Intel's Dual-core strategy, 75% by end 2006

DigitumDei writes "Intel is moving ahead rapidly with their dual core chips, anticipating 75% of their chip sales to be dual core chips by the end of 2006. With AMD also starting to push their dual core solutions, how long until applications make full use of this. Some applications already make good use of multiple cpu's and of course multiple applications running at the same time instantly benifit. Yet the most cpu intensive applications for the average home machine, games, still mostly do not take advantage of this. When game manufacturers start to release games designed to take advantage of this, are we going to see a huge increase in game complexity/detail or is this benifit going to be less than Intel and AMD would have you believe?"

Comments

Memory latency is the limiting factor

[rastan]

AFAIK memory latency/bandwidth is currently the limiting factor in conmputation speed. Dual core processors will not change this, but make the gap even bigger.

Re:Memory latency is the limiting factor

[philipgar]

The idea of using a second core to prefetch is not really a new idea, and actually (at least for Intel chips) is not really a smart idea.

A more useful practice is the use of speculative prefetching on SMT (i.e. Hyper-Threading) cpus, where one thread runs the code, and the other thread speculates ahead issuing prefetch instructions. Of course to really support this well you need to have a compiler optimized for generating a speculative thread to run ahead of the primary thread.

All this makes programming much more difficult. My approach to software prefetching (Im currently involved in research in using SMT and software prefetching in databases) follows a different model, and shows that using a software-pipelining approach works remarkably well to hiding main memory latency due to cache misses.

Of course this is in a database world where things are ordered. .. . However this approach could also be applied to a game where instead of iterating over objects running every detail for the object you do work on object 1, prefetch object 1's next memory reference, do the second stage of object 2 etc etc. It needs special consideration to be done properly, but if its the core of your algorithm, it can have dramatic effects on performance (particularly if your objects do not fit entirely within L2 cache, and require different parts of them to be loaded each iteration).

While a speculative thread can prefetch these objects the performance benefits of this thread yield roughly the same performance boost as simply using software-pipelining techniques, and leave the resources for a second thread to run simultaneously with the first. Although if a sufficient speculative prefetch thread compiler is created then that approach is much easier to use for everyday applications.

Phil

Re:Memory latency is the limiting factor

[Lonewolf666]

Not as much as you imagine.
Compare an Athlon64/Socket939 to an Athlon64/Socket754 with the same clock speed. The Socket939 version has twice the memory bandwidth, but on average only 10% better performance according to AMD's P-Rating.
Now consider a dual core Athlon64/Socket939 with the same clock speed, where the two cores share the higher memory bandwidth. I would expect this chip to be as fast as two Athlon64/Socket754, or 80% faster than a single core Socket939 model.
Actually, clock speed will be a greater limitation:
AMD has announced that the dual core versions will run at 400-600MHz less to reduce the heat output.

relevant article

[antonakis]

I don't know if it has been referenced here before, a very interesting and enlightening article : http://www.gotw.ca/publications/concurrency-ddj.ht m

Re:relevant article

[prezninja]

I don't think the parent did enough to sell this article to the masses reading through, although it is an excellent reference.

The article linked to by the parent ("The Free Lunch Is Over: A Fundamental Turn Toward Concurrency in Software") should be read, and is of particular interest to developers.

The article draws a very good picture of how the trend towards mutli-core systems will require developers to rethink the way they design their applications if they want to continue taking advantage of future increases in processing power.

I was referred to this article yesterday, and it is so good and motivating that I imagine it will be the feature of or featured in future Slashdot articles.

It will be appearing in Dr. Dobb's Journal later this month.

Re:Dual Core Gaming

[mcc]

The XBox2 and Gamecube are both already known to be using POWER/PowerPC derivatives. Besides which, chip contracts for new consoles are the sort of thing that get worked out an amount of time in advance measured in years, and they're usually not bought from quite the same stock that PC OEMs are buying from. Intel's plans for their mass market "by late 2006" lineup really couldn't have any impact on the console world at all at this moment.

Re:dual cores

[sbryant]

all else equal.. two cores, two times the power, two times the heat..

You haven't been paying attention! Go back and read this article again (about AMD's demo of their dual core processor). While you're at it, read the related /. article.

The dual core processors use nowhere near double the power and produce nowhere near double the heat.

-- Steve