PDA

View Full Version : 500ghz Processor is here!!!!



Lee Bartholomew
15th-August-2007, 10:40 AM
BBC NEWS | Technology | Chilly chip shatters speed record (http://news.bbc.co.uk/1/hi/technology/5099584.stm)

They recon they will be able to get it running at 1 Terahertz at room temp!!!

Come along way since the Spectrum.

Dreadful Scathe
15th-August-2007, 10:59 AM
speed is a fairly meaningless rating for a cpu :)

Lee Bartholomew
15th-August-2007, 11:02 AM
Of course it depends on bus speeds etc but I disagree that speed is meaningless you just have to have the technology available to support the chips speed.

Also I think that the technology they are refering to using this chemical in the chip would probly transfer over in to Gpu's, memory etc.

bigdjiver
15th-August-2007, 11:10 AM
Now we can do the wrong thing a zillion times faster ...

Dreadful Scathe
15th-August-2007, 11:10 AM
well meaningless in any sort of useful way, like to describe its capabilities. But purely as a number to represent how fast it is, its fine ;)

Dreadful Scathe
15th-August-2007, 11:11 AM
Now we can do the wrong thing a zillion times faster ...

or we can do a zillion wrongs things before we even notice ;)

Lee Bartholomew
15th-August-2007, 11:15 AM
well meaningless in any sort of useful way, like to describe its capabilities. But purely as a number to represent how fast it is, its fine ;)


Ah I get what you mean now. Agreed

Wobbly Dave
15th-August-2007, 11:18 AM
Ultimately the I/O will always be the greatest bottleneck. So even with the fastest processor in the world - will it make the system faster? I doubt it

Lee Bartholomew
15th-August-2007, 11:29 AM
Depends on what your doing. Posting on the forum or rendering the latest CGI film.

Will def affect games etc. Will take a long long time befor we see this chip in a desktop machine but for things like Seti etc it has its use.

ducasi
15th-August-2007, 11:32 AM
Ultimately the I/O will always be the greatest bottleneck. So even with the fastest processor in the world - will it make the system faster? I doubt it
:yeah:

Current system buses can't get near the speed of this CPU, so it would spend most of its time waiting for memory.

I'm also curious what CPU design they have used – is it a full-blown one, like one of the POWER series, or is it just some noddy processor, equivalent to an Intel 8080.

I figure though that even if they do get the speed up to 1 THz at room temperature, it'll take longer to get the cost down to something affordable.

Lee Bartholomew
15th-August-2007, 11:50 AM
I remember very similar things being said when AMD anounced a 1ghz processor. Memory wouldn't cope, Hard drives to slow and that data would be flying off the buss.

The solution they came up with was better memory and a better FSB (still waiting for decent HD solution)

David Franklin
15th-August-2007, 11:54 AM
:yeah:

Current system buses can't get near the speed of this CPU, so it would spend most of its time waiting for memory.Of course, current CPUs spend most of their time waiting for memory too...

The fundamental problem here would seem to be propagation delays: Speed of light is 3x10^8m/s, so light travels 0.3mm in a pico-second.

If you look at a Core 2 Duo (Conroe), the chip is about 12mm square. So you'd be talking about 40 clock cycles for a signal to get from one end of the chip to the other. The packaging is more like 50mm square, so 200 clock cycles just to get a signal to one of the pins on the chip. Delays to any external components are going to be much much bigger, obviously.

This assumes electical signals travel at the speed of light (not true), and even more erroneously, no gate delays. So the actual situations going to be a lot worse than that. (Actual current CPUs already have to deal with clock skew because it takes time for the clock signal to get to different parts of the chip).


I'm also curious what CPU design they have used – is it a full-blown one, like one of the POWER series, or is it just some noddy processor, equivalent to an Intel 8080.Given the above comments, I'm guessing noddy processor.

[Disclaimer: all above based on what I've read on comp.arch over the years; I'm have no special knowledge of my own].

Wobbly Dave
15th-August-2007, 12:02 PM
I thought the main block to raising the frequency was interference?

Lee Bartholomew
15th-August-2007, 12:23 PM
And as I think DF was saying, the speed at which data can travel.

They ob have some way of making it run that fast and it prob offers no practicality in its current form.

Im sure they will find a way to adapt the chip to run in special machines. I doubt they would be appearing in desktop pc's any time soon though.

David Franklin
15th-August-2007, 12:26 PM
I thought the main block to raising the frequency was interference?No. Basically, a CPU is made up of a load of transistors that can be thought of as a switches. These switches are connected by "wires".

There are two sources of delay: how fast a switch can change state, and how long it takes for a signal to travel along the wires. Currently the main block is the switching time. A transistor can switch far faster than 3Ghz, but the problem is that a CPU operation is a lot more complicated than just a switch changing state; you end up needing about 30 transistor switches to do anything useful, so for a 3Ghz cpu, the transistors are already switching at around 100Ghz.

The thing is, there are lots of theoretical things you can do to improve switching time (even if we don't know how to do them yet), but there's no theoretical way of improving the speed along the wires (above light speed). So wire delay is going to become a bigger and bigger issue in the future.

You can improve wire delay by making things smaller, but there's only so much you can do, particularly when there are practical reasons for wanting to build a machine out of separate parts instead of having everything in some incredibly expensive CPU the size of a sugar-cube that has to be completely replaced when something goes wrong.

And of course, things like the internet are becoming increasingly important, and there's no real way of shortening the "wire" between a computer here and one in Australia.