Different speed grades of the same processor

[Xalista]

What is involved in this? I'm pretty sure architecturally nothing changed between i.e. a P4@1500 and a P4@1900. But what did change? Is it just that intel is optimizing the fabrication process which results in a higher yield and faster chips, or are they also changing things on the die itself? Are there just minor layout changes, or are the changes more rigorous? Do these changes result in differently sized dies for the different speed grades? Thing like that, who can shed some light on this for me?

 

[Evadman]

PM is the man on this one 🙂

Every stepping is different. The EE guys change the actual die. Reorent this transister, do this do that...

It depends on what you concider minor. The die sizes are all the same size. the dies are not changed that significantly.

 

[rummyPPG]

It is also possible to have two chips with no layout changes between the two which are available at different speeds. This is because the fab process isn't totally exact; i mean they're good, but they're not good at say a level of atoms.

For example, when they hit the masks with light in etching process, its hard to get the light to disperse totally evenly across the whole wafer, meaning some chips experience a different angle of light penetrating through the openings in the masks, meaning they have slightly (usually very slightly) different physical characteristics.

This and other similar fabrication "errors" lead to chips that come back being operational at slightly different speeds. A company then takes those chips and tests them to see what they [the company] feels they will work at comfortably (and perhaps there's internal political and marketing influences too) and then set their chip multiplier to achieve the target speed.

That's why people can overclock chips cause they "can" work at faster speeds, the company (intel, motorola, etc.) just doesn't support it. And that's also why overclocking depends on the chip generation and some luck, because depending on better layouts in later generations (or maybe even earlier generations) and/or the exact happenings during your chips fab, it may be able to withstand the faster switching speeds and be overclocked high, or it might produce errors because some circuitry has currently leakeage, lockup, or other electrical quality problems, or it could theoretically fry the chip through something like electromigration.

I'm not very knowledgable of intel's definition of stepping, but i'd assume its a change on one or more of the metal masks for the same architectural chip.

Side Note: You COULD have the die size change if you took a design and shrunk it to a new process, without adding functionality. I'd consider this the same chip. However, its usually a fair amount of work to just shrink a chip because of the new process properties (i.e coupling, rail bounce, current leakage, etc.) so its usually not much more work to redesign parts of you chip at the same time thus leading to a new chip. But i think the pentiums go through die changes when change process to get more speed out a current line.

 

[pm]

RummyPPG has a good description of it. I don't have much that I'd add to his points.

We had a big discussion about this in HT ages ago. It's in this thread. Unfortunately I can't seem to view any archived messages (but I know this is the right thread from the title when I search for it). I'm not sure if this is a problem with my Mozilla browser on HP-UX 11i or if it's AT's database.

 

[Den]

How odd, I can see that thread (and the links in it to even older threads) just fine.

 

[pm]

Thanks for the feedback, Den. It's a HP-UX Mozilla issue then. Which helps. I get this error: "An error occurred while evaluating the expression: FTVAR_CUSTOMTEXTTMP = ReplaceNoCase(FTVAR_CUSTOMTEXTTMP, vchemoticonreplace, vchemoticonstring, "All") Error near line 6, column 8." I've actually been getting it for months on all archived threads. I thought it was a subscriber thing at first, but then I noticed it worked at home. It appears to have something to do with the way that the browser returns information to the database since I'm getting the error from the AT servers. In any case, this is OT and I'll send feedback to the forums issues, or to Zuni.

Now that I can view archived threads at home... the thread that I linked above links to another even older thread that goes into more detail here

Edit: Now I'm sporatically getting it on my home Windows XP IE 5.5 and Mozilla 1.0 browsers as well. I wonder if it's something to do with my account or my login or something. Anyway... I hope everyone else can see them.

 

[Sunner]

Strange, I can view it just fine using both Opera 6.01 and Mozilla 1.0 on Debian/Linux 2.4.18.