AMD Manufacturing

[Viditor]

An excellent article by Charlie D with the Reader's Digest version of the differences between Intel's and AMD's manufacturing. Only about a 2-3 on the "technical knowledge required" scale...

Snippet:
"Intel has a strategy of copy exact. Basically, whatever the technology is, it is rolled out to the relevant fabs exactly the same way. AMD has a more "on the fly" approach and experiments in real time on the production lines"

 

[jiffylube1024]

Neat! Effing complicated process, but that's necessary given the fact that transistors are less than 1/10 the thickness of a human hair and there's ~100M of them per chip.


It seems to make good sense - AMD chips do seem to just "creep" up in overclockability/ability to run with lower volts as the weeks and months go by. I've noticed that the average chips with the same label (eg 3800+ X2, etc.) from a later week is a better chip based on overclocker criteria than early ones.

With Intel, most of the big speed gains are because a new run gets created where they tweak something (more cache, die shrink of course, etc.) or implement a new technology (like a new gen of SOI or other things...).

 

[harpoon84]

Yeah I think you're on the right track jiffy.

With AMD overclocks just steadily increase over time as the process matures.

With Intel you often get sudden increases in headroom due to new 'steppings'.

 

[BobDaMenkey]

Interesting stuff. Good read, thanks for the link Viditor.

 

[zephyrprime]

Originally posted by: jiffylube1024
Neat! Effing complicated process, but that's necessary given the fact that transistors are less than 1/10 the thickness of a human hair and there's ~100M of them per chip.

Transistors are alot smaller than that. More like ~1/100 as wide as a human hair.

 

[jiffylube1024]

Originally posted by: zephyrprime

Originally posted by: jiffylube1024
Neat! Effing complicated process, but that's necessary given the fact that transistors are less than 1/10 the thickness of a human hair and there's ~100M of them per chip.

Transistors are alot smaller than that. More like ~1/100 as wide as a human hair.

Are they? You may be right - my reference point was years ago. I may be thinking of like the Pentium days (?) when chips were .80um not .065um.

 

[Viditor]

Originally posted by: harpoon84
Yeah I think you're on the right track jiffy.

With AMD overclocks just steadily increase over time as the process matures.

With Intel you often get sudden increases in headroom due to new 'steppings'.

I tend to agree...
It explains some of the differences we keep seeing in much of the OC data for the X2 as well. OC levels for X2 keep getting better for the same models over time (causing many disputes on this board).
This is most likely a function of APM/STT/CTI...

 

[Viditor]

Originally posted by: jiffylube1024

Originally posted by: zephyrprime

Originally posted by: jiffylube1024
Neat! Effing complicated process, but that's necessary given the fact that transistors are less than 1/10 the thickness of a human hair and there's ~100M of them per chip.

Transistors are alot smaller than that. More like ~1/100 as wide as a human hair.

Are they? You may be right - my reference point was years ago. I may be thinking of like the Pentium days (?) when chips were .80um not .065um.

Actually, the average human hair is ~105nm...so we're talking about half the diameter of a human hair.

 

[RichUK]

From reading bits of that, it reminds me of an old friend that worked in some fabrication plant. Apparently he left the wafers in some chemical bath for too long and cost his company around £15,000. EDIT: The guy only worked at a simple chip plant, nothing to do with the top tier people.

 

[ZOXXO]

Originally posted by: Viditor

Originally posted by: harpoon84
Yeah I think you're on the right track jiffy.

With AMD overclocks just steadily increase over time as the process matures.

With Intel you often get sudden increases in headroom due to new 'steppings'.

I tend to agree...
It explains some of the differences we keep seeing in much of the OC data for the X2 as well. OC levels for X2 keep getting better for the same models over time (causing many disputes on this board).
This is most likely a function of APM/STT/CTI...

Or perhaps it's a figment of your imagination.

 

[Viditor]

Originally posted by: ZOXXO

Or perhaps it's a figment of your imagination.

Reading is fundamental...Google is your friend!

 

[ZOXXO]

Originally posted by: Viditor

Originally posted by: ZOXXO

Or perhaps it's a figment of your imagination.

Reading is fundamental...Google is your friend!

Reading is indeed fundamental but leaping to conclusions is int he realm of fanboyism.

You see what you want to see.

Google is a search engine.

 

[myocardia]

Originally posted by: Viditor
Actually, the average human hair is ~105nm...so we're talking about half the diameter of a human hair.

Um, no, the average human hair is ~80 µm: link. A µm, also known as a micron, is one millionth of a meter: link. A nm is one billionth of a meter: link. That means that a nm is 1/1,000th of a µm.

So, an "average" human hair, using my size (80 µm), is 80,000 nm, or 1,230.769 times as wide as the process in the new Intel chips. Also, jiffy, the original Pentiums were supposedly a 3µm process, but I don't really care enough to do the research on Intel's site.

 

[oztrailrider]

Interesting article, thanks for the link.

 

[pm]

Originally posted by: myocardia
Also, jiffy, the original Pentiums were supposedly a 3µm process, but I don't really care enough to do the research on Intel's site.

The original Pentium 60MHz and 66MHz was on a 0.8um process. The Pentium 200MHz was on a 0.35um process.

 

[Viditor]

Originally posted by: myocardia

Originally posted by: Viditor
Actually, the average human hair is ~105nm...so we're talking about half the diameter of a human hair.

Um, no, the average human hair is ~80 µm: link. A µm, also known as a micron, is one millionth of a meter: link. A nm is one billionth of a meter: link. That means that a nm is 1/1,000th of a µm.

So, an "average" human hair, using my size (80 µm), is 80,000 nm, or 1,230.769 times as wide as the process in the new Intel chips. Also, jiffy, the original Pentiums were supposedly a 3µm process, but I don't really care enough to do the research on Intel's site.

My apologies...This damn laptop screen isn't the best and I misread the full stop in the ref page I went to. Found another at the FBI...
Interestingly there is a wide size range for human hair (sorry about the OT)

"Fine Hair - 50 microns, Medium Hair - 60 to 90 microns, Coarse Hair - 100+ microns"