http://arstechnica.co.uk/business/2016/10/qualcomm-to-buy-nxp-semiconductor/

NXP worth more than ARM?

Au revoir PowerPC......

> First numbers for the X-Gene 3 place it at Xeon performance levels.

X-Gene future may just have become somewhat uncertain:

"MACOM intends to divest the well-positioned but non-strategic Compute business within the first 100 days of closing. [...] MACOM expects to improve the profitability of AppliedMicro by divesting the Compute business [...] MACOM intends to divest AppliedMicro’s well-positioned Compute business within 100 days from closing the transaction, as the business does not strategically align with MACOM’s long-term focus. AppliedMicro has been exploring strategic options for the Compute business and there is known strategic interest among several potential buyers and investors. MACOM will continue to support Compute customers and partners during this transition."
https://www.apm.com/news/macom-announces-definitive-agreement-to-acquire-appliedmicro/
https://www.macom.com/about/news-and-events/press-release-archive/row-col1/news--event-archive/macom-announces-definitive-agr-4

Regarding the Embedded Processing business (which the PPC-based SoCs and the ARM-based HeliX SoCs belong to), page 11 of this presentation reads like MACOM is going to keep it ("Solid Cash Flow Business"), as opposed to the Compute business comprised of X-Gene ("Active Sale Process to Divest").
As for how they could sell the X-Gene line but keep the HeliX line, this is somewhat of a mystery to me considering both lines are based on the same core IP (Potenza).

John Croteau, MACOM president and CEO, on the other hand has to say:
"the embedded PowerPC part of the portfolio may be part of the divestiture; it may not be. Depends which buyer, frankly. It’s a cash cow either way, to be honest."

Where would this leave the embedded ARM part (HeliX) of the Embedded Processing portfolio?

Statement from 6th December:
"employees working on Connectivity or embedded PowerPC products will continue performing their roles. Employees working on ARM-based server products may be tied to any ARM-based Compute business divestiture. There are a number of “shared” employees working on both Connectivity and ARM products. [...] AppliedMicro and MACOM management will work to identify the skills, experience, strengths and weaknesses of each of the shared employees [...] in order to determine which business group they should be assigned to post-closing. [...] MACOM currently plans to maintain and support our embedded PowerPC products following divestiture of the ARM-based Compute business, including sustaining PowerPC engineering, applications engineering and QA. MACOM and AppliedMicro management will work together to identify the individuals to be assigned to these ongoing efforts. These plans could change depending on the buyer of the ARM-based Compute business."
https://www.sec.gov/Archives/edgar/data/711065/000119312516786489/d306596d425.htm
https://www.sec.gov/Archives/edgar/data/711065/000119312516786496/d301947d425.htm

Statement from 26th January by John Croteau:
"MACOM plans to promptly engage with previously identified potential buyers toward a near-term divestment of AppliedMicro’s well-positioned Compute business. As previously stated, this portion of the business does not strategically align with our long-term product focus, but we feel confident a successful transaction can be consummated."
https://www.apm.com/news/macom-successfully-completes-acquisition-of-appliedmicro/
https://www.macom.com/about/news-and-events/press-release-archive/row-col1/news--event-archive/macom-successfully-completes-a-1

[ Edited by Andreas_Wolf 01.12.2018 - 21:32 ]

Update:

> in terms of 64-bit ARMv8 (AArch64) cores, there is announced so far: [...]

In terms of 64-bit ARMv8 (AArch64) cores/microarchitectures, there exists or has been announced so far:

- Cortex-A35, Cortex-A53, Cortex-A57, Cortex A-72 and Cortex-A73 from ARM Ltd.
- Denver and Denver2 from nVidia
- Potenza, Potenza+ and Potenza++ from Applied Micro
- ThunderX and ThunderX2 from Cavium
- Cyclone, Typhoon, Twister, Hurricane and Zephyr from Apple
- Vulcan from Broadcom (probably cancelled)
- K12 from AMD
- Kryo 100, Kryo 200, Kryo 280 and Falkor (new) from Qualcomm
- Xiaomi FTC660 and Xiaomi FTC661 from Phytium
- Mongoose M1 and Mongoose M2 from Samsung


Edit: added Kryo 280
Edit2: added Mongoose M2

[ Edited by Andreas_Wolf 30.03.2017 - 00:44 ]

Quote:

- You missed the A35 and the A32 (though the A32 is a 32-bit only v8).


Qualcomm are going after Intel with their server chip. Unlike the others, they're going straight to a leading edge process, they're starting at 10nm a year before Intel.


In other news Windows 10 will run on ARM, complete with x86 emulator. ARM is heading back to the desktop!

[ Edited by minator 09.12.2016 - 19:08 ]

Quote:

X86 emulation via an ARM processor?
Sounds ugly.
Will the decedents of the processors in the Acorn RISC Machine ever make it back to the desktop?
Anybody's guess, but the real question is will they be competitive?

Quote:

No different to running 68k binaries on PPC with Trance really.

Photoshop works great.

Quote:

Pretty big difference, Nick.
The PPCs we run Trance on run many times faster than a 68K processor.
The ARM processors will be running at comparable speeds (or slower) to an X64.
Which, with the overhead, will mean X86 apps will run slowly on ARM.

Quote:

Have you seen the video?

It's like it's running native ARM code. It's very impressive indeed.

https://youtu.be/A_GlGglbu1U

[ Edited by Intuition 09.12.2016 - 19:47 ]

Quote:

Admittedly, pretty neat.
Surprisingly fast for emulation.
I guess I no longer have to worry about Intel's cancellation of the latest Atom cpus.
If X86 can be emulated this well on an open platform...

> You missed the A35

Yes, thanks for the note. I had it included in my Dhrystone list but forgot it here.

> and the A32 (though the A32 is a 32-bit only v8).

A pure 32-bit core/microarchitecture isn't relevant for my list of 64-bit ARMv8 (AArch64) cores/microarchitectures.

> Qualcomm are going after Intel with their server chip. [...] they're starting
> at 10nm a year before Intel.

Not just with their server chip. Already the Snapdragon 835 (MSM8998) with Kryo 280 core will be built using this process node.


Edit: changed Kryo 200 to Kryo 280

[ Edited by Andreas_Wolf 03.01.2017 - 17:14 ]

>>>> X86 emulation via an ARM processor? Sounds ugly. [...] the real question is
>>>> will they be competitive?

>>> No different to running 68k binaries on PPC with Trance really.

>> Pretty big difference, Nick. The PPCs we run Trance on run many times faster than a
>> 68K processor. The ARM processors will be running at comparable speeds (or slower)
>> to an X64. Which, with the overhead, will mean X86 apps will run slowly on ARM.

> It's like it's running native ARM code.

Even if this was true (which I don't think it is as there simply must be emulation overhead), it's still different to "running 68k binaries on PPC with Trance" for the reason Jim mentioned.

> https://youtu.be/A_GlGglbu1U

No side-by-side comparison with native x86 speed (or native ARM speed for that matter).

> Intel's cancellation of the latest Atom cpus

Intel Atom processors for smartphone and tablet launched in Q4'16:
Intel Atom x3-C3205RK
Intel Atom x3-C3265RK
Intel Atom x3-C3295RK

Intel Atom processors for embedded announced for Q4'16 launch:
Intel Atom x5-E3930
Intel Atom x5-E3940
Intel Atom x7-E3950

Intel Atom processors for server launched in Q1'17:
Intel Atom C3338

Intel Atom x3 SoCs (SoFIA) use 28nm Silvermont/Airmont cores, while x5-E39xx and x7-E39xx SoCs (Apollo Lake) and C3338 SoC (Denverton) use 14nm Goldmont cores. What Intel cancelled is Goldmont-based Atom SoCs for smartphone and tablet (Broxton).


Edit: added Atom for server

[ Edited by Andreas_Wolf 21.02.2017 - 09:20 ]

Quote:

Cool, I rather like the low draw cpus.

Quote:

Like means ‘similar to’. We often use it with verbs of the senses such as look, sound, feel, taste, seem:

My sister is like my mother. (My sister and my mother are similar)

I think this tastes like coconut.

That looks like Marco’s car.

He seems like a nice man.

When we use like to mean ‘similar to’, we can put words and phrase such as a bit, just, very, so and more before it to talk about the degree of similarity:

It’s a bit like skiing but there’s no snow.

Isn’t that just like the bike we bought you for your birthday?

That smells very like garlic.

The car was more like a green than a blue colour.Vv

Quote:

It's very much like Trance. The x86 binary is dynamically recompiled and calls native ARM libraries as though they were x86 libraries.

Quote:


Your entire post is based on your misunderstanding of the meaning of the word 'like' as used in this context.

See the dictionary definition of the word as quoted above:

https://dictionary.cambridge.org/grammar/british-grammar/common-verbs/like

>>>>>> X86 emulation via an ARM processor? Sounds ugly. [...] the real question is
>>>>>> will they be competitive?

>>>>> No different to running 68k binaries on PPC with Trance really.

>>>> Pretty big difference, Nick. The PPCs we run Trance on run many times faster than a
>>>> 68K processor. The ARM processors will be running at comparable speeds (or slower)
>>>> to an X64. Which, with the overhead, will mean X86 apps will run slowly on ARM.

>>> It's like it's running native ARM code.

>> Even if this was true (which I don't think it is as there simply must be emulation overhead)

> Like means ‘similar to’.

I know. Jim's point you were replying to was about competitiveness of this emulation solution, aka relative execution speed. And I doubt that the execution speed of x86 code on ARM is in any way similar to the execution speed of native ARM code on the same ARM CPU for the simple reason that there must be significant emulation overhead.
The fact that Microsoft's video only shows situations where this overhead doesn't seem perceptible is not surprising at all, after all that's the very purpose of that video. I'm sure eventually there will be independent side-by-side comparisons with native ARM code on the same ARM CPU and with the same x86 code on current x86(-64) CPUs. That's when we will know. Until this point I will maintain my opinion (which is shared by knowledgable people, btw).

>> it's still different to "running 68k binaries on PPC with Trance" for the reason
>> Jim mentioned.

> It's very much like Trance. The x86 binary is dynamically recompiled and calls
> native ARM libraries as though they were x86 libraries.

As I see it, Jim was talking about the anticipated speed of that solution, not about the implementation concept.

>>> https://youtu.be/A_GlGglbu1U

>> No side-by-side comparison with native x86 speed (or native ARM speed for that matter).

> Your entire post is based on your misunderstanding of the meaning of
> the word 'like' as used in this context.

I think your entire post is based on your misunderstanding of what Jim and I have been referring to. And I'm going to spare you the link to the Wikipedia article about hardware emulation ;-)

Thanks Andreas, at least you seem to understand my point that there will be overhead that will negatively affect ARM running X86 code.
How much I can't guess, although if it is converting OS calls to native ARM calls that will help.
But we are still dealing with cpus of comparable power as opposed to the disparity between the PPC and the 68K.

However, until more Windows packages are ported over to ARM, this will be a very useful functionality.

I could see using it.
At least on some hardware, as my future seems to be primarily tied to X64.

Snapdragon 835

http://www.twitlonger.com/show/n_1spgd0k

>> Snapdragon 835 (MSM8998) with Kryo 200 core

> Snapdragon 835
> http://www.twitlonger.com/show/n_1spgd0k

So it's the Kryo 280 core in the Snapdragon 835, not the Kryo 200 core. Thanks for the link.

Some slides:
http://cdn.videocardz.com/1/2017/01/Qualcomm-Snapdragon-835-1.jpg
http://cdn.videocardz.com/1/2017/01/Qualcomm-Snapdragon-835-2.jpg
http://cdn.videocardz.com/1/2017/01/Qualcomm-Snapdragon-835-3.jpg
http://cdn.videocardz.com/1/2017/01/Qualcomm-Snapdragon-835-4.jpg
http://cdn.videocardz.com/1/2017/01/Qualcomm-Snapdragon-835-5.jpg

Quote:

Apparently Microsoft themselves designed the hardware x86 emulator inside the 835.

http://www.pcworld.com/article/3148191/windows/microsoft-qualcomm-deal-finally-puts-windows-10-and-win32-apps-on-arm-devices.html

Quote:

Quote:

"Why this matters: Though the PC may be in decline, two growth segments have been low-cost (priced between $200 and $300) and two-in-one laptops."

"Microsoft said Thursday morning that it's partnered with Qualcomm to enable new, low-cost PCs. These are intended to replace Windows tablets built around Intel’s Atom, a chip the company's essentially discontinued."

IMHO, Microsoft is really showing persistence with their ARM ambitions! It didn't play out well the last time, now they are back with a new strategy. Who knows, maybe x86 emulation may just be the stepping stone the doctor ordered, on the way to "native". It has worked before, for others.

It's already looking very much usable (https://youtu.be/A_GlGglbu1U) although it's obviously not the speed they are aiming at, since the CPU's are simply not here for about another year.

Microsoft's persistence and ambitions (and their x86 emulator of course) may very well inspire a broader range of ARM CPU manufacturers to develop chips more suitable for PC's and laptops than "just" cellphones etc? Apple's designs is for sure a source of inspiration, and 10nm will of course play its part!

ARM for the future?

Quote:

I don't think this is correct?

Maybe I missed something in the article (and/or elsewhere), but

1. "hardware X86 emulator" (your words) and "X86 hardware emulator" (the article) is a difference in semantics that actually matters

2. JIT translation in SW makes sense. A JIT is a program that can look through a binary stream ahead of execution and "recompile" and optimize it to native binaries (using the entire palette of available native registers etc). Can a CPU do the same thing as efficient at runtime? JIT in SW is a proven concept. "Built-in emulation" in CPU's are AFAIK not. SW JIT's can be updated easily, CPU silicon not so, and the investments needed to make a CPU design is kind of huge, measured by any wallet.

3. "Microsoft released a video showing off how the current Snapdragon 820 runs Windows 10 Enterprise, via its emulator." The emulator is obviously already running, meaning it's SW (most certainly JIT).

4. The author of that article is perhaps a bit confused himself, mentioning Transmeta, etc for no obvious reason? Wasn't Transmeta more of an op-code thing than something comparable to a modern SW JIT?

The purpose of this thread has from the beginning been to try to predict how the ARM architecture is scaling upwards and growing into new markets, [constantly] a few years down the road.

To summarize a bit:

The x86 PC market has been in decline for a couple of years. Other kind of devices appeal, and desktop CPU performance has kind of been "enough" for many people for a long time, meaning less need for new HW.

But as the closest above article mentions, "Though the PC may be in decline, two growth segments have been low-cost (priced between $200 and $300) and two-in-one laptops."

The PC is mainly consisting of Core i7 (expensive, mainly targeting the most demanding rendering applications and workstations where multi-threading is key), Core i5 (essentially like i7, less multi-threading but more suitable for high end gaming and "normal" power users) and the low-cost Core i3 (that is probably is enough for the general usage of some 90% of PC users today). Then there is various Core m CPU's, with ambitious marketing labels aiming to resemble the desktop market segments, but not really being so. Do you agree with this analysis?

With the general decline of the PC market and the new and renewed ARM efforts from Microsoft (including x86 emulation, IDE/framework/compiler development) and CPU manufacturers, with the drop of Atom, etc, do you agree that ARM could very well be taking over the current Core i3 and Core m segments kind of "soon"?

In this very thread, the first of the "ARM NG" CPU from Apple easily compared to a laptop Core m7 in that chess game application, its design and execution model surely compared to a Core i7, and the generations that followed from Apple has proven comparable or even surpassing the existing lower-end laptops and older desktop PC's, even while running battery powered and passively cooled!

It has been said that the Sam460 port of MorphOS forced a general cleanup and modification to the source code of MorphOS that made it possible to bring up MorphOS running on the X5000 in a matter of hours instead of a matter of weeks. I believe that a drop of Amiga Legacy compatibility and 64-bit adaptions etc will make the sources even more clean and platform agnostic, no matter what the target platform currently is. This too is certainly a good thing.

My main platform has been x86 for years, my next computer will be x86 as well (very soon). But as x86 continues to evolve, far beyond the needs of general people, maybe x86 will be more heading into "POWER" market territory a few years from now (not that it isn't already there technologically, I'm talking about *market* focus), while ARM will conquer the broader, general computer needs for general people?

Maybe the focus of future MorphOS versions should be "Agnostic" rather than "x64"? Would put some demands on development environment etc to make a single source run on multiple platforms (like Linux). But maybe they are already going there?

>> the hardware x86 emulator inside the 835.

> I don't think this is correct?

Me neither.

Quote:

I think it's more likely that the 835 has added instructions that aid the software emulation than and actual "hardware emulator" in the vein of the Crusoe.

> I think it's more likely that the 835 has added instructions that aid the software emulation

Sounds unlikely to me. So far I've not seen or read any evidence in favour of that assumption.