Update:

> New list of DMIPS per MHz and core figures for ARM cores, taking the Swift core
> of the Apple A6 SoC into account (only with an estimation unfortunately):
> [...]
> Apple Swift: 3.3...3.5 (estimated)
> [...]

Figure for Swift substantiated and figures for Cortex-A53 and Cortex-A57 added, according to these sources:
http://en.wikipedia.org/wiki/List_of_ARM_microprocessor_cores#Designed_by_ARM
http://en.wikipedia.org/wiki/List_of_ARM_microprocessor_cores#Designed_by_third_parties
http://www.arm.com/products/processors/cortex-a50/cortex-a53-processor.php?tab=Performance

ARM Cortex-A7: 1.9
ARM Cortex-A8: 2.0
Qualcomm Scorpion: 2.1
ARM Cortex-A53: 2.3
Marvell Sheeva PJ4: 2.4
ARM Cortex-A9: 2.5
Marvell Sheeva PJ4B: 2.6
Qualcomm Krait: 3.3
(Broadcom Brahma15: 3.5 (unsure whether or not this core is just ARM Cortex-A15))
Apple Swift: 3.5
ARM Cortex-A15: 3.5
ARM Cortex-A57: 4.1...4.8 (depending on implementation)


Edit: Added Broadcom Brahma15 core.

[ Edited by Andreas_Wolf 26.12.2013 - 21:28 ]

Some more benchmarks for the Chromebook:

http://openbenchmarking.org/result/1211246-RA-EXYNOS5DU63

and some to compare it to:
http://openbenchmarking.org/result/1202030-AR-MERGE774596

A bit confusing but it comes in close to a Core Duo.

minator,
Quote:

I appreciate that you're a strong advocate of ARM cpus, but if you're going to make a comparison like that could you give us a bench that shows the two cpus together?

Edit - Andreas, can you come up with some figures that would allow us to compare Cortex A15 with e6500 core cpus (dmips, etc)?

[ Edited by Jim 25.11.2012 - 03:26 ]

> Andreas, can you come up with some figures that would allow us to
> compare Cortex A15 with e6500 core cpus (dmips, etc)?

DMIPS figures for both those cores to compare are already on MorphZone for 3.5 months. The figure for Cortex-A15 can be found in posting #420 of this very thread, and the figure for e6500 can be found there:

https://morph.zone/modules/newbb_plus/viewtopic.php?forum=3&topic_id=7001&start=743

Sorry, I should have just searched.
Surprisingly high for the A15.
But then I remembered the e6500 figures as being pretty good too.
Is that quote on the A15 for one, two, or four cores?
And is it fair to assume that an e6500 based product with far more cores would produce even better results?

Basically, would 12 cores dual threaded equal 72 (or close to that)?

> Surprisingly high for the A15.

The DMIPS figure for the Cortex-A15 has been known for about 2 years already :-)

> Is that quote on the A15 for one, two, or four cores?

Quote from that posting: "DMIPS per MHz and core" :-)

> is it fair to assume that an e6500 based product with far more cores
> would produce even better results?

Yes, it is, as the Dhrystone benchmark scales quite linearly with the amount of cores.

> Basically, would 12 cores dual threaded equal 72 (or close to that)?

Yes, 12 dual-threaded e6500 cores would deliver 72 DMIPS per MHz.

So, to reinforce my less then startling grasp of the obvious, a best case scenario leaves us with a 14 DMIPS per MHz vs 72 DMIPS per MHz comparison.
And even with comparable numbers of cores we got an advantage if dual threaded, and not much of a disadvantage if not.

So, except for price, what's the advantage to ARM?

> with comparable numbers of cores we got an advantage if dual threaded,
> and not much of a disadvantage if not.

That'd be the case with an SMP-capable OS, yes. And what we didn't regard so far is the clock rate comparison which turns DMIPS per MHz into pure DMIPS.

Quote:

HW that isn't routers/switches?

If you don't want to care about price (hence disregard the things that makes ARM cheap and "powerful enough" for the price point, like the on chip accelerators and controllers for all kinds of graphic/audio handling that's usually there), then why not look at the x86 instead?

MorphOS (and Amiga) is single core only. If this would change some day, it would happen at the expense of Amiga compatibility, and if that ever happens, then there is no reason whatsoever to look at PPC ever again anyway, so...

> They did have several different ISAs (ARM, Coldfire, 68K, PPC, 8-bit)
> until fairly recently but they have been replacing their designs with
> ARM based designs for some time now.

On that topic, there's an interesting article from 3 days ago that reads:

"Kaivan Karimi, head of strategy and business development for Freescale's microcontroller business, explains. "When I joined the company in 2003, we had 21 different chip architectures to support. The goal then was to bring it down to 14," Karimi said. But now that number is three. [...] The two most important of its remaining chip architectures are Power and ARM. [...] Freescale has dropped a lot of its dead-end chip architectures in favor of ARM [...]"

The third of the 3 architectures to be kept besides Power and ARM is probably ColdFire. Do you know what the other 18 architectures are that were still being actively supported by Freescale a decade ago? I didn't know it were that much.

Some comments on other statements from that article:

"The more recent change has been to ARM, which Freescale has embraced with full force after Motorola fought it for years. [...] Freescale hasn't always been in love with ARM."

Odd statement. Motorola acquired its ARM license in 2000, so 4 years before it spun out Freescale, and released its first ARM chips (i.MX1) in 2001.

"Geoff Lees, Freescale's new general manager for microcontroller products, says his company is well-suited to play in the ARM world, which is moving rapidly toward more muscular 32-bit processing cores. [...] "We still do have products based on other architectures, but virtually 100 percent of our research and development is directed toward 32-bit ARM-based (chips)," Lees said."

That surely doesn't sound too good for the future of Power Architecture at Freescale. Am I the only one who senses somewhat of a contradiction here to the previous "two most important of its remaining chip architectures are Power and ARM" statement made in that same article?

Quote:

AFAIK Coldfire only had one major customer and they switched to ARM a while back. At a guess I'd say it'll be a DSP.

Quote:

They still sell quite a few even today but I'd guess at:

8 bit
16 bit
68K
88K
Mcore
Coldfire
ARM
PPC
Several DSP variants...

Quote:

I'd take it to mean they'll do new chips while they have customers for them but they'll not develop any new cores. IIRC the new 64 bit core is based on a G3. Apart from that they haven't done anything really new in a very long time, just tweaks.

They could co-develop PPC and ARM SoCs so all they have to do is drop in whatever cores the market requires.

iminator,
Quote:

G3? Hardly, you need to take a closer look.
Many of the new cores outperform G4s and compete well with G5s.

And ARM's development, while impressive, still does not outperform our current ISA.
So, if we switch, X86 would still make more sense then ARM.

AND, I'd like to state once more that i am not in favor of moving to X86.

AND,

> AFAIK Coldfire only had one major customer and they switched to
> ARM a while back. At a guess I'd say it'll be a DSP.

The latest release of new ColdFire chips was announced in May 2011. But as that was almost two years ago, you may be right that there won't be any new ColdFire development, and that the third architecture to be kept besides Power and ARM could be the StarCore DSP family. It's a pity that the article isn't clear about that.

> 8 bit
> 16 bit
> 68K
> 88K
> Mcore
> Coldfire
> ARM
> PPC
> Several DSP variants...

This would be 13 DSP variants then. Besides, I have my doubts about 88k being actively supported in 2003 ;-)
I did some research and think this may be how Freescale arrived at its "21 different chip architectures" figure for 2003:

PPC, 68k, ColdFire, ARM, MCORE, DSP56300, DSP56600, DSP56800, StarCore MSC711x, StarCore MSC8100, M68HC05, M68HC08, M68HC11, M68HC12, M68HC16, HCS12, C-Port, MC4472xA, S1 family (MPC1xx)

Okay, that's just 19, not 21 ;-) And of course it's highly debatable if that's really all *different* architectures or more like subarchitectures in some cases. But I really see no other possibility than including subarchitectures for Freescale having actively supported the number of "chip architectures" they supposedly have.

> I'd take it to mean they'll do new chips while they have customers
> for them but they'll not develop any new cores.

That would be a strange meaning as they surely won't develop their own ARM cores, yet "virtually 100 percent of [their] research and development is directed toward 32-bit ARM-based (chips)". R&D-wise, developing new Power Architecture chips around their own existing Power Architecture cores is surely not too different from developing new ARM chips around existing ARM cores licensed from ARM Ltd. Thus, I'm still of the opinion that Freescale's ongoing development of Power Architecture chips (which I'm sure we can consider to be fact) contradicts the above-quoted statement ...unless it's really referring solely to microarchitecture development and Freescale is secretly developing its own ARM core right now, which I doubt.

> IIRC the new 64 bit core is based on a G3.

No, you have said that before, but it's not. The e6500 is (via e5500, e500mc and e500v2) a descendant of the e500v1 (called just e500 back then), which was no descendant or derivative of the 74x/75x microarchitecture but a from-scratch implementation of the Book E spec done around 2001.

> they haven't done anything really new in a very long time, just tweaks.

I think the e6500 and its ancestor e500v1 are really worlds apart, even if it were "just tweaks" that led from the one to the other.

> They could co-develop PPC and ARM SoCs so all they have to do is
> drop in whatever cores the market requires.

Yes, that's what the announced QorIQ Layerscape product line will be all about.

http://www.eetimes.com/electronics-news/4375606/Freescale-adopts-ARM-cores-in-QorIQ-line

"ARM will not replace Power cores for Freescale, at least in the foreseeable future. The company said it has long-term plans to continue designing and supporting Power cores."


So ARM for the future? Why>
PPC still looks very attractive to me.

Andreas_Wolf,
Quote:

Fair enough.

However my point is they haven't done a completely new core design in a long time.
That said this is a perfectly valid way to do things, Intel used this strategy to great effect with the Pentium Pro.

Quote:

Where the G5s are concerned, I don't think that's going to be true for much longer (see Nvidia and Samsung's latest announcements).

--

Quote:

Do the prices the OS4 guys pay for hardware look attractive to you?
You're going to pay through the teeth for anything newer than the G5s if you want new hardware.

The availability of PPC CPUs has never been a problem.

Update:

> APQ8064 got degraded to 1.7 GHz, which means 2.5 GHz Krait will have to wait until
> APQ8094 in second half of 2013.
> https://developer.qualcomm.com/discover/chipsets-and-modems/snapdragon
> http://vr-zone.com/articles/qualcomm-reveals-future-plans-we-give-you-the-details/13581.html

Still no 2.5 GHz, but Krait 400 core got announced running up to 2.3 GHz in the MSM8974 SoC (and Krait 300 core up to 1.9 GHz in the APQ8064T SoC):

http://www.qualcomm.com/media/releases/2013/01/07/qualcomm-announces-next-generation-snapdragon-premium-mobile-processors
http://www.qualcomm.com/media/blog/2013/01/07/snapdragon-800-series-and-600-processors-unveiled
http://www.anandtech.com/show/6568/qualcomm-krait-400-krait-300-snapdragon-800

In the process, the original Krait core got renamed to Krait 200.

Update:

>> 28nm Cortex A-15 (Eagle) i.MX73 is proposed. [...]
>> i.MX Roadmap

> Seems this may take longer than anticipated:
> http://semiaccurate.com/2012/11/01/freescale-layoffs-target-one-particular-product/
> http://www.eetimes.com/electronics-blogs/the-engineering-life-around-the-web/4401885/Freescale-s-i-MX-7-canned-as-firm-refocuses-
> http://www.linleygroup.com/newsletters/newsletter_detail.php?num=4917

Freescale going Cortex-A53 for i.MX (and Cortex-A57 for QorIQ):

http://media.freescale.com/phoenix.zhtml?c=196520&p=irol-newsArticle&ID=1789227

Maybe they'll simply skip Cortex-A15 for i.MX?

Nice to hear that they are going to license something more powerful then the A8.

>Maybe they'll simply skip Cortex-A15 for i.MX?

"Because the company’s remaining target markets can be addressed with existing processors, Freescale was able to significantly reduce headcount in its i.MX unit."

Sounds to me like ARM is losing ground at Freescale.

> Nice to hear that they are going to license something more powerful then the A8.

Huh?

01/2011: Freescale announces i.MX6 with Cortex-A9 cores
06/2011: Freescale demonstrates i.MX6 with Cortex-A9 cores
10/2011: Freescale licenses Cortex-A15 core for future i.MX chips
01/2012: Freescale extends i.MX6 series based on Cortex-A9 cores
06/2012: Freescale announces QorIQ Layerscape with Cortex-A15 cores
11/2012: Freescale announces volume availability of i.MX6 chips with Cortex-A9 cores

> Sounds to me like ARM is losing ground at Freescale.

And yesterday's press release sounds like ARM is gaining even more ground at Freescale. So which one is it?

Quote:

Who knows, I guess it depends on what markets they are trying to reach with their products, as well as *when*...?

@Jim

Quote:

It was a long time they expanded their product portfolio to A9. (ARMv8 != (cortex) A8 )


Quote:



Huh?!

Too me it sounds exaclty the opposite, when they introduce high performance ARM cores into the QorIQ family, where previously the PPC ruled almighty, and they say things like:

"This agreement expands Freescale’s long-term strategic technology partnership with ARM and gives us access to the latest ARM technology for future generations of our i.MX applications processors."

"This agreement is part of a new multiyear subscription license with ARM that demonstrates Freescale’s commitment to the ARM architecture and its intent to further expand its ARM Powered® portfolio – one of the industry’s broadest range of solutions built on ARM technology."

I'd say that ARM is getting more important than ever for Freescale, and if any architecture is going to be down-prioritized due to Freescales poor financial performance, it will hardly be ARM, more likely the PPC.

>And yesterday's press release sounds like ARM is gaining even more ground at Freescale. So which one is it?

Heck if I know Andreas.
I've paid so little attention to this part of Freescale's portfolio that I wasn not even aware that they had licensed the A9.

By dropping the A15, I'm guessing they intend to focus on low end and high end ARM (leaving products like the A15 to other companies).
I've never understood Freescale's interest in ARM in the first place.
It a low profit, high volume product that virtually anyone with a license can pump out.

At the really low end, the Chinese have flooded the market.
Cheap (a not very good) tablets are everywhere.

Even fabless companies like VIA are shipping this stuff.

Although I am impressed with the specs on this latest product.
Then again, what do we need 64bit for?

> what do we need 64bit for?

I'm not sure who "we" is in your question, but that's what ARM Ltd. responds:

"In the desktop environment, a larger virtual address space is important for modern desktop software APIs that may come to rely on having a vast virtual address space for techniques such as memory mapped file I/O and sparse addressing (e.g. for persistent objects). AArch64 also enables efficient 64-bit immediate generation meaning less need for literal pools. A large program counter relative addressing range (+/-4GB) for efficient data addressing is helpful within shared libraries and position-independent executable."
http://www.arm.com/products/processors/cortex-a50/?tab=Why+64-bit

"Position independent executable"

Ah, I never needed a large address space to do that.
I was using position independent code on 6809, 68000, and 68020 systems in the late '80's.
Although you did manage to link an argument that sways me greatly.

Basically, it helps support the kind of coding I was trying to talk everybody into decades ago.
You have no idea how frustrating it was seeing Amiga coders using what ever bizarre coding style suited them (including, the seriously irritating, self modifying code).

Keep it up Andreas, you m,ight actually get me to take this architecture more seriously.

[ Edited by Jim 28.02.2013 - 05:11 ]