A few more published Mac + Intel info

Qaxpla said:

Maybe I don't understand this completely. But are lots of people over-reacting here and having hissy fits?

I don't really think Apple would chuck away decades of hardwork to get them where they are easily. Apple have been making OS X to run on intel processors for the last 5 years, i'm pretty sure that they won't just chuck away their reputation without valid justification. I think that them being sure OS X runs better on Intel Processors is enough proof that they know what they're doing here. Seriously.

Or am I missing something?

Click to expand...

Some people have an irrational hatred of Intel and the x86 architecture.

If you don't like the way apple is going, stay with ppc, good luck running any apps in 5 years...
Your insolence will gain you nothing but laughs from your peers.


Incidently sevenhelmet:
'Besides, won't this create security problems for the mac? Windows spyware, windows viruses, windows worms, windows security gaps...'

- Think again, how does linux run on intel and AMD systems without issues? Spyware works on amount of privileges the system gives its programs, not its hardware.

Alex

dirtydog said:

Some people have an irrational hatred of Intel and the x86 architecture.

Click to expand...

It's not "irrational hatred" at all, can you actually show me anyone who likes the architecture, I don't think even Intel like the architecture and have tried several times to get rid of it but at the moment legacy rules.

Microsoft certainly aren't that keen on it otherwise they wouldn't be going the PPC route for the new XBOX, ditto for Sony and Nintendo are staying with PPC.

Personally I have nothing against Intel itself only the x86 range, I've never liked them from day one, from the very first 8086 assembler I wrote.

Amen-Moses

Amen-Moses said:

Why wouldn't it be a good general purpose CPU?

The cores will indeed be PPC 970 based, maybe minus the AltiVec, so should have virtually the same instruction set. OS X runs on G3s and they don't have AltiVec so I can't imagine why it wouldn't run on the new PPC chips.

That may be an interesting exercise when the XBOX 360 arrives, try to get the Darwin core running on it and maybe even full OSX although the display drivers would probably be a big problem, wouldn't it be really really funny if an XBOX ran OS X faster than a Mac. :spook:

Amen-Moses

Click to expand...

Each of the processors have half the floating point units, half the int, an unkown Vector processor, and no out of order execution. Branch prediction is minimal at best but instead relies on software optimization for performance. The cores are Power PC based yes, not 970 based though, all public information on the structure points to a much less complex core. How exactly if IBM can not mass produce a dual core 970 presently do you think IBM could accomplish a 3 core system in months?

The XBox without a doubt will be faster in some tasks, especially with software that has a predictable execution order and has been optimized for it. Most programs for OS X do not optimize well for Altivec and that is simple comparitively.

Also a valid point was raised elsewhere, with this CPU in the console, IBM's goals going forward will be to make it cheaper, not faster. Console generations have a nice static requirement for speed.

Kokopelli said:

How exactly if IBM can not mass produce a dual core 970 presently do you think IBM could accomplish a 3 core system in months?

Click to expand...

IBM are currently going down a single core path for the entire range of processors, they are calling it "Power Everywhere", it is the basis behind their Power 6 processor and as far as I can make out all three of the major console customer (i.e Sony, MS and Nintendo) devices as well.

In theory the only difference between processors in future will be how many cores per processor and how many processors per chip.

As far as the instruction set is concerned I can't see any reason why they would not also have a single instruction set across all models as well, it would certainly make a lot of sense to do that otherwise they have to support variant versions of AIX, Linux and developer tools.

The one confusion in my mind is where the cell processor fits in as they initially stated that it would be 45nm yet their Power 6 seems to be 65nm so unless all they are doing with the cell is making the same core smaller I'm not sure how they retain the single core strategy.

Unfortunately I don't work for them anymore so don't get the insider view of all this. neye:

btw, "single core" doesn't mean that there won't be different sized/speed buses, caches, pipelines and vector units.

Amen-Moses

Do you have a pointer to an architecture book with the instruction set?

Kokopelli said:

Each of the processors have half the floating point units, half the int, an unkown Vector processor, and no out of order execution. Branch prediction is minimal at best but instead relies on software optimization for performance. The cores are Power PC based yes, not 970 based though, all public information on the structure points to a much less complex core. How exactly if IBM can not mass produce a dual core 970 presently do you think IBM could accomplish a 3 core system in months?

The XBox without a doubt will be faster in some tasks, especially with software that has a predictable execution order and has been optimized for it. Most programs for OS X do not optimize well for Altivec and that is simple comparitively.

Also a valid point was raised elsewhere, with this CPU in the console, IBM's goals going forward will be to make it cheaper, not faster. Console generations have a nice static requirement for speed.

Click to expand...

mmoy said:

Do you have a pointer to an architecture book with the instruction set?

Click to expand...

No. Sentence added to bypass lameness filter.

From the IBM website:

"CELL is not limited to game systems. IBM has announced a CELL-based "blade" leveraging the investment into the high-performance CELL architecture. Other future uses include HDTV sets, home servers, game servers, and supercomputers. Also, CELL is not limited to a single chip, but is a scalable system. The number of attached SPUs can be varied, to achieve different power/performance and price/performance points. Also, the CELL architecture was conceived as a modular, extendible system where multiple CELL subsystems each with a Power core and attached SPUs, can form a symmetric multiprocessor system."

That was dated late last year and the core was stated as 64 bit 90nm Power core running at 4Ghz with 8 SPUs (basically little cut down processors that only do math, integer, FP and vectorised SIMD). Since then I've seen several news items referring to a projected 45nm cell probably since they got their new fabrication facility up and running (the one that will be supplying MS as well).

Amen-Moses

How thrilling. (Looks around) So where are these 45nm chips? I don't doubt IBM has a plan and is in motion to follow through on the plan. This relies somewhat on everything falling in place though.

I freely admit I could be wrong in my assumptions and inferences. But even a slight amount of research on what is publicly available shows (from the leaked documentation among other locations) that the Xenon is a custom PowerPC CPU built based upon Microsofts spcifications and requirements. This says nothing on relative performance versus a stock G5 or other processor but does tend to suggest that,assuming the documentation is not a lie, the CPU is a custom chip.

What's really odd about CELL is the lack of concrete information on it. Usually you put out architecture documents
and instruction sets so that other companies can prepare for
the actual product. All I've seen so far is block diagrams and
vague descriptions.

It's hard to judge a chip without the architecture and software optimization documents from my point of view as everything just seems to be speculation right now. With the K8 and Pentium 4 docs, you know what you're getting. With CELL, it seems as if IBM can say whatever they want to but not have to provide details to the world.

Amen-Moses said:

From the IBM website:

"CELL is not limited to game systems. IBM has announced a CELL-based "blade" leveraging the investment into the high-performance CELL architecture. Other future uses include HDTV sets, home servers, game servers, and supercomputers. Also, CELL is not limited to a single chip, but is a scalable system. The number of attached SPUs can be varied, to achieve different power/performance and price/performance points. Also, the CELL architecture was conceived as a modular, extendible system where multiple CELL subsystems each with a Power core and attached SPUs, can form a symmetric multiprocessor system."

That was dated late last year and the core was stated as 64 bit 90nm Power core running at 4Ghz with 8 SPUs (basically little cut down processors that only do math, integer, FP and vectorised SIMD). Since then I've seen several news items referring to a projected 45nm cell probably since they got their new fabrication facility up and running (the one that will be supplying MS as well).

Amen-Moses

Click to expand...

Kokopelli said:

How thrilling. (Looks around) So where are these 45nm chips?

Click to expand...

2 or 3 years away most probably, at present everyone is gearing up for 65nm but existing designs from all manufacturers are at 90nm (with the possible exception of the very lates AMD chips just released at the beginning of this month).

As mmoy said above all anyone is really going on is the patents that have been issued and already known architectures like 970 and Ghecko.

Maybe IBM is just blowing a lot of hot air but they seem to have MS and Sony convinced which is good enough for me.

Amen-Moses

Yes but the needs of a console are different from the needs of a general use computer. Or at least that is my contention.

The cell and Xenon are well designed to handle streaming data. It is dead on for processing HDTV, streaming video, and (at a guess) number crunching. That does not necessarily make it dead on for other types of computing and does not give any indication of how easy or hard it will be to get good performance out of any particular problem.

If a problem can be made into a set of linear parallel problems so that all 3 processors are continually fed, then results are likely to be good. If not, my supposition is that performance might be less impressive. I am sure it will still be fast, the question is how fast and how much developer effort will be required to get this speed versus development on a plaform with more extensive branch prediction.

I like PPC, but I do not worship it and I would classify cell and xenon as more of an unkown quantity than Intels current 3 year strategy.

Finally I have said it before Apple made this announcement I will say it now. PPC and the current crop of x86 CPUs have different strengths, neither is the clear winner in all performance categories. Given the need for water cooling on the 2.7 GHz systems I am not too sure the G5 wins in the Watts for performance category.

Kokopelli said:

Given the need for water cooling on the 2.7 GHz systems I am not too sure the G5 wins in the Watts for performance category.

Click to expand...

Which is why Jobs invented it!

Mind you he didn't specify what exactly he was comparing it too so we can't tell if it is just another piece of snake oil salesmanship from the undisputed master. :ninja:

A fairer comparison in somethings per watt would be a dual 1.5 Ghz G4 (at 15 watts) and a single 2.0 Ghz Pentium-M at almost 25 watts.

Amen-Moses

Amen-Moses said:

Which is why Jobs invented it!

Click to expand...

No this was invented long before the slide, which is nothing more than spin. i am talking about the factor that people expect Macs to be fast and quiet. Not just one or the other.

In the past Intel just wanted to be fast. Silence was not necessary. Nut eventually even the hardocre started to object to the noise. Thus the trend in Intel moving to a variation of the Pentium-M on the desktop and C&Q in the AMD line. Performance per watt was around long before the WWDC keynote.

For Apple to compete with the (granted louder and hotter) PIV systems they went to exotic cooling techniques. While I am not privy to the inner workings of either company it is a commonly held opinion that IBM promised Apple that G5s would be faster and cooler than they are currently. This is very likely at least part of the friction between the two. Intel has taken steps to fix the problem, IBM has taken a different path.

Interesting read:

freezes over; it must've been the liquid cooling: Hannibal on the Apple-to-Intel transition