CPU microcode, flash memory, and longevity?

[VirtualLarry]

Just thought of something. If the microcode on a CPU is stored in flash memory, and flash memory is only good for storing info for 5 years, then does that mean that most modern CPUs, will eventually start to fail after 5 years, regardless of voltage-induced degradation?

 

[GammaLaser]

AFAIK, Microcode patches are stored in BIOS NVRAM, and is loaded into the CPU during startup. The CPU loses patches every time it is powered down. In other words, within the CPU, the patches are not stored in Flash but in a volatile memory like SRAM.

Source: Intel System Manual, Section 9.11 "Microcode Update Facilities" http://www.intel.com/Assets/PDF/manual/253668.pdf

 

[VirtualLarry]

AFAIK, Microcode patches are stored in BIOS NVRAM, and is loaded into the CPU during startup. The CPU loses patches every time it is powered down. In other words, within the CPU, the patches are not stored in Flash but in a volatile memory like SRAM.

Source: Intel System Manual, Section 9.11 "Microcode Update Facilities" http://www.intel.com/Assets/PDF/manual/253668.pdf

The patches aren't stored in flash, but isn't the core microcode stored in some sort of writable memory? Isn't that how Intel differentiates between the Core i3-2100T and the G620T, both being Sandy Bridge dual-cores, certainly the same exact CPU die, but the G620T lacks SSE4.1, 4.2, and possibly HyperThreading.

 

[Topweasel]

The patches aren't stored in flash, but isn't the core microcode stored in some sort of writable memory? Isn't that how Intel differentiates between the Core i3-2100T and the G620T, both being Sandy Bridge dual-cores, certainly the same exact CPU die, but the G620T lacks SSE4.1, 4.2, and possibly HyperThreading.

Yeah but I thinking you are missing something. Microcode update is like a driver or patch for your CPU. In the past this is stored in the emprom for the BIOS. Lets say its stored in flash on for the new EFI BIOS's. It still located on the motherboard and not the CPU and is only loaded onto the CPU during the bootup process and is lost during power off. In this type of setup, the CPU will never go bad, well at least for that reason. Of course it could be an issue for the motherboard. But isn't that almost always the way it is motherboard dies before CPU?

 

[Wizlem]

I think the core microcode isn't in any sort of easily writable memory.

http://en.wikipedia.org/wiki/Read-Only_Memory

Edit: Better Article

http://en.wikipedia.org/wiki/Mask_ROM

 

[Topweasel]

I think the core microcode isn't in any sort of easily writable memory.

http://en.wikipedia.org/wiki/Read-Only_Memory

True. Well at least with dealing with older BIOS's that are written to an Eeprom. But I don't know if that still applies to nand flash Bios's of today. Could be that its still stored in a more resilient storage facility on the board. (well looking over it nand flash is still considered an eeprom, but the point being the older chips where of a much longer lasting (and harder to write to, and horribly space inefficient tech).

Also I believe that the longevity has nothing to do with how long the material will last but calculations on write limits of the sectors and expected amount of use. If that is the case then its still going to last a long time because even the worst at 3000 writes isn't going to hit that limit unless hitting that limit was the goal (like changing a bios setting a 3k times).

 

[Voo]

I think the core microcode isn't in any sort of easily writable memory.

http://en.wikipedia.org/wiki/Read-Only_Memory

Edit: Better Article

http://en.wikipedia.org/wiki/Mask_ROM

If that was true the microcode couldn't be updated, which we know is possible for modern Intel CPUs - Intel describes the process in detail in their SPG (but without discussing internal details). They could store the updates in a special area and make lookups first in the patch table to see if some code was overwritten, or they could load the original code in SRAM and then overwrite the changes, or they could just store the data in EEPROM, who knows. I don't think Intel would publish such internal details.

 

[tweakboy]

As long as its not a moving part, ur longevity will be,10 years or more...
gg and gb

 

[Topweasel]

As long as its not a moving part, ur longevity will be,10 years or more...
gg and gb

Well that's wholly inaccurate and quite a generalization. If that was the case my batteries would last for more then 2 years.

 

[CTho9305]

Microcode is stored in a ROM (usually programmed by the presence/absence of vias or contacts), and patches are stored on the motherboard and loaded at boot time into an SRAM on the processor (which is always lost on poweroff and may be lost on reset depending on the specific implementation). There are patents that disclose a LOT of detail if you're interested. The ROM won't meaningfully degrade; if the motherboard flash degrades, well, you can't boot anyway.

As for differentiating products made from the same piece of silicon, I believe that's usually done by blowing fuses (basically vaporizing or otherwise damaging a small wire/conductor, by running a high current through it).

 

[Topweasel]

Microcode is stored in a ROM (usually programmed by the presence/absence of vias or contacts), and patches are stored on the motherboard and loaded at boot time into an SRAM on the processor (which is always lost on poweroff and may be lost on reset depending on the specific implementation). There are patents that disclose a LOT of detail if you're interested. The ROM won't meaningfully degrade; if the motherboard flash degrades, well, you can't boot anyway.

As for differentiating products made from the same piece of silicon, I believe that's usually done by blowing fuses (basically vaporizing or otherwise damaging a small wire/conductor, by running a high current through it).

Well in case of the Athlon, it was literally cutting lines in caseing in between fuses with high powered lasers. Loved the pencil trick to death. Didn't think I would ever have a use for good old #2 after I was done with school.