Phenom VS. Intel Quadcores (Q6600) ?

Have you used any Phenom system and encountered any stability issue yourself? I have built two Phenom systems and never encountered the issue so far, Manny has built 3 Phenom systems, ask him whether he has encountered the issues? Pepsiennis also uses a Phenom system, has he encountered this issue? Kyle also said before that the TLB bug is not what most user will experience.


I have not run into it.Not with any of the systems I built,or used or tested.
 
The 9600 and the Q6600 have similar performance at 2.4Ghz, but run both at 3Ghz or higher, and the Q6600 runs away.

If AMD were to revise the Phenom so that when it reached the speeds in the range of 3Ghz, peformance scaled proportionally to the clockspeed increase, AMD would have a winner.


I firmly believe it goes beyond AMD having to only fix the NB and L3 so it runs at the same speed or faster then the cores themselves,as was the original plan for K10.

Core has more execution/better units then K10,and a little something called macro fusion

Architecurally speaking even with the many 'improvements' on K10 versus K8,it still pales in comparison to the Core architecure of near 2 years ago.NM the Penryn updates.I dont think adding another 4Mb of L3 cache,for a total of 6Mb L3 (as they claim to plan to do in the late Summer of 2008),will offer the huge bost in IPC we all want to see in K10.

Its not enough ! As the chip in my eyes in its current state,is somewhat hobbled by the speed of its L3 and associated NB.
 
Regarding to the TLB bug, this is what a person who worked with AMD said:

The issue is not that the bug is extremely rare (I never argued that... I don't think anyone in this thread did).
The issue is that some people here claim that the bug ONLY affects virtualization. In other words: if you don't use virtualization, your system is bugfree.
This is NOT true. The bug is still there, rare as it may be. AMD said so.
Again, this quote from morfinx says nothing about the bug only occuring with virtualization.
No matter how rare the bug is... it's still NOT bugfree. Just a matter of principle.
Saying that it's bugfree is a lie. I don't tolerate lies.
 
Core has more execution/better units then K10,and a little something called macro fusion

Actually that macro-op fusion doesn't do much for performance. In fact, it only works in 32-bit mode, yet Core2 maintains its advantage in 64-bit mode.
I also doubt that the 4th ALU does much... The Pentium M had only 2 ALUs, and still racked up impressive IPC. Compared to the 3 ALU Athlons (and I believe original Core?), the 3rd ALU didn't seem to add much to the package, I doubt that a 4th ALU adds anything at all... but hey, when you're way ahead in production technology, why not? :) You have to use that transistorbudget for something...

Architecurally speaking even with the many 'improvements' on K10 versus K8,it still pales in comparison to the Core architecure of near 2 years ago.NM the Penryn updates.I dont think adding another 4Mb of L3 cache,for a total of 6Mb L3 (as they claim to plan to do in the late Summer of 2008),will offer the huge bost in IPC we all want to see in K10.

I think AMD is mainly losing out because of their inferior cache. I think the execution core itself is about as fast as the one in Core2... the main issue is that the Core2 pipeline is longer, which should allow the architecture to scale to higher clockspeeds, without sacrificing IPC... But Intel hasn't quite flexed Core2's muscles in that direction yet.
 
I don't tolerate lies.
Yeah right ;):
I don't need proof. The TLB errata from AMD simply don't specify anything relating virtualization only.
It wouldn't make sense either. If it was only virtualization, they could simply disable that feature and let the CPU run as if it didn't have hardware virtualization at all, but run all other software with no performance loss.
If you look into what the workarounds actually do, you'll understand that it's a general problem with a race condition on some bits in the TLB.
It just affects virtualization most, because it uses the TLB more extensively than regular software does (I think you just misinterpreted that).
However, the TLB is used with regular software aswell, so the bug can and will still occur there, just less frequently.
So if anyone wants to prove otherwise, go ahead and try, but AMD itself says otherwise.



Re-read the thread. I already covered this numerous times.



Probably less than 2 8800GTs in SLI.
 
Too bad 16x/4x on P35 boards hinders performance significantly vs 8x/8x the MSI offers. You also linked to an open-box. We are going off-topic btw.
http://www.bit-tech.net/hardware/2007/10/12/crossfire_comparison_intel_x38_versus_p35/3

Seems to depend a lot on the application.
In some cases the 16x/16x solution is the fastest, in other cases it's the slowest.
In one case even the 16x/4x solution is actually the fastest.
I think this has more to do with driver issues at this time, than with the actual performance of the PCI-e slots. These results are far too irregular to draw any solid conclusions.
 
I don't know, what I understand from "less frequently" is that it is still frequent but not as frequent as virtualization. I also just wanna quote Kyle:

No, less frequently is just less frequent than frequently. It could be much less frequent.
I never said how less frequent, because I don't know, and AMD never said anything about it. They just said that it can and will occur in other situations, it's just 'extremely rare', as Saucier put it.
I don't care what Kyle says, really. It's not likely that one person will experience such an extremely rare bug. So the fact that he hasn't experienced it yet is exactly what is to be expected. It doesn't prove that there is no bug.
Bottom line is: AMD says the bug exists. Want to argue with AMD? You'll lose. They created the bug themselves.
 
No, less frequently is just less frequent than frequently. It could be much less frequent.
I never said how less frequent, because I don't know, and AMD never said anything about it. They just said that it can and will occur in other situations, it's just 'extremely rare', as Saucier put it.
I don't care what Kyle says, really. It's not likely that one person will experience such an extremely rare bug. So the fact that he hasn't experienced it yet is exactly what is to be expected. It doesn't prove that there is no bug.

How about 4x HD3850 would be slower than 2x 8800GT? Can you tolerate that?

I don't know, I think that there is a huge difference between the phase
the bug can and will still occur there, just less frequently
and
it's just 'extremely rare'

What do you guys think :D ?
 
How about 4x HD3850 would be slower than 2x 8800GT? Can you tolerate that?

Sure, I used the word 'probably', to point out that it is my expectation, not a fact. Nobody knows for sure yet. Your guess is as good as mine.
You are trying to put it across as a fact that Phenom doesn't suffer from bugs when you do not use virtualization, even in the face of AMD's claims to the contrary.
That is just an outright lie.

I don't know, I think that there is a huge difference between the phase "the bug can and will still occur there, just less frequently" and "it's just 'extremely rare'".

What do you guys think :D ?

I think you're trying to put in your own interpretation there, to suit your agenda.
Grammatically both phrases mean exactly the same. The terms 'less' and 'extremely' are not quantatitive terms. You can read any meaning into them that you'd like.
What AMD considers 'extremely rare' could be every few weeks. That's billions and billions of instructions, so from a technical point-of-view it indeed is *extremely* rare... from a practical point-of-view it would be quite often.

As for 'less frequently'... Well, once in a century is less frequently than once every microsecond. Can't argue with that.

So let's not get into playing word games, you'll lose.
 
The 9600 and the Q6600 have similar performance at 2.4Ghz, but run both at 3Ghz or higher, and the Q6600 runs away.

If AMD were to revise the Phenom so that when it reached the speeds in the range of 3Ghz, peformance scaled proportionally to the clockspeed increase, AMD would have a winner.



I apologize, I thought were talking about equivilant priced platforms such as the 9600 and the Q6600. I missed the part about the 9500. My bad.

Actually that macro-op fusion doesn't do much for performance. In fact, it only works in 32-bit mode, yet Core2 maintains its advantage in 64-bit mode.
I also doubt that the 4th ALU does much... The Pentium M had only 2 ALUs, and still racked up impressive IPC. Compared to the 3 ALU Athlons (and I believe original Core?), the 3rd ALU didn't seem to add much to the package, I doubt that a 4th ALU adds anything at all... but hey, when you're way ahead in production technology, why not? :) You have to use that transistorbudget for something...



I think AMD is mainly losing out because of their inferior cache. I think the execution core itself is about as fast as the one in Core2... the main issue is that the Core2 pipeline is longer, which should allow the architecture to scale to higher clockspeeds, without sacrificing IPC... But Intel hasn't quite flexed Core2's muscles in that direction yet.



Anand and Johan paint those feetures on C2D in a very different light,and no offense,but I give them and thier opinions,more credence then yours. :) Maybe your right and they are worng,but...

Of course the cache on K10 is very much inferior,and I have posted that several times,but I dont think its,the one main thing holding AMD back here.
 
Sure, I used the word 'probably', to point out that it is my expectation, not a fact. Nobody knows for sure yet. Your guess is as good as mine.
You are trying to put it across as a fact that Phenom doesn't suffer from bugs when you do not use virtualization, even in the face of AMD's claims to the contrary.
That is just an outright lie.



I think you're trying to put in your own interpretation there, to suit your agenda.
Grammatically both phrases mean exactly the same. The terms 'less' and 'extremely' are not quantatitive terms. You can read any meaning into them that you'd like.
What AMD considers 'extremely rare' could be every few weeks. That's billions and billions of instructions, so from a technical point-of-view it indeed is *extremely* rare... from a practical point-of-view it would be quite often.

As for 'less frequently'... Well, once in a century is less frequently than once every microsecond. Can't argue with that.

So let's not get into playing word games, you'll lose.
All your statements are here, people can judge themselve :cool:

I don't know what you are up to but Manny, me, Pepsiennis and probably Kyle have used Phenom ourselves and didn't come across with the bug so I think that the bug is not what most users should be concerned with but you want to make the bug issue seems bigger than what it really is.

Btw, I can't recall claiming that the CPU is bugfree, if I remember correctly, what I said before was the bug won't affect a normal user so now I don't know who is the liar here :D
 
Anand and Johan paint those feetures on C2D in a very different light,and no offense,but I give them and thier opinions,more credence then yours. :) Maybe your right and they are worng,but...

Because they have their own website? :)
It's not about credibility though. This is technology. I write my own testroutines to figure out what different CPUs can and cannot do. Ofcourse the problem there is that I only aim at the sort of software that I'm writing.
I'm not sure if they do any kind of testing with actual software at all... but if they do, it could be that they focus on something completely different, and get completely different results... but then we'd still both be right :)
I tend to just write routines that depend entirely on registers, to rule out caches and memory, so I can see what the 'bare' core can do. Then when you add cache and memory, you know how much those affect the complete picture.

Of course the cache on K10 is very much inferior,and I have posted that several times,but I dont think its,the one main thing holding AMD back here.

What do they consider the main thing that holds AMD back then?
I think we can rule out macro-op fusion anyway, as I said before.
 
I don't know what you are up to but Manny, me, Pepsiennis and probably Kyle have used Phenom ourselves and didn't come across with the bug so I think that the bug is not what most users should be concerned with but you want to make the bug issue seems bigger than what it really is.

It's entirely in your head. I never said anything about the frequency of the bug (I don't know the frequency within or without virtualization usage, because AMD never specified that).
All I did was point out that the bug DOES affect people who don't use virtualization, as AMD itself stated.
Your problem is that you feel the emotional urge to downplay this bug, so you somehow mistake me for claiming that the bug occurs often, or is very severe or anything... But I never said that. In fact, I have stated the opposite many times, in this and other threads. But you seem to be blind to that. You just want to attack me in your irrational rage.

Btw, I can't recall claiming that the CPU is bugfree, if I remember correctly, what I said before was the bug won't affect a normal user so now I don't know who is the liar here :D

Let me refresh your memory then... Page one:
alg7_munif said:
We have ran two Phenoms so far and never encountered any instability due to the TLB bug because the bug will only affect the virtual machine usage which most the user will not use anyway.

That pretty much says the bug doesn't exist outside virtual machine usage. Which is not true.
 
It's entirely in your head. I never said anything about the frequency of the bug (I don't know the frequency within or without virtualization usage, because AMD never specified that).
All I did was point out that the bug DOES affect people who don't use virtualization, as AMD itself stated.
Your problem is that you feel the emotional urge to downplay this bug, so you somehow mistake me for claiming that the bug occurs often, or is very severe or anything... But I never said that. In fact, I have stated the opposite many times, in this and other threads. But you seem to be blind to that. You just want to attack me in your irrational rage.



Let me refresh your memory then... Page one:


That pretty much says the bug doesn't exist outside virtual machine usage. Which is not true.
Playing the same word game? I didn't said that it is bugfree, it means the same thing that most user won't be affected by the bug anyway. Please give me a prove that shows the bug has affected a user a part from using virtualization.
 
Playing the same word game? I didn't said that it is bugfree, it means the same thing that most user won't be affected by the bug anyway.

No, you literally said that "the bug *only* affects virtual machine usage". That is simply not true. No amount of word games can change that, the usage of the word 'only' in that sentence is simply wrong. You should have been using the word 'mainly'.
 
No, you literally said that "the bug *only* affects virtual machine usage". That is simply not true. No amount of word games can change that, the usage of the word 'only' in that sentence is simply wrong. You should have been using the word 'mainly'.

Fair enough, so do you agree that the bug is so rare that most users won't be affected by it and therefore it is not much of a concern for most users?
 
Fair enough, so do you agree that the bug is so rare that most users won't be affected by it and therefore it is not much of a concern for most users?

Yes, but why do you ask? I already said that multiple times in this thread, and I don't think I've ever said anything otherwise anyway?
(By the way, in that sense I interpret 'concern' as "Even if they may experience a crash because of it, it's no big deal"... I won't go as far as saying most users won't ever actually get a crash that is caused by this bug... just that they'd probably get more crashes for other reasons, so it doesn't make a difference)
 
Yes, but why do you ask? I already said that multiple times in this thread, and I don't think I've ever said anything otherwise anyway?
(By the way, in that sense I interpret 'concern' as "Even if they may experience a crash because of it, it's no big deal"... I won't go as far as saying most users won't ever actually get a crash that is caused by this bug... just that they'd probably get more crashes for other reasons, so it doesn't make a difference)

So you think that most user would get a crash that caused by this bug?
 
So you think that most user would get a crash that caused by this bug?

You can get silent errors and data corruption.To a casual gamer it might look like an artifact in a scene , but most people use computers to do real work.I bet someone wouldn't be happy to experience a crash or data corruption while working on Excel tables.

It's silly to put out a CPU which has a known error and say : it's for gamers and websurfers only.I mean , wtf , :rolleyes: ?!
 
I'm NOT saying that most people won't get a crash.
They might not, but I'm not going to guarantee it.

So are you saying that most user would get a crash that is caused by the bug or are you saying that it is out of your expertise to say whether most user would or would not get a crash that is caused by the bug?
You can get silent errors and data corruption.To a casual gamer it might look like an artifact in a scene , but most people use computers to do real work.I bet someone wouldn't be happy to experience a crash or data corruption while working on Excel tables.

It's silly to put out a CPU which has a known error and say : it's for gamers and websurfers only.I mean , wtf , :rolleyes: ?!
Nothing other than system instability related to the erratum 298 was mentioned here:
http://techreport.com/discussions.x/13724

So Scali, can you tolerate this :p ?
 
So are you saying that most user would get a crash that is caused by the bug or are you saying that it is out of your expertise to say whether most user would or would not get a crash that is caused by the bug?

Yea, I can't say anything about it... It depends on so many things... The quality of CPU dies is pretty much variable. Some CPUs may overclock higher than others, some CPUs may suffer from this bug more severely than others.
Then there's the software people run on their systems.... how long people use these systems... and all that.
The only thing I can say is this:
Since AMD pulled the Opteron line altogether, and has instructed BIOS developers to enable the fix by default, AMD doesn't seem to think that the bug is so rare that most people won't ever see it.
AMD is not taking chances, and given AMD's current financial state, I think that speaks volumes. The last thing they need is their flagship CPU being a problem child.
If this bug really wouldn't affect most users, why would they even make the bug public? They could just silently RMA the few CPUs that users have problems with, and pretend it's just a production flaw, not a structural design problem. Nobody would ever know.

Nothing other than system instability related to the erratum 298 was mentioned here:
http://techreport.com/discussions.x/13724

So Scali, can you tolerate this :p ?

He's right, actually. The bug means that the TLB can return outdated addresses to the core logic. In some cases this may cause access violations or even crashes. In other cases these addresses are accidentally valid, just not the addresses you were expecting. In that case the CPU would just continue to work with the wrong data, so you'd get silent errors, data corruption and all that.
 
Yea, I can't say anything about it... It depends on so many things... The quality of CPU dies is pretty much variable. Some CPUs may overclock higher than others, some CPUs may suffer from this bug more severely than others.
Then there's the software people run on their systems.... how long people use these systems... and all that.
The only thing I can say is this:
Since AMD pulled the Opteron line altogether, and has instructed BIOS developers to enable the fix by default, AMD doesn't seem to think that the bug is so rare that most people won't ever see it.
AMD is not taking chances, and given AMD's current financial state, I think that speaks volumes. The last thing they need is their flagship CPU being a problem child.
If this bug really wouldn't affect most users, why would they even make the bug public? They could just silently RMA the few CPUs that users have problems with, and pretend it's just a production flaw, not a structural design problem. Nobody would ever know.
Companies are just conservative, it's the same like overclocking, what they guarantee is the chip would run at the stock speed problem free, they don't guarantee that it would run faster, but it depends on the user to take the risk. The same thing applies to the bug, they want the fix to be turned on by default just to cover their own ass.

He's right, actually. The bug means that the TLB can return outdated addresses to the core logic. In some cases this may cause access violations or even crashes. In other cases these addresses are accidentally valid, just not the addresses you were expecting. In that case the CPU would just continue to work with the wrong data, so you'd get silent errors, data corruption and all that.
Can you back this up?

Btw a good read regarding the bug:
http://www.lostcircuits.com/cpu/amd_phenom/2.shtml
 
Companies are just conservative, it's the same like overclocking, what they guarantee is the chip would run at the stock speed problem free, they don't guarantee that it would run faster, but it depends on the user to take the risk. The same thing applies to the bug, they want the fix to be turned on by default just to cover their own ass.

Yes, because there is a need to cover up their ass :)
As I say, if they were confident that only one in every few thousand CPUs may have a problem... they would just RMA individual CPUs.

Can you back this up?

Why? If you understand the official AMD errata docs and the explanation of the fix implementation, it should be obvious.
If you don't understand what a TLB actually is for, and what the bug changes about the TLB behaviour... then I don't really see what else I could do to back it up.
Really, I wouldn't just trust whatever hardware site writes up about bugs such as these, because most of the time these people don't understand it either. It's something that only OS/driver developers and CPU designers are familiar with. And usually such people can get better jobs than online journalist, no offense :)
Just stick to the official AMD info and official AMD quotes... and only believe things that make sense based on that.
I've seen a ton of nonsense on CPUs on internet sites over the years... A classic one I recall was one where somehow sites claimed that a CPU had improved FPU performance because of better branch prediction. Since the FPU has nothing to do with branching, the whole statement makes no sense... yet nearly every big review site parroted this. I'm not sure which CPU this was... I think the Pentium 4 Northwood?
 
Why? If you understand the official AMD errata docs and the explanation of the fix implementation, it should be obvious.
If you don't understand what a TLB actually is for, and what the bug changes about the TLB behaviour... then I don't really see what else I could do to back it up.
Really, I wouldn't just trust whatever hardware site writes up about bugs such as these, because most of the time these people don't understand it either. It's something that only OS/driver developers and CPU designers are familiar with. And usually such people can get better jobs than online journalist, no offense :)
Just stick to the official AMD info and official AMD quotes... and only believe things that make sense based on that.
I've seen a ton of nonsense on CPUs on internet sites over the years... A classic one I recall was one where somehow sites claimed that a CPU had improved FPU performance because of better branch prediction. Since the FPU has nothing to do with branching, the whole statement makes no sense... yet nearly every big review site parroted this. I'm not sure which CPU this was... I think the Pentium 4 Northwood?
So instead of explaining and proving that this is true, I guess that it is much easier to say that others couldn't understand it so that it would make you look more educated and people would accept that this is true without the need of any explaination.
 
So instead of explaining and proving that this is true, I guess that it is much easier to say that others don't understand it so that it would make you look educated and people would accept that this is true without the need of any explaination.

What exactly do you want me to explain then?
I don't see it... Take the literal text from AMD:
The processor operation to change the accessed or dirty bits of a page translation table entry in the L2 from 0b to 1b may not be atomic. A small window of time exists where other cached operations may cause the stale page translation table entry to be installed in the L3 before the modified copy is returned to the L2. In addition, if a probe for this cache line occurs during this window of time, the processor may not set the accessed or dirty bit and may corrupt data for an unrelated cached operation. The system may experience a machine check event reporting an L3 protocol error has occurred. In this case, the MC4 status register (MSR 0000_0410) will be equal to B2000000_000B0C0F or BA000000_000B0C0F. The MC4 address register (MSR 0000_0412) will be equal to 26h.

I already made the relevant part of this bug bold.
Take this, and what I said above about the consequences... and 2+2=4?
I don't see what there is to explain about it?
In fact, even the article you linked says it can result in data corruption, on page 3:
Since the TLBs are used to find the task-specific data within the virtual memory address space, this could lead to updating data in system memory with data that do not pertain to the task at hand but to another cached operation. This is generally referred to as data corruption.

So don't just take my word for it. Just read the articles you link to.
 
You seriously expect an answer to that question?
You don't understand how completely random the behaviour of this bug is?

Yes I expect an answer for that, so please. What most people will think when they read the post is that they will have problems with Excell.
 
Yes I expect an answer for that, please.

I already answered it... The behaviour is completely random.
If the bug occurs you are trying to read or write 'some data', and because of an error in the address translation you are actually reading or writing 'some other data'.
Anything could happen, just depends on where and when the bug decides to occur. Completely unpredictable.
 
I already answered it... The behaviour is completely random.
If the bug occurs you are trying to read or write 'some data', and because of an error in the address translation you are actually reading or writing 'some other data'.
Anything could happen, just depends on where and when the bug decides to occur. Completely unpredictable.

Yet it is rare, you don't know whether most user would be affected or not and you also don't know what could happen :D
 
Yet it is rare, you don't know whether most user would be affected or not and you also don't know what could happen :D

I know exactly what happens. Problem is, it happens at the hardware level, and you're asking about consequences for the software that is running at that time.
It's the same as asking "What happens in Excel when I'm using a faulty stick of memory?".
Can you answer me that one?
 
Yet it is rare, you don't know whether most user would be affected or not and you also don't know what could happen :D

It would most likely cause a system hang because the hardware is attempting to access data that doesn't exist or has already been rewritten. It can and has been reproduced on Linux environments (thus the hotfix) without the use of virtualization.

Quite frankly, it doesn't matter for consumers because no one gives a damn if your system crashes once every month of hardcore gaming (in which you'd blame on Microsoft anyways).
 
Quite frankly, it doesn't matter for consumers because no one gives a damn if your system crashes once every month of hardcore gaming (in which you'd blame on Microsoft anyways).

Funny enough that article he linked to already took a dig at Vista :)
Does any of this affect the standard Desktop user? Sure, when hell freezes over. Especially in any situation where Microsoft Vista is used, the entire thing is a completely moot point since the OS will crash a gazillion times before the “Erratum 298” bug is encountered.

Could they be experiencing Phenom crashes and just blaming Vista? :)
Why don't we ask the Vista-users here how stable Vista is? I've seen the various Vista threads on this forum, and it seems the Vista users here generally are tired with this kind of FUD, and Vista is actually received relatively well here. Haven't seen many complaints about stability.
 

Which is of course because of I had more nVidia builds than ATi (so please put the sword back). Second on my list is Silicon Image sata controller on my nForce2 mobo, next is IRQ_NOT_LESS_OR_EQUAL, I've also had some ATi driver flash video crash before but they are nowhere near the number of crash due to a Phenom bug which I never had it yet. I won't put the blame on M$ because for me, since Win2K, M$ Windows can be considered stable already. WinME and former are hardly stable for my liking.
 
IBTL +1

I recently upgraded from a 939 x2 3800 to a Q6600, in my situation it made good sence as i had to purchase a new systemboard, ram, and CPU so I took the path that at this point has more longevity. At the time of my upgrade the Phenom was about 200$ compared to the Q6600 oem at 260$ so for 60$ more I was able to put together a system that should have a few years of good productivity in it.
 
Back
Top