Slashdot Mirror
90nm 3GHz PPC 970FX by Summer
dmdimon sent in linkage to a Forbes story on the upcoming PPC chips and notes "IBM is said to be ready to deliver a new version of its PowerPC processor to Apple by the end of this year in from sizes of 130 nanometers to 90 nanometers...
Apple CEO Steve Jobs has already gone on the record saying that the G5 computer will contain PowerPC chips that run at 3 GHz by the summer of 2004. A mid-step between the current systems, which top out with two chips running at 2 GHz, and systems with chips as fast as 2.6 GHz would be a logical move come January..."
Does anyone even care about the leetness of their speed with Apple stuff?
Why my friend converted: Final Cut Pro, he's in the movie/TV biz.
You can't take the sky from me...
Most of the people editing the audio and video you're encoding on your x86 are using macs. Well, maybe not most, but a hell of a lot of 'em.
"If you're thinking what I'm thinking, you're right." -
This guy trolls for these types of posts, hoping to get modded up. His sig is the most disgusting thing ever, please mod down.
IBM's PPC compiler is XLC.
Who do you get to be an expert to tell you something's not obvious? The least insightful person you can find? -J Roberts
Since no one gets my username, I must educate the world] Dave Cutler, NT 1.0's chief architect (and a rabid unix hater
btw, there was no NT 1.0. The first version of NT was NT 3.1, magically version-synced with Windows 3.1.
cpeterso
From the article:
Where 32-bit chips are limited to addressing only 2 gigabytes of memory, 64-bit chips can theoretically address thousands of gigabytes of memory, though Apple's G5 boxes are limited to 8 gigabytes. Secondly 64-bit chips can perform complex calculations in fewer steps than 32-bit chips.
So far Apple's machines can see all the memory, they can't yet do 64-bit calculations. Present it with a 64-bit calculation, and a Mac with a G5 chip still breaks it into two 32-bit pieces.
First of all 32-bit chips can access 4 GB of memory, not 2 GB. And second, he's got it backwards. Apple's machines CAN do 64-bit calculations, but they can't do 64-bit addressing.
As others have mentioned, there's no great benefit in changing from 32-bit pointers to 64-bit. The optimizations for the G5 that you see in today's apps are from handling complex 64-bit calculations. In this regard Mac OS X is fully up to speed.
infested with jello like fishes no melotron wishes
... that they can fit more chips on a wafer. Which means that the price per chip is reduced. That's the REAL reason for die shrinks and moving to processes with smaller feature sizes.
Not that cheaper PPC970s are a bad thing, mind you...
The two things you quote are very mundane and ordinary ways to get more performance from a CPU. Barring redesign, miniaturization and voltage drops are the ways to make hardware faster, and compiler optimizations are the way to make software faster. These are the bread and butter of performance improvement, and you give Apple/IBM entirely too much credit for doing these things. (And this is coming from someone who works on an IBM compiler.)
Having said that, the PPC compiler team's work has been amazing, and congratulations are due for the sheer magnitude of the performance boost. In a field where a 2% improvement is an achievement, 50% is incredible.
Patrick Doyle
I mod down every jackass who puts his moderation policy in his sig. Oh, wait a sec....
You don't need a "fully 64 bit OS" to get the speed improvements (although there still are memory limitations. see below). It would break a LOT of things, and really wouldn't be any improvement over the current offering. Unlike x86, PPC was designed for the 64bit transition from the start (even though it remained unused in Apple's product line for almost a decade), and so there is no speed penalty, whatsoever for running 32bit PPC code on a PPC970.
Specifically, the article states:
So far Apple's machines can see all the memory, they can't yet do 64-bit calculations. Present it with a 64-bit calculation, and a Mac with a G5 chip still breaks it into two 32-bit pieces. That's because, Glaskowsky says, Apple doesn't have a 64-bit operating system
Among it's other inaccuracies, it claims that a 32bit machine can only address 2GB.
They fault Apple for only allowing 8 GB of RAM in a desktop enclosure, even though this is still a significant improvement. This limit is still physical, there are 8 slots, and the largest capacity chips are 1024MB right now. They will work when 2048MB chips are released, increasing the max capacity of the existing line to 16GB.
As for the 64bit calculation bit, that's also incorrect. If the binary is compiled for the g5, then 64bit calculations will not be split as they are on 32bit architectures. The downside is that this binary will no longer work on 32 bit machines, for the obvious reasons. For best performance/compatability, two binaries, one 32bit, one 64bit, can be compiled and placed in the same Application bundle, making the difference between the two irrelevant to the user (only a single icon to click on, works on both systems, full 64bit calculation on the g5)
The biggest limit of the G5s at the moment is (and it's quite severe), to my knowledge, a single processes can still only address 4gb, because the size of void* is still 4bytes. Apple will need to duplicate all the libraries in 64bit form to make this work seamlessly, which will probably have to wait until 10.4.btw, there was no NT 1.0. The first version of NT was NT 3.1, magically version-synced with Windows 3.1.
It went like this
MS + IBM OS/2 ver 1.x
IBM OS/2 ver 2.x (my OS/2 ver 4.5 internally reports ver 2.45 as ver 3 was)
MS OS/2 NT ver 3
Windows NT ver 3.1
https://en.wikipedia.org/wiki/Inverted_totalitarianism
YOUR SIGNATURE IS OBSCENE.
ummm, hate to rain on your parade but apple uses industry standards in newer boards: OpenFirmware, USB, Hypertransport. To do what you are suggesting would mean dropping standards compliance, something that runs counter to their profit model.
The policy of the United States is worse than bad---it is insane. -- Ludwig von Mises, Economic Policy(1959)
No, it didn't. OS/2 and NT are completely different products with a completely different code-base. As the parent said, Windows NT 3.1 was version-synced to fit with Windows 3.1. OS/2 was never called "OS/2 NT", and still had releases after OS/2 3.0.
You are mistaken. The problem is that the upper half of 64-bit registers are not guaranteed to be safe across system calls. So while you _could_ use 64-bit instructions to, e.g., add two large numbers together using a single instruction, your program might be trashed if a context switch occurs between the time you compute this addition and the time you use the result. There are ways and means around this but they're painful, not the kind of thing realistic for a software project of any typical size.
The Power4 is not the same chip as the PowerPC 970. IBM will probably begin selling PPC 970 machines eventually, but they haven't begun to ship them yet.
I don't mean alpha particles. I mean neutrons. One source of the neutrons is generated from cosmic ray interaction with our atmosphere.
I do know the industry uses Neutron Flux as a unit of measurement:
A measure of the intensity of neutron radiation in neutrons/cm2-sec. It is the number of neutrons passing through 1 square centimeter of a given target in 1 second. Expressed as nv, where n = the number of neutrons per cubic centimeter and v = their velocity in centimeters per second.
The level of measurable Neutron Flux fluctuates with changes in altitude and latitude. Also solar flares can cause a 10 fold increase in measureable flux levels.
For soft error rates, the neutron particles are the most important.
I don't have a doctorate on this subject and cannot claim to be an expert in heavy ions. I would not be able to comment about half life and decay. A nice article by Dr. Eugene Normand is my source.
Holland
The heat is addressed not only with the 90NM process but with the new PowerTune technology which scales the voltage along with the frequency providing a quadratic decrease in power and heat whereas frequency only scaling only provides a linear decrease. Too, if IBM is including their "voltage island" techniques under the PowerTune umbrella, then IBM can use lower voltage components for none core processor sections. Too the efuse technology can dynamically switch and turn off redundant components that may become defective improving the life of the processor and of course more importantly the yield.
Nintendo is already going to use PPC processor for it's next console which is slated to debut in Japan late next yr. Microsoft was the last to go with IBM because both Sony and Nintendo had already been signed on with them for months and months in advance. ATi is also slated to be used in all 3 consoles as well.
1) POWER4 and PowerPC 970 is two different chips. 970 was derived from POWER4 for the light server/desktop market.
2) Linux runs just fine on POWER4 hardware as does it on 970 hardware. IBM's first machines using 970 (the blade module JS20) will run soley on Linux for half a year until AIX comes.
- Henrik
- when the Shadows descend -
I think you're missing part of the story here - megahertz for megahertz, the Pentium III outperforms a Pentium 4! The P4's "NetBurst Microarchitecture" or whatever is the ultimate incarnation of the megahertz myth - Intel made a bunch of design decisions to boost the clock speed at the expense of slowing down the number of instructions that could be processed per clock cycle compared to the P3. Why? Because Joe Consumer goes to Best Buy and looks at "how many megahertz it does" to make the decision about what to buy. Joe Consumer doesn't know what SPEC is. Heck, Joe Consumer probably doesn't know what a benchmark is.
The new Pentium M processors (used in Centrino laptops), based on the old Pentium III design, clock for clock outperform the Pentium 4. An AMD Athlon 64 3000+ runs at 2.17 GHz, yet AMD claims it runs like a 3 GHZ Pentium 4. AMD uses these model numbers because of the "megahertz myth." An Opteron or Athlon 64 at 2 GHz performs on par with a P4 at 3.2 GHz, thanks in part to a better cache design and higher memory bandwidth. The world's #2 supercomputer runs on 1.25 GHz Alpha processors. (ok, there's 8192 of them) and the #8 and #9 computers run on (8192 and 6656) Power3s at 375 whole megahertz!
The megahertz myth is alive and well. Trying to sum everything up with a single number is arguably a bad thing, but Joe and Jane Consumer want their single number. Look at digital cameras - before they became a mass market consumer item, companies said their camera had XGA resolution, or 1280x1024 resolution, or 640x480 VGA resolution, or whatever. Now you have megapixels.
But don't worry, by next summer, you'll be seeing 3 GHz Athlon 64s and Opterons along with the 3 GHz G5s. (AMD says they are going to start volume production of 90nm chips in Q3 2004 - i.e. the end of next summer.) If patterns hold, these will probably all perform about the same as a 4.5 GHz P4 (extrapolating from today's numbers.) And who knows what Intel will have by then? Rumors of an Intel 32/64 bit CPU a la Opteron have been flying lately. And even if they just could scale the Pentium M design up to around 3 GHz, it could be quite formidable...
You know of at least 200 reports? Where? Did you see 200 posts in Apple's tech forum and assume that there are 200 different problems with G5s? Are you claiming that of the 500,000 machines shipped, you have personal knowledge of 200 defective units? What is it?
I ordered a dual G5 2GHz for the office and I've only noticed 2 issues...
1) It used to hang when the plastic cover was removed and replaced while the machine was running (Apple specifically says I shouldn't do this so it's my own fault). I think this was fixed with the recent firmware patch.
2) There is a very subtle high frequency noise when the machine becomes active from rest. It isn't noticable in a normal office, but in a very quiet room I imagine it could be annoying. You can, however, stop this with a command line argument (can be run from startup script if I prefer)
For a machine based on a completely new architecture, the G5 has been the model of great design. It's stable, it's blazingly fast, and I really have zero real problems to report as I consider the above to be incredibly minor issues.
I have had Zero issues with fan noise, termerature sensors, sleep mode, or fan failure. I've not had any service calls to repair any issues with G5s in my division either and I've seen new G5s poping up.
So, I know it's easy to come on here as an anonymous coward and claim there are hundereds of problems with "beta" G5s, but where's the pudding?
Put up or shut up.
I'm not feeling witty so bite me