> Applied Micro's new offering still look at little weak

You mean Titan/APM83290, which might or might not be dead? Or rather the yet to be announced new core/APM86xxx at 40nm? If the latter: do you have specs for this?

Quote:

That would be something indeed.

> That would be something indeed.

http://media.freescale.com/phoenix.zhtml?c=196520&p=irol-newsArticle&ID=1474744

I can only congratulate Freescale on this exceedingly sensible decision.

More detail:

"Freescale [...] will incorporate its AltiVec vector processing technology into its multicore QorIQ microprocessor family later this year [...]. [...] Freescale will announce specific QorIQ microprocessor products announcements with AltiVec technology within the next few months"
http://www.militaryaerospace.com/index/display/article-display/6971313341/articles/military-aerospace-electronics/online-news-2/2010/9/altivec-floating-point.html


Edit:

Another nice article:

"new demands from multiple 4G users are now enabling wireless basestations to utilize SIMD. To meet the need, Freescale Semiconductor Inc. is upgrading its venerable AltiVec vector processing unit for its QorIQ family of Power Architecture-based communications processors. "Customers using QorIQ to deal with all the data being processed by 4G basestations are probably telling Freescale that the Altivec SIMD processor could be very useful at performing the same processing task [single-instruction] on multiple users [multiple-data]," said Linley Gwennap, principal analyst at The Linley Group (Mountain View, Calif.) [...] Freescale already sells legacy AltiVec processors to industry, military and aerospace customers performing traditional tasks such as image processing, pattern recognition and forward kinematics to position robotic arms as well as for emerging applications such as smart analytics on multiple surveillance cameras. But by adding AltiVec to its leading edge QorIQ processors, Freescale hopes to expand further its penetration of the basestation processor market. "We have taken a proven technology, enhanced it, and moved it to the multicore QorIQ family of processors," said Glenn Beck, marketing manager for aerospace, defense and single board computing within Freescale's Networking Processor Division. "We see a broad spectrum of applications that are now needing vector-caliber performance, ranging from networking and telecom to the classic ones in aerospace and defense." [...] With an AltiVec inside each QorIQ core, an additional eight parallel floating-point and integer processing units execute vector algorithms using 180 instructions."
http://www.eetimes.com/electronics-news/4208818/Vector-processing-accelerates-QorIQ2

[ Edited by Andreas_Wolf on 2010/9/28 19:16 ]

Update:

>> Applied Micro's new offering still look at little weak

> You mean Titan/APM83290, which might or might not be dead? Or rather the yet to
> be announced new core/APM86xxx at 40nm? If the latter: do you have specs for this?

I guess we know the specs by now. From today's press release:

"The AppliedMicro PacketPro family features performance of up to 2 GHz per core, 32KB L1 I/D & 256KB dedicated L2 cache per core, support for full symmetric multiprocessing (SMP) and ultra flexible asymmetric multiprocessing (AMP). Memory and bus architecture supports 16/32/64-bit DDR2/3 up to 1,600Mbps and beyond with ECC option. Connectivity features include PCI-e Gen 2 controller, GE, 10GE, SGMII, RGMII, IEEE1588 Rev2 on all Ethernet ports, USB 2.0 - H/D, OTG, all with integrated PHY, USB 3.0, SATA ports and SDHC. The PacketPro family is manufactured on a 40nm TSMC(R) CMOS process and is available in both wire-bond and flip-chip packaging. The first PacketPro device begins sampling in November."
http://investor.appliedmicro.com/phoenix.zhtml?c=78121&p=irol-newsArticle&ID=1474906

From the product page:

"PACKETpro is AppliedMicro's second-generation of embedded multi-core processor SoCs and the first to feature expandability from one to eight 32-bit PowerPC 465 cores ranging in performance from 600 MHz to 1.5 GHz."
http://www.appliedmicro.com/products/process.html#EmbeddedProcessors

From the product info PDF file:

"PACKETpro is AppliedMicro's second-generation of embedded processor SoCs and the first to feature expandability from one to eight 32-bit PowerPC 465 cores ranging in performance from 600 MHz to 2.0 GHz."
http://www.appliedmicro.com/products/apm_pp_web.pdf (page 2)

Page 4 shows an interesting chart listing three CPU generations:
- past: single core, 32 bit, 1.5 GHz (PPC464 core based 1.4 GHz PPC460SX/PPC460GTx)
- today: multicore, 32 bit, 2.0 GHz (PPC465 core based PacketPro processor)
- future: multicore, 64 bit, 2.5 GHz (Viper?)

Titan/APM83290/"Gemini" doesn't seem to fit any of these generations, indicating that it's rather dead than alive.

My observations/questions:
1. Does PacketPro have an FPU?
2. They don't seem to be sure if PacketPro is 1.5 or 2.0 GHz.
3. PacketPro is PPC465 based, which explains the "465" being listed as "32-bit Commercial Core" of Applied Micro in Power.org's latest Power Architecture roadmap version.
4. PacketPro obviously doesn't use a "40nm variant of their Titan core".
5. What is Applied Micro's "Next-gen" "32-bit Commercial Core" that is listed in Power.org's latest Power Architecture roadmap version? Is it Titan?
6. What is Applied Micro's "APM86XXX" "32-bit Commercial Processor" that is listed in Power.org's latest Power Architecture roadmap version? Is it PacketPro?

What's also a nice destail in this news item is that Freescale does not list the network and telecomunications as target only but also medical, military, printing, robotocs and video systems.
My impression was that QorIQ was yet marketed as powerQUICC replacement which more or less targeted the network/telco market only.
Let's see how this develops...

> What's also a nice destail in this news item is that Freescale does not
> list the network and telecomunications as target only but also medical,
> military, printing, robotocs and video systems.

I guess that networking and telco is the target application of present QorIQ (i.e. without AltiVec), while coming QorIQ with AltiVec will (also) target the other fields (like MPC86xx to date).

> My impression was that QorIQ was yet marketed as powerQUICC replacement

As I told you already, Freescale's Power Architecture roadmap has explicitly been painting the QorIQ P4 and P5 chip families as replacements for the MPC86xx chip family. QorIQ P1 to P3 is supposed to replace PowerQUICC I to PowerQUICC III. At least that's what the roadmap presentation depicts.
It seems that Freescale has come to realizing that chips without AltiVec aren't quite adequate as replacements for chips with AltiVec. Future QorIQ chips with AltiVec will be regarded by 3rd parties as the real de-facto MPC86xx successors, as opposed to the presently announced/available QorIQ P4 and P5 chips without AltiVec which are regarded by 3rd parties only as "de-jure" successors to the MPC86xx, if at all.

> which more or less targeted the network/telco market only.

By far not all PowerQUICC chips even have the QUICC engine. I guess that's part of Freescale's intention to deliberately confuse (potential) customers ;-) PowerQUICC as a whole not only targets network/telco but also imaging/video, (multifunctional) printing, industrial control/robotics, measurement, storage and even military.


I wonder if Freescale will implement AltiVec in QorIQ
a) as a sub-unit of e500/e5500 core (like with e600) and thus create new core(s) or
b) as a coprocessor unit to e500/e5500 core or
c) simply by taking e600 core and continuing its development (i.e. e600 based QorIQ) ;-)

Yes, Freescale's marketing can be confusing, but its not nearly as vexing as Applied Micro's product information.

And I'm completely confident that Freescale will release the products they announce. Applied Micro on the other hand seems to regularly announce products before they're even assured that they will be produced.

Think about it, who has more credibility Freescale who inherited Motorola's CPU production technology or Applied Micro who licensed low end IBM designs?

> I'm completely confident that Freescale will release the products they announce.

You mean like the MPC75xx or the MPC87xx (e700 core)? ;-)

> Applied Micro on the other hand seems to regularly announce products
> before they're even assured that they will be produced.

The most confusing thing is that they said that Titan/APM83290 samples had been in the hands of (selected) customers who successfully validated the chip for their products, which is for about a year now:

------------------------------
Paramesh Gopi: I am extremely pleased to report that we are sampling 1.5 GHz multi-core SMP devices in an extremely cost optimized TSMC 90 nm node. [...] Our current TSMC core has garnered a tremendous amount of interest in the document processing and control processor segments and we look forward to reporting upcoming wins at major tier 1 OEMs as they materialize. [...] We are sampling the world's first TSMC-based dual core SMP 1.5 GB [sic!] power architecture SoC.
[...]
Robert Gargus: [...] I just want to caution you that this is a chip that we have just begun sampling. So it's not yet really into production and generating revenue. It's probably still a couple of quarters, two to three quarters away from starting to generate any kind of meaningful revenue on our P&L. But the amount of interest--and if you want, customer interaction on the chip is very encouraging at this point in time.
[...]
Allan Mishan: Hey Guys, quick question on the Titan products. You specifically mentioned document processing and control plane as areas where you're targeting now. Are you also offering Titan to all the other applications where you've historically worked with the IBM products as well or really are those two the priorities in the near term?
[...]
Robert Gargus: I think the ones we mentioned are just the ones that we can point to in terms of design activity that's going on currently. But over time, we expect to have that chip find its way into many, many more places.
------------------------------
http://www.sec.gov/Archives/edgar/data/711065/000119312509223997/dex992.htm (transcript of Applied Micro's second quarter 2010 earnings conference call in late October 2009)

------------------------------
Paramesh Gopi - Applied Micro Circuits Corporation - President, CEO
[...] we are extremely pleased with the reception of our TSMC-based 1.5 gigahertz multicore SMP device. We have secured the first of a series of SoC design wins in Tier 1 enterprise document processing systems. The integration of power-efficient cores in an established mature TSMC process node, coupled with our industry-leading packet processing IP and DSP subsystems, enable highly efficient implementations of image pipes, as well as network document and video processing functions. This has firmly established us as the leader in providing multicore, power efficient, multi-gigahertz cores to a wide customer base. We expect to announce meaningful revenue in 2011 resulting from these designs. Additionally, we are pleased to report that the validation of our cores at major customers sets the stage for our migration to 40 and 28 nanometer process nodes, and ensures that our core technologies are fully scalable. [...] Power architecture has never had the same accessibility as other architectures in very normal vanilla CMOS processes. So a lot of the initial barrier was nobody believed us. There was a huge amount of skepticism as to whether we would even get a gigahertz core at a process node that was 90 nanometer, much less anything that is 65 or 40. The uniform feedback was, you guys have shown us slides for three years, make us believe it. So I think I would say the last six months have proven beyond doubt that there is no more uncertainty in us being able to develop, sample, validate and drive our internal core architecture into the market. [...] So we have pretty much proven that there is a secret sauce and it has been a lot of work, but I think very few companies in this world can boast of a gigahertz plus architectures, especially in the PowerPC space in a TSMC process. So I think where we have established that beyond doubt that we can execute that. So now it is a matter of essentially driving that into real business. And that process has started very, very nicely in earnest with our key customers. As they see performance and they see cost and they see the scalability to do the 4 to 6 to 8, I mean, it is a full SMP-capable core and interconnect subsystem, so it is all there.
------------------------------
http://www.sec.gov/Archives/edgar/data/711065/000119312510020952/dex992.htm (transcript of Applied Micro's third quarter 2010 earnings conference call in late January 2010)

------------------------------
Paramesh Gopi - Applied Micro Circuits Corporation - President, CEO
[...] we also had success in the enterprise market where we won a Tier 1 cost-optimized wireless access point designed for the China and India markets. This was enabled once again by our cost-optimized, customized, power-efficient multicore technology being brought to a new cost point. Overall, I'm extremely pleased with our penetration into our target segment and our overall design execution.
------------------------------
http://www.sec.gov/Archives/edgar/data/711065/000119312510109166/dex992.htm (transcript of Applied Micro's fourth quarter 2010 earnings conference call in late April 2010)

So the question is: What went wrong? Is it really due to Intrinsity's acquisition by Apple as some suspect? On the topic of Apple/Intrinsity I only found the following from the April 2010 conference call:

------------------------------
Robert Gargus - Applied Micro Circuits Corporation - SVP, CFO
[...] Following the end of the quarter, a company where we had an equity investment was acquired by Apple. As a result of this transaction, we realized the full value of our investment, which was approximately $5.4 million. We received this cash earlier in the month. [...] The $206.6 million cash balance excludes [...] the $5 million received in the first week of April upon the sale of our equity investment to Apple.
------------------------------

> who has more credibility Freescale who inherited Motorola's CPU production
> technology or Applied Micro who licensed low end IBM designs?

I think they both have plenty of skeletons in their respective closets.

Personally, I can't really hold Freescale's failure to pursue the e700 core against them. It reminds me of Motorola and the 68K. I was really into that family line (and when the 68030 came out I was convinced it was going places). Motorola did forge ahead with the 68060 (an incredible processor) even after losing a huge part of its market.
If it doesn't sell, you've got to move on. The e600 series did not sell well and like the 68060 that spelled doom for a future successor.
Freescale's current shift in marketing merely reflects what they think will be profitable.

When Applied Micro actually lets someone see some of this mythical silicon, then I'll start to place more faith in their press releases.

> The e600 series did not sell well and like the 68060 that spelled doom
> for a future successor.

The difference is that a 68060 successor was never announced (AFAIK), but the e700 was.
So far, we've yet to hear Applied Micro's rationale for discontinuing Titan/APM83290 (if they did, that is).

> When Applied Micro actually lets someone see some of this mythical silicon,
> then I'll start to place more faith in their press releases.

Do you think they lied to the analysts when they claimed that Titan silicon is sampling and in the hands of customers?


Edit:

Insightful article:

"After a false start, Applied Micro Circuits Corp., or AppliedMicro, is taking another shot in the multicore processor arena. [...] The company has announced multicore products in the past. But to date, it failed to release them. "The company announced a multicore processor, but never released it as an official product, and used it more as a demonstration platform," according to the company. It hopes to have better luck with PacketPro."
http://www.eetimes.com/electronics-news/4208848/AMCC-takes-another-shot-at-multicore-semiconductor

And from a comment to this article:

"AMCC is trying again at multicore. Last year, it rolled out a CMOS-based, 32-bit processor, built around IBM Corp.'s Power Architecture. What's different is that the codenamed Gemini multicore processor from AppliedMicro [...] will be made using a 90-nm, bulk CMOS process from Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) [...]. Gemini never flew, however. Here's what AMCC said: "So after announcing Gemini at 90nm last year, the company decided to wait until the 40nm in order to provide true differentiation in the market, and that's what you see announced today with PacketPro. The decision was made some time after the Gemini announcement in Sept. 2009. Gemini is a working processor and it was demonstrated at ESC Silicon Valley this year. It's been used mostly as an evaluation platform with customers. The innovative capabilities demonstrated with Gemini at 90nm led directly to strong customer feedback and the development of PacketPro at 40nm, which is being demonstrated today at the Linley Processor Conference in San Jose.""

My conclusions, considering the statements in that comment:
1. Titan/APM83290/"Gemini" silicon works and is in the hand of customers.
2. It won't be pursued any longer. The existing customers will all be switched to PacketPro.
3. The reason for killing Titan/APM83290 seems to be a mind change to go with 40nm instead of 90nm.

#3 is puzzling because Applied Micro claimed that the Titan design could easily be migrated to smaller process nodes like 40nm(*). So why was there the need to replace the Titan core by the PPC465 core? And why was the Titan/APM83290 road presented as still being pursued in late July (conference call) and early August (PDF presentation)?
It still doesn't make sense.

(*) "The Titan core [...] is designed to be portable to a variety of standard bulk CMOS geometries of 65 and 40nm."
http://www.sec.gov/Archives/edgar/data/711065/000119312509165561/dex992.htm (transcript of Applied Micro's first quarter 2010 earnings conference call in late July 2009)

[ Edited by Andreas_Wolf on 2010/9/28 19:35 ]

While Freescale indeed announced the e700 and didn't deliver they may get excused because their main customer for chips incorporating teh e700 core moved on - Apple. But it would have been nice if Freescale had given an official statement hat the e700 got abandoned. Anyway there was (and probably is) a lot of internal movement/restructuration inside Freescale and maybe some department does not know what the other department is doing or who is responsible for tasks of a closed department.

> it would have been nice if Freescale had given an official
> statement hat the e700 got abandoned.

Indeed.

> there was (and probably is) a lot of internal movement/restructuration
> inside Freescale and maybe some department does not know what the
> other department is doing or who is responsible for tasks of a closed department.

While that may be true it doesn't speak in favour of the Freescale company as a whole. It's just that I don't find it fair to nag at Applied Micro for similar things that Freescale is excused for (you didn't do that, I know).

Very valid point about the end of the 68K.

And, yes, they appear to have given some early Titan products to select customers. You and I will never see one.\

Your last point is the one that baffles me too. Why not migrate Titan to a finer process? Wouldn't that afford the possibility of higher clocking?

Does the loss of their design partner to Apple mean that some part of the intellectual property behind Titan can no longer be used?

And I'm further confused by their references to IBM design components.

Does either Titan or the PPC465 use significant design elements from their older IBM licensed products?

Edit - I do want to add a final note on both your and Zylesea's distress at Freescale's failure to communicate its intentions.
They must have lost, fired, or transferred a lot of employees when they shifted focus.
My contact at the company is no longer working there and even though I've received product samples since Motorola introduced the 6829MMU, no one had been given any of my representatives contact files.
To summarize, I had to reintroduce myself to a company I'd formed a relationship with in the '80's.
Frankly, you wonder at times how dis-coordinated Freescale's various endeavors are.

[ Edited by Jim on 2010/9/28 20:54 ]

[ Edited by Jim on 2010/9/28 20:57 ]

> yes, they appear to have given some early Titan products to select
> customers. You and I will never see one.

Fair enough, but I don't consider me to be any kind of gold standard. It's funny to imagine only the things existing that I've seen with my own eyes ;-)

> Why not migrate Titan to a finer process? Wouldn't that afford the possibility
> of higher clocking?

Yes, that's the rule of thumb. Seems there's more to the story than Applied Micro's telling.

> Does the loss of their design partner to Apple mean that some part of the
> intellectual property behind Titan can no longer be used?

That would certainly be the perfect excuse. But remarkably, they don't use it.

> I'm further confused by their references to IBM design components. Does either Titan
> or the PPC465 use significant design elements from their older IBM licensed products?

I can't say for Titan, but PPC465 is IBM IP the same way the other cores (PPC405, PPC440, PPC464(FP)) used by Applied Micro are. And considering that both PPC464 and PPC465 belong to the same core family (PPC460) I'd think that they share some significant design resemblances.

So you've got a situation where there are some obvious questions left unanswered?

I'm not worried, If there's anyone here who's more likely to figure this out then you, I haven't met him.

I look forward to what ever you come up with (unless this remains a mystery and thus totally shatters my sense of the universal scheme of things).

[ Edited by Jim on 2010/9/29 1:22 ]

> So you've got a situation where there are some obvious questions left unanswered?

Yes.

> I look forward to what ever you come up with

I'll keep you posted.

Update:

> I wonder if Freescale will implement AltiVec in QorIQ
> a) as a sub-unit of e500/e5500 core (like with e600) and thus create new core(s) or
> b) as a coprocessor unit to e500/e5500 core or
> c) simply by taking e600 core and continuing its development (i.e. e600 based QorIQ) ;-)

Freescale decided for option 'a' according to this:

"According to Freescale, AltiVec requires about as much die area as a data cache. As processors integrate more features, the portion of a die used by the CPUs is diminishing. Thus, the cost of adding AltiVec is fairly minimal, even when multiplied by several CPU cores. Once Freescale implements AltiVec into its CPU cores, we expect the company to offer it in all five tiers of the QorIQ family."
http://www.linleygroup.com/Newsletters/LinleyWire/wire101001.html#3

And according to this it seems that it will be members of the recent QorIQ P series that will get AltiVec (not just the future QorIQ T series as I suspected as a possibility at first).

Addendum:

> They don't seem to be sure if PacketPro is 1.5 or 2.0 GHz.

One more (external) source supporting the '1.5 GHz' version:

"PacketPro [...] will feature one or two PowerPC CPU cores running at up to 1.5GHz."
http://www.linleygroup.com/Newsletters/LinleyWire/wire101001.html#1

Some more from that same article:

"PacketPro is essentially AppliedMicro's Plan B. The company set out to design its own PowerPC-compatible CPU, code-named Titan, but that project ran well behind schedule and was ultimately canceled. From its ashes, and amidst a change in corporate management, the PacketPro product line was born."

That "amidst a change in corporate management" part is interesting. The Titan core based APM83290/"Gemini" processor was announced in October 2009. Logically, the switch from Titan core based APM83290 to PPC465 core based PacketPro must have happened somewhen after that. Applied Micro's CEO transition was announced in late January 2009 to be completed in early June 2009. In August 2009 the CEO transition was presented (page 3) by the company as being completed. So if I'm not mistaken then there must have been a time period of at least about half a year between the completion of the "change in corporate management" and the birth of PacketPro (and thus the abandonment of Titan). (Not to mention that as recent as July/August 2010 Applied Micro presented Titan as still going well.)

From that information, it would appear that Applied Micro hasn't been that honest about its latest products in development.

> From that information, it would appear that Applied Micro hasn't been that
> honest about its latest products in development.

You refer to PacketPro's clock frequency discrepancy?

> Freescale's marketing can be confusing

Absolutely. Look at this:

"AltiVec SIMD has long been recognized for its performance in the PowerQUICC processor line"
http://www.freescale.com/webapp/sps/site/homepage.jsp?code=QORIQ_HOME

AltiVec in PowerQUICC? Huh? While it's true that they present QorIQ P1 to P3 as PowerQUICC replacements and QorIQ P4 and P5 as MPC86xx replacements that surely doesn't make MPC86xx a PowerQUICC, does it? Are they now trying to retroactively subsume MPC86xx (or even MPC74xx) as PowerQUICC?

Quote:

No, although that is confusing. What I meant was that it seems as though Applied Micro came to the decision to change course awhile ago, but until recently has still been touting Titan.

The more I compare the two companies, the more I lean toward Freescale (and QorIQ).

>>> From that information, it would appear that Applied Micro hasn't been that
>>> honest about its latest products in development.

>> You refer to PacketPro's clock frequency discrepancy?

> No [...]. What I meant was that it seems as though Applied Micro came to the
> decision to change course awhile ago, but until recently has still been touting Titan.

But that's not what you could have extrapolated "from that information" in my posting you replied to, is it? After all, the "amidst a change in corporate management" statement, which made me wonder, stems from the Linley Group, not from Applied Micro. As far as I can see, Applied Micro never claimed that the switch from Titan core based APM83290 to PPC465 core based PacketPro happened or was decided "amidst a change in corporate management". To me it seems that with spreading misinformation like this the Linley Group is contributing to the confusion created by Applied Micro more than it's helping to resolve it, which makes it even harder to tell fact from fiction.

It doesn't help that Applied Micro withholds a great deal of information. In a vacuum information like the stuff you quoted from Linley Group stands unchallenged.
And Applied Micro is even worse than Freescale at this, but they both often allow their working partners to release information before they do themselves.

The dates you've mentioned for the likely shift in development are close to when I approached Applied Micro about an NDA. No one I talked to was anything but positive about the Titan core. But at the time, their reason for no giving me an NDA (you don't maintain a website) was pretty lame (considering I wasn't interested in promoting my endeavors on the internet).

Finally, I still don't understand why Applied Micro is harder to establish a relationship with than IBM (which is in itself pretty close mouthed, but not that bad).
I don't get my suspicions about Applied Micro's solely from your postings. It's the entire way they've dragged out this situation and the incomplete information a half-truths we've been getting the whole time.

> In a vacuum information like the stuff you quoted from
> Linley Group stands unchallenged.

It's not quite a vacuum. Some information are fixed, like for instance the fact that Applied Micro publically announced the APM83290 in October 2009. Do you really consider it possible that they had scrapped Titan and the first CPU to be based on it *before* they publically announced that CPU?

> The dates you've mentioned for the likely shift in development

What's to be considered likely in this case strongly depends on which source of information is followed, and beyond that at which point in time it is followed, as even one and the same source doesn't shy away from changing the story back and forth as time passes.

> are close to when I approached Applied Micro about an NDA.
> No one I talked to was anything but positive about the Titan core.

When was that exactly?

> their reason for no giving me an NDA (you don't maintain a website)
> was pretty lame

I remember you mentioned that in early July, a mere three days before the Linley Group reported on July 5th that Titan had been killed. Now do you retrospectively suppose that your contacts at Applied Micro only pretended to be positive about Titan, hiding their true feelings and/or knowledge in that matter?

> I don't get my suspicions about Applied Micro's solely from your postings.

I very much hope so. After all, you had closer contact to them than me (which isn't that hard considering that I had absolutely no contact with them to date ;-) I merely quote and combine statements I can find on the freely accessible part of the Internet here.

> the incomplete information a half-truths we've been getting the whole time

...not to forget the mutually contradicting information.