4 Jun
2013
4 Jun
'13
8:49 a.m.
On Jun 4, 2013, at 10:56 AM, Liam Proven <lproven at gmail.com> wrote:
What, like TileEra?
I don't actually remember if Tilera is MIPS or if they have a more
proprietary instruction set. But Cavium makes (or made, a few years
ago) the OCTEON line, which combines loads of MIPS64 CPUs with some
custom processors which are really good for processing discrete
finite automata (great for pattern matching, among other things).
NetLogic (now owned by Broadcom, which means all the info is quickly
disappearing from the web) has a line of heavy hitters based on lots
of MIPS64 cores at high speeds and some really serious I/O stuff.
You tend to see things like these (and the high end PowerPCs) in the
routing core of telecom and wireless infrastructure. It's not the
sort of thing you usually see broadcast (so to speak) because the
products containing them are usually basically blackboxes, so you
don't need to know what kind of CPU is running inside. It's also on
a layer hidden from most of the public; when was the last time your
local telco told you what they were installing in their NOC? :-)
But they do buy a LOT of them at a time.
- Dave
On 4 June 2013 12:41, David Riley <fraveydank at
gmail.com> wrote:
Er, no. It was (if you really must frame it in those terms, with their
strong implicit biased interpretation) because IBM "wasn't interested"
in developing the PPC further for the *notebook* market.
I suppose I should have said "consumer, general-purpose", but the
desktop aspect was there; IBM wasn't putting much effort into
bringing up clock speeds on the G5, either, which was a major
embarassment for Apple, since their roadmap had shown 3 GHz way
earlier than it ever came about (which is why clock speeds are a
dumb metric). IBM was pretty busy at the time, anyway, since they
were developing or co-developing PPC architectures for all three
major game consoles.
And it
didn't work. Apple led the exodus, jumping to its arch-enemy
Intel, soon dropping PPC code compatibility and offering a free,
built-in tool to dual-boot with Windows, while there is a /choice/ of
free hypervisors to run Windows under Mac OS X.
Apple did this largely because Motorola and IBM weren't interested in
developing the PPC farther for the desktop market. The G5 was a good chip, but too hungry and way too hot
for portables -
just as the market was swinging decisively to portables. Apple was
left with a competitive big-desktop offering, but limited to the
by-then-underpowered G4 in its notebook and notebook-tech based
machines (i.e. the laptops and the Mac mini).
It was not about the desktop. IBM is still cheerfully developing POWER
derivatives for servers and in the later days there wasn't all that
much difference between POWER and PowerPC any more, not as I
understand it.
The difference between POWER and PPC has always primarily been one
of size; the 601 was actually a lot more like the POWER of the time
than the later PowerPCs (it supported a number of POWER instructions
that the 603/604 didn't). There are definitely some instruction set
differences between the two, but the big difference has always been
that POWER was intended to run in big machines with all the cooling
you could want.
Whereas the original PowerPC 6xx chips used the
Motorola 880x0 bus
(IIRC), the G5 didn't - it's not really a PPC at all but a
desktop-scale single-chip POWER.
Up until the G4 (PPC74xx), that was true; the 60x bus was essentially
a renamed 88110 bus. The G4 added the MPX bus, but there were still
backwards-compatible models, which was quite handy for CPU upgrades.
The 970 was essentially a scaled-down POWER chip, largely due to the
shift from Motorola to IBM.
Meantime, MIPS is coming back, amazingly, as an
even-lower-power-than-ARM offering from Imagination, the company that
makes PowerVR GPUs.
There are also some very high power MIPS64 chips from other vendors
which are primarily used in networking equipment.