If Macs Get Real POWER, Who Needs a Hammer?

A while ago I wrote a short article on a rumour about Apple potentially using a POWER4 derived CPU. Since then there has been a lot of talk of Apple switching to an x86 based solution and now a desktop POWER4 derivative has been announced. Some people suggest this will be too expensive and in it’s cut down form slower than a x86. What is Apple more likely to go for?

Editorial notice: All opinions are those of the author and not necessarily those of osnews.com


The speculation about Apple going x86 is based mainly around a comment by Apple CEO Steve Jobs in Apple’s Q3 Financial Analyst Meeting Q&A here.

Then we’ll have options and we like to have options

What seems to be missing is the rest of his comment and this puts things into much clearer perspective:

The roadmap on the PowerPC actually looks pretty good and there are some advantages to it. As an example, the PowerPC has something in it called AltiVec, we call the Velocity Engine — it’s a vector engine —
it dramatically accelerates media, much better than, as an example, the Intel processors or the AMD processors… so we actually eke out a fair amount of performance from these things when all is said and done. And the roadmap looks pretty good. Now, as you point out, once our transition to Mac OS 10 is complete, which I expect will be around the end of this year or sometime early next year and we get the top 20% of
our installed base running 10, and I think the next 20 will come very rapidly after that. Then we’ll have options, then we’ll have options and we like to have options. But right now, between Motorola and IBM,
the roadmap looks pretty decent.

Thats the full quote and to me it doesn’t sound very much like there’s going to be an Apple with x86 flavouring anytime soon.

We know Motorola are working on a G5 PPC processor and the most recent rumours peg it at 2.5GHz. Motorola’s roadmap states it as being 64 bit and having a 13 stage pipeline. The jump from 1GHz to 2.5GHz
may seem unduly vast however the longer pipeline and 0.13um manufacturing technology will allow such clock speeds so 2.5GHz does not sound unreasonable. We can expect to see similar clock rates from AMDs hammer when it arrives using as it has a similar length pipeline and the silicon process is the same (Motorola and AMD are process partners) – The current rumour is Hammer will start shipping around 2.6GHz (or in sales speak “3400+”). Apart from that Motorolas roadmap doesn’t give much away although we can expect higher bandwidth busses and backward compatibility.

However IBM has not been inactive and have not only announced a “cut down” version of the POWER4 but will be giving out details at Octobers Microprocessor
Forum
, an industry leading event where new microprocessors and technologies are often announced.

Notably IBM are pegging the new CPU as a “desktop” processor and I don’t know of many other companies in the desktop PowerPC space other than Apple. I think however we can safely assume that IBM has not produced this CPU for the soon to be resurrected Amiga platform (but thats not to say they couldn’t utilize it).

Another notable feature of the new processor is the addition of vector instructions. I pointed in the previous article out one of the problems of the existing POWER4 CPU was the lack of support for Altivec
instructions. It would appear that this problem has already been taken care of although we don’t at this point know it they are the same instructions as used in Altivec but I for one would be more than surprised if they were not.

It will be interesting to see if dual precision floating point instructions are featured in the vector unit, these are featured in Intel’s SSE2 but not in Altivec, Altivec added 162 instructions whereas IBM have mentioned 160 so this may not be the case. That said the POWER4 FPU is a mighty beast already capable of issuing 4 floating point operations per cycle so it’s questionable if dual precision vector instructions would actually be necessary.

Performance and price.
Some people get the idea POWER4 is a horribly expensive CPU and oddly enough they’d be right. However they also assume that it will remain expensive and even if it were to be “cut down” it’s performance will
plummet to levels below that of the x86 processors.

I disagree on both counts:

Firstly POWER4 in it’s current form is horribly expensive yes, but thats probably because it’s shipped as an 8 way SMP system. Yes thats right, the smallest POWER4 has 8 CPUs. Cutting that down in price is going to be easy, dump 6 of those CPUs and the large Multi chip packaging and you’ve just cut a great chunk off the price.

It also has 128MB cache, 32MB per CPU pair so you can dump 96MB of this as well. This will impact performance but probably by not a significant margin. Like everything the law of diminishing returns applies so going from 32 to 128MB cache isn’t likely to have a massive impact although this very much depends on the application in use.

There will be a big change in the memory interface however. The POWER4 having 8 CPUs needs a large fast memory system to keep those processors fed with data and this is implemented in a special memory controller attached to the L3 Cache chips. This controller has multiple 400MHz ports going out to main memory.

This isn’t going to be necessary for a single processor and removing it will also cut costs further. IBM however have not skimped on the memory system however as the new PowerPC has a bandwidth 6.4GB per second. 50% higher then the current top Pentium 4, and 15% higher than the forthcoming AMD Opteron. Apples professional target markets (audio/visual) need bandwidth by the bucket load and this is sure a great improvement over the current 1GB per Second (although this figure looks like it might go upwards rapidly in a couple of weeks).

Performance.
Cutting the Cache and memory bandwidth will have an effect on performance but as above the law of diminishing returns applies, I don’t think the performance drop will be very great and certainly nothing a die shrink to 0.13um wont fix. Also current SPEC marks for the POWER4 are for a single processor, the single chip version may keep the dual cores of the full POWER4. A single core may be behind the Itanium 2 in SPEC marks but once the second core kicks in I can see it showing Itanium 2 a clean pair of heels.

A G5 will bring performance levels back up to those delivered by the x86 CPUs, a POWER4 CPU will go straight past. Yes they will still not be cheap but who says Apple hardware is cheap? I can see a POWER4
based CPU going into the PowerMacs at least at the top end (or possibly a new, more expensive Super-PowerMac). You wont however see them appearing in your iMac anytime soon and they’ll have problems fitting them into a laptop unless they cut the clock speed whilst on batteries.

As to the question of which CPU Apple will use next, read the CEOs comments:

But right now, between Motorola and IBM, the roadmap looks pretty
decent
.”

Note he mentions Motorola and IBM, he never mentioned AMD or Intel producing CPUs. Some people have mentioned x86 as a future CPU for Macs, Others have suggested Itanuim. Switching the CPU to POWER4 –
even just at the top end – will not only be a great deal easier but compared to x86 IBMs new POWER4 derivative will outgun them, How? The current SPEC marks for POWER4 are for a 0.18um chip, whereas Intels are for 0.13um expect them to leap upwards when POWER4 transitions to 0.13um.


About the Author:
Nicholas Blachford is a Software Engineer / Architect currently living in Amsterdam. He has numerous geeky interests including CPUs. He does not own a Mac but wants one.

75 Comments

  1. 2002-08-09 5:00 am
  2. 2002-08-09 5:23 am
  3. 2002-08-09 5:36 am
  4. 2002-08-09 5:48 am
  5. 2002-08-09 6:08 am
  6. 2002-08-09 6:12 am
  7. 2002-08-09 6:24 am
  8. 2002-08-09 6:42 am
  9. 2002-08-09 7:34 am
  10. 2002-08-09 8:24 am
  11. 2002-08-09 8:39 am
  12. 2002-08-09 9:30 am
  13. 2002-08-09 9:48 am
  14. 2002-08-09 11:27 am
  15. 2002-08-09 11:38 am
  16. 2002-08-09 11:56 am
  17. 2002-08-09 12:23 pm
  18. 2002-08-09 1:03 pm
  19. 2002-08-09 1:21 pm
  20. 2002-08-09 1:30 pm
  21. 2002-08-09 2:23 pm
  22. 2002-08-09 2:31 pm
  23. 2002-08-09 2:35 pm
  24. 2002-08-09 2:40 pm
  25. 2002-08-09 2:54 pm
  26. 2002-08-09 3:09 pm
  27. 2002-08-09 3:22 pm
  28. 2002-08-09 3:34 pm
  29. 2002-08-09 4:16 pm
  30. 2002-08-09 4:16 pm
  31. 2002-08-09 4:27 pm
  32. 2002-08-09 4:38 pm
  33. 2002-08-09 4:38 pm
  34. 2002-08-09 4:39 pm
  35. 2002-08-09 5:01 pm
  36. 2002-08-09 5:19 pm
  37. 2002-08-09 5:40 pm
  38. 2002-08-09 6:01 pm
  39. 2002-08-09 6:09 pm
  40. 2002-08-09 6:16 pm
  41. 2002-08-09 6:40 pm
  42. 2002-08-09 6:44 pm
  43. 2002-08-09 6:45 pm
  44. 2002-08-09 7:00 pm
  45. 2002-08-09 7:00 pm
  46. 2002-08-09 7:05 pm
  47. 2002-08-09 7:53 pm
  48. 2002-08-09 8:20 pm
  49. 2002-08-09 8:53 pm
  50. 2002-08-09 9:16 pm
  51. 2002-08-09 9:23 pm
  52. 2002-08-09 10:52 pm
  53. 2002-08-10 12:26 am
  54. 2002-08-10 12:36 am
  55. 2002-08-10 12:50 am
  56. 2002-08-10 2:58 am
  57. 2002-08-10 3:12 am
  58. 2002-08-10 3:14 am
  59. 2002-08-10 3:23 am
  60. 2002-08-10 6:06 am
  61. 2002-08-10 8:44 am
  62. 2002-08-10 4:44 pm
  63. 2002-08-10 7:10 pm
  64. 2002-08-10 7:25 pm
  65. 2002-08-10 10:02 pm
  66. 2002-08-10 11:33 pm
  67. 2002-08-10 11:54 pm
  68. 2002-08-11 1:37 am
  69. 2002-08-11 3:58 am
  70. 2002-08-11 11:07 pm
  71. 2002-08-12 2:07 am
  72. 2002-08-12 6:58 am
  73. 2002-08-12 9:08 am
  74. 2002-08-12 11:44 pm