Phenom Exposed! Shipping with flaky 3rd cores.

[endyen]

Must...defend...AMD...at...all...costs...

Nope.
AMD has it's good points, and it's bad.
Do they make a resonable chip? It's fine for e-machines and servers. Also works quite well with linux.
Does Intel make a better chip? Oh ya!!
Would I rather have an Intel quad than a B2 phenom. You bet.
Do I think gang bangers are scum? You figure it out.

 

[endyen]

Remember, the fastest K8s are still 90nm 😉

That's normal. How long did it take for the FX chips to move to 90nm?
When AMD moves to a new node, they need at least a couple of steppings to get up to speed.

 

[endyen]

I think you're simplifying this issue too much.

It could be a process / masking issue, since the mask used in making Phenom is completely different than X2s.

It could be a quad core design problem, as there are fairly decent number of such report across the internet. Apparently they couldn't get the 3rd core to get above 2.1Ghz, and that is the speed of the highest Opteron is currently offered.

It could be a motherboard problem, only IF all of those encountering this problem are using the same chipset, or the same board, or the same mobo manufacturer. So far I haven't been able to determine that.

On the other hand, I think you're getting too edgy about this.

I certainly do not think of you as an I I I O.
I'm not sure if you are aware of the problems boardmakers are having with phenom bios.
The main reason that most AM2 boards do not have phenom bios is because of cmos space. There is just too much information to go into the old small bios. Some companies have found a workaround, but it is tight.
It's a whole lot of information being crammed into a small space. Easy enough to make a mistake.
Since it's a "borrow what you can" universe, a simple error can easily spread around.
That's why I think it's a mobo problem.
When you consider that there are now hundreds of thousands of barcelonas and phenoms out there, and probably as many that have been designated as dual or tri cores, the chance that a masking problem has gone unnoticed is very slim indeed. A design , or process problem making it this far is just not creditable. After all, they are stockpiling tri-cores so they are doing an active search for exactly that type of problem.

 

[endyen]

<rant>
Sure is also easy to tell the guys who *work* in semiconductor process design and manufacturing and understand these issues from those that don't.

Tell me, how would a motherboard BIOS issue create a flakey indivual core problem?
</rant>

It certainly can be a mask+process issue. The thing of it is that, even with a defect, it might not appear the same on every wafer. Believe it or not, the biggest issue facing those CPU manufacturers is variability. And the variability between lots, wafers, and even withing a wafer itself can be quite large. I'm more familiar with variabilty that can come from the patterning process (scanner dose, defocus, topography ("flatness" of the underlying features), scanner hotspots, temperature control, developer puddle, random etch residuals....)

Especially when given a portion of a design which is particularly prone to failure under process variability.

And it's not true that, just because you don't see it in the Athlons doesn't mean it's not a overall process problem. Certain design layout geometries are particulary poorly suited to handle the variabilities. No single design will explore all possible design space. AMD is at a particular disadvantage here becuase of the larger disconnect between design and process development.

I am leaning toward, not the design itself, but what it on the reticle itself that's the problem -- assuming the reports of the problem being dominant in core #2. And assuming the design itself is symmetrical. You see, even though the design of the four cores might be the same, it is NOT a guarantee that the data on the reticle will be. There are a number of data processing steps while taping out the reticle which subtly alter the design, and if the error is in there.

but hey, what do I know. I don't work for AMD and I don't know the details of the issue.

I've never heard of an NDA that precribes from telling people who you work for, or your tittle, so who is it, and what are you? Where's the plant?

 

[Reynod]

Language is appropriate don't you think?

Hey ... I'm just a janitor ...

 

[yomamafor1]

I certainly do not think of you as an I I I O.
I'm not sure if you are aware of the problems boardmakers are having with phenom bios.
The main reason that most AM2 boards do not have phenom bios is because of cmos space. There is just too much information to go into the old small bios. Some companies have found a workaround, but it is tight.
It's a whole lot of information being crammed into a small space. Easy enough to make a mistake.
Since it's a "borrow what you can" universe, a simple error can easily spread around.
That's why I think it's a mobo problem.
When you consider that there are now hundreds of thousands of barcelonas and phenoms out there, and probably as many that have been designated as dual or tri cores, the chance that a masking problem has gone unnoticed is very slim indeed. A design , or process problem making it this far is just not creditable. After all, they are stockpiling tri-cores so they are doing an active search for exactly that type of problem.

Even when OP's board is the new AM2+ 790X board?

My configuration is:

AMD Phenom 9600 Black Edition cooled by a Zalman 9700
Gigabyte GA-MA790X-DS4 Socket AM2+ motherboard
4GB Gskill 5-5-5-15 DDR2-800 Memory (4x1GB Sticks)
3x74GB WD Raptors in RAID 0, Primary Drive
500GB Seagate PMR Hard Disk
eVGA 8800GTS 320MB
Vista Ultimate 64bit

I'm just curious, how did you receive information like this?

EDIT:
Seems like those who are having problems also have Gigabyte board with 790 chipset (790X / 790FX). I'll keep look around if this only happens in Vista w/ Overdrive.

 

[yomamafor1]

I've never heard of an NDA that precribes from telling people who you work for, or your tittle, so who is it, and what are you? Where's the plant?

 

[homerdog]

That's normal. How long did it take for the FX chips to move to 90nm?
When AMD moves to a new node, they need at least a couple of steppings to get up to speed.

Brisbane has been out for over a year and it isn't showing any signs of ever outpacing Windsor.

Even if it could match the clockspeeds of Windsor, it is actually slower clock for clock (half the cache and slightly higher latency).

 

[zenmaster]

Depends on what CPU they swapped in there. If they swapped one Phenom 9600 in there for another Phenom 9600 and the problem went away, the problem was clearly with the first Phenom 9600.

And that is what they did.
They swapped one Phenom 9600 for another 9600 of the same stepping and the issue went away.

 

[ahslan]

wow...this got carried way too far...

 

[jimmysmitty]

Did anyone happen to read the other threads that he linked? This is not just happening to him but to other people. I would be a mad customer though if it only happened on Vista x64 especially since AMD touts better Vista compatability and superior 64bit processing.

I still think it is a possibility that some have a bad core. But who knows?

 

[spongebob]

Why do I say it's a mobo problem? Because mobo makers are having problems getting bios out for thier old AM2 boards.

It was pointed out on these forums that there might not be much financial incentive for them to offer upgrades for their old boards. Upgrading an old board generates no income, I’m sure they’d rather sell a new board. They’re part of a talented industry that knows how to turn product, and I suspect that any ‘problems’ may be more motivational than technical in nature. I'm reluctant to believe that this is all a BIOS issue.

That is, unless you know something about the mobo industry that I don’t (very well possible :lol.

 

[ragemonkey]

Did anyone happen to read the other threads that he linked? This is not just happening to him but to other people. I would be a mad customer though if it only happened on Vista x64 especially since AMD touts better Vista compatability and superior 64bit processing.

I still think it is a possibility that some have a bad core. But who knows?

I've tested this on both Vista 32-bit and Vista x64

The problem exists in both.

 

[jimmysmitty]

I've tested this on both Vista 32-bit and Vista x64

The problem exists in both.

So now we just need a test in XP. That way people will stop saying its the OS. I doubt its Vista since Vista just feeds code to the CPU. Its the CPU that sends that back to Vista to say "Hey its not working right".

 

[spongebob]

So now we just need a test in XP. That way people will stop saying its the OS. I doubt its Vista since Vista just feeds code to the CPU. Its the CPU that sends that back to Vista to say "Hey its not working right".

It's nowhere near that simple. Vista doesn't 'feed code to the CPU' - to a large degree an OS defines the context in which an applications runs - threading models and such.

 

[ryman554]

I've never heard of an NDA that precribes from telling people who you work for, or your tittle, so who is it, and what are you? Where's the plant?

<pm>

 

[johnnyq1233]

Referring to a previous thread about Intel stealing Amd's technologies.....
I don't think Amd has anything to worry about this time..lol !!!!

 

[jimmysmitty]

It's nowhere near that simple. Vista doesn't 'feed code to the CPU' - to a large degree an OS defines the context in which an applications runs - threading models and such.

My basic meaning is that Vista puts the thread through a processor and the processor of course processes it. If the program could not be completed it will show as an error. If the error is very bad it results in a BSoD or freeze up or restart.

 

[TechnologyCoordinator]

AMD should rename them to something different.


AMD BSoD 9600 Black Edition



It actually sounds kind of bad ass if you say "Bee-Sod" instead of saying the letters.

 

[jimmysmitty]

AMD should rename them to something different.


AMD BSoD 9600 Black Edition



It actually sounds kind of bad ass if you say "Bee-Sod" instead of saying the letters.

LMAO

How about AMD Pre-Production Phenom 9600BE X3?

 

[TechnologyCoordinator]

LMAO

How about AMD Pre-Production Phenom 9600BE X3?

Don't see why not. Intel had the balls to name their processors C2D (Core 2 Duo, Crash to Desktop)

 

[jimmysmitty]

Don't see why not. Intel had the balls to name their processors C2D (Core 2 Duo, Crash to Desktop)

Only difference is that C2D works.

 

[ragemonkey]

AMD should rename them to something different.


AMD BSoD 9600 Black Edition



It actually sounds kind of bad ass if you say "Bee-Sod" instead of saying the letters.

I think it flows better simply as "AMD Phenom 9600 BSoD Edition"

Its funny in a sad sort of way (hears sound of money spent on a Phenom, rates it as *worse* than blowing the same amount of money on a stripper)

 

[MU_Engineer]

And that is what they did.
They swapped one Phenom 9600 for another 9600 of the same stepping and the issue went away.

Then that looks like a problem with that first particular Phenom 9600, not a product-wide defect like some have been saying. If all of the Phenoms were crapped up, then swapping in one Phenom 9600 for another would not have made one lick of difference.

 

[croc]

Then that looks like a problem with that first particular Phenom 9600, not a product-wide defect like some have been saying. If all of the Phenoms were crapped up, then swapping in one Phenom 9600 for another would not have made one lick of difference.

And, as Yipsyl and others have already pointed out, it looks as if QC / binning testing is not doing the job. When was the tri-core announced? Did AMD know that there was a problem way back then?