nm?

[Atrye]

I just read part of an article about how Intel is planning on moving from 90nm to 65nm and then eventually to 32nm. What are they talking about?

Also.. They're supposedly testing new dual-core technology but the only clock speed I saw was 1.7GHz. Is that supposed to be 1.7GHz per core?

 

[xonik]

"nm" is the SI unit of measure for nanometer, or 10^(-9) meters. This is the "feature size" of individual circuit elements, a rough average. You would expect a transistor, for example, to be about 90 nm in size when under the 90 nm process.

Moving to smaller feature sizes allows chip designers to make ever-complex circuits that do more, while keeping costs at bay. Once a fabrication process is stable, the main cost factor is physical size of the die containing all of the circuits. Using smaller components naturally makes the entire circuit smaller, or you can make the circuit more complex but the same physical size.

To answer your second question, yes, 1.7 GHz applies to both cores. The frequency is like a tempo in music. If you had two instruments playing at a tempo of 120 beats per minute, would you ever think to say that the music overall was playing at 240 beats per minute? The frequency just sets a predictable, stable tempo for simple operations to execute in a useful manner.

 

[Atrye]

Would it seem to run approximately twice as fast as a single 1.7GHz processor? Given that it would be able to run four (with HT) processes at a time..

 

[(cf)Eclipse]

no... at best it would be 50-75% faster because of issues with getting data from the cache and memory, and quite often, it won't be even that much faster cause of all the programs not written for multi-threading. though it should improve multitasking a lot. just look at dual cpu benchmarks of the xeons and opterons, and you'll know what kind of scalability you'll have (and the a64 chips will definitly scale better with more cpu's cause of how the memory architecture is set up)

 

[bountyhunter]

yeh, but what Atrye is refering to is Itaniums. They'll probably be more sufficient because they have their own [whatever x86 is]-(IA-64?), so all software built for that [" " "] is most likely bulit to take advantage of multiple cpu's.

 

[xonik]

yeh, but what Atrye is refering to is Itaniums. They'll probably be more sufficient because they have their own [whatever x86 is]-(IA-64?), so all software built for that [" " "] is most likely bulit to take advantage of multiple cpu's.

Still, the EPIC/Itanium architecture uses the inferior SMP infrastructure that relies on a shared pool of memory, managed by an off-die controller. That will continue to limit multiprocessor performance scalability, making it hard to match such ambitious targets as 50% and 75% increases.