PowerPC Evolution

thttht
Posted:
in Future Apple Hardware edited January 2014
Quote:

Originally posted by trilobita

I would like to see that cladogram



Per your request:



Code:




RISC ISA Class

|

|

+--------------------------------------------------- Others

|

+--------------------------------------------------- Motorola 88000

|

|

IBM Power1

|

IBM Power2

|

IBM P2SC

|

|

PowerPC ISA

Order

|

|

IBM ---+---------------------------------------------- Power3

Family | |

| | ------------------------------------------ Power3-II

| | |

| | -------------------------------------- Power3-III

| | |

| | ---------------------------------- Power4

| | |

| | +------------------------------ Power4+

| | | |

| | | -------------------------- Power5

| | | |

| | | ---------------------- Power5+

| | |

| | ------------------------------ PPC 970

| | |

| | -------------------------- PPC 970fx

| | |

| | ---------------------- PPC 970mp/gx

| |

| +---------------------------------------------- RS-64 I

| | |

| | ------------------------------------------ RS-64 II

| | |

| | -------------------------------------- RS-64 III

| | |

| | ---------------------------------- RS-64 IV

| |

| +---------------------------------------------- Xenon

| | |

| | +------------------------------------------ Cell

| | |

| | ------------------------------------------ Broadway?

| |

| ---------------------------------------------- PPC 405

| |

| ------------------------------------------ PPC 440

|

|

Somerset ----------------------------------------------- PPC 601

Family |

| +------------------------------------------ MPC 603

| | |

| | +-------------------------------------- MPC 603e

| | | |

| | | +---------------------------------- MPC 740/50/55

| | | | | |

| | | | | -------------------------- PPC 750cx/cxe

| | | | | |

| | | | | +---------------------- PPC 750fx

| | | | | | |

| | | | | | ------------------ PPC 750gx

| | | | | |

| | | | | ---------------------- Gekko

| | | | |

| | | | ------------------------------ MPC 7400

| | | | |

| | | | -------------------------- MPC 7410

| | | | |

| | | | ---------------------- MPC 7440/50

| | | | |

| | | | ------------------ MPC 7445/55

| | | | |

| | | | -------------- MPC 7447/57

| | | | |

| | | | ---------- MPC 7448

| | | | |

| | | | +------ MPC 8641D

| | | | |

| | | | ------ e600 embed

| | | |

| | | ---------------------------------- e500 embed

| | |

| | -------------------------------------- MPC 52xx

| |

| +------------------------------------------ e300 embed

| |

| +------------------------------------------ PPC 602 embed

| | |

| | -------------------------------------- MPC 5xx?

| |

| +------------------------------------------ PPC 604

| |

| +-------------------------------------- PPC 604e

| |

| -------------------------------------- PPC 620

|

Exponential --------------------------------------------- x704

Family









If you see anything wrong, please provide input so it can be corrected.



I also have a tree available as well, but I want to turn it into a PowerPC technology tree and subsequently needs work.



[2nd edit: slight changes.]
«1

Comments

  • Reply 1 of 26
    rraburrabu Posts: 239member
    I may be wrong but didn't the 7400 root from the 604 family as the 750 was developed from the 603 family?
  • Reply 2 of 26
    wmfwmf Posts: 1,164member
    Power3 (aka 630) is presumably a derivative of the 620. POWER4 is not a derivative of Power3.
  • Reply 3 of 26
    thttht Posts: 3,168member
    Quote:

    Originally posted by rrabu

    I may be wrong but didn't the 7400 root from the 604 family as the 750 was developed from the 603 family?



    This is a myth that never dies. The G4 genus of CPUs are direct descendents of the 750.



    The 7400 microarchitecture is nearly identical to the 750. Motorola's stated purpose for the 7400 was "to change from the MPC750 base only where required to gain compelling multimedia and multiprocessor performance. MPC7400's core is essentially the same as the MPC750's..." Same front end, same instruction fetch and issues (2 + 1 branch), same integer units, nearly the same FPU, same completion unit, same branch unit, same pipeline depth. "Same" here means "for all intents and purposes".



    The only major difference between the 750 and the 7400 is the AltiVec unit. All other changes were to accomodate AltiVec. The MPX bus is essentially the 60x bus modified for less latency, less dead cycles, and more out-of-order, concurrent transfers to accomodate the higher bandwidth demands from the AltiVec unit. The physical difference between the 60x and MPX buses is actually quite small.



    The root of the myth stems from the a misunderstanding that 7400 has the FPU unit of the 604. The only major difference between the 750 FPU and the 7400 FPU is that the 7400 FPU does double precision with a latency of 3 cycles while the 750 does the same operation with a latency of 4 cycles.
  • Reply 4 of 26
    thttht Posts: 3,168member
    Quote:

    Originally posted by wmf

    Power3 (aka 630) is presumably a derivative of the 620. POWER4 is not a derivative of Power3.



    The 620 and 630 has no relationship to each other. Going be the names alone we may think they are related, but I don't think they are. Perhaps IBM wanted to make the Power3 a commercial chip at the time, and thus named it the 630.



    The 620 is a 64 bit implementation of the 604. Same number of integer units, 1 FPU unit, branch unit, other units, 4 instruction issue, and same pipeline depth.



    The Power3 is a descendent of the Power2 (P2SC actually) by virtue of 2 integer units, 2 FPU units, and 2 load store units common in both. If you haven't noticed, the Power4 also happens to have 2 integer units, 2 FPU units and 2 load store units. Power4's big evolution was the pipeline extension, instruction ordering, grouping schemes, dual-core, and MCM packaging with lots of memory bandwidth.
  • Reply 5 of 26
    henriokhenriok Posts: 537member
    Excellent!

    I really like the included embedded cores. Saves me a lot of research Thanks!
  • Reply 6 of 26
    splinemodelsplinemodel Posts: 7,311member
    Quote:

    Originally posted by THT

    This is a myth that never dies. The G4 genus of CPUs are direct descendents of the 750.



    Yes indeed. The 7400/7410 is very similar to the 750 series. The weird thing is that the 7450 is a pretty big jump, and in my opinion it is more different than the 7400 than the 7400 is to the 750.



    It's also funny how IBM doesn't even acknowledge the existence of the 7400 series. It was all Motorola.
  • Reply 7 of 26
    benroethigbenroethig Posts: 2,782member
    Quote:

    Originally posted by Splinemodel

    Yes indeed. The 7400/7410 is very similar to the 750 series. The weird thing is that the 7450 is a pretty big jump, and in my opinion it is more different than the 7400 than the 7400 is to the 750.



    It's also funny how IBM doesn't even acknowledge the existence of the 7400 series. It was all Motorola.




    The early 7400s were even pin combatible with the 750s. I could have stuck a 7400 G4 in the ZIF socket of my B&W G3.
  • Reply 8 of 26
    programmerprogrammer Posts: 3,409member
    Quote:

    Originally posted by THT

    The root of the myth stems from the a misunderstanding that 7400 has the FPU unit of the 604. The only major difference between the 750 FPU and the 7400 FPU is that the 7400 FPU does double precision with a latency of 3 cycles while the 750 does the same operation with a latency of 4 cycles.



    The 750 only has a throughput of 1 double precision operation per 2 clock cycles (like the 603's FPU, which makes sense because the 750 was a follow-on to the 603 simpler architecture with a whole lot more cache). The 7400's FPU was improved to have a throughput of 1 double precision operation per clock cycle, which is the same as the 604's FPU -- whether they actually used any circuitry design from the 604 is unknown.
  • Reply 9 of 26
    thttht Posts: 3,168member
    I will concede that the 7400 FPU, as far as I can tell, is identical to the 604e FPU in terms of execution latencies and througput. It could very well be that they even have the exact same circuitry but the nagging problem is that the 604 was an IBM design. It may be one of the reasons Motorola (or even Apple) couldn't just evolve the 604 on their own without IBM's approval.



    In an alternate history where the 604e would be enhanced with backside cache, expanded pipeline depth, on-chip L2, more integer units, AltiVec, higher bandwidth bus and such, Apple could have been very competitive to Athlon and Pentium 3, 4 CPUs. A 4 instruction issue processor with lots of execution resources would have performed quite well. The 604e was 47 mm^2 at 250 nm. IBM could have had a 120 mm^2 dual-core 604e derived processor at 180 nm. That would have been sometime around the year 2001.
  • Reply 10 of 26
    wmfwmf Posts: 1,164member
    The enhanced 604 was called 760/770. I don't know how far it got.
  • Reply 11 of 26
    programmerprogrammer Posts: 3,409member
    Quote:

    Originally posted by THT

    I will concede that the 7400 FPU, as far as I can tell, is identical to the 604e FPU in terms of execution latencies and througput. It could very well be that they even have the exact same circuitry but the nagging problem is that the 604 was an IBM design. It may be one of the reasons Motorola (or even Apple) couldn't just evolve the 604 on their own without IBM's approval.



    Actually it was designed and developed right along side the 603 and after the 601 at Somerset with all three companies participating. The 601 was essentially an IBM design as well, after all.



    Quote:

    In an alternate history where the 604e would be enhanced with backside cache, expanded pipeline depth, on-chip L2, more integer units, AltiVec, higher bandwidth bus and such, Apple could have been very competitive to Athlon and Pentium 3, 4 CPUs. A 4 instruction issue processor with lots of execution resources would have performed quite well. The 604e was 47 mm^2 at 250 nm. IBM could have had a 120 mm^2 dual-core 604e derived processor at 180 nm. That would have been sometime around the year 2001. [/B]



    If I recall there was a project to build a 604 follow-on, but it failed due to excessive complexity, project mis-management, disagreement among AIM, and the eventual falling apart of the AIM group. Fortunately the 750 project turned out remarkably good results, and the VMX project was also very successful... allowing the G4 to come into existence. The G4 didn't scale as hoped, but at least it existed! I had high hopes at the time for a 604 follow on as well, but it just wasn't to be. The result was a drought of new designs until the 970 arrived. Hopefully we'll soon start to see the fruits of IBM's Power initiative made available to Apple.
  • Reply 12 of 26
    smalmsmalm Posts: 656member
    Henrik Wannhedens has a fantastic chart
  • Reply 13 of 26
    henriokhenriok Posts: 537member
    Quote:

    Originally posted by smalM

    Henrik Wannhedens has a fantastic chart



    Thanks! Based on THT's first attempt to sort out the PowerPC family tree a couple of year ago I might add. He was the one who got me started!

    This new attempt of his is a new fresh view that I really like. Much cleaner.
  • Reply 14 of 26
    thttht Posts: 3,168member
    Quote:

    Originally posted by Programmer

    Actually it was designed and developed right along side the 603 and after the 601 at Somerset with all three companies participating. The 601 was essentially an IBM design as well, after all.



    The 603 is derived from the 601? Or, the 601, 603, and 604 were independent designs? Who would be the ancestor among the 601, 603, 604 CPUs?



    I only put the 601 first because it shipped first, and henced should have been the ancestor. The 603 and 604 shipped later, and therefore, had the 601 to design from and to learn from. If the 601, 603 and 604 are siblings, I will have to change the cladogram. Any info on the 602?



    My impression as well is that the 604 was mostly an IBM design even though it was a Somerset product. Motorola likes to ignore it as much as they can. They even manufactured it for awhile, but they certainly seemed to have forgotten about it after the 750 shipped. AIM was collapsing already by then, and Moto put in the money for the 7400 and 7450 which made it more 604-like. But it was still 3 to 4 years lost in a time when a 604-derived CPU would have made Apple more competitive.
  • Reply 15 of 26
    thttht Posts: 3,168member
    Quote:

    Originally posted by Henriok

    This new attempt of his is a new fresh view that I really like. Much cleaner.



    I'm still working on that technology tree. It's just really difficult to make clean looking with text.
  • Reply 16 of 26
    programmerprogrammer Posts: 3,409member
    Quote:

    Originally posted by THT

    The 603 is derived from the 601? Or, the 601, 603, and 604 were independent designs? Who would be the ancestor among the 601, 603, 604 CPUs?



    The 601 was the original PowerPC design and it was largely a nearly finished internal IBM project that was donated to Somerset to form their first product.



    The 603 and 604 were done side-by-side at Somerset, and were "from scratch" (as near as such things ever are, anyhow).



    Quote:

    I only put the 601 first because it shipped first, and henced should have been the ancestor. The 603 and 604 shipped later, and therefore, had the 601 to design from and to learn from. If the 601, 603 and 604 are siblings, I will have to change the cladogram. Any info on the 602?



    I used to know something about the 602, but I can't remember at the moment. It never showed up as a real product.



    Quote:

    My impression as well is that the 604 was mostly an IBM design even though it was a Somerset product. Motorola likes to ignore it as much as they can. They even manufactured it for awhile, but they certainly seemed to have forgotten about it after the 750 shipped. AIM was collapsing already by then, and Moto put in the money for the 7400 and 7450 which made it more 604-like. But it was still 3 to 4 years lost in a time when a 604-derived CPU would have made Apple more competitive.



    I think Motorola fabbed (or tried to fab) the 604, but wasn't particularly successful. They were more interested in the 603 since it was a low power design and more appropriate to the embedded market that they have been interested in for a long time. IBM used the 604 in their workstations for a long time. The Somerset design teams were composed of members from all 3 companies, but each company was more interested in different uses of the resulting chips.
  • Reply 17 of 26
    thttht Posts: 3,168member
    Quote:

    Originally posted by Programmer

    I used to know something about the 602, but I can't remember at the moment. It never showed up as a real product.



    Found the info on the IBM website. Seems very similar to the 601, same number of units, but single issue and emulated double precision FPU.
  • Reply 18 of 26
    splinemodelsplinemodel Posts: 7,311member
    Quote:

    Originally posted by THT

    Found the info on the IBM website. Seems very similar to the 601, same number of units, but single issue and emulated double precision FPU.



    I remember that the PPC602 was used in the 3D0 video game console. That may have been the only thing.
  • Reply 19 of 26
    programmerprogrammer Posts: 3,409member
    Quote:

    Originally posted by Splinemodel

    I remember that the PPC602 was used in the 3D0 video game console. That may have been the only thing.



    No it wasn't -- the 3DO used a 12 MHz ARM processor. What a letdown that hardware was...
  • Reply 20 of 26
    splinemodelsplinemodel Posts: 7,311member
    Hmmmm. I remember that SOMETHING in the game console world from about, erm, 1994 used the PPC 602. It may never have seen the light of day, but I'm sure it wasn't the Apple/Bandai thing.



    Edit: http://www.gamezero.com/team-0/whats.../m2-press.html



    INTERESTING. The 602 was 64bit, and was the heart of the M2 add-on module to the 3D0.
Sign In or Register to comment.