[Eug Wanker]

PowerPoint presentation with IBM sekrit roadmap

970GX planned for 2005: Single core with 1 MB L2
970MX proposed for 2006: Dual core with 2 MB shared L2, 3+ GHz. (The 970MP has 2 x 1 MB dedicated L2, and maxes out at less than 3 GHz.)

[Todd Madson]

Veryyy interesting. I wonder who leaked that information and what it could lead to?

[Link]

OMG! Geez, these things wipe the floor with those crappy Pentium Ds and "yonah" crap that Intel is coming out with!

[zed57]

Originally Posted by Link

OMG! Geez, these things wipe the floor with those crappy Pentium Ds and "yonah" crap that Intel is coming out with!

what makes you say that?

[Catfish_Man]

Originally Posted by Link

OMG! Geez, these things wipe the floor with those crappy Pentium Ds and "yonah" crap that Intel is coming out with!

If you actually read the leaked presentation you'll see that IBM is not *nearly* as confident as you are.

[blackwind]

Originally Posted by Catfish_Man

If you actually read the leaked presentation you'll see that IBM is not *nearly* as confident as you are.

The graph above shows that IBM knows that the 970FX uses more power for each "performance point" than the Pentium M, but if you believe the x-axis, then a 2-GHz PowerPC 970FX beats a 2-GHz Pentium M in pure performance (by a slight margin).

Of course, this means that the Pentium M (and its successor, Yonah) will be better laptop processors than the 970 lineage.

For the Power Macs, though, this means that the 970 family had life left in it, at least from what it looked in the fall of 2004.

In late 2006 or early 2007, a dual-core 3-GHz PowerPC 970MX with 2 MB of shared L2 cache would likely still be competitive with Intel and AMD offerings if it ever comes into being in an Apple Power Mac.

[Eug Wanker]

Originally Posted by Todd Madson

Veryyy interesting. I wonder who leaked that information and what it could lead to?

IBM leaked it. The link is from IBM itself.

Originally Posted by Catfish_Man

If you actually read the leaked presentation you'll see that IBM is not *nearly* as confident as you are.

I agree, although I will point out that those power specs don't take the newer low power 970FX into consideration. That said, Dothan is still the performance/Watt winner in anything that would work in a laptop.

Originally Posted by blackwind

The graph above shows that IBM knows that the 970FX uses more power for each "performance point" than the Pentium M, but if you believe the x-axis, then a 2-GHz PowerPC 970FX beats a 2-GHz Pentium M in pure performance (by a slight margin).

That's with the DMIPS bench.

Note that the G5 is very good with FP compared with Dothan, but not integer.

[Link]

Yeah, but "Dotham" sounds geeky"

[blackwind]

I assumed that the DMIPS benchmark refers to Dhrystone, which was indeed a test in integer performance. As usual, a synthetic benchmark like Dhrystone is not truly indicative of real-world performance, as there is almost no way that the PowerPC 970FX can match a Pentium M at the same clock speed in integer performance.

Indeed, though, the PowerPC 970FX does have better floating-point performance than a Pentium M.

Additionally, upon thinking about it further, I somehow doubt that a PowerPC 970MX will see the light of day anymore. For Apple's needs, a PowerPC 970MP will already suffice for the interlude before the Power Mac finally switches over to Intel. By the time a PowerPC 970MX would hypothetically be in Power Macs (MWSF 2007), it would not really be worth the effort to develop it.

[Eug Wanker]

Originally Posted by blackwind

I assumed that the DMIPS benchmark refers to Dhrystone, which was indeed a test in integer performance. As usual, a synthetic benchmark like Dhrystone is not truly indicative of real-world performance, as there is almost no way that the PowerPC 970FX can match a Pentium M at the same clock speed in integer performance.

Indeed, though, the PowerPC 970FX does have better floating-point performance than a Pentium M.

Additionally, upon thinking about it further, I somehow doubt that a PowerPC 970MX will see the light of day anymore. For Apple's needs, a PowerPC 970MP will already suffice for the interlude before the Power Mac finally switches over to Intel. By the time a PowerPC 970MX would hypothetically be in Power Macs (MWSF 2007), it would not really be worth the effort to develop it.

Sorry, I didn't mean to say that DMIPS was FP. I was just saying that the G5's forte is FP, not integer. But yeah, in other non-Dhrystone integer tests, I do get the impression that Dothan does noticeably better than the G5.

I don't think we'll see the 970MX in a Mac either. Too late... 970MP most definitely though.

[osxisfun]

Originally Posted by Link

OMG! Geez, these things wipe the floor with those crappy Pentium Ds and "yonah" crap that Intel is coming out with!

Not.

[Catfish_Man]

Originally Posted by blackwind

The graph above shows that IBM knows that the 970FX uses more power for each "performance point" than the Pentium M, but if you believe the x-axis, then a 2-GHz PowerPC 970FX beats a 2-GHz Pentium M in pure performance (by a slight margin).

Of course, this means that the Pentium M (and its successor, Yonah) will be better laptop processors than the 970 lineage.

For the Power Macs, though, this means that the 970 family had life left in it, at least from what it looked in the fall of 2004.

In late 2006 or early 2007, a dual-core 3-GHz PowerPC 970MX with 2 MB of shared L2 cache would likely still be competitive with Intel and AMD offerings if it ever comes into being in an Apple Power Mac.

I'd have been shocked if the 970fx couldn't beat Dothan at pure performance. What would have been more telling is if Intel made a desktop variant of Dothan (bump the power target from 30ish watts to, say, 80, and tweak the chip based on that), and the 970fx beat that. It might, at FP. The 970fx certainly isn't slow

About the dual core 3GHz 970MX... I'm sure it would be competitive, but where exactly are these power savings coming from that make it practical/possible?

[Link]

Originally Posted by Catfish_Man

About the dual core 3GHz 970MX... I'm sure it would be competitive, but where exactly are these power savings coming from that make it practical/possible?

OH COME ON! This is like someone in a glass house shooting stones!

The P4 is a power hungry hot beast! You guys are sitting on a high horse and calling the g5 power hungry? There are P4s that suck a whole &#*($#@$ing 130 watts!

[mousehouse]

ok, i must admit that i didn't read the whole presentation (trying to get some work done while browsing )

what would be a good estimate for general availability of the dual-core 970MP's ? and given that, what is our experience with apple integrating chips into their designs if no big architectural changes are nessecary within the machine? also, are all current powermac's 970FX's ? even though they are only listed up to 2.5GHz? or is that because the presentation is from last year and they upped the speed? then there could also be an improvement if apple switched to the GX series (power wise).

[Eug Wanker]

Originally Posted by Catfish_Man

I'd have been shocked if the 970fx couldn't beat Dothan at pure performance. What would have been more telling is if Intel made a desktop variant of Dothan (bump the power target from 30ish watts to, say, 80, and tweak the chip based on that), and the 970fx beat that. It might, at FP. The 970fx certainly isn't slow

Well, as you suggest Dothan does very well at integer, even at current speeds.

About the dual core 3GHz 970MX... I'm sure it would be competitive, but where exactly are these power savings coming from that make it practical/possible?

65 nm most likely.

Originally Posted by mousehouse

what would be a good estimate for general availability of the dual-core 970MP's ?

I don't know, but I'm guessing a 970MP Mac make will appear within 6 months.

Originally Posted by Link

The P4 is a power hungry hot beast! You guys are sitting on a high horse and calling the g5 power hungry? There are P4s that suck a whole &#*($#@$ing 130 watts!

What's with the continued anti-Intel trolling? As has been pointed out before, most of us don't think a P4 derivative will be used in a retail Mac. Maybe you should read the graph where even IBM states Dothan has superior performance/Watt to the 970FX. The first retail Intel Macs will likely use derivatives of Dothan.

[Todd Madson]

The IV is a dead end, they're already banking their futures on derivatives of the III
(which, arguably, with some modifications could have really been the IV they were
hoping for).

I still would like to see a CPU upgrade someday for my G5 tower to get that clockspeed
thru the roof but I don't expect to see it. Face it, the G5 tower would be a killer case for
a future computer of some kind once the G5 I have becomes moribund (4-5 years).

[ReggieX]

Real artists ship. I'll believe all this crap when I can see it on a shelf. An internal presentation from 2004 isn't indicative of much.

[mhuie]

New rumors off a presentation from August 2004???

[Eug Wanker]

Originally Posted by mhuie

New rumors off a presentation from August 2004???

Yes, it's called a roadmap. It confirms the long-rumoured 970GX, and does give a glimpse of IBM ideas for the future (970MX).