadrillion transistors at this point, which is just a little off from where we are.

[Lethal Seraph]

I've read an article related to this before... Somebody used the PS3s hardware and reprogrammed it to do simulations like a supercomputer would do. I think it was some researchers from a university.

[mackncheesiest]

I've read an article related to this before... Somebody used the PS3s hardware and reprogrammed it to do simulations like a supercomputer would do. I think it was some researchers from a university.

Probably thinking of this airforce compute cluster or, if it was recently, this physicist doing black hole simulation things.

[Bouzoo]

The geek inside me enjoyed this

[Delusional Phil]

Very surprising that an S6 has the same processing power of an PS2 while the Apple Watch equals 2 Iphone4 ... That's amazing ... 

[LukaP]

 

I personally like this one because according to cnet.com Sony has sold 20.2 million PS4s by March 2015 so all units combined would have more than 1000 times the processing power of the fastest supercomputer right now. Also the Tianhe-2 costs about 390 million USD whereas 18400 PS4s would only burn a 9.2 million USD hole into your pocket. Also 24MW vs. 2.76MW (source Wikipedia 1, 2)

 

So an Intel 8086 (father of the x86 architecture) introduced in 1978 has a transistor count of 29,000, a fabrication process of 3000nm (3µ), a die size of 33mm² and a whopping 4.77MHz clock speed.

Comparing it to Linus new favorite CPU, the E5-2699v3 introduced in 2014 has a transistor count of 5,560,000,000, a fabrication process of 22nm (0.022µ), a die size of 661mm² and runs up to 3.6GHz.

Separated by 36 years equals 2^18 or a multiplier of 262144 so 29,000x262144=7,602,176,000. Not quite Intel

 

(source: Wikipedia again)

Tianhe is much smaller than a 20k PS4s would be. As well as uses a supercomputer grade cooling and power delivery, as well as has a team ready to service it 24/7. It also has a custom OS made for a supercomputer, and so on

 

the moore thing is there actually. it says it will double every 2 years. if it were to not be there, than it would have to be less than 3 800 000 000 transistors Physics only cares about the ballpark, not the exact figure and this is in the same power of 2 ballpark

[Sharkyx1]

Probably thinking of this airforce compute cluster or, if it was recently, this physicist doing black hole simulation things.

Did no one remember the big push for folding at home on ps3s? Before PC's used largely gpgpus the ps3 had a sizeable advantage and it's wide network of consoles working together are exactly what some people are thinking of in the thread