@KimmoK

PPC is dead :D

_________________

@WolfToTheMoon

Could be, but that article isn't really helping to clear up any confusion. ARM and PPC have not had a lot of overlap in funtions so far. There just isn't any ARM SoC that can easily jump in and take over for those (mostly military, I'll grant you) who are using the data processing power/bandwidth of e.g. P4080 and the upcoming T4240. As long as ARM == handheld devices then ARM == just as dead on desktop as PPC has been for a while.

Now, since the article says they've reduced from 21 down to 14 arches and we here haven't been affected by that, I'm not entirely convinced we're gonna immediately notice the next change either.

But it does look like Hyperion might need to start look outside PPC. Funnily enough I'm wrapping up a PPC design here and have the Freescale distributor on the phone all the time, even went to a seminar in November and noone mentioned anything about shutting down the line

Last edited by olegil on 08-Jan-2013 at 01:28 PM.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@KimmoK

Argh!

Without a proper market anc communication division, we (amigans) actually can't even ask informations to freescale.

I hope to be wrong.

A contact channel is needed for freescale, as well as IBM.

_________________
retired

@olegil

Freescale isn't what it used to be. They're ridden with debt, unprofitable and with a falling market share.

In that situation, it is entirely possible that there is no money to support and develop 2 distinctive ISAs. And it is blatantly clear which of those two has a bright future, and which doesn't. PPC is going down.

_________________

@WolfToTheMoon

It would be wrong for Freescale to ditch their industrial/military customers.

They are saying they're shrinking down to 3 arches, and :

Quote:

The two most important of its remaining chip architectures are Power and ARM.

And the article starts off with talking about "internet of things", as in small devices that don't need processing power but need to be networked. This is an awesome application for ARM but it doesn't really help the guys who are doing ICBM targetting systems with the P4080

So it's sadder for those who are using Coldfire than those who are using POWER, really

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

From what I read Freescale anyway is looking forward to 10 years ahead with PPC co-existing with ARM. And if it still delivers more per watt in high end after that, it might still have future beyond 10 years.

And they have indicated that they had also next gen after AMP/T series on their roiadmap.

We'll see...

Some numbers from article:

Type Revenue ($m) Share % in 2012
PC ------- $41.1 ---- 43.6% x86
Embedded - $22.3 ---- 23.6% PPC & ARM & others
Server --- $16.5 ---- 17.6% x86 & PPC (ARM coming??)
Network -- $7.4 ----- 7.8% PPC & MIPS (ARM coming)
Graphics - $5.9 ----- 6.3% GPUs?
Phone/Tablet $1.1 --- 1.2% ARM


Type Revenue ($m) Share % in 2017
Embedded -- $38.0 --- 28.0% PPC & ARM & others
PC -------- $36.1 --- 26.6% x86
Server ---- $24.4 --- 18.0% x86 & PPC (ARM?)
Phone/Tablet $13.2 -- 9.7% ARM
Network --- $12.2 --- 9.0% PPC & MIPS & ARM
Graphics -- $11.7 --- 8.6% GPUs


So, they estimate:
-Embedded market grows. ( most likely ARM gains more share from coldfire and PPC etc.)
-Server market stays the same. (most likely ARM enters there)
-Phone/Tablet market grows fast. (dominated by ARM)
-Network market grows a little (ARM is the newest competitor there)

So ARM is heading towards growth while PPC is facing more competition. +ARM is already available in simple media PCs for home for 80eur.

I only hope that while PPC has strong market position NOW, they (power.org) use it to widen their scope, before ARM gets strong everywhere.
Some companies like Verisilicon has shown that good GPU can be integrated with PPC core. Hopefully some end product materializes from that.

+even if they estimate that x86 will drop from it's nbr1 position, I would not yet stop AROS x86 developments.

How it was some years ago:
http://www.embeddedintel.com/standards_watch.php?article=605
Communications processors in 2009:
Graph.

Last edited by KimmoK on 08-Jan-2013 at 02:59 PM.
Last edited by KimmoK on 08-Jan-2013 at 02:57 PM.
Last edited by KimmoK on 08-Jan-2013 at 02:47 PM.
Last edited by KimmoK on 08-Jan-2013 at 02:41 PM.
Last edited by KimmoK on 08-Jan-2013 at 02:40 PM.
Last edited by KimmoK on 08-Jan-2013 at 02:40 PM.
Last edited by KimmoK on 08-Jan-2013 at 02:39 PM.
Last edited by KimmoK on 08-Jan-2013 at 02:38 PM.
Last edited by KimmoK on 08-Jan-2013 at 02:31 PM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

Intel just announced a multicore Haswell chip at 7W TDP this year. Hard to see how anyone can compete with that.

A dual core Exynos 5 with 2 A15 cores tops 8W in Anandtech's testing.

edit: Actually, it's an Ivy Bridge 7W platform. Got confused reading various articles

Last edited by WolfToTheMoon on 08-Jan-2013 at 03:14 PM.

_________________

@WolfToTheMoon

got any performance specs on those? Also, does it have any peripherals or do you still need a chipset?

Cause that was the biggest problem for me with the 2W Atom when it first came out. I couldn't justify connecting a 2W CPU to a 5W chipset, no matter what the software guys were saying (3W total power budget for our card).

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

Ivy Bridge is not a SoC, it's primarily aimed at thin ultrabooks.

Mobile Haswell will be a SoC design.


Also, AMD presented a 4 core 28 nm Temash sub 6W TDP SoC running Dirt 3 in 1080p.


Great times for x86/x64.

Last edited by WolfToTheMoon on 08-Jan-2013 at 07:39 PM.

_________________

@WolfToTheMoon

Yes, it will be interesting to see...

Some room for improvement.

The embedded board that I work with has 6 core MCU that consumes 15W and the whole board consumes 17A of 5V (without RF Amplifiers etc.). Cooling needs some 15kg of metal. Nasty beast. I imagine with coming freescale chip it would use only half of the power.

UPDATE/ADDITION:
The challenge for x86, IMO, is that it needs something like 2x amount of transistors per core to do the same as some simple RISC core. And when RISC chips continue to gain business income they afford to spend even more in R&D. In the end it will make x86 harder and harder to keep up with performance per watt race. (+for some uses the "wrong" endianess does not help at all)

@ARM on servers

As current ARM chips are no good for servers (32bit), it seems futile to develope new ARM64 instruction set & chips for servers where good PPC, MIPS etc. chips already exist.
But perhaps 64bit PPC is so expensive to licence that also APM/AMCC switch to ARM.
ARM for servers -development.

The good thing is that if PPC dies one day, ARM might one day become serious high end alternative.

Last edited by KimmoK on 09-Jan-2013 at 09:40 AM.
Last edited by KimmoK on 09-Jan-2013 at 09:00 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

Now you lost me. Your MCU uses 15W but the board uses a total of 85W. Shrinking those 15 to 7.5 surely won't reduce the total consumption by more than 7.5W?

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

There are also some DSP's that consume similar 15W each. Freescale is going to have chips with DSPs integrated. After that we might only have MCU with DSPs + one hot RIO switch & few hot ASICs.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@WolfToTheMoon

Quote:

Great times for x86/x64.

I agree. ARM in ultra-portables and hand-helds has really pushed advances in x86/x64. I would rather see OS4 moved to x86 than to ARM, because for the foreseeable future ARM just isn't an improvement over PPC _other than for handheld devices_. To get sensible performance out of an A15 you would need a heck of a lot more than two cores. The beagleboard and the Raspberry Pi are awesome toys, but they are just toys nonetheless. I'm not saying you can't hack it into solving an industrial use case, but it wouldn't STAY a Pi if you did

If Haswell brings more peripherals, thus reducing the number of support components needed, then I can see more embedded use of x86 in the future. But Intel are currently completely misunderstanding the meaning of "embedded". Their "embedded processors selection guide" consist of processors and motherboards, not SoCs

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@KimmoK

Quote:

The challenge for x86, IMO, is that it needs something like 2x amount of transistors per core to do the same as some simple RISC core. And when RISC chips continue to gain business income they afford to spend even more in R&D. In the end it will make x86 harder and harder to keep up with performance per watt race. (+for some uses the "wrong" endianess does not help at all)

2x the amount of transistors... hardly. Compare high end PPC chips with x86.

For instance, the first PPC in Macs, the 601, had 2.8 million transistors, in 0.6 µm vs 3.1 million for P5 Pentium in 0.8 microns and 2.5 million for 68060 in 0.6 microns.

First PPC G4(7400) had 10.4 million transistors vs 9.5 for first Pentium III that debuted a few month before G4

ARM has fewer transistors because it was designed primarily for power efficiency and embedded use.


BTW, according to Anandtech extensive testing, Atom saltwell core in 32 nm compares very favorably with qualcomms Krait in 28 nm, in fact it's more power efficient. That's a 5 year old in order Atom core based on the Pentium P5 architecture from 93'. Next Atom will be out of order 22 nm TriGate chip... say bye-bye to ARM.

_________________

@WolfToTheMoon

That "7W TDP" is actually 13W. It's not the first time marketing has abused common terms at Intel.

And it also uses a ~3W chipset.

The 7W figure comes from something called a "Scenario Design Power" - what the chip will draw in normal use. Whether that scenario includes gaming is not specified.

Naturally, whilst the latest high-end ARM SoCs can quite a bit of power under load, they are really really good when the load drops. And remember that the AnandTech article was entirely driven by Intel.

Also, the ARM SoCs don't cost $200.

Last edited by Hattig on 09-Jan-2013 at 11:13 AM.

@Hattig

Driven or not, the data given by Anandtech is very clear. x86 can be competitive with ARM even on a older process. And once the new Silvermont SoC is out, on 22 nm TriGate process, I think we can safely say that Intel will have clear performance and efficiency edge over ARM.

And for ARM fabs the problems will only get bigger since going past 28nm and to finFET will be very, very expensive indeed(as Intel found out), thus negating their price advantage somewhat. Intel has already done that and can afford higher margins on the high end.

I'm also pleasently surprised by AMD's Temash SoC. If they can really deliver a 4 core SoC in under 6W, they got a winner on their hands. I'd much rather have that in a Win 8 tablet than Tegra 4 and Win RT or even worse, Android.

Looking even further, if a 4 core Temash is 6W, than AMD is also a generation away from a competitive smartphone chip. And their prices will be even lower than Intel's, who has already done the heavy lifting in porting Android to x86. Rising tide lifts all ships.

_________________

Observing CPU prices in local shops is also interesting. AMD can offer a chip for 39eur while intel is far above 100eur.

(if I now bought x86 for dekstop, it would be 4.2Ghz 4 core AMD as I see no point in intel's pricing)

@transistors
"So, we have 200 M transistors for the PA6T vs. 291 M for the Core 2 Duo T7600. Considering that the PA6T had on-board dual channel memory controllers and the Core 2 did not...."
from http://markstechchat.blogspot.fi/2010/03/ipad-sage-week-6-apples-mobile-device.html
Core2 has 2x more cache, but PA6T had memory controller etc. onboard.

But I admit I was mistaken (again). x86 is not as bloated in transistor count as I thought.
(and absolute numbers are hard to get, for example I found 1.6M for PPC603 and 3.1M for Pentium1)

Last edited by KimmoK on 09-Jan-2013 at 12:35 PM.
Last edited by KimmoK on 09-Jan-2013 at 12:34 PM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

Quote:

@transistors "So, we have 200 M transistors for the PA6T vs. 291 M for the Core 2 Duo T7600. Considering that the PA6T had on-board dual channel memory controllers and the Core 2 did not...." from http://markstechchat.blogspot.fi/2010/03/ipad-sage-week-6-apples-mobile-device.html Core2 has 2x more cache, but PA6T had memory controller etc. onboard.

I hope you realize that caches is what consumes most transistors nowadays? Memory controllers take more die space (due to bigger sized I/O transistors) than transistors and that applies to most I/O technologies.

The overheads of x86 are mainly concentrated in two areas: instruction decode and microcode storage while other features are used in most high-performance processors (e.g. generously sized load/store buffers with store->load forwarding). Assume a maximum of ~15% extra transistors (many of which are "cold" that is: not activated often like in microcode paths).

The first x86 processor was the 8086 which had ~30k transistors using NMOS logic (NMOS uses transistors for pull-up, if those is included in the transistor count I don't know but if so the active number of transistors are considerably lower).
The 68000 processor is commonly claimed to have ~68k transistors.
The ARM2 processor uses CMOS and had less than 30k transistors and was (unlike the previous two) pipelined.
Here it's obvious that the ARM2 is a better design but as we go further in time we begin adding caches, deeper pipelines and out of order execution. The architectural differences shrinks for each new performance enhancing feature and the skills of designers plus the process technology gets more and more important. Intel leads in both those areas.

@megol

"Intel leads in both those areas."

And only way to compete with that is to get more money for R&D than what intell has. It seems some predict that ARM chips might have more money around in 2017 than intel. But ARM business is split between 10? companies (& their R&D), while intel gets 90% of x86 CPU business? So intel should stay ahead after all. .. ...


UPDATE:
Roadmap picture that I had not yet seen: From Trevors blog.

Good amount of companies working with PowerPC.

Last edited by KimmoK on 10-Jan-2013 at 07:52 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

instruction decoder is a few % of a die size. Nothing deciding.

_________________