[zoetrope]

'Nuff said

[Maflynn]

I don't think your alone in this, a lot people expect to see new equipment, at MWSF and WWDC.

[Hi I'm Ben]

August till their pro machines come out? I highly doubt it. And it will most likely because called a "PowerMac" seeing as how their whole reason for changing "PowerBook" was to get the word "Mac" in it.

Why would apple say their pro software would be universal in a few months from the iMac announcement but release the machines that software runs on so far down the line? That would make no sense.

[zoetrope]

Originally Posted by Hi I'm Ben

August till their pro machines come out? I highly doubt it.

I don't doubt it at all. In fact a lot of other people think the very same thing, that the "Power Mac" or "Mac Pro" or whatever it's called will be announced at this year's WWDC and it will complete the Intel transition. For Jobs, image and marketing is of the utmost importance, and what better way to announce the completion of a major hardware transition than to announce it at the WWDC a little over a year later after the initial announcement at last year's WWDC.

You really think Apple is just going to slap a couple of Intel's current Dual Core chips on a logic board and call it their Power Mac (I still think they will change the name to Mac Pro, but whatever). No! It will be Intel's Conroe chip, its next generation chip, that will be put inside the Pro desktop Mac, and that chip is due to be released at the end of Q3 this year. Why would Apple push back the date of WWDC from June to August?

Intel to 'bring forward' Conroe release date

[mduell]

With Intel already showing off Clovertown, could it be, an 8-way Mac?

Base price would be $6k easy, but it would be "the ultimate Mac."

[zoetrope]

Originally Posted by mduell

With Intel already showing off Clovertown, could it be, an 8-way Mac?

The quad-core Xeon. I thought that was for servers, but they might be very nice for a high end desktop. They would be great for the Xserve, of which Apple has made no mention of its transition.

[Ken_F2]

The next-generation Intel desktop processor, Conroe, is expected for release between July 31 and August 31. Next-generation Powermacs using this processor are a lock for WWDC. Apple moved back WWDC with Conroe in mind -- even Intel executives suggested as much during today's developer conference.

Today at its bi-annual developer conference (IDF), Intel officially unveiled the details of its next-generation mobile, desktop, and server processors shipping in the second-half of 2006. These are the processors we will see in PowerMacs and revised MacbookPros later this year.

The processors included the next-generation version of the Core Duo known as "Merom" -- a processor with identical battery life to the Duo, but with 20% greater performance and support for 64-bit execution. Merom is slated to ship late summer or early fall.

Intel also unveiled two next-generation processors for the desktop to replace the Pentium 4. The first was "Conroe", which will use 40% less power than Intel's current top-of-the-line P4, but offer 40% greater performance. Intel claims "Conroe" will be 20% faster than any chip available from the competition (AMD) in 2006. Conroe is slated to ship early this summer, just in time for the WWDC in August.

The second chip unveiled for the desktop was "Kentsfield", which integrates two Conroes onto a single chip (4-core processor) using less power than some of today's P4 chips. The quad-core "Kentsfield" processor is slated for release in January, 2007. A server version of the chip, dubbed "Clovertown," was also shown in a dual processor configuration (8 cores!) running Guild Wars at ridiculous speeds.

Just how fast will these processors be? The current Intel Core Duo @ 2.167MHz -- the fastest chip available in the new iMac and MacBook Pro -- comes in at 1748 SPECint_base2000 and 1580 SPECfp_base2000. By comparison, the 2.8GHz version of the Conroe is said to come in at 2800 SPECint_base2000 and 2500 SPECfp_base2000 -- a 60+% performance improvement. And this is only the "mid-range" version of the chip.

Vaporware? Not quite. Anandtech ran benchmarks on the new dual-core "Conroe," which you can see here. In Anandtech's tests, the Conroe outperformed the higher-clocked, dual-core Athlon64 by up to 40%, and did it while using substantially less power, according to Intel. Note those benchmarks are only for the mid-range 2.66GHz version. According to Intel, the processor will be available in 2.93Ghz and 3.33GHz versions at release.

Anand's first articles on the conference (still in progress) are here:
General info: http://www.anandtech.com/tradeshows/showdoc.aspx?i=2711
Benchmarks: http://www.anandtech.com/tradeshows/showdoc.aspx?i=2712

Here's another interesting report from the developer conference:

We saw a very quick demo of Battlefield 2 being played on Intel's Conroe chip, today.

We watched a quick demo of two gamers loading up a Battlefield 2 map. Anyone who's played Battlefield 2 knows that it takes an absolute age to go from 'Enter map', through 'Verifying data' all the way to frag-o-matic.

With Robson working, and what was called 'Intel's special sauce' (leading to quips amongst the audience of 'Intel wants to feed gamers its special sauce'), the load times for a Battlefield 2 were cut, roughly in half.

This led to an amusing demo where the chap playing on the Conroe system was in the map, had grabbed a plane and was able to frag his opponent with a hail of missile fire just as the poor chap was spawning.

[mduell]

The "special sauce" is NAND flash being used as a disk cache.

I saw a link on it earlier today showing the performance difference. Opening Office files and running macros in 3s instead of 15s. Booting the OS faster. All sorts of good stuff.

Coming with Santa Rosa in Q1 2007.

[Todd Madson]

Announcing is very different than shipping.

I suspect they'll show off some prototypes - and have a date when
they can be bought.

If the specs on those machines are as good as they say it could
be a very interesting fall 2006.

[Maflynn]

Originally Posted by Todd Madson

Announcing is very different than shipping.

Agreed and apple has a history of announcing then a delay (sometimes significant) in shipping.

[zoetrope]

Originally Posted by Maflynn

Agreed and apple has a history of announcing then a delay (sometimes significant) in shipping.

I agree with your assesment of Apple's history w.r.t. announcements vs. shipping, except since the transitiion has begun that delay has been nullified. The Intel iMacs shipped the day announced, MacBook Pros shipped a week later after estimated date announced (but with a speed increase), and the Mac minis announced and shipped same day. I expect the PowerMac-MacPro announcement to be similiar to the MacBook Pro, Apple will make the announcment at WWDC and ship about a month later. They will have demo units on the floor, and Jobs will demo a working unit on stage at his keynote. I don't expect any delays however. The Intel switch has provided at least this one (if not many other) positive benefits and on time shipments of new hardware seems to be one of them.

[zoetrope]

Originally Posted by mduell

The "special sauce" is NAND flash being used as a disk cache.

You think Apple will be using these hybrid hard drives in their new Intel PowerMacs?

[P]

The "special sauce" is NAND flash being used as a disk cache.

Actually, Robson is the NAND cache thingy. It mentions Robson AND the special sauce. I read it as Intel recompiling the source to get performance out of Conroe.

[golfer099]

Most likely they will call it a Mac Pro because the PowerMac came out with the PowerPC processor. They can't call it a Macintosh II cause that well is an old (and imho ugly) computer.

[P]

Macintosh X?

[power142]

Interesting. 40% less power consumption than 100+W .... wow, doesn't do much for the heat production compared with a G5, but if the SPEC results have any effect on real world performance, at least Steve will be on the right side of the truth as far as performance per watt is concerned.

[olePigeon]

Originally Posted by P

Macintosh X?

Macintosh eXtreme Edition

[SalBaker]

I don't think Apple will announce the new Intel Pro Macs until after the very last iteration of the G5 towers have been out for 5 - 6 months. If the current PowerMacs are the end of the line, with no updates coming at all, August launch of new machines seems possible. Or, the current G5s would need to have their final speed bumps this month otherwise August updates would be too soon IMO.

[Eug Wanker]

Originally Posted by Ken_F2

The next-generation Intel desktop processor, Conroe, is expected for release between July 31 and August 31. Next-generation Powermacs using this processor are a lock for WWDC. Apple moved back WWDC with Conroe in mind -- even Intel executives suggested as much during today's developer conference.

AFAIK, Conroe per se does not support dual dual-core configs.

[mduell]

Originally Posted by Eug Wanker

AFAIK, Conroe per se does not support dual dual-core configs.

It doesn't, but when you put two Conroes on one piece of ceramic and call it Kentsfield it does.
Since Intel is sticking with a shared bus for now, I think Apple will go with single-socket systems (higher FSB speed than having multiple sockets).

[yaro]

To most of us these are not pro machines. They are towers with pci slots for extra goodies, ala G4 quicksilver, so why can"t they use the current core duos which have been accorded rave reviews and are on par with the G5s?

[Ken_F2]

To most of us these are not pro machines. They are towers with pci slots for extra goodies, ala G4 quicksilver, so why can"t they use the current core duos which have been accorded rave reviews and are on par with the G5s?

The existing Core Duos are notebook cpus. They are intended for low-power, low-heat mobile applications, not desktops. They aren't supported in multiprocessor configurations.

Core Duo is a 10-20 watt processor. The PowerPC 970MP 2.5GHz is a 100 watt processor.

A PowerMac system based on the existing Core Duo @ 2.16GHz would not outperform Apple's current dual-processor, dual-core PowerMac systems. In contrast, systems based on the Conroe will outperform these G5 systems by a wide margin.

According to this page:

2.5GHz 970MP reaching a score of 1,438 in the SPECint test, with a 2,076 score in the SPECfp test. Addressing the "low-power" side of his title, Rorher added that the 970MP will max out at 100W when both 2.5GHz cores are running.

In contrast, the Conroe 2.8GHz is said to come in at 2800 SPECint_base2000 and 2500 SPECfp_base2000, while requiring 35% less power (max 65 watts). Thus, a mid-range Conroe system will be almost twice as fast as the fastest dual-core G5 in the integer ops used by most applications. The Conroe will also be available in higher-performance 3.33GHz version this year.

The quad-core "Kentsfield" processor coming in January 2007 combines two Conroes into a single chip -- for up to twice that performance, or four to five times the performance of today's 2.5GHz dual-core PowerPC 970MP.

[krove]

Makes one wonder what dire straights the Mac community and Apple would have been in if we had stuck with PowerPC processors and not transitioned to Intel.

<-- hugs his MacBook Pro

[Catfish_Man]

Originally Posted by Ken_F2

The existing Core Duos are notebook cpus. They are intended for low-power, low-heat mobile applications, not desktops. They aren't supported in multiprocessor configurations.

Core Duo is a 10-20 watt processor. The PowerPC 970MP 2.5GHz is a 100 watt processor.

A PowerMac system based on the existing Core Duo @ 2.16GHz would not outperform Apple's current dual-processor, dual-core PowerMac systems. In contrast, systems based on the Conroe will outperform these G5 systems by a wide margin.

According to this page:
In contrast, the Conroe 2.8GHz is said to come in at 2800 SPECint_base2000 and 2500 SPECfp_base2000, while requiring 35% less power (max 65 watts). Thus, a mid-range Conroe system will be almost twice as fast as the fastest dual-core G5 in the integer ops used by most applications. The Conroe will also be available in higher-performance 3.33GHz version this year.

The quad-core "Kentsfield" processor coming in January 2007 combines two Conroes into a single chip -- for up to twice that performance, or four to five times the performance of today's 2.5GHz dual-core PowerPC 970MP.

Although Conroe does look exceptionally cool, one thing to remember is that some of the main Mac "pro" apps use altivec pretty heavily. Conroe improves SSE a lot, but it's still not up to par with Altivec. Basically, expect "faster" but not as much faster as non-vector benchmarks like SPEC would indicate. The improved graphics cards that will inevitably be included will also skew the comparison, although in the other direction.

For compiling, though, it will be an absolute beast

[Ken_F2]

Although Conroe does look exceptionally cool, one thing to remember is that some of the main Mac "pro" apps use altivec pretty heavily. Conroe improves SSE a lot, but it's still not up to par with Altivec. Basically, expect "faster" but not as much faster as non-vector benchmarks like SPEC would indicate. The improved graphics cards that will inevitably be included will also skew the comparison, although in the other direction.

The Altivec on the dual-core 970MP @ 2.5GHz helps it to achieve that score of 2079 on SPECfp_2000. Compare that to a score of 1978 for a dual-core Opteron 185 @ 2.6Ghz and 1836 for a dual-core Xeon @ 3.2GHz. The 970MP is a SPECfp monster thanks to Altivec.

Now compare those scores to the 2500 attained by the Conroe @ 2.8GHz. As impressive as the 970MP's score is with those Altivec-optimized tasks, it can't compare to those same tasks on the Conroe. Intel achieved this performance by expanding SIMD FP capability to 4 DP flops/s cycle, which is twice that of the K7/K8, Athlon, P4, and 970MP.

Conroe's SSE/SSE2 implementation is so fast that it exceeds the performance of Altivec on the 970MP. I'm sure there will still be some special cases where an Altivec optimized app will exceed the performance of a SSE optimized app running on a Conroe, but I expect those cases to be few and far between.

[rhashem]

Originally Posted by Ken_F2

The Altivec on the dual-core 970MP @ 2.5GHz helps it to achieve that score of 2079 on SPECfp_2000. Compare that to a score of 1978 for a dual-core Opteron 185 @ 2.6Ghz and 1836 for a dual-core Xeon @ 3.2GHz. The 970MP is a SPECfp monster thanks to Altivec.

The 970MP does well in SPECfp because it has a good amount of memory bandwidth, dual load/store units, and dual symmetric FPUs. Altivec is almost a complete non-factor, since the SPEC benchmark source cannot be modified to include custom AltiVec code, and autovectorization is of limited use in most code. Beyond that of course is the fact that SPECfp measures mostly double-precision performance, and Altivec cannot handle that.

Regarding Altivec versus SSE, there are a couple of points:

1) Altivec has traditionally had a large performance advantage over SSE. That has been not so much due to the superiority of Altivec as an instruction set, but due to the superiority of the Altivec implementations in the G4 and G5. Conroe is the first x86 chip that will actually perform 128-bit SSE ops per cycle. Both the P4 and the Opteron had 64-bit SSE units that did a 128-bit operation in 2 cycles. This will close a large portion of the performance gap.

2) Vector processing is relatively unimportant in the PC world. That's why Intel and AMD have been able to get away with such crappy SSE implementations for so long --- the market really never clamored for anything better. The model the PC world has moved to is pushing single-precision operations to the GPU, and letting the CPU handle double-precision ops (where the advantage of a 128-bit vector unit over dual 64-bit FPUs is nil). This works well for a lot of the traditional uses of Altivec (eg: image processing via CoreImage), because the GPU has far more memory bandwidth and single-precision vector throughput than any CPU could hope for. This doesn't work so well for some other problem domains (eg: signal processing, which is what Altivec was designed for), so chips like the new dual-core G4 derivative will probably still have a foothold in such markets.

[Catfish_Man]

rhashem is, as usual, correct, although I'd point out the existence of the select and permute instructions in altivec rather than just attributing its speed to superior implementations.

I have to say though that I wrote my previous comment on SSE before learning more about the details of Conroe's SSE implementation. Altivec is still probably 'better' (more flexible certainly), but I think the situation is much less clear cut for a lot of apps than it used to be.

Originally Posted by Paul DeMone on realworldtech.com

128 bit packed mul + 128 bit packed add + 128 bit
packed load + 128 bit packed store + cmp/jcc per cycle.

That goes a looong way towards evening the odds, and is definitely an about-face from previous SSE implementations. This should be an interesting 6-12 months coming up

[turbopants]

I was thinking this morning that the Mac Pro's would be announced at the WWDC. That and 10.5.