AMD CPU bug

http://www.fudzilla.com/home/item/26187-amd-confirms-minor-cpu-bug

The fud article makes it seem far less threatening than does your post.

The specific compiled version of
the fill_sons_in_loop code, through a very specific sequence of
consecutive back-to-back pops and (near) return instructions, can
create a condition where the process or incorrectly updates the
stack pointer."

<<endquote

We exchanged a few emails to try to come up with a good test case.
Owing to the difficulty of reproducing the bug...

It'll be interesting to see just how many chips are affected. Considering the PII 840 is listed there might be a good chance that the Deneb family is included which would put my 955 in that list. I've never experienced any random BSODs or strange behavior, though.
 
Looks like someone found a fairly serious bug affecting some AMD processors. There are two confirmed CPUs with the bug: the Opteron 6168 and Phenom II X4 820. More are probably affected. The bug causes programs or the OS to crash under heavy loads.

I really hope a simple microcode update can fix this.

Or a workaround in the kernel.
 
Let's look at some language....


The specific compiled version of....

through a very specific sequence of consecutive....

can create a condition....

where the processor incorrectly updates the stack pointer.

....That is such a rare occurrence, you almost have to MAKE a circumstance for it

Summary: Nothing different from every other CPU out there.
 
pretty much, i've never once seen this issue nor have i seen a single person mention anything like this happening on this forum.. and given the fact that we have about 10-15 people on the folding team that run multiple 2p and 4p magny cour opteron systems full load 24/7 and have never seen this issue i really wouldn't worry about it.
 
now if they can only find a bug in the ps3 cell to let us get lv0 access.
 
Relax, I'm not trying to crap on AMD. And yes, I'm am aware that there are errata for CPUs. :rolleyes:
 
"AMD has taken your example and also analyzed the segmentation fault and the fill_sons_in_loop code. We confirm that you have found an erratum with som e AMD processor families. The specific compiled version of the fill_sons_in_loop code, through a very specific sequence of consecutive back-to-back pops and (near) return instructions, can create a condition where the process or incorrectly updates the stack pointer. "

"Engligh motherfucker, do you speak it?!"
seriously though...can someone translate?
 
Translation: the guy found a specific order of CPU instructions, which results in something else than it should. AMD will now document in a PDF document called CPU errata that this specific order of instructions at this specific set of conditions will end with error, and the compiler should not create such set of instructions in that specific order. Same as with any other dozens of errors in any CPU on market - the only real difference is that this time it was found by outsider, not the CPU manufacturer itself.
 
Translation: the guy found a specific order of CPU instructions, which results in something else than it should. AMD will now document in a PDF document called CPU errata that this specific order of instructions at this specific set of conditions will end with error, and the compiler should not create such set of instructions in that specific order. Same as with any other dozens of errors in any CPU on market - the only real difference is that this time it was found by outsider, not the CPU manufacturer itself.


translation of the translation.. it has absolutely no effect on every day system users but i wanted my 15 minutes of fame..

that pretty much sums everything up in a nut shell. like faugustztin mentioned this so called error effects maybe .1% of programmers/compilers. otherwise the problem would of been found a long time ago. there's no need for AMD to fix the problem.

he found the pot at the end of the rainbow but there wasn't any gold in it.. lol
 
"AMD has taken your example and also analyzed the segmentation fault and the fill_sons_in_loop code. We confirm that you have found an erratum with som e AMD processor families. The specific compiled version of the fill_sons_in_loop code, through a very specific sequence of consecutive back-to-back pops and (near) return instructions, can create a condition where the process or incorrectly updates the stack pointer. "

"Engligh motherfucker, do you speak it?!"
seriously though...can someone translate?

Short story is someone found a certain sequence of instructions that causes an incorrect result returned, AMD was able to reproduce the finding, so now that knowledge can be added to the errata documentation, that way compilers or bios updates etc can avoid it.

Errata is common in processor design, it's usually not a big deal for the average Joe, but it's nice to discover them so they get handled - regardless of how rare the chance of actually encountering the scenario triggering them may be.
 
Just to ask a question:
How many actual REAL applications and programs were EVER Affected by the Pentium F0 0F bug? I mean real world programs you can buy, not programs written for nuclear scientists trying to recreate wormholes with the Haldron Collider....
 
Just to ask a question:
How many actual REAL applications and programs were EVER Affected by the Pentium F0 0F bug? I mean real world programs you can buy, not programs written for nuclear scientists trying to recreate wormholes with the Haldron Collider....

Here you go. The main reason it was a problem is that anything could attempt to execute the command and cause the computer to freeze up.

The Pentium FDIV bug was reproducible in Excel or calculator programs that use floating-point numbers.
 
Considering the complexities of modern processors, be they CPU or GPU or whatever, I'm surprised more discoveries like this aren't made. That's really a testament to how well these things work and the effort involved in finding an actual reproduce-able flaw.
 
Let's look at some language....


The specific compiled version of....

through a very specific sequence of consecutive....

can create a condition....

where the processor incorrectly updates the stack pointer.

....That is such a rare occurrence, you almost have to MAKE a circumstance for it

Summary: Nothing different from every other CPU out there.

Nothing different...EXCEPT that all of their AMD machines fail to complie the code they have to compile. In other words it interferes with their work and the work around is to use Intel boxes.

But relax, I am sure it will be the least of your concerns, I mean, playing whatever your favorite game is and all. Besides, a BSOD to you is probably just normal anyway.
 
Nothing different...EXCEPT that all of their AMD machines fail to complie the code they have to compile. In other words it interferes with their work and the work around is to use Intel boxes.

But relax, I am sure it will be the least of your concerns, I mean, playing whatever your favorite game is and all. Besides, a BSOD to you is probably just normal anyway.

Intel has its own long sheet of errata, fixed and not fixed, for each processor they make. It's common in any complex processor design.

The serious issues are found through extensive testing and taken care of before the products even ship, benign ones like this are found before or after product design and often get handled through things like the BIOS etc; they're pretty much transparent fixes to the users.

Quite a non-issue for the average Joe, it's actually pretty cool that a researcher was able to find this one, can't imagine it was an easy task. It's the sort of thing NASA might care about, not people playing a video game.
 
Nothing different...EXCEPT that all of their AMD machines fail to complie the code they have to compile. In other words it interferes with their work and the work around is to use Intel boxes.

But relax, I am sure it will be the least of your concerns, I mean, playing whatever your favorite game is and all. Besides, a BSOD to you is probably just normal anyway.

...... and the solution to all of the Intel errate is to buy AMD.

You fail miserably at logic. Get a sales job at Apple, ok?
 
translation of the translation.. it has absolutely no effect on every day system users but i wanted my 15 minutes of fame..

Reporting a legitimate CPU bug to the hardware vendor is not just wanting "15 minutes of fame".
 
Back
Top