Thanks Matt :-)

What's about the AltiVec-Unit? Equal to the unit in the 7448?
Which AltiVec uses the PWRficient and POWER6 processors? More like 7400 or more like 7448 or something different?

No informations about the PWRficient und POWER6 VMX available :(

What model iBook is it?

i have 2 ibooks, 1066 Mhz and 1333 Mhz G4, both have the maximum ram in them, 1.25 and 1.5 GB.

Apple never used a bus speed faster than 200MHz...

yes i know the ibooks dont have a fast bus speed. i know that will help alot. i was thinking that 2 Ghz would be better than 1 Ghz, and dual core, and 64 bit.

i know 64 bit isnt needed, i thought it would be nice to have.

matt

is there any current info for a new efika besides cpu? what else it will have ? for example usb, firewire? would it have openbios ? would it have dvi and support high def wide screen or only have a pcie slot?

matt

The 7448, 8610 and 8641 use freescale's e600 core (the 7448 was the first to use the e600 core), Altivec is part of the e600 core. But they differ in their additional integrated components (L2 cache size, mem controller, network, etc.) and the production process.

AFAIK, the heart of the 8610 is essentially a 7448 (with smaller cache though, probably to make room for all the other goodies), i.e. an e600 family processor, but the 8610 is the latest and the best of them all.

Thanks
What's about the VMX unit in PWRficient and POWER6?

Is a 8610D in the works?

MPC7410, MPC7447, MPC7457 and MPC7447A are other examples of CPU's using the e600 core other than those you mentioned.

Pre 7448 CPUs aren't e600 cores.

e600 cores have:
- bigger L2 cache
- OoO AltiVec
- manufactured in 90nm

Besides this they are virtually identical to 7447, 7445 and so on.
FreeScale

You probably mean "pre 86xx CPU's", right?

Post 74xx CPU's *aren't* e600 cores either, they *contain* e600 cores (as does the chips in the pre 86xx series *in a technical sense*; the label didn't exist pre 2004, but the cores did).

The "G4" CPU (with Altivec) debuted in late summer of 1999 with the 7400 CPU. Back then the focus was desktop CPU's in the traditional sense of the chips didn't contain anything else but the core. There weren't any need for system builders to mentally separate the core from the physical chips, since it basically was one and the same, and nothing more in addition. From their perspective a 7447 = G4 = 7447, and any label of the actual core inside was never that important. All controllers were outside of the chip, and accessed through a bus.

The desktop CPU label "G4" (that again describes a chip with nothing but a core) lost its purpose back in 2004 when Freescale shifted focus from desktop CPU's to SoC's as seen in the picture below. They now needed a way to distinguish the cores from physical chips in a marketing context, hence they introduced the embedded label "e600", which is a label for *a component* (one of several) inside a chip.

The 7448 was the last "G4" CPU, from then on (86xx) the chips based on the e600 core contained *additional controllers* that traditionally hasn't been part of a desktop "stand-alone" CPU, but is common in SoC's.



For example, look here at the "Core Type" column, and you will see that Freescale lists all the chips I mentioned as containing the "e600" cores.

Different chips have (and will continue to have) variations when it comes to cache size, bus speeds and bus technology, the level of CMOS manufacturing technology, etc, things are evolving and improving and adapted according to the various purposes of the SoC's, but the *fundamentals* of the core has pretty much been the same in this family of cores all along.

The "e600" is a *marketing label*, a result coming from the need to mentally separate the CPU core from the old school desktop CPU *chips*, in order to be able to discuss it (the same technology) in the context of a SoC.

That's all I'm saying and that's all FreeScale says :-)

Sadly, in a few years AltiVec fans -like myself- will become irrelevant or will have to find other architectures to work on, as 8610 and 8641D are the last cpus from Freescale with AltiVec on them, as Freescale focus on multi-core designs based on e500 core (no AltiVec). And we'll have to see if 8641D will ever see the light as a cpu in a consumer product. The only other *new* CPUs using AltiVec are Power6, PWRficient and Cell. Oh and PPC970(G5), well this is not new exactly but it's not entirely dead either. Anyway, none of these CPUs are in a consumer system (the PS3 is nice but it doesn't count as a developer's system, sorry). And Intel just released SSE 4.2.

My point is, AltiVec maybe be great, and I really *love* it, but it's not perfect (nothing is). Unless it will be developed further, it will be left behind by other architectures. I'm not pessimistic, just a realist. For what it's worth though, I intend to keep on working on AltiVec for as long as I can find a computer to do AltiVec work on :)

Konstantinos

PS. To be fair to Freescale, they've produced some amazing hardware while focusing on really important stuff -like power performance/efficiency. And I do understand that they need to run a business and it's totally sane to prefer to sell multi-core cpus based on e500 rather than e600-based. AltiVec takes a significant part of the die and the production process is quite more expensive compared to the e500. In the end, if it means a 32-core e500 will cost less than a 2-core e600 then I'm all for it, it's just that I feel sad about such a beautiful architecture...

For firmware, it'd be ours (open source firmware is NOT a credible development path). When you see what Aura is capable of (and the other new features in the 2.0 firmware) you won't be disappointed.

As for the rest, the development goal is to expose as many relevant features as possible - this may or may not include Firewire, but it would include support for digital flat panels and high definition displays, enough expansion capability to keep everyone happy.

_________________
Matt Sealey

It'll be faster than those at those speeds. Much faster. You might notice under some workloads that it acts a little slower; it has less cache (256kb vs. 512kb or 1mb) and the cache latency is a little higher (1 cycle). But the bus bandwidth makes up for a lot. The integration with the PCI controllers on-chip makes up for a lot.

I don't think you would notice the difference in dual-core or 64-bit nature of a chip - does anyone really notice this when running single-threaded applications and only have less than 4GB of RAM? I think you would have to have software which truly took advantage of it. I can only think of some high-performance filesystems which require 64-bit atomic operations which actually are easier to code and better to run on real 64-bit systems.

Let's do a little thought experiment. Most of the time, your system is idle. It's not running heavy application load, browsing files or the web, maybe coding in a text editor or doing some compiling. These tasks rarely need to soak up 100% CPU time for extended periods. If you are not using 100% of one CPU core, there is no advantage in moving it to another; in fact, this will slow things down due to the thread locking and coherency checks involved.

Software which traditionally did better in multi-core is things like media decoding and encoding - but we have seen, that media decoding is not exactly a huge problem for the MPC8610. Gaming is another; but name a high-performance, commercial game newer than Quake 3 that supports true multi-threading and runs on Power Architecture anyway, and I will buy you a cookie.

You could do compiles in parallel, but most compilations don't take advantage of having many threads at once; if you have a single processor, adding -j2 to your "make" command line makes it compile two source files at once. If you have dual core, you could set this to -j4 or -j6 and get more done. On an Efika, Pegasos, PC, multi-processor server, this does not do much except mean it can spawn two or more gcc processes (which actually spawn another process which then spawns another process by indirection, startup time is atrocious), neither of which use more than 50% of the CPU, most of which are constantly I/O bound for most of their run (collecting headers, preprocessing, piping data between backends). There are times when compiles will block each other - either by I/O scheduling in the kernel or at some point the make application needs to execute a different rule or has to take care of a dependency which throttles the multithreading back - you cannot compile two files in parallel if one of them needs the other. You cannot link (by far the slowest part) in parallel.

As I said before and will continue to express, I would love to see an MPC8641D board. I think the larger caches, the second core, and the built-in ethernet controllers have advantages in some environments. Doing something with the Assymetric Multi-Processing would be cool (running two OS's side by side with a dedicated core).

But right now, that is a long way off from being actually supported in Linux, a long way off from being useful to customers.

_________________
Matt Sealey

Is the development of the e700 core canceled?

No, Freescale also say:

"MPC7410 contains e600, MPC7447 contains e600, MPC7457 contains e600, MPC7447A contains e600"

The footnote in the 7448's product page you are quoting was merely a way back then of telling potential 7448 customers, that were worried and confused about this "G4 -> e600" transition, that: "Hey, calm down, the e600 core is the same thing as found in the G4 chips, but *this* implementation of it is the best and most evolved this far!"

There is *always* improvements in performance and manufacturing technology when CPU's evolve, otherwise - what would be the point? Enhancements to cache performance and Altivec performance has probably taken place in every single new CPU model, ranging from the 7400 to the 8610.

- The core family we are talking about was introduced with the "G4" Desktop CPU chip in 1999.
- The term "e600" was introduced in 2004, when focus shifted to SoC and highlighting the cores became important
- Chips "pre SoC" = G4
- Chips "post SoC" = e600 SoC platform
- The 7448 was the last of the G4/desktop chips
- The 8641 was the first(?) of the e600 platform chips

Or am I wrong?