Phenom Exposed! Shipping with flaky 3rd cores. - Page 5

We don't know that as they may run into thermal issues on their 45nm. On thig is for sure they wont have HK/MG for their first 45nm stepping and is probably going to be SOI thus the thermal issues they may run into.

Intel already has 32nm SRAM chips working and will have the 32nm shrink of Nehalem released in November of 2009 codenamed Westmere. AMD wont have 32nm probably until 2010 since thats when IBM plans on having it and thats where AMD gets most of their process technology from.

I always love how this is brought up every time. I always find it funny how in a two year period (2004-2006) AMD gained quite a bit of market share especially considering the manufacturing capabilities. But still AMD has had their faults too same with IBM and every other company as there is no perfect company. Just some hide it better than others.

ragemonkey I am sorry that both of your BE's turned out flakey. But what are you going to do when they have a horrible QA department? Well maybe the B3's will get better but whos to say they wont have the same problems?

mathos, ragemonkey is using a Gigabyte GA-MA790X-DS4 Socket AM2+ motherboard. So its the same chipset as yours minus the "F". So in reletivity its the same chipset and you think it might be his setup when the BSoD is clearly stating a CPU problem?

It is possible to get bad parts more than once. Or it could just be that there are more bad Phenoms out there than you would think.

I think its just that Phenom wasn't really focused on and they are having more problems than they want to admit thus why they publisized the errata to cover up their QA messups.

True, but there are a lot of reasons why one core could be broken.
It may come from the outside of the wafer. A bit of dust may have landed on that spot. There may be a slight imperfection in the wafer itself. Damage can occur in cleaning, masking or heating.
If they are selling 3X as many barcelonas and phenom quads, as they are tri or dual core phenoms, everything is fine.
We will have to wait and see.

Please elaborate on exactly what to type into the search engine to get all these results of many people having this problem? The ones that I find link back to here, and to extremesystems. And like I said, give me exact settings that you're using, as far as bios goes. I've seen a few people talking exclusively about Windows Vista 64bit problems, and a few about Vista 32bit, which could likely be a chipset driver issue.

Hey my Machine check specifically points out CPU, Mainboard, Power supply, Memory, but it didn't start happening until after installing the the HD3870, so obviously it couldn't be the video card.

Mathos, you bring up some very valid points. However, i will point out that my success rate has not been 0%, except for the 9600 Black Editions specifically. For work I have purchased 9600s and 9500s that have worked fine, save for one which was returned for a working replacement which has been fine. Although one of my systems at work has begun to randomly BSoD... and I am attempting to isolate why.. it could end up being the proc, yet it very well could not. I simply do not know yet. As for what I've seen online recently, the amount of posts regarding this issue have not grown significantly since I created this post. Also, other new posts I have seen simply refer back to this one, which calls credibility into question as it creates the "circular reference" problem that wiki tends to suffer from hence why it is no longer considered a valid academic source.

To adequately confirm failure rate I would need to conduct some actual valid reseach purchasing 9500s, 9600s, and 9600BEs from multiple vendors and test them under identical conditions to find out from a statistical standpoint if it is a particular model has a problem, or whether all of them have a problem. Personally, I don't have the money nor the time to conduct such an experiment. That is why I created this thread in an attempt to get THG's attention... or somebody else's that may endeavor onto such a task. I take no credit for these findings, I have simply confimed some other people's findings with my own personal experience. That, in no way, suggests a statistical liklihood that *all* phenoms are busted. I do not have the means, or the time, to come to such an encompassing conclusion.

As for my personal failure rate, it could be for several reasons. My primary suspicion is that since the first was purchased from Newegg, and the second one an RMA within two weeks from the same place (obviously)... it is possible I got a chip from the same potentially bad batch of chips. Bad luck... thats all.

As for my current configuration, I am happily running a Phenom 9500 at 2.62 ghz. It will not run Vista x64 stable at 2.62ghz, but will at 2.5ghz. Vista 32 runs completely stable at 2.62ghz and I have run the AOD stability test for 4 hours twice at this speed. I haven't yet seen a BSoD at 2.62 in Vista 32 running this 9500.

CPU Reference Clock set at 238
CPU Multiplier at 11
Memory Controller Multi set at 8
HTT locked at 1.8 ghz
Memory divider set at DDR2-667, final memory speed of 792mhz ddr
Stock memory voltage
Northbridge Voltage +.05v
CPU Voltage at 1.40v
C & Q Disabled
TLB Erratum fix disabled

The CPU is cooled via a Zalman CPS9700 running at full steam
Idle temp is 27C, consistent 100% load is 38C

As for other stuff:

Gigabyte GA-MA790X-DS4 motherboard running F3 BIOS update
Phenom 9500 running at 2.62 ghz
4GB (4x1) Gskill 5-5-5-15 DDR2-800 running in ganged mode
2x150GB WD Raptor in RAID 0 -- OS and program volume
1x500GB Seagate Barracuda 7200.10 PMR -- Storage and Encoding volume
eVGA 8800GTS 320MB
Sound Blaster Audigy 2 ZS Platinum
Antec Neopower 550w PSU

Everybody seems to be getting their undies in a bunch about the 3-core Phenoms being "broken." Disabling a portion of a CPU die has been done for many years for various reasons and is currently being done by BOTH Intel and AMD on no small number of CPUs. I'll bring up a few notable examples:

1. For the older ones among us, the Intel 486SX was a 80486DX with a disabled FPU.
2. Many chips have some L2 cache disabled- Celerons, most Semprons, some Athlon 64s, Turions and X2s (such as every 2x512 KB Windsor), Core 2 Duos (such as every B-stepping E6300/6400 and T5500/5600), every Pentium Dual Core chip manufactured and every Celeron Dual Core chip manufactured.
3. Some chips even have a core disabled. There were a few 939 Athlon 64 3200s and 3500s made from X2 Manchester and Toledo dies with a core disabled. Many mobile Celerons are Core Duos or Core 2 Duos with a core disabled. If you want proof, look at the package specifications in the Celeron and Core/Core 2 data sheets. Note the physical dimensions of the processor's die surface- it is the same between the two.

I haven't heard anybody complaining about any of the chips above being "broken" so I don't understand why AMD is taking heat over the 3-core Phenoms other than some people haven't taken a close enough look at how CPUs are actually made and sold before hitting the "Reply" button.

Got proof?
So far, all I've seen is a few noobs who cant figure out how to make the chip work.
If it were true, Intel would have thier private tech sites all over it.
People having problems? Sure, just like 8400 buyers. Does that mean it's the chip's fault? No.
Quit being such an IIIO, and stop spreading FUD.
( All phenom users are noobs, the chip isn't old enough to have any veterans)

damn that sucks and man my B3 Q6600 can go up to 3.6 GHz and phenoms cant?? damn

Wow... how hard is it to get a computer to work?

Put on anti-static wrist thingy (does anybody actually use those?)
Open ZIF socket.
Put in CPU.
Close ZIF socket.

The OP sure doesn't look like a noob to me. How many servers have you built?

(and if the Phenom is harder to install than that, then you'd have to say that it's not ready for primetime....)

What??!? The point is that the OP and many other had tried OCing this and had come to the same conclusion.

Proof?:
http://www.siliconmadness.com/2008 [...] blems.html
http://www.evga.com/forums/tm.asp?m=251136
http://messages.finance.yahoo.com/ [...] so=R&frt=2
http://www.xtremesystems.org/forum [...] p?t=175878

dcfan60 @ Xtream Systems

So if the chip wasn't bad why are all these people coming to the same conclusion?

@endyen

You make some very valid points, and yes... we are all newbs at OCing Phenoms. I will concede that. This is not a matter of ego, therefore, I will openly admit if my findings are skewed in one direction, or outright incorrect. You are also correct in saying that OCing a Phenom is a matter of luck. However, I think your insulting manner of saying this is a bit out of line, and degrades somewhat... the overall validity and credibility of this thread. The only thing many of us are trying to do is to get to the bottom of what may or may not be a problem. Because, in the end, that's what this is really about... getting to the root of a problem. It has nothing to do with "drinking kool aid", "fanboi-ism", "noob-ism", or anything else for that matter. Therefore, there is no need to be insulting, patronizing, or anything else. If you have nothing worthwhile to add to this thread, then I please ask you... don't contribute anything at all.

Now... this is where I will point out where you are wrong. My first Phenom 9600BE, as well as one other user on this thread, could not get the chip to run properly at *stock* speed. This had nothing to do with an OC. My second Phenom, as I posted, would run x86 Vista with no problems, but would BSoD when trying to install Vista x64. Some of the other external forum posts have shown the same problem. The problem is not simply that of not being able to OC a Phenom, that is a matter of luck or a buyer beware type of deal. The *real* problem is that some, or many, chips are shipping and not able to run reliably at stock speeds due to, in large part, the core 3 problem.

This is tiring, and I have no further interest in defending a thread whose validity speaks for itself...

Millions eh care to show a link to millions of K8L's since there has been a stop ship on them since last year?

Word, Playa.

There has only been a stop ship on the Barcelona based Opterons that were supposed to go to retailers and channel partners. Other things such as super computer builders have been getting 10s of thousands of Barcelona and Agena base quad cores, regardless of the TLB errata.

The phenoms seem to sell halfway decently on sites like Newegg, and other etailers, regardless of what some intel fanatics may lead you to believe, not to mention whatever Phenoms some OEM companies are using.

I have yet to see AMD state anything about this. Nothing, zip, nada, ziltch.

Funny thing about this endyen guy is that I think that your first chip wouldn't run at the stock 2.3GHz stable.

So it is that some chips are defective. Even if it was a QA problem they were labled as quad cores that were certified by AMD to work at 2.3GHz all 4 cores, not just three. But for some strange reason their QA missed it.

What does this all mean for enthusiasts? Well first off the Phenom has no OC guarantee. Neither does the C2Q6600 but most can be OC'ed to 3GHz on air easily. So for enthusiasts it means Phenom, for now, is not the best choice. For those who got one that gives them the bad core2 it means they paid quad prices for a tri core. Thus got ripped off and should either RMA them once B3 is out(if able to) or AMD should bite the bullet and replace them.

Yeah, I'm thinking that if the 3rd core is flawed, it's probably a run of B2s that had this problem during the lithography process, and they didn't catch that early enough, had them available as a quad core. Then they may have figured out the problem, and then disable it and turn it into a Phenom 8000 series.

Before I turn up the fanboyism, I would say that if they do knowingly sell bad quad cores, shame on them, and they should recall them if they want to salvage their reputation. It is a rather deceptive practice.

However, I don't think they're hiding behind the TLB bug. The TLB and "core 2 flaw" _is_ independent, and the BIOS fix (turning off the TLB) would not fix the core 2. People who can't run their Phenom 9000 series should RMA the chip. Whether they would cop to a problem... I don't know. IIRC Intel didn't really cop to the FDIV problem until publicly disclosed anyways, so if they don't cop to it until proven, well they would be doing no worse than Intel.

Now, I think that even the core 2 problem, requiring a downclock, isn't exactly a show breaker. I can hear the Intel fanboy sharpening knives now, but hear me out. With the latest Tom's article on what the Phenom 8000 series could be, there are plenty of benchmarks where the 3 core and 4 core scores are indistinguishable, or at least statistically insignificant. Majority of games aren't optimized beyond 2 cores anyways. This will only manifest as a problem when you use a properly multi-threaded app, and it's pegging all 4 cores.

Secondly, the core is downclocked, not knocked out. This means that the cores are performing differently, and there are ongoing research on heterogeneous systems, where the cores perform differently (mind you "real" heterogeneous systems tend to perform differently due to different functional unit, etc etc). So, a scheduler could use the performance counters of the CPU, find out whichever process that's memory bound and stick that process on the slow core. These processes will not suffer significantly because they tend to wait on memory anyway.

I know, cause I'm doing the graduate course on this type of research, and using a Phenom 9600 BE for simulating heterogeneous systems. When I'm done with it though, I'll probably tinker around with AOD to see whether this is the case.

You're the expert in this field. I know quite a bit about PC's but I'm a hobbyist who works with mainframes in a data center. I know that IBM thought well of SOI a couple of years ago. It's one thing for AMD to go HK/MG at 32nm, but if they're going SOI at 45nm then turning around and redoing a second generation as HK/MG, then that makes me wonder about the first Shanghais.

So far, all we have are rumors. I'll wait and see. I've read somewhere that the SOI implemented by IBM and AMD for Shanghai will differ from the version Intel ditched, so that could very well involve thickening the areas to avoid linkage. plus increasing pipelines for higher clocks. What didn't work on Prescott might work on Phenom. I won't say it's the best route, but it might make Shanghai viable until it's all HK/MG.

You cannot compare different cpu voltage requirements for different designed CPUs being made with different processes

http://www.tomshardware.com/2008/03/18/amd_phenom/

I might have said that......

I think selling the processors with the bad cores shut down is good for AMD. However, I agree with others that this product isn't innovative or earth shattering, it's just going to be a cheaper Phenom with a core disabled.

If they clock higher, they'll do better in single threaded apps, that's the only hope I have for the tripple cripple, I mean core.

Why do people think the only reason for a triple core is because 1 particular core is probelmatic/shot and this is a way of getting rid of it? surely the first explanantion that would come to mind is that the triples are just a way to make use of chips that don't get through the tests, but it could be any core at random playing up, not always the same one.

"You're just kidding yourself if u think AMD is putting out triple cripples just because its a cool idea." excuse me, but may I ask how do you KNOW that they arn't. Oh wait, it's heresay I forgot.

Gosh this thread still chugging along.. hehe..

AMD will come back in approx 2 years time and will be on par with Intel I believe..

INTEL ARE EVIL!!!

No, only the lawyers who approved the OEM rebate program.

I can forgive MBA's who did. Whenever anything like that is floated in a corporation, the suits go to the lawyers for insight into whether it violates any laws. Intel's lawyers dropped the ball on this one. I bet their memos are a hoot to read.

My Phenom runs fine at stock voltage and clocks or how it is now at 2.5GHz (200 X 12.5) at +.05V core. Prime 95 X 4 stable. I have never gotten a BSOD either. I am also running the GIGABYTE GA-MA770-DS3, but I am looking into getting a 790FX for better overclocking my limit on this board has been 2.54 stable.

Arise Necrothread. Arriiise!!