Computers to Become More Powerful Than Expected

[CaptainRAVE]

The Semiconductor Industry Association (SIA) has announced that it believes that processor chips can become smaller even faster than expected. Today, most processor chips use 180 nanometer or 130 nanometer circuits.

 

Earlier predictions had estimated that in 2005, circuits would use 100 nanometer technology, and in 2014, they would be down to 35 nanometers. Now, they are calling for 90 nanometer circuits available by 2004 and 22 nanometer circuits in 2016.

 

The plan published by the SIA lets companies know what to plan for over the next 15 years. By bumping up their expectations, the SIA has indicated that computers will become much more powerful than previously anticipated.

[thrEEpaGe]

the massive amount if the increase of transistors equates to teh shrinking of the die size..but this doesnt equate to an increas of performance

 

here's an excellent article about moores law telling how the SIA's 'prediction' isnt necessarily as profound as they would like you to believe

 

http://www.pcmag.com/article/0,2997,s%253D1510%2526a%253D2110,00.asp

 

The Meaning of Moore's Law

By Richard V. Dragan

 

 

 

By now, many people have memorized Moore's Law: The number of transistors on CPUs doubles about every 18 months. Intel chairman Gordon Moore's law has been accurate for decades, for CPU evolution. What about total system performance, though? PC Magazine Labs recently decided to see whether Moore's Law also means that systems run twice as fast every year and a half. Joel Santo Domingo, project leader for our PCs and Devices team, headed up the testing effort.

 

"Moore was originally observing that the number of transistors on a chip doubles every 18 to 24 months," Santo Domingo says. "But people believe the law implies that performance doubles every 18 months. Moore's Law is very, very misquoted."

 

So Santo Domingo and his team tested the assumption. The team made use of a brand-new Gateway system with a 1.5-GHz Pentium 4 CPU. This system was compared with a four-year-old NEC machine with a 300-MHz Pentium II CPU.

 

The Pentium II CPU was released in May 1997 and had 7.5 million transistors. Three years and six months later, in November 2000, the Pentium 4 was released. For this 42-month period, Moore's Law declares that the number of CPU transistors should jump to about 40 million. In fact, the new Pentium 4 chip we tested had 42 million—slightly above the Moore's Law prediction.

 

Did total system performance jump by the same amount over that period? To find out, the team ran the Ziff Davis Business Winstone 2001 benchmark test program. Both the old and the new system had 128MB of RAM and ran Windows 2000 SP1. The results were enlightening. The old system scored 13.9. The new system's score of 43.5 represented a significant increase in speed—but was well short of the 74.1 score that would result from a doubling of total system performance every 18 months.

 

If CPU speeds and transistor counts are still doubling at a fast clip, what's holding us back? Neither memory speeds nor bus speeds are growing at the pace Moore's Law predicts for transistor count, but the main culprit seems to be disk I/O speeds.

 

Disk capacities have grown from megabytes to gigabytes in recent years, but disk speeds have not kept up with CPU speeds. Santo Domingo speculates that "a solid-state silicon drive or a nonmechanical optical storage device might follow Moore's Law."

 

Intel's recent advances in chip manufacturing technology (using Extreme Ultraviolet, or EUV, lithography) should ensure that Moore's Law itself continues to hold through at least the year 2010. But boosts in CPU transistor count and speed don't necessarily equal advancements in total system performance.

 

 

 

/end article

 

by this reasoning, even though the chip is decreasing in size at a faster rate than previously predicted (and thus a faster transistor increase), it doesn't mean that performance will increase at the same rate.

 

as the article said, I/O speeds cannot currently keep up with the processor speeds. in short, the processor is pumping faster than the pipe can be pumped through. even if we do get those faster mhz, it doesnt matter!

 

until we get cost-effective i/o solutions (ie faster fsb, solid-state storage, etc.) we wont see the actual sheer performace increase indicated by those faster mhz..

 

 

that's my 2 cents

[StormHammer]

I agree with thrEEpaGe.

 

There are some inherent bottlenecks in MoBo design. RAM used to be a culprit until DDR came along, and that's still not perfect. But as thrEEpaGe said, the whole I/O system badly needs to be addressed.

 

The problem with that, of course, is getting people like Microsoft to incorporate new standards into their software out of the box. I mean, Win XP has no native support for USB 2.0. It seems that a lot of manufacturers/developers are dragging their heels over new standards...which is probably why we haven't seen too many radical redesigns of MoBo's and slot types.

 

I believe SGI did some major redesign once, which showed some performance gains at the time, though it wasn't too impressive.

 

Solid state storage would certainly help considerably, but I don't think it's the only culprit. Many more changes have to be made before the system really works in a unified way, rather than a jumble of disparate parts.

 

Just IMHO.

[Rancor]

Stormhammer is right. Bus speeds and i/o have not increased at a proportional rate and so the CPU and GPU are probably not even being fully pushed to what they could do in an optimal situation.

 

Related to that, load times were pretty fast in JK. If you look at the size of many of their textures, you've got mainly 64x64 and 128x128 textures. Some are larger, and some are smaller, but the generally seem to be in that size range.

 

Now if the Sof2 posts are correct, and they are using mostly 1024x1024 and 512x512 textures. You've got textures that are 4 to 8 times as large in one dimension. In two dimensions, you end up with a JK2 texture that take up to 16 to 64 times as much memory as a JK texture.

 

We all know that there are a limited number of textures you can have in memory at one time, so textures will have to be loaded from disk between levels. Now, CPU speeds may be 4 or more times faster, but i/o speeds aren't. Even if I/O operations were 4 times as fast as they were in the JK days...if a texture took 'X' seconds to load in JK, we can expect it to take about 'X' x 4 seconds, or even 'X' x 16 seconds in the worst case.

 

So, the point I'm getting at is that, we can expect load times to be worse in JK2. If this proves to be true, it's not the developers fault--it's just the simple fact that i/o speeds really haven't improved to keep pace with CPU and GPU speeds.

 

So....maybe we should make a mod with 64x64 textures for JK2, heh. Load times would be super fast, but it will look awful compared to any game that has come out in the past 4 years. Guess that's the price we pay for wanting games that look awesome. And if we don't like that, we can just blame the hard drive manufacturers.

 

[ December 02, 2001: Message edited by: Rancor ]