Explain to me the whole 90nm 45nm 32nm thing

[Maximilian]

Arent these just sizes? Ive heard time and time again "Intels 90nm sucks" and AMD's was good, and that Intel have had breakthroughs on 65nm.... Arent these just sizes? Like centimetres/metres etc? How can Intels process be worse than AMD's, isnt it simply their architechure that isnt up to scratch?

Also, somthing thats loosely related, how big is a micrometre compared with a micron/nanometer?

Google came up with a lot of crazy stuff about fab 36 and dresden

 

[JAG87]

by process size, you can basically classify CPUs in generations. What anybody means by intel's 90nm sucked, is that compared to AMDs 90nm generation, intel's 90nm processors are behind.

and by behind someone can mean a lot of things. behind in pure performance, behind in performance per watt, behind in degrees per watt.
intel's 90 nm (prescott) pentium 4 was a horrible generation, and intel got smashed by AMDs 90nm (venice, san diego, toledo and such) in terms of everything. performance, power, and heat.
now it looks like intel barely caught up to AMDs 90nm processors, with their 65nm processors. speed is more even, power is still totally on AMDs side, and heat is more even but still leaning towards AMD.

I personally think that AMD can stay one generation behind, and still beat intel on all the fronts.

 

[dmens]

The lenghts are talking about feature sizes. The given number is the smallest granularity that can be printed with certainty onto a silicon wafer. That is a basic guideline.

1nm is 0.001 micron, or 10^-9 m.

How can Intels process be worse than AMD's, isnt it simply their architechure that isnt up to scratch?

A common mistake is to link process to architecture, like JAG just did. The quality of a process has nothing to do with the design it will be used to produce. They are two totally seperate issues. I can use intel 45nm to print a 386 and it will still be a terrible performer... although it would be very cool.

When merom comes out, you'll be able to see the difference between a good design and a mediocre design on the same process by compraing it against presler.

 

[Quinton McLeod]

Originally posted by: dmens
The lenghts are talking about feature sizes. The given number is the smallest granularity that can be printed with certainty onto a silicon wafer. That is a basic guideline.

1nm is 0.001 micron, or 10^-9 m.

How can Intels process be worse than AMD's, isnt it simply their architechure that isnt up to scratch?

A common mistake is to link process to architecture, like JAG just did. The quality of a process has nothing to do with the design it will be used to produce. They are two totally seperate issues. I can use intel 45nm to print a 386 and it will still be a terrible performer... although it would be very cool.

When merom comes out, you'll be able to see the difference between a good design and a mediocre design on the same process by compraing it against presler.

I'll have to disagree just a little.
The reason why Intel's 90nm process had issues with AMD's is partly due to the size. The bigger the process, the more heat it generates. However, the smaller, the less heat it generates. Also, since the process is smaller, it takes less time for the electrods to reach from one end of the process to the other.

Intel's 65nm process competes with AMD 90nm now is largely due to that. Since Intel is now able to reduce the heat, they are able to clock their processors higher. And since the process is smaller, it's just a bit faster. However, the designs are still the same.

Also, concerning your 486 example... If that design was made into a 45nm process, it would run slightly faster and much much cooler. However, it would still be slow

 

[stevty2889]

Originally posted by: Quinton McLeod

Originally posted by: dmens
The lenghts are talking about feature sizes. The given number is the smallest granularity that can be printed with certainty onto a silicon wafer. That is a basic guideline.

1nm is 0.001 micron, or 10^-9 m.

How can Intels process be worse than AMD's, isnt it simply their architechure that isnt up to scratch?

A common mistake is to link process to architecture, like JAG just did. The quality of a process has nothing to do with the design it will be used to produce. They are two totally seperate issues. I can use intel 45nm to print a 386 and it will still be a terrible performer... although it would be very cool.

When merom comes out, you'll be able to see the difference between a good design and a mediocre design on the same process by compraing it against presler.

I'll have to disagree just a little.
The reason why Intel's 90nm process had issues with AMD's is partly due to the size. The bigger the process, the more heat it generates. However, the smaller, the less heat it generates. Also, since the process is smaller, it takes less time for the electrods to reach from one end of the process to the other.

Intel's 65nm process competes with AMD 90nm now is largely due to that. Since Intel is now able to reduce the heat, they are able to clock their processors higher. And since the process is smaller, it's just a bit faster. However, the designs are still the same.

Also, concerning your 486 example... If that design was made into a 45nm process, it would run slightly faster and much much cooler. However, it would still be slow

Dothan was on the same 90nm process and used less power and ran cooler than the 130nm Banias..but Prescott, due to drastic design changes, used more power, and produced more heat, than the 130nm Northwood..it was more the changes made to Prescott than it was the proccess..Cedar Mill on the other hand is just a die shrunk Prescott with no major changes..and it gained the usual benifits of a die shrink, runs cooler and uses less power.

 

[dmens]

The reason why Intel's 90nm process had issues with AMD's is partly due to the size.

You mean die size. Even that is not an accurate trendsetter, total gate width would be better. The netburst situation is the result of a decision made long ago. I guess ultraspeculation is no longer in vogue.

 

[Quinton McLeod]

No, not die size. Process size. The process size is basically the size of the electrical paths in the die. It's not the die itself.

 

[Furen]

Like dmens said, the process is just the feature size, the only way it could suck more or less compared to another company's is perhaps because of yields. AMD also used SOI while Intel went for bulk silicon, which is part of the reason why AMD had a leakage advantage.

 

[dmens]

Originally posted by: Quinton McLeod
No, not die size. Process size. The process size is basically the size of the electrical paths in the die. It's not the die itself.

Not necessarily. It's just the finest level of detail that can be printed with certainty. I don't like thin wires. They're slow.

So how is intel's 90nm bigger than amd's?

 

[Furen]

Intel had problems because it had far more logic than AMD (though the cache is denser on the P4 the die-size is about the same as an A64), it worked at far higher clocks and it didnt have the benefit of being SOI (which lowers current leakage quite a bit). Dothan doesn't suffer from this because the CPU core is very small compared to either the A64 or the P4, it works at lower clock speeds than either and is heavily tuned to save power.