Come on AMD!!

[Dug]

From Anand's 1st day Comdex report-

Combine this with what we heard a few weeks ago and it does seem that the motherboard manufacturers are definitely waiting on AMD to provide CPUs for Hammer's launch; the chipsets and boards are basically ready.

We want some Barton's and Hammer's. Now!
What's the hold up? How hard is it to put another 256k cache on the chip?

Why would AMD shut FIC down from demonstrating a Hammer working on their board?

 

[dowxp]

Originally posted by: Dug
From Anand's 1st day Comdex report-

Combine this with what we heard a few weeks ago and it does seem that the motherboard manufacturers are definitely waiting on AMD to provide CPUs for Hammer's launch; the chipsets and boards are basically ready.

We want some Barton's and Hammer's. Now! What's the hold up? How hard is it to put another 256k cache on the chip?

probably because its hard. why dont you try it

 

[Dug]

probably because its hard. why dont you try it

Yeah, I already did, but I got too much glue on it.

 

[Windogg]

Originally posted by: Dug
From Anand's 1st day Comdex report-

Combine this with what we heard a few weeks ago and it does seem that the motherboard manufacturers are definitely waiting on AMD to provide CPUs for Hammer's launch; the chipsets and boards are basically ready.

We want some Barton's and Hammer's. Now!
What's the hold up? How hard is it to put another 256k cache on the chip?

Why would AMD shut FIC down from demonstrating a Hammer working on their board?

You do realize it can take 12 or more weeks to roll a wafer through the manufacturing process? Not only that but you need to work on equipment calibrations, test wafers, and ramp up.

 

[Maggotry]

12 weeks?!? What fab has a 12 week cycle time? That sounds WAY too long.

 

[Macro2]

RE:"We want some Barton's and Hammer's. Now!
What's the hold up? How hard is it to put another 256k cache on the chip?
Why would AMD shut FIC down from demonstrating a Hammer working on their board?"

Indications are that they ran some rocket lots and the Hammer die would not yield or bin well.
In the case of yielding, thats "good die" this probably had something to do with trying Low k and SOI right from the get go. At that point Clawhammer in 2002 was shelved because even if they could get some chips to bin well enough the poor yields wold have made them too expensive for the desktop market. Opteron on th other hand will garner a higher price and offset any poor yield problems.
Binning. Different from yields in that yields are % of functional die where as bins are speed grades. Clawhammer was originally set to be released at 2.0 ghz. Indications are that it's having trouble at that speed.

So...why is Hammer delayed? AMD had to go back to the drawng board so to speeak on the process used to manufacture the chip.
"Likely", Clawhammer got the axe because of yields. Opteron is still on recent schedule but that could change.
At this point it also sounds like speed binning is a problem too.

 

[Macro2]

RE:"12 weeks?!? What fab has a 12 week cycle time? That sounds WAY too long"

Usually 10 weeks or so. 12 is not too long.
This is from wafer starts through "final testing and packaging". Usually these two things take place in different locations as well.

 

[CraigRT]

Originally posted by: Dug

probably because its hard. why dont you try it

Yeah, I already did, but I got too much glue on it.

You too? crap.. gotta respect those guys across the street at AMD..... they got a NICE workbench and good lighting..

 

[Windogg]

Originally posted by: Maggotry
12 weeks?!? What fab has a 12 week cycle time? That sounds WAY too long.

From the time its a wafer to the time it works its way through the disty channels. Time is also dependant on how many masks are needed.

Windogg

 

[Maggotry]

Well, I probably shouldn't reveal too much detail, but I'll just say that our fab's cycle time is "significantly" lower than 12 weeks from wafer start to probe...not including packaging.

Edit: And that's 100+ processes from laser scribe to probe. Add another week in probe and time for packaging and you're still way below 12+ weeks. I'm not saying you're wrong, it's just vastly different from what I've experienced after 10+ years in the semiconductor industry.

 

[Tbirdkid]

In a mainstream memory manufacturing plant (not processors but similar) a wafer makes it through all of the processes including dice, test and shipment in a matter of 40 to 50 days..not 10 weeks. At least that is how it is in a memory manufacturing plant... I am not too worried about the next cpu amd puts out because I am happy where I am with my old 1600 aroia. CPU power is needed in certain applications but man it eventually is going to stop.. Just my opinion.

 

[Macro2]

RE:"Well, I probably shouldn't reveal too much detail, but I'll just say that our fab's cycle time is "significantly" lower than 12 weeks from wafer start to probe...not including packaging"

Are we talking about an AMD or Intel cpu fab?

RE:"In a mainstream memory manufacturing plant (not processors but similar) a wafer makes it through all of the processes including dice, test and shipment in a matter of 40 to 50 days..not 10 weeks."

I guess your FAB is faster than AMDs.

Mac

 

[Wingznut]

Comparing how long it takes to manufacture memory to how long it takes to manufacture a CPU, is apples and oranges. And there are well over 100 processes in the manufacturing of a 9-layer CPU.

 

[smp]

Originally posted by: Macro2
RE:"Well, I probably shouldn't reveal too much detail, but I'll just say that our fab's cycle time is "significantly" lower than 12 weeks from wafer start to probe...not including packaging" Are we talking about an AMD or Intel cpu fab? RE:"In a mainstream memory manufacturing plant (not processors but similar) a wafer makes it through all of the processes including dice, test and shipment in a matter of 40 to 50 days..not 10 weeks." I guess your FAB is faster than AMDs. Mac

hmmmm .. what fabs are located in houston texas?

 

[Tbirdkid]

sorry wingz.... not trying to be rude about what I had said about memory comparing it to a cpu. i just know there are differences and similarities. time in a fab is definitely different i am sure just for that reason but it is also not manufactured like memory in a hurry up type of basis because the more you make in memory the cheaper it is. With there only being a couple of cpu suppliers in the world you dont have to worry about that. we all know there are tons of memory manufacturers trying to survive and make a good product at an awesome price... thus comes less time to create a product...

 

[Wingznut]

Well, it's not so much that... But CPU's are vastly more complex and involve more steps to manufacture.

It's not an economic thing... Just a reality.

 

[Maggotry]

hmmmm .. what fabs are located in houston texas?

Two, actually. Texas Instruments and Input/Output. Saying I work for TI is not a problem, as I've done several times here in these forums. I just didn't want to get into specifics about what we build and how long it takes. Our main product here are DSP's. Just about eveyone in these forums has TI technology in their PC's. Firewire, hard drive controllers, modems, etc. If you don't have our stuff there, then you probably have it in your cars...anti-lock brakes, air bags, monitoring systems, etc. Got a cell phone? Good chance we're in that too.

Most of our stuff is triple level metal and as I said a few posts up:

"And that's 100+ processes from laser scribe to probe. Add another week in probe and time for packaging and you're still way below 12+ weeks. I'm not saying you're wrong, it's just vastly different from what I've experienced after 10+ years in the semiconductor industry."

Windogg's original comment that prompted my post was that it takes 12+ through the manufacuring process which again, is vastly different than my experience.

 

[Adul]

Originally posted by: Wingznut
Well, it's not so much that... But CPU's are vastly more complex and involve more steps to manufacture.

It's not an economic thing... Just a reality.

i guess you can say how long it takes, but is the 10 week time a good est?