New Low Power POWER processors - not IBM

Posted:
in Future Apple Hardware edited January 2014
I wonder if those guys were too stealthy or if Apple really didn't know about them when they decided to stop playing POWER games (pun intendend):

PA-Semi
«1

Comments

  • Reply 1 of 35
    heinzelheinzel Posts: 120member
    ... they actually go on to compare their floating point power efficiency to the Intel's projected future Sossaman processor here

    Crazy times! (Or timing?!) The Microprocessor Forum will be interesting...
  • Reply 2 of 35
    jedhajedha Posts: 24member
    well.. lets hope its true
  • Reply 3 of 35
    tubgirltubgirl Posts: 177member
    apple could always return to ppc or mix the lineup if steve wants to.

    i think/hope steve have learnt his lesson and will keep macos portable to have a easy way out if he faces another 'motorola situation'.
  • Reply 4 of 35
    neat! nevertheless it's nothing but paperware at the moment and will stay so for quite some time.



    besides, how often have we heard about some "new, fantastic powerpc-processors from <insert your favourite scapegoat here> with <insert your favourite hair-raising spec here> being just around the corner" now?
  • Reply 5 of 35
    This will be good to keep a fire under Intel in the future - especially if Apple keeps their combined PPC/Intel approach to software and the OS.



    Personally I wish them luck - it would be nice to see someone being able to move the PPC forward at a reasonable speed.
  • Reply 6 of 35
    The product is apparently 2 years away, and they plan to focus on the embedded market. No telling how it will compare to other chips at that point. Performance or power.
  • Reply 7 of 35
    programmerprogrammer Posts: 3,458member
    Quote:

    Originally posted by nowayout11

    The product is apparently 2 years away, and they plan to focus on the embedded market. No telling how it will compare to other chips at that point. Performance or power.



    This is an important point to remember. There is a large possible "design space" for processors, and this chip likely doesn't address the portion of that space Apple is interested in. If these guys thought it might you can bet they would have talked to Apple long before any public announcement.



    Also, 2 GHz doesn't really tell us much about performance as there are many other factors involved.
  • Reply 8 of 35
    heinzelheinzel Posts: 120member
    There's a nice overview of the PA-Semi architecture and some of the design decisions involved at Real World Tech by David Kanter. Seems like they have a different take on what useful things to do with the increasing transistor budget than Intel.
  • Reply 9 of 35
    Quote:

    Originally posted by heinzel

    There's a nice overview of the PA-Semi architecture and some of the design decisions involved at Real World Tech by David Kanter. Seems like they have a different take on what useful things to do with the increasing transistor budget than Intel.



    That's quite a good article (typical of RWT). I'm very surprised by how beefy that core is going to be! Other aspects of the chip are clearly oriented for embedded systems though. We'll also have to see how Intel's future cores turn out -- 2 years is going to see a lot of changes from Intel.
  • Reply 10 of 35
    Quote:

    Originally posted by Programmer

    That's quite a good article (typical of RWT). I'm very surprised by how beefy that core is going to be! Other aspects of the chip are clearly oriented for embedded systems though. We'll also have to see how Intel's future cores turn out -- 2 years is going to see a lot of changes from Intel.



    It finally made me understand what the use of the e.g. TCP offload is supposed to be that the different chip manufacturer keep touting for their next generation designs. To me, their design seems like a very simplified offshoot of the EV8 design, multiple cores, multiple on-die memory controlers, programmable offload engine, and everything highly modularized to suit all possible needs and bring it to market rapidly. Wow! Hopefully the marketing is better than DIGITAL's... (and they don't get bought by Intel and laid to sleep!)
  • Reply 11 of 35
    ptrashptrash Posts: 296member
    Quote:

    Originally posted by nowayout11

    The product is apparently 2 years away, and they plan to focus on the embedded market. No telling how it will compare to other chips at that point. Performance or power.



    Two years from now we'll be using our ipods to download SW to our Apple branded home entertainment systems. Or is that 5 years away?



    OK, I'm talking out of my ass but I suspect things will look radically different in 5-10 years, with PCs a thing of the past and everything working via distributed systems.
  • Reply 12 of 35
    henriokhenriok Posts: 537member
    Wow! Seems like a really cool processor and the people behind it seems like the real deal. It will be interessting to follow this development. From what I see this processor will certainly give Freescale a challenge with their e600 and e700 platforms. IBM's processors and CELL have completely different focus.
  • Reply 13 of 35
    programmerprogrammer Posts: 3,458member
    Quote:

    Originally posted by Henriok

    IBM's processors and CELL have completely different focus.



    Not entirely -- IBM wants to use Cell in high performance computing (i.e. supercomputers) and these guys talk about that as an application as well. This things double precision performance, computation:bandwidth, and more traditional programming model might give them the advantage there.
  • Reply 14 of 35
    rickagrickag Posts: 1,626member
    So, reportedly a third party fabless company will be using VMX and IBM will not be the manufacturer, interesting.



    Wasn't IBM kinda pulled grudgingly toward VMX(aka Altivec and Velocity Engine). Isn't VMX IBM's name for Altivec/Velocity Engine.



    Actually, for me, this is the most interesting fact from these announcements.



    Could this mean that Intel may be able start incorporating aspects of Velocity Engine in their SSE, if they so choose or are licensed by Apple? Or am I wayyyyyy out in left field here?????
  • Reply 15 of 35
    imiloaimiloa Posts: 187member
    Does seem like a good processor, made by some experienced and sharp folks. I remember the Alpha being way ahead of its time on release, just not having a market foothold (like Betamax).



    That said, from the literature, this companies focus seems to be (wisely) on embedded and supercomputer models. And no doubt, the relatively low volume of CPUs Apple actually buys every year will not give PA-Semi much incentive to focus on desktop-specific designs.



    So if their chips are suitable for Mac hardware down the road, it will probably be more due to coincidence than strategic alliance.



    As many have stated, Apple is moving to Intel because Intel has a mass market focus on desktop and notebook chips. Volume-wise, Apple remains a small fish in the chip demand pond. So it makes complete sense for them to align themselves with the tastes of the bigger fish.



    I honestly don't know the params, but I expect that Intel's success in fab technology is largely a result of the huge influx in capital resources they gained over the past couple decades due to the proliferation of PCs.



    With Vista on the horizon, there will be a huge consumer market for their next gen chips as well (perhaps something of a risk without Vista). And Apple (and we Macaphiles) can now come along for the ride.
  • Reply 16 of 35
    imiloaimiloa Posts: 187member
    Quote:

    Originally posted by rickag

    Could this mean that Intel may be able start incorporating aspects of Velocity Engine in their SSE, if they so choose or are licensed by Apple?



    Would this even be feasible hardware-wise? I'm not a chip designer, and haven't written assembler since 8086, so I have no idea. But intuitively, it seems the vector architecture instruction set may not be compatible with x86 in general.



    Also, I doubt Mot & IBM would be keen on licensing the tech, and moreover, that Intel would prefer to roll their own, given they have the cash flow to do, and the opportunity to try to make something even better, given the new hardware paradigms that have evolved since the original Altivec design.



    Programmer, your insight on this?
  • Reply 17 of 35
    wmfwmf Posts: 1,164member
    Quote:

    Originally posted by rickag

    So, reportedly a third party fabless company will be using VMX and IBM will not be the manufacturer, interesting.



    The manufacturer hasn't been specified AFAIK.



    Quote:

    Wasn't IBM kinda pulled grudgingly toward VMX(aka Altivec and Velocity Engine). Isn't VMX IBM's name for Altivec/Velocity Engine.



    More like VMX was the original name, and then Moto decided to call it AltiVec, and then Apple decided to call it yet another thing.



    Quote:

    Actually, for me, this is the most interesting fact from these announcements.



    Why? VMX benefits some embedded apps, Freescale has VMX, PA wants to compete with Freescale, thus PA will have every feature that Freescale has, plus more. I'm not sure why you're focusing on IBM, given that (by your own admission) IBM doesn't really care about VMX.



    Quote:

    Could this mean that Intel may be able start incorporating aspects of Velocity Engine in their SSE, if they so choose or are licensed by Apple? Or am I wayyyyyy out in left field here?????



    Intel has been copying ideas from VMX for years; that's how they got SSE, SSE2, and SSE3. At this point there's very little left to copy. And (as has been said over and over) Intel cannot afford to build stuff that only Apple wants (actually, no company can afford to build stuff that only Apple wants).
  • Reply 18 of 35
    thttht Posts: 5,444member
    Bah. Not impressed.



    A 19 stage execution pipeline, 3 issue PowerPC processor with 1.5 integer units (branch unit can handle integer instructions), 1 FPU unit, a simple SIMD unit and a complex SIMD unit. Dare I say it? Freescale and the 8641D ain't got nothing to worry about from this chip.



    That SPECfp2k score must have been simulated with an infinitely perfect memory system, otherwise, I don't believe it.



    They are going to have a tough time entering a market. They will be late, maybe Spring 07 if their 65 nm manufacturer is good. Not that many candidates out there: Intel? no. IBM? no. AMD? maybe. Sony-Toshiba? maybe. By that time the 8641D will be entrenched and who knows what sort MIPS-based processors will be in the market with competitive specs.
  • Reply 19 of 35
    rickagrickag Posts: 1,626member
    Quote:

    Originally posted by wmf

    The manufacturer hasn't been specified AFAIK.



    I read it on one of the news websites, can't remember which. They supposedly quoted a PA representative as saying IBM would not be the manufacturer.



    Quote:

    Originally posted by wmf

    More like VMX was the original name, and then Moto decided to call it AltiVec, and then Apple decided to call it yet another thing.[/B]



    All evidence has indicated that the original break up of the Somerset design team(AIM) was largely caused by Apple/Motorola pushing for SIMD and IBM stubbornly refusing to accept it. IBM only recently added VMX to any cpu, that being the G5.

    As I understand it one of the key players in designing and/or supporting the development of Altivec/VMX/Velocity Engine in the original Somerset team (AIIM) was an Apple employee. I believe his name was Dierdorf(sorry if I misspelled the name) I've even read that without his presence VMX/Altivec/Velocity Engine may have been shelved.



    Quote:

    Originally posted by wmf

    Why? VMX benefits some embedded apps, Freescale has VMX, PA wants to compete with Freescale, thus PA will have every feature that Freescale has, plus more. I'm not sure why you're focusing on IBM, given that (by your own admission) IBM doesn't really care about VMX. [/B]



    It's not the fact that it benefits embedded apps. It's the fact that IBM is allowing/licensing the technology to a third party, again. Neither Motorola nor Apple, both of which have patents concerning Altivec/Velocity Engine have allowed or licensed this technology.

    I didn't say that "IBM doesn't really care about VMX". Reread what I said. I used the word "wasn't" which is past tense. Finally after what 12 years IBM has seen the light and accepts SIMD.



    Quote:

    Originally posted by wmf

    Intel has been copying ideas from VMX for years; that's how they got SSE, SSE2, and SSE3. At this point there's very little left to copy. And (as has been said over and over) Intel cannot afford to build stuff that only Apple wants (actually, no company can afford to build stuff that only Apple wants). [/B]



    Where did I say or even imply that Intel would build stuff that only Apple wants?? You really need to take the time to read before you post and put words in my mouth.

    SSE is similar only in the fact that it is SIMD. There are still significant differences. I believe one of the more significant one is the FP unit is shared in SSE, in Velocity Engine the FP units are separate.



    Let me clarify a little.

    #1. Until recently no third party has been allowed to or been licensed any part of VMX/Altivec/Velocity Engine.



    #2. Now we have Sony and Microsoft using VMX in their game boxes, no real competition in the general purpose computing field.



    #3. Now we have PA designing a general purpose cpu to be used in supposedly low power high performance laptops, blade servers and networking. IBM probably will use this chip in blades and possibly networking applications, but IBM no longer makes laptops. Could be this was intended for Apple, we may never know.



    #4. Altivec/VMX/Velocity Engine have features that many consider to be superior to SSE 3, and when compared they outperform SSE significantly.



    #5. If possible, may Apple be able to license features of VMX/Altivec/Velocity Engine to Intel to improve the performance of SSE.
  • Reply 20 of 35
    Check out this link

    http://www.theinquirer.net/?article=27233



    The architecture is impressive.

    I hope Jobs & Apple to take a good look into these chips.

    More choices won't hurt.



    gc63hk
Sign In or Register to comment.