From the Freescale "Design News: Volume 8, Issue 19":
anyone knows more about this and its difference to 8641D? Both Google and a search on Freescale site didn't show up anything.

Now that's interesting!

I wonder if it will come in a single-core version as well (suitable for, say, MorphOS)? Probably, since AFAIK the single core version is the same dual core chip where one of the cores are defunct/turned off, right?

And I wonder what the price will be for that one?

The original design documents for the MPC8641D mentioned that a single core design - one with a true, single core - would be made, but in the end it turned out that they would just turn off broken cores in validation and testing and ship those as single core chips. This actually made single core designs just as expensive as dual-core ones, and as such, I don't think anyone has even bothered making them.

The MPC8640D would probably be much the same; it's not worth making a single-core design out of it, that's a significant resource drain and would take time to engineer, especially now the MPC8610 is out, there is no need for a single-core MPC864x design. The e600 product roadmap is pretty complete.
Cheaper.

I can only assume the lower power is from less L2 cache (since this is a hell of a lot of space and the MPC8641D has 2MB L2 in total - I would guess the new design would have 256kb per core as with the MPC8610) and a lower top end clock frequency (same as the MPC8610 - 1.3GHz, rather than the 1.5-1.8GHz they specced for the MPC8641D)

The problem with an MPC8641D or MPC8640D design then, is graphics. There is no DIU in either of these designs (or at least we do not know yet for the new one) so you have to add external graphics adapter. We're going back to the Radeon card kerfuffle of old with that chip...

_________________
Matt Sealey

Wasn't this also the case back in the 680x0 days? When some parts of a CPU didn't work they were simply turned off and the chip rebadged with a new version number and sold at a lower price. Hence the 68LC040 didn't have a FPU and the 68EC040 had neither FPU nor MMU, despite the actual silicon came from the very same production batch, made from the same 68040 blueprints?
That's a bit weird pricing philosophy IMHO. OK, I can see how the production costs are indeed identical, but in the 68k case they tried to recover some of it by selling the "faulty" ones as feature reduced Low Cost versions instead of throwing them in the trash (which would have meant $0 in sales for those units). But who would have bought a 68EC040 at the same price as a *full* 68040?
I can see how none of these chips would fit in an "Efika" kind of design, i.e. something that's ultra mobile and tightly integrated. But in a "Pegasos" context it would be a different matter.

With a PCI-e slot (8x lanes routed to a 16x slot) for an external graphics card and an AMD SB750 southbridge you would reach some *very* impressive system system specs! And supposedly the 8640, 8640D, 8641 and 8641D are all pin compatible, giving additional flexibility in configuration option.

Yes, but it's also possible (as with Intel Core Solo and Intel Core Duo) to make chips with the same core complexes but just one less CPU part and less cache, which is a lot cheaper in terms of silicon wafers (you can usually get 3+ single core chips for the price of 2 duals on the same size silicon wafer).

However that involves designing the core around a single core first, and then manufacturing and testing it, which may cost more in the long run. I can only assume they decided it would be better to design a new chip around a single core, and fix some problems in the old design while they were at it.
When they first came out? Everyone, as far as I recall the pricing only came down when full 68040s hit some kind of magical plateau of production where yields were high enough, or when the 68060 came out, and the 68040 was cheaper, and they had a bunch of chips in a trashcan waiting to be recertified for lower cost sales :D
Except there is no MPC8640 or MPC8641, and if you were opting in a design to use the MPC8640D, it means the MPC8641D didn't meet your needs already (too hot, too broken?)

_________________
Matt Sealey

Or too expensive to be included in a product aimed for a certain target audience...

Anyway, it will be interesting to see more info about this one! :-)

We know the target audience will accept and buy a full-featured workstation totaling many thousands of dollars (see Apple Quad G5), so cost is not an issue so much as what can you do once you have thrown your wallet to the moths.

However we didn't think people would even pay $799 for a dual-core workstation of that spec, given the problems we had with the chip. At 1.5GHz and above the MPC8641D had a far greater power dissipation than Freescale intended. At lower speeds, we didn't think it would be worth doing from a cost:performance ratio point of view.

We do not play the clock speed numbers game, much to some customers' chagrin, but we think we can produce a much better product if we only consider how much bang you get for your buck and not just how fast we can clock a chip.

At the end of the day, dual-core Power Architecture chips would be a boon, but we still haven't gotten all of the performance out of the single-core ones yet - and we can do single-core, especially with the MPC8610, at a much lower power rating (a whole system at 1.3GHz taking as much power as the MPC8641D chip did at 1.6GHz) and with reasonable, and very pleasing performance characteristics (just from a look-and-feel point of view, it's 2x-3x faster than your Pegasos G4/ODW)

_________________
Matt Sealey

What about putting a lot of MPC5200's in a motherboard? Would it be viable? If they were 8610, performance could be very serious. Oh, well, there's this little thing called software that is missing.
Seriously, Genesi has never talked about multiple processors in a product, at least not since those "dual CPU" blurbs about the original Pegasos.
What about clstering? The current Efika 5200 is very suitable for that isn't it. Hasn't anyone tried clustering these ever?
GIMME!

No, really, when will this nice 8610 (with unexpected performance) will get soldered to a Genesi board? It's a pity you can no longer make computers just for the sake of it (like the original Pegasos). You have a buyer for the Efika 8610 yet?

AFAI understand, the 8610 is currently under software adjustment. I.e. first there must be an optimized Linux for the development board, and then Carda et al. can go to work and create an own board. W/o an optimized Linux, evaluation of the own design is hardly possible.
The thread where Neko got the 2-3 times faster impression from was promising, because Linux does run reasonable now, but still needs some adjustments.

We always have thought about clustering..
That's because the cost/performance ratio was also pretty bad. It would have cost literally more than twice the money for single processor, require some very heavy cooling, and wouldn't amount to more than a 33-50% speed increase (which does not justify double the cost at all)

We have spoken briefly about resurrecting the High Density Blade, or using many MPC5121E or MPC5123 chips and the same communication/chassis architecture..

It's possible to link processors through PCI Express or RapidIO or even directly connected Ethernet on a single board - I am sure someone made a 8-processor MPC8641D board at some point. However you have to find customers for this.
You can really only do that once before you need to pay the bills..

_________________
Matt Sealey

The kernel patches from Freescale are still being reviewed. It's possible to bootstrap an MPC8610 board with firmware, and get a broken Linux on it, but what would be the point in that?

We will wait a little bit for the patches to stabilize (it seems it's going to be my job to patch wrangle for Peter since it's a horrible mess right now, I may make a status report here or on the SuSE thread about it), and concentrate on the firmware and drivers in firmware, on the development boards :)

_________________
Matt Sealey

In any case, there is an 8610 Development System on the way to Konstantinos. We will have a couple of meetings with the e600 FSL folks in the weeks ahead (864X is part of that). We will see what we can find out about the '8640' if there is such a thing.

R&B :)

_________________
http://bbrv.blogspot.com

Source: Freescale

that would replace my x86 junk nicely.

@Karl

Thanks! :-)

Well, up to 27% less power consumption (and heat) and costing up to 37% less, this one is quite interesting IMHO. :-)

Couple this chip with a AMD SB750 southbridge, and you could have a quite nice machine:

- One (or two) e600 core with altivec running at 1.25GHz
- PCI Express x16 slot for graphic cards (8 lanes connected)
- 4 Gigabit Ethernet connectors
- Up to 6 SATA 3.0 Gbit/s hard disk drives, with RAID 0, 1, 5, 0+1 support
- eSATA
- 1 IDE channel for HyperFlash module
- Up to 14 USB ports (12 USB 2.0 and 2 USB 1.1)
- HD Audio
- Infrared receiver/transmitter port compatible with IrDA standards
- Super I/O (not really needed with that many USB's?)

Make it as a Micro-ATX motherboard. A perfect little developer desktop! :-)

With the dual core version, you could even have two operating systems running at the same time! :-)

(Edit: with regards to the discussion previously in the thread, it seems like *there is* a price difference between the single core and dual core now; the dual core is 33% more expensive than the single core)