[rhashem]

Originally Posted by iMouse

The PowerPC platform is a VERY capable one with more expansion options than you can shake a stick at. P.A. Semi's project is a good start if you want information on what the PowerPC Mac could have been.

Vaporware? Yep, that's exactly what Apple needs.

Moreover, PA Semi wouldn't be an impressive chip even if it were out now. At 2.0 GHz, it's projected integer performance is about 1/3 the performance of Core 2, while its FP performance is about 2/3s.

Of course, PA Semi knows that just fine. The PA Semi chip is intended for networking gear and blade servers that need a lot of FP power with minimal cooling requirements. In a workstation, where you have a much larger power envelope to work with, the Core 2 is going to offer a lot more real performance.

[rhashem]

Exactly how bad does IBM have to suck at making CPUs before you'll embrace the cold hard reality?

IBM is actually very good at making CPUs. Their fabrication tech is good, and their design is relatively good as well. It's just that IBM doesn't have a lot of incentive to make chips for Apple. Power5+ is excellent at the high-end, but that's because IBM can afford to stick it into a system with tons of reliability/RAS features, tens of gigabytes per second of memory bandwidth, a few dozen megabytes of L3 cache, and multiple 10+GB/sec I/O links everywhere all nestled into an extremely expensive multi-chip module. There is no way Apple could afford to ship a real POWER chip in a PowerMac, and very little profit-motive for IBM to make a custom chip designed for the ground-up for desktop/workstation use, like the Core 2 is.

[mduell]

Originally Posted by rhashem

IBM is actually very good at making CPUs. Their fabrication tech is good

That's not the impression I get. The early (and possibly continuing) 970 yields were poor and the current Cell yields are quite poor (10-15% fully functioning with 8 SPE, another 10-15% with 7 functioning SPEs).

[rhashem]

Originally Posted by mduell

That's not the impression I get. The early (and possibly continuing) 970 yields were poor and the current Cell yields are quite poor (10-15% fully functioning with 8 SPE, another 10-15% with 7 functioning SPEs).

Early 970 yields were likely poor because the 970 never got as much attention as its POWER bretheren, being a fairly low-profit chip as far as IBM was concerned. 970 yields on 90nm were reportedly quite good, and problems on specific processes are commonplace in the industry anyway. The IBM/AMD 90nm process was excellent, like Intel's 65nm process, while Intel's 90nm process started out with a lot of problems. It happens.

As for Cell yields, figures for that are not widely available. The IBM engineer, when referring to the 10-20% figure was talking in general terms, not about Cell specifically. The fact that Cell has low yields at 90nm is really not surprising --- its a 235 mm^2 chip. It's only 4% smaller than Power5+, and is a full 60% larger than Woodcrest. Cell at 90nm is more about getting to market in the fall 2006 timeframe than about manufacturability.

On the other hand, POWER5 and POWER5+ have been very good chips. They're pretty much the fastest chips in their class, beating out Itanium most of the time, and initial indications suggest that POWER6 is shaping up very well, and will meet its 4-5GHz target. The biggest problem for PPC in the context of Apple is that it makes very little sense for IBM to spend the type of design effort they did on POWER6 on a desktop/workstation chip.

Apple folks like to slag on IBM's chip designers, but the simple fact is that Apple just wasn't an important enough market for IBM to put in the effort. Apple ships what, 5 million PCs a year? Of those, maybe 2 million have a G5-class chip? IBM would have to make profits of hundreds of dollars per chip to actually make it worthwhile for them to actually build a design to compete head-on with Intel in the desktop market. Apple couldn't afford that.

[mduell]

Originally Posted by rhashem

Early 970 yields were likely poor because the 970 never got as much attention as its POWER bretheren, being a fairly low-profit chip as far as IBM was concerned. 970 yields on 90nm were reportedly quite good, and problems on specific processes are commonplace in the industry anyway. The IBM/AMD 90nm process was excellent, like Intel's 65nm process, while Intel's 90nm process started out with a lot of problems. It happens.

March 2004: Apple will now not ship dual-CPU Xserve G5 servers until April, two months later than the company originally promised, it admitted this week.

IBM was due to launch the 90nm version of the PowerPC 970 - the chip Apple calls the G5 - during that month, so presumably server shipments were set to follow the launch.

April 2004: One of the problems is ongoing yield issues within the company's 300-mm fab in East Fishkill, N.Y. For months, IBM has been struggling with yields in the fab, a 130- and 90-nm plant.

June 2004: IBM admitted earlier this year that 970FX yields were not as high as it had anticipated.

July 2004: ... but the [iMac/Xserve] supply issue suggests it's the 90nm PowerPC 970FX, which has been causing IBM such yield problems of late.

August 2004: ... while IBM continues to struggle with the 90nm yield issues that are causing Apple so much grief.

August 2004: IBM Corp.'s chip business was profitable in the second quarter, but its major customer is still plagued by delays in obtaining a sufficient quality of the company's new 90-nanometer processors, signaling that IBM isn't all the way out of the woods just yet.

Originally Posted by rhashem

As for Cell yields, figures for that are not widely available. The IBM engineer, when referring to the 10-20% figure was talking in general terms, not about Cell specifically. The fact that Cell has low yields at 90nm is really not surprising --- its a 235 mm^2 chip. It's only 4% smaller than Power5+, and is a full 60% larger than Woodcrest. Cell at 90nm is more about getting to market in the fall 2006 timeframe than about manufacturability.

Here's the interview. It's not with some random engineer at IBM, it's with the VP of semiconductor and technology services.
"With a chip like the Cell processor, you’re lucky to get 10 or 20 percent."

Are you saying POWER5/5+ is also seeing 10-20% yields?
235 puts it at about the same size as the original P4 (217).

Originally Posted by rhashem

On the other hand, POWER5 and POWER5+ have been very good chips. They're pretty much the fastest chips in their class, beating out Itanium most of the time, and initial indications suggest that POWER6 is shaping up very well, and will meet its 4-5GHz target. The biggest problem for PPC in the context of Apple is that it makes very little sense for IBM to spend the type of design effort they did on POWER6 on a desktop/workstation chip.

Hardly. POWER5 wins in SPECint and SPECfp, but Itanium wins in SPECint_rate and SPECfp_rate. POWER5 is tops in TPC-C, but POWER5 and Itanium each have one win in TPC-H. Itanium has 7.4% of the Top500 list, while POWER5 has just 4.4%.

POWER6 is an interesting departure from the path IBM had been following. Instead of a big complex core and accepting a low clockrate, they're making it very simple, building a bunch on a die, and clocking the piss out of them. We'll see how it turns out.

[Catfish_Man]

Originally Posted by mduell

Hardly. POWER5 wins in SPECint and SPECfp, but Itanium wins in SPECint_rate and SPECfp_rate.

Does it? I can't find any 64 core Montecito results, but the 128 core Montecito score doesn't indicate that it would be able to beat POWER5's score with a similar core count.

Power5, 32 socket, 64 core: 1752 SPECfp_rate
Itanium 2 (Montecito), 64 socket, 128 core: 2837 SPECfp_rate

That's even given the temporary advantage the I2 has in comparing the brand-new Montecito to the somewhat older POWER5 (I couldn't find any large-system POWER5+ results).

[Cross]

I would have pref amd over intel though..... I also would have liked to get a G5 in a Powerbook. I don't see how it would be a bad thing to let us choose the processor we want.

Both PPC and Intel code is present why not let us choose what we want. It would be something apple has never done before *shrug*

Everyone would be happy. I never got to try the G5 on anything I do only the G4. I have tried the Intel chips and they are nice but I always have pref AMD's offerings. *shrug* again another personal choice. They seem to push things further. Intel just really got back on the game with core 2. I personally though don't think everything is going to be able to use the full dual core setup for awhile. So again why not be able to choose a Core 2 or G5? (etc)

[joe]

Originally Posted by Cross

I never got to try the G5 on anything I do only the G4.

In my experience the G5s are noticeably faster than G4s, but also noisier too. It's not just that my G5 fans got louder than my previous G4 tho, my G5 seemed to constantly windup/down even when doing light web surfing (these forums for example). So if I hated the noise so much why didn't I dump my dual 2.5Ghz G5?

I came across this AWESOME utility called "min-o-max." Essentially, it puts the power savings setting from the Preference pane right in the menu bar. I keep it set to "Reduced" 99% of the time which keeps the fans from spinning up. And it's still fast enough for most anything - even playback of 1080i HD files. But if I'm going to run something cpu intensive I click "Highest." It's not a perfect solution since the fans still windup when running in "Highest" and "Automatic" modes. But at least I can keep the fan noise to a minimum most of the time.

Originally Posted by Cross

I have tried the Intel chips and they are nice but I always have pref AMD's offerings. *shrug* again another personal choice. They seem to push things further. Intel just really got back on the game with core 2.

Until recently I'd have preferred AMD too. But the Core 2 didn't just put Intel back in the game, it put them squarely in front and by a wide margin. Perhaps that's not good enough just yet. It remains to be seen whether Intel will keep up the pace over time.

[Todd Madson]

I liked the G5's performance, but I didn't like its electrical consumption (power angel
readings in non-nap mode were well in excess (>200 watts) of any other computer in
the house), as well as the truly astonishing amount of heat these machines produce.

I suspect I could actually SAVE MONEY in long term electrical use by selling my G5
at a loss and using the proceeds to get one of the new boxes.

Although I'd miss my toasty room on cold winter days....