Hi,

I'm wondering how a MPC8640 could be distinguished from a MPC8641. I couldn't find anything in the datasheets about differences in registers.
I heared, that the MPC8640 is actually the same processor as the MPC8641 but only manufactured in a different process and that they can't be distinguished by software.
So I'd appreciate if someone could post how both processors can be distinguished.

Thanks,
Bertram

Why would they be indistinguishable? That doesn't make sense. Every PPC CPU model has an unique id. Sure there are classes of CPU, and 86xx CPUs fall in the e600-type of CPUs, but everyone of these has an id. The kernel will definitely know the type, even if it's not reported in /proc/cpuinfo, then you could definitely write a small program that prints out this id.
The current list of supported CPUs can be found in the kernel tree, in

/arch/powerpc/kernel/cputable.c

Hi Konstantinos,

yes, I agree with you. I'd expect that they can be distinguished by different values in the PVR and/or SVR. But I couldn't find different register settings for the MPC8640.

Btw, both MPC8641 and MPC8640 are not listed in /arch/powerpc/kernel/cputable.c (I checked the latest 2.6.28.8 source).

Bertram

Hi Bertram,

There is no PRACTICAL difference. The MPC8641 and MPC8640 are absolutely identical in every sense, other than that the MPC8640 runs at significantly lower power and has a lower top clock speed - it is essentially an optimized design.

You should not need to tell if you're running on one or the other.

The CPU table references the processor core inside; in this case, it is identical to the MPC7448 and any other e600-based core - with the obvious exceptions of features such as DFS (these are all documented in the MPC8641 reference manual).

_________________
Matt Sealey

Hi Matt,

thanks for the clarification. We don't want to make any difference in the code for the two processors; we just want do display the correct CPU we're running on. It might sometimes be of interest if the board is equipped with a MPC8641(D) or with a MPC8640(D) since this makes a difference in the thermal behavior.

Bertram

Well I'd not go so far as use the processor model to determine how aggressively you power manage or run fans or so. That's what thermal diodes are for :)

I think the best you can hope for is to say you're running an MPC864x - add a D if it's got two processor cores (SVR will read 809001xx where that 1 is whether it's dual-core or not. xx is the revision).

If it's really that important, take into account the speed of the processor. Freescale don't sell MPC8640D past 1.25GHz - so if you have a 1.5GHz chip then you're either over-clocking or undoubtedly on an MPC8641D. If you have a chip lower than revision "0x21" - 99% chance it's an MPC8641 and not an MPC8640..

_________________
Matt Sealey

Agree. Thermal management must not rely on the model string. Nevertheless it would be fine if the correct CPU string could be displayed.

Bertram

I think you will have to live with saying "MPC8640D/MPC8641D" (and drop the D if the SVR does not have bit 9 set) or resolve to work this out based on the hardware revision of the board around it, if this is available.

_________________
Matt Sealey

I was told that the 8640 is a low-power binning of the 8641 die. It is not even an optimized design or different fab technology. It is identical in every way, coming from the same wafers. The chips in the 8640 bin just happened to use less electricity than those selected to be the 8641 on the same wafer.

A real pity that the technology will be left to rust.

I think there's a little difference; the MPC8640D represents the very final revisions of the MPC8641D which was better all round. Bin sorting plus *all* the stuff they did between the first tapeout and "Revision C" makes a lot of improvement.

All in all though it's more marketing than anything. It is an awesome chip; I'm proud to have one here. It beats the absolute crap out of most systems I've used just on processor performance (and it's only 1.5GHz).

I just can't think of a use for it anymore, and I quit trying to run SUSE on it when we got the i.MX boards in.. it will still, if I manage to get Debian installed perhaps, make a wicked awesome Scratchbox/OpenEmbedded compiler.. except for that fan. Which I can't find a replacement for. I need to keep my office door open so the wireless can get through it, and you can still hear it *outside*.. poor component choice on the part of the Freescale reference designers (it's a PWM fan too, why didn't they put a power management chip on there and scale the fan speed?)

_________________
Matt Sealey