Chips too fast for programmers?

[Bernie64]

This is from a cnnfn article.
Stanford professor!!!!!!
Didn't find his name in search.
http://money.cnn.com/2008/08/13/technol ... 2008081405

[7im]

The PPL has been in the news a lot the last few months, ever since the PPL opened in May. Borderline sensationalist journalism. OMG! My computers are going to slow down because "they" can't program the software to run on multi-cored computers. Big deal. They still run just fine on a single core of that multi core system.

However, the PPL is still needed to help develop methods to take fuller advantage of that multi-core future.

http://ppl.stanford.edu/wiki/index.php/ ... Laboratory

[divery4eyes]

did you read the section of the article where he states that sales might be hurt by the future multi core processors might actually run processes slower
correct me if i am wrong, but last time i checked processors get more efficient not less. and another thing, he states that multi core processors dont perform any faster than their single core brethren? what?
a P4 running just as fast as a C2D

whomever wrote that article should do their research

[7im]

What they probably have done is fallen in to the GHz speed trap. Intel pushed the "GHz is king" in performance for so long that most average users think a 2.8 GHz P4 will do things faster than a 2.1 GHz Allendale. (Okay, there are a few select things the P4 will still do faster, but for the most part...)

And Intel is pushing the GHz speed down even more as it lumps more cores together on the same chip to keep the heat and power down, but Intel is also increasing the processor efficiency at the same time. So a 1.6 Nehalem is probalby about a 2.4 Conroe, which is about a 3.2 P4.

[Evil Penguin]

I think you are giving the P4 too much credit, 7im.
A Conroe/Penryn would stomp all over Prescott like there was no tomorrow.
But I get where you are going with this...

[DjSoulshot]

Lemme copypaste a mail i just sent him:

Hey Michael

First of all thanks for a good article, it's a serious look at a problem that's all too real for a lot of developers to face. However i have to say that you are incorrect in the assumption that anything will go slower at all. People who are claiming this do not understand the significant boosts that single cores have gone through in the last few years. Even though their "speed", how fast they move, as denoted by the megahertz denominator, has actually decreased, the chips themselves are a lot quicker. While chips might only be running at 2400mhz today compared to the late pentium chips in the mid-3000mhz categories, they do a lot more work per hertz. This often amounts to an increase in actualy work being done per core of almost 100% compared to that time, making the 2400mhz chip run equally fast as a 4800mhz chip would have years ago. This is also the reason why Intel have been able to make slow chips such as the Atom chip, being used by the biggest computer companies in the world right now to create cheap computers. The Atom chip runs extremely slowly, but is often more than enough to get really simple everyday tasks done as quickly as you would need them. This would have been unthinkable a few years ago, and drives amongst other things the new version of the Asus EEE PC.

The real issue really is that things are just not getting faster as quickly as they should - a lot of programs are, as you nicely put, only using one chunk of processing power. But this does not have to equal a problem, as the programs that do not fully utilize multicore processors often really don't need that power. Scientific applications, computergames, video decoding and other heavy tasks are already using as many cores as you can have in a computer effectively.

Hopefully you'll be able to add some of this to the article or maybe even do a followup article at another point, cause i have been amazed at the amount of work i'm able to achieve compared to a few years ago. Also the initative by Mr. Olukotun's stanford collegues known as folding@home has already proven how effectively it can use a variety of different computer processing options, including both single and multicore processors, and exciting new technologies as the very effective Playstation 3 chips based on the cell architecture and the massively parallel computer graphics cards.

Thanks for the time

Niels Madsen

And Intel really isn't pushing ghz down as you say, there are some fantastically quick quadcores already, and the new dualcores are reasily hitting 4+ ghz overclocked, and shipping at 3+ ghz stock speeds. This will happen with quad cores over time too, and the octocores and xxx cores to come.

[DjSoulshot]

What they probably have done is fallen in to the GHz speed trap. Intel pushed the "GHz is king" in performance for so long that most average users think a 2.8 GHz P4 will do things faster than a 2.1 GHz Allendale. (Okay, there are a few select things the P4 will still do faster, but for the most part...)

And Intel is pushing the GHz speed down even more as it lumps more cores together on the same chip to keep the heat and power down, but Intel is also increasing the processor efficiency at the same time. So a 1.6 Nehalem is probalby about a 2.4 Conroe, which is about a 3.2 P4.

Actually i almost think it's a 1.6 Penryn beating a 3.2 P4 by now, i believe the C2D chips were a good 25-30% more effective than the X2 chips, which were in turn 50% more effective than the late P4 chips. Penryn is even more effective than that. Nehalem, if true to it's promises, will keep this going, at which point they could ship a fantastic laptop with a 2ghz chip running on 32nm on sandy bridge with some fantastic batterytime potential, gonna be a fun couple of years the next ones, SSDs, Nehalem, DDR3, GDDR5 and all

[Beberg]

Well I said this over a year ago...

http://www.interesting-people.org/archi ... 00024.html

It's barely even difficult anymore, and I think OpenCL will make is easy enough for anyone (Brook and CUDA are still rather difficult).