Separate names with a comma.
Discussion in 'CPUs and Overclocking' started by CHOPPER GOD, May 4, 2005.
WTF is 'Speed binning' in laymans terms?
When you make CPUs, some come out better than others.
Let's say you make Athlon 64s with 512k cache at .9 process.
Some come out better than others. The ones that come out best clock higher and get called A64 - 3500+. The ones that don't turn out so well get called A64 - 3000+.
The reason we can overclock them is that the 'bin' they put the finished product into often is far below its maximum capacity.
AMD also has to keep a steady supply of chips for each rating(3000+,3200+,3500+, and so on)
Sometimes when yields get low for one rating they fill up the bin from yields from one bin higher...thus some chips clock better than others.
They've said it all.
They have a bunch of bins with cartoon stickers on them. The fastest chips goes into a bin with a roadrunner sticker on it. The slowest go into the bin with cecil the turtle on it. The VERY slowest (and the chips that don't work properly) go into the bin with an Intel sticker.
thats funny..thx guys
so basically you would want a 3500+ that got called a 3000+? You order a 3000 and it is secretly a 3500 which explains my 3000 oc's to 2.6 but my friends 3000 craps out lower. hmmm Is this why you guys worry about week such and such and CBBID or whatever? If this is the case Please tell me what you all deem the best week/code whatever for 3200 Winnies..thx
My 3200 Winnie is CBBFD 0441 TPAW week 41, 2004 it will prime95 forever at 2588mhz and is benchable up to 2628mhz, not the best 3200 I've seen but a good one IMHO. I haven't kept up with the current chips but the week 38-48 chips from 2004 all seemed to hit 2.6ghz.
It's almost worth going to Fry's electronics and having the Zit faced kid look through all of them for a certain week since the retail package is see-through!
If you are buying an A64, just get a Venice (later revision than Winnchester) from monarchcomputer.com. It's almost guaranteed that you'll hit more than 2.6 with one.
nothing is ever guaranteed....
I always knew about binning, but it just occured to me that the multiplier must be locked after the cpu is finished and tested, so whatever locks it almost has to be an exterior mod. Seems there should be an easy way to unlock it in that case.
I believe they lock the multipliers by laser cutting part of the chip, ie there's a small piece that allows the chip to do an 11 multi, for example, and so if they want that chip to only have a 9 multi, they will laser cut the part that lets it do the 11 multi and the 10 multi off.
anyway, getting good chips and OC's are still partly determined by your luck (and of course your skill at OC too). what you can do is control as much of he variables as you can (FSB, mem, PSU, Mobo, week of manufacture of CPU, cooling) and cross your fingers. you can also follow Zebo's uber guide up there.
i agree... overclocking is partially luck (from the chip you get) and then the skill parts come in from tweaking your system to run at a target speed...
Binning gets done prior to cutting up the wafer, so after they test every chip, they bin them, laser cut them to the correct multiplier, and then cut up the wafer, and they get packaged.
I've never seen or read about the complete process, but I would never have thought the cpu was tested before the waffer was cut up and the chip packaged. In fact, I'd bet pretty long odds that it doesn't happen that way.
The speed binning is done prior to the wafer being cut up. I use to run the tools that did the testing, so trust me I know..the chips are tested again after being packaged, but every die on the wafer is tested and binned prior to the wafer being cut, it's much more efficient that way..
If it didn't happen that way, then what would you do if you made a 3500+ and it's only good for a 3200+? You'd have to throw it away. That would be very bad for AMD
That makes some sence anyway. But whats the reason for double testing? It seems that they have to test the finished cpu, whats the logic behind not speed testing at the same time?
Because they have to laser cut the die to the correct multipliers prior to packaging it..they can test it's speed prior to seperating it from the wafer, while they can test it's functionality after it's been packaged by putting it in a test system..
EDIT: Also it would kinda be a waste of money to package it first, and then find out it didnt' work..
I love the AMD love here :laugh:
Interesting stuff, and also answers a question I had about locking the multipliers. I had been wondering how that worked. Thanks for the info.