Tech topic: historic change in computing performance growth

69 posts / 0 new
Last post
Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
Tech topic: historic change in computing performance growth

This topic is not directly related to the main topics of this site, but in my opinion it does have economic implications, so I am creating a thread here off main street. Also since I am one of them weekend warriors here it may take me a while to respond and it will be convenient to have it all in one place. I enjoyed the responses to my original postings and I hope to continue the discussion here.

This is what I posted: http://www.tfmetalsreport.com/comment/71447#comment-71447 and http://www.tfmetalsreport.com/comment/71506#comment-71506.

Edited by admin on 11/08/2014 - 06:27
Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
links

I will repost the links and highlights so people can have some background.

This is the story that reported the study: http://www.eetimes.com/electronics-news/4214473/Report--Computing-has-hit--power-wall-

This is the home page of the study report: http://books.nap.edu/catalog.php?record_id=12980 and by clicking into the first link on the table of contents http://www.nap.edu/openbook.php?record_id=12980&page=R1 you can read the entire report online using the buttons on page.

This is another story that says the approach of "just add more cores" will not work: http://www.eetimes.com/electronics-news/4215708/Panel-Wall-ahead-in-multicore-programming

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
Title page of the report

This is the title page of the report. Why did they end the title with a question mark?

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
Authors of the study

Here we can see who is saying this. Hardware and software, industry and academia, frienemies are coming together to say... that they are not sure what the future of computing performance will be like?  Hmmm...

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
More authors, reviewers, etc.

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
The growth trend changed in 2004

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
Why does this even matter at all

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
(No subject)

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
Relevance, multiple cores

OK so I posted page images from the report straight from their website, so hopefully people will be able to read about topic relevance (economy) and multicore approaches before commenting. If you have a different opinion than this expert committee please feel free to (re)state it below.

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
Re: Moore's Law good for 10 years easy - top Chipzilla boffin

http://www.tfmetalsreport.com/comment/71527#comment-71527

Moore's Law good for 10 years easy - top Chipzilla boffin

Seems to me, the faster they make the chips, the more they load up on eye candy, bells & whistles and background operations and the computer still has lousy response to the user. I'm on a QuadCore MacPro and I am still spending way too much time staring at the ol' hourglass and spinning wheel! ;-(

From "The Register" (U.K. Tech Site with a wry sense of humor... Check out the B.O.F.H. column. ;-)

http://www.theregister.co.uk/2011/10/14/intel_predicts_continued_life_fo...

"There's always physical limits to everything," Steve Pawlowski, senior fellow and head honcho on exascale research, told The Register at the European Research and Innovation Conference in Ireland.

"But you can always come up with clever ways... for example, there’s nothing that says I can’t take two dies and stack them on top of each other so I can grow Moore’s Law in the third dimension," he added.

Very interesting article, thank you for the link Bos. Intel says "we're good for a while" but they also are represented in the committee authoring the report shown above in this thread. So which one is it Intel, are you sure you're good for the next 10 years, or not? And what about after that?

cpnscarlet
cpnscarlet's picture
Offline
Joined: 08/07/2011
Hat Tips: 11082
Posts: 1996
Computing Power - my thoughts

When I was a teen (1976), an episode of "Space: 1999" has one character using a video camera that was about the size of two packs of cigarettes. Some geek detractors at the time claimed that a video camera could never be that small for various reasons. Of course we now know what was possible due to miniaturization. And my group of fellow computer geeks in 1976, in our wildest dreams, had no idea how far digital memory would go. But the one thing wrong with that 1976 hypothetical camera was that the LENS was too small. There are some things that can be made smaller, but physics demands certain aperture characteristics to get the resolution you want.

With computers, the physical limits appear to be two-fold - thermal control and clock cycle. In the past few years we have seen great advances in thermal transport to get the heat out of these chips and perhaps there are still advances to be made. However, the main reason clock speed has stalled is the physical fact that the clock cycle must be at the same state across the computer's processing topology (pc board). At ~2GHz, that can only occur across the dimension of a typical desk-top box. It seems to me that we need a basic architecture advance so that a single clock cycle may exist across a single topology at higher speeds. to do this, the clock cycle may have to be propagated from multiple sources on the board. This would require strict synchronization. Another solution may be asynchronous comm across a board where only the most basic parts of the processor share address and memory buses. Past these thoughts, I'm stumped.

I still think multi-core processors still have promise, but programmers need to get much smarter about how to use them.

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
Re: Computing Power - my thoughts

Thank you cpn for the thoughtful reply. IMO you are "getting warm" with your thoughts. The current architecture has stalled in its performance growth and the limits appear to be fundamental in nature. It was a good run, but it is over. The big question is if something better will be invented. Given that these experts are stumped, I think we can reasonably expect that a better solution would be significantly different than the currently conceived approaches. I would also expect it to emerge from new players unconcerned by the burdens and the mindset of maintaining their cash cows in good milking condition. It will take some major "thinking different" to overcome these walls.

If something better is not invented, we are looking at a wall for single thread speed. This limit has been partly sidestepped by cloud computing, but at a cost: the network stack of hardware and software (with their own limitations) becomes part of the computation architecture. Plus you need a number of boxes each consuming so much power, dissipating heat. This approach can result in higher speed but it looks to me like these extra costs will impose their own limits as well.

cpnscarlet
cpnscarlet's picture
Offline
Joined: 08/07/2011
Hat Tips: 11082
Posts: 1996
Just one more thought -

I think it needs to be said (and easily refuted for sure), how much more speed do we need? As long as data rates over the lines stay sufficiently high (and that can always be increased with more fiberoptic lines and switching stations), the average person can get pretty decent video real-time. With that, the needs of the "residential computer" are satisfied. Business, dealing mostly with numerical data and forms, has had their computing needs satisfied for some time. The only group I can see that NEEDS absolutely faster clock speeds are the theoretical mathematicians who are trying to model extremely large and extremely small physical phenomena.

The delays we see from time to time in our home computers are due to, IMHO, bad comm hardware/infrastructure and poorly written software. Truly efficient operating systems wouldn't hurt either (Bye Bye Vista).

In my former business of aerospace engineering, we needed far more computing power in the design phase (running high end CAD/CAM and simulators) than we ever needed in the air or in orbit. In general, the best desk-top boxes were finally giving us what we needed after 2008 (2GHz multi core).

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
The "need for speed" is a

The "need for speed" is a funny thing. A review of the history of computing reveals many cases where the availability of higher computer speeds enabled completely new application domains that were not feasible or even conceivable at lower speeds. So I don't buy into the notion of "we don't really need faster computers". (I recall exiledbear made a similar argument.) People seem to have always come up with innovative apps as faster machines became available. Also, faster cpu's kept the pc replacement cycle going. It becomes harder to keep selling the boxes over time if the buyers don't see the new ones are faster than their current ones. As an aside, I wonder if it's not a coincidence that a certain computer company sold their hardware business in 2005. Anyway, it seems to me that faster computers are economically advantageous. For example, I bet Hollywood would love to have new boxes that could significantly cut down the rendering time for their movies, and so on.

Today the only way to make tasks go faster with cheap normal boxes is to parallelize and/or break up the app as much as possible and have the ability to run it concurrently over a set of networked boxes - the cloud. Companies that were building their apps that way before Peak CPU came about were very lucky to be ready for the new reality.

As the report says, multi core and parallelism are the keys to speeding up microprocessors from now on. Two big problems now are: 1) The sequential stored program model of computation is not really suited to the expression of highly parallel programs. 2) Multiple cores do not make much of a difference unless there are about a thousand of them on one chip/board/box/whatever. I will find the reference for that and post it later. At this level the memory wall begins to affect performance, as all those cores vie for access to memory; not even considering the thermal stuff.

IMHO what we have here is the ultimate limit of the sequential stored program architecture that has been the basis for mainstream computing since the beginning. Any speedups other than faster switching technology will have to come from an architectural change. And to me this indicates that the arts of both hardware and software as practiced today will have to change significantly if any new architecture is to become adopted. I also see that it would be highly disruptive to some of the established players' businesses, so such a thing would likely emerge from a startup.

The funny thing is, so many computer companies that have tried to make "parallel computers" have gone out of business that they have been nicknamed the Dead Computer Society.

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
More peak pictures

http://www.nap.edu/openbook.php?record_id=12980&page=87

http://www.nap.edu/openbook.php?record_id=12980&page=88

http://www.nap.edu/openbook.php?record_id=12980&page=91

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
After many decades of

After many decades of dramatic exponential growth, single-processor performance is increasing at a much lower rate, and this situation is not expected to improve in the foreseeable future.

http://www.nap.edu/openbook.php?record_id=12980&page=90

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
The growth in the performance

The growth in the performance of computing systems—even if they are multiple-processor parallel systems—will become limited by power consumption within a decade.

http://www.nap.edu/openbook.php?record_id=12980&page=97

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
Software is limited only by

Software is limited only by human ambition and expectation. We will always find new algorithms, new applications, and new users.

http://www.nap.edu/openbook.php?record_id=12980&page=107

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
The problem is that much of

The problem is that much of the innovative software is sequential and is designed to execute on only one processor, whereas the previous chapters explained why all future computers will contain multiple processors. Thus, current programs will not run faster on successive generations of hardware.

http://www.nap.edu/openbook.php?record_id=12980&page=108

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
There is no known alternative

There is no known alternative to parallel systems for sustaining growth in computing performance; however, no compelling programming paradigms for general parallel systems have yet emerged.

http://www.nap.edu/openbook.php?record_id=12980&page=109

Thieving Corp.
Thieving Corp.'s picture
Offline
Joined: 07/14/2011
Hat Tips: 538
Posts: 147
...there is an ever-widening

...there is an ever-widening gap between processor and memory performance.

http://www.nap.edu/openbook.php?record_id=12980&page=112

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
Topic locked
Syndicate contentComments for "Tech topic: historic change in computing performance growth"