Slashdot Mirror
PowerPC 970 Running at 2.5 GHz
kuwan writes "IBM has just released a press release that indicates they have the new PowerPC 970 running at 1.8 to 2.5 GHz making it 'the fastest PowerPC so far.' IBM's original estimates were to have the chip running at 1.4 to 1.8 GHz at introduction, so this is very good news for those of us hoping Apple will use this as their next-generation chip."
I wonder how they managed to up the clock so dramatically? Is it just SOI and other techniques, or did they lengthen the pipeline significantly.
If it's just a pipeline lengthening scheme, well, meh, but if they kept the same execution pipeline and are now at 2.5ghz operating range, they're going to kick some ass.
I just hope Apple has their motherboards ready for 2.5GHz. The original spec of 1.8GHz with 6+GB bus was a little heady compared to Apple's current technology (no thanks to Motorola). I'm hoping they know how to build motherboards with the best of them to take advantage of IBM's new 970 chip. Pushing the envelope from 1.8GHz to 2.5GHz just makes the whole motherboard engineering issue more challenging. Let's hope Apple hardware design it up to the task (and then some).
Does this chip match the power consumption and low heat dissipation that we have all come to know and love from the PPC arch? Does anyone know?
Fnord.sig
I'm also wondering how they measured the clock speed. Mac fanatics are always saying "MHz is a myth" because their chips will do more per clock cycle. I'd be really interested in seeing ALL the specs (flops, and mips, etc) along with clock speed.
Two things struck me as odd. First, there's an inconsistent use of a comma as a decimal seperator:
6,4 GB/s I/O system bus throughput
Second, they direct you to a German IBM site for more info:
Further information in the Internet: www.ibm.com/de/entwicklung
Which leads me to think this was originaly a German IBM press release, which was quickly translated.
What does this mean? I have no idea. Is IBM's PowerPC development done in Germany?
One man's -1 Flamebait is another man's +5 Funny.
Not really. Intel intends to speed up the P4 by increasing the length of the pipeline. This offsets quite a bit of the performance benefit of the higher clockspeed. The 4Gz P4 will however be better for building toasters and blast furnaces.
Now that Centrino is out, how will Intel keep up the facade that clock speed matters for than MFLOPS or other (imperfect, but far better that clock speed alone) benchmarks ?
Dean G.
Comment removed based on user account deletion
I certainly don't need the speed. I wasn't schpieling against Apple--I've used macs for 15 years and type this response on an iBook500 happily running 10.2.4.
This is a great processor for Apple, and I'm sure they know it. It's fast, it's made by a reliable vendor with an interest in continued development, it's (basically) fully compatible with their existing OS, it's altivec compatible, it provides a great incentive for people to upgrade/switch.
So yes. Like I said--no doubt Apple is chomping at the bit for the 970.
(and so are we users looking to buy a new computer in the upcoming year or two)
sig my booty, check my website
The processor has a 32-bit mode. If the OS wants to switch to 32-bit mode to run an app, it can, with no (significant) performance penalty.
The bitness of the OS and the processes it runs are mostly orthogonal concepts.
From what I have read (mostly at macspeedzone.com), I would think that Apple will NOT be ready to show a 970-based computer at the July Macworld. Notice that I said "show", as opposed to "ship".
I hope I'm wrong though.
So this is my question: what do they do as a stopgap? Ship a G4 with four processors? Punt and simply lower prices until the 970 is ready?
Steve Jobs has been dealing with Motorola since, when.. 1983 or so? Might be high time for a divorce.
The G3 (PPC750) is a development of the PowerPC 603. It's got superb integer performance, decent FP, no Altivec and extremely low power draw.
The G4 (PPC74x0) is a development of the PowerPC601 and 604. Integer Performance is about the same as the 750, but it has a much faster FPU and Altivec. Moderate power draw and a much more powerful CPU overall.
There are more differences between the PPC750 family and the PPC74x0 Family than just Altivec, although that's the most notable difference.
All of these CPU's are descendants of the Power CPU line. Theoretically Mac OS X could run on teh Power4 with some minor work. Now that would make a killer system, at the expense of cost (A single Power4 CPU package, with multiple cores, costs as much as a PowerMac.)
"You've got an invalid haircut" -Warren Zevon - Life'll Kill Ya
With 3GHz CPUs on the horizon, i am kinda confused as to how these cpus can function. I mean, assuming electricity travels in copper/semiconductor (or whatever they use in ICs these days) as fast as light in vaccum (in fact it travels at a fraction of 'c'):
in one 3GHz cycle, a signal in the the CPU can only travel less than 10cm (~4 inches for those still stuck with the imperial units). With CPU dies sizes of a similar magnitude (~4cm), and with all the routing inside the CPU, why dont we get some very serious race conditions? are the intel engineers actually going in and laying out the chip keeping this speed in mind? as the speed will vary with the CPU temperature, its even more difficult.
And doesnt that impose a HARD LIMIT on how many MHz can be squeezed out? I mean, a comment below mentions intel demoing 5GHz CPU (this is the first time i'v heard about this, so i dunno how true this is), and that means the signal only travels 6cm, which means it cant even traverse the whole die in one cycle.
Is there something i am forgetting here? Can someone in the know please shed some light?
Ghoul2
Sigura Non Grata
Fair enough. Right now, the fastest processor in the world is the Pentium 4 3.06ghz: 1130/1103 (int/fp). For pure floating-point horses, it's the Itanic 2 743/1427 (int/fp).
So a 2.5ghz 970 would be close in performance to both of today's fastest shipping processors. It's likely that the P4 and Itanic will be 15-20% faster in six months, so IBM will still be lagging in the performance hunt. However, it's striking how much closer to the peak performers this chip will move IBM -- and, by extension, Apple.
'jfb
To spur "enterprise Linux," Big Bang, the distributed two-phase commit.
They shrunk the size of the gates on the transistors, basically trading reliability for performance. Considering that one of the main selling points of Apples is their longevity and ability to hold value due to it, I can't help but wondering if this is the right move.
Apparently, in order to increase the reliability of the Power4 for the high-end server market, IBM used much thicker gate oxides on the chip's transistors. The trade-off for this decreased failure rate and improved reliability was that the Power4's transistors have slower switching speeds, so even with process shrinks it's harder to push the design to higher clock speeds. Since the 970 is made for the desktop market, there's no need for such measures and therefore the new chip's clock speed will scale much higher than the Power4's. In sum, the 970 is made to be faster, cheaper, and significantly less reliable than the Power4. (Of course, when I say "significantly less reliable than the Power4," you have to understand that this puts the 970's product life and failure rate on par with other mainstream CPUs, since the Power4's increased gate oxide thickness makes it significantly more reliable than most mainstream CPUs.)
ArsTechnica overview
It's a given that Apple enthusiasts will be happy as can be once they fire up a brand new powerfull box, the question is how they will feel when they find out it has the lifespan of a typical Intel or AMD CPU.
It wasn't very long ago that IBM decided to add Moto's Altivec engine to their chips. Earlier, IBM had rejected Altivec because IBM wanted to target the embedded controller market, which they felt really didn't require Altivec enhancements.
In persuing new markets, adding Altivec made sense. Not to mention that it had a good strong successful demonstration as part of Apple's hardware. So now it is included. I know there are more details to it than all that, but at the moment I'm just too tired and lazy to look up the articles...
Whew! This water sure is cold!
http://www.intel.com/home/desktop/pentium4/faq.
Q: What is Streaming SIMD Extensions 2?
A: Streaming SIMD Extensions 2 extends Intel® MMX(TM) Media-enhanced technology and the Streaming SIMD Extensions. Single Instruction Multiple Data (SIMD) allows a single instruction, such as addition or subtraction, to operate on more than one data set concurrently. The 144 new cache and memory management instructions enhance performance to accelerate the most-demanding Internet and computing applications. SIMD double-precision floating point accelerates demanding content creation, 3D rendering, financial calculations and scientific applications. In addition, 64-bit MMX technology (SIMD integer) instructions have been enhanced and extended to 128-bits, accelerating video, speech, encryption, imaging and photo processing.
I believe and they run circles around the 3Ghz P4.
Not that I'd heard - if the Itanic ran circles around anything, Intel wouldn't be saying nobody needs 64-bit yet. They'd just double the clock speed (with no performance change) for marketing reasons. No, Itanic is slow and has been doomed to fail for some time now - thus the name.
Anyone know who came up with that name first?
$x='S24;r)>63/* h@<5+oZ)32"5cz';$me='phroggy'x$];
$x=~y+ -xz+\0-Tx+;print$_^chop$me for split'',$x;
Your comparison only goes to show how much both of those above companies are gouging their customers. You can build a faster system for a lot less that will probably have higher quality components. Let's see:
Pentium 4 2.4GHz CPU: $160
Asus P4S533 mainboard: $100
512MB memory: $70
Enlight case: $47
nVidia GeForce 4 4600: $220
120GB Western Digital HD: $128
Plextor CD RW/DVD combo drive: $100
NEC / Mitsubishi 17" LCD flat panel: $450
Microsoft OS Tax: $200
Mouse & keyboard & floppy & network card: $100.
$1575 total.
The only problem is that you have to supply an afternoon's worth of assembly with a screwdriver and have a little knowledge about what you're doing. And I could save $2000 over your estimate, more if I installed Linux (or my prefered OS, FreeBSD - and it would scream on such a machine).
You'd end up with a fully-functional system that will blow both your above specced machines out of the water. If you don't believe me, go price up the components yourself on NewEgg.com. Sure, the components I've picked above aren't top-of-the-line, but just off the top of the price/performance curve.
You can't really do-it-yourself with Apple machines - too much is proprietary and there's certainly not a choice of components. This is a real turn off for those who have knowledge and are comfortable building their own system.
At the same clock speed, and for short sequences of instructions, a Z80 can beat a P4. The problem is... they don't make them at the same clock speed.
It's irrelevant how many times per second the chips clock says "tic-tac", what matters is how fast real chips can get real jobs done. For real-world purposes, you can compare the best (ie, the fastest chips) or the most valuable (ie, the ones with the best speed/price ratio).
So you see, Mr. Anonymous Coward, comparing the performance "per clock cycle" is irrelevant. It's like comparing the performance "per instruction length", or "per transistor count". It might be interesting from a theoretical point of view, but if a chip that does a lot of work per cycle cannot do more than a couple of cycles per second, it's still a terribly slow chip. The P4 was designed to do less work per cycle, but work at higher frequencies. The Athlon, on the other hand, does more work per cycle but cannot reach such high frequencies. In the end, they're more or less matched. So, in that situation, which one do you buy? Perhaps you buy the one with better "performance per clock cycle". I buy the one that's cheaper (funnily enough, in this case they would be the same).
I thought Macs were competitive with PCs. Or are you saying that anyone who buys a Mac is totally clueless? It all depends on the market you're talking about. When this chip is finally released, PC processors will be twice as fast than they are now, and will probably cost half what they cost now. Anyone buying a Mac for raw number-crunching is an idiot, just as anyone using Windows for a firewall or a quad Xeon for an office machine is an idiot. It doesn't matter is something is faster or slower, as long as it's fast enough.
To use a car metaphor (that most people seem to understand), not everyone needs or wants to drive a Lamborghini. It's expensive, it's hard to park, it's hard to drive, it's cramped and it drinks like a fish. Most people are better off with a "normal" car, that's fast enough and powerful enough for them, is easy to drive, and has room for the kids and the dog.
Having said that, if you spot someone selling a metallic-gray Lamborghini Diablo Roadster (convertible) for less than 15K, let me know, will you?
RMN
~~~
Whodathunk that one day we'd be reading a story titled "Apple: ..." with an IBM icon? Maybe I'm getting old, but I think it's kinda cool.
When I pried the heatsink off my brand-new Power Macintosh 7100/66 (um, nine years ago?) I found a gleaming blue plate with IBM in big white letters on top. It's no biggie.
Really we should just come up with one icon for M$ and one for everyone else. (shall we call them the rebel alliance?)
> The following is a simplistic view of things, but we are talking
> about a 64bit processor. Remember the Itaniums Intel is selling
> are running at around 1GHz - 1.5GHz I believe and they run
> circles around the 3Ghz P4.
That's overgeneralized. The 3GHz P4 is very much faster at most tasks than the 1.0GHz Itanium II, which is the fastest instance of the chip that has been given entries at spec.org. The reason why the Itanium II appears much faster is that you only see benchmarks that relate to its very narrow field of marketing. It's a server processor. You won't see it tested in areas more suited to general purpose computing (games, office suites, etcetera). And, hell, the Itanium sucks in specint, one half of the single processor version of the most prolific server benchmark suite in the world. The 900MHz (fastest speed submitted -- for some reason, they only gave specfp scores for 1000MHz, unless I missed an entry or few) Itanium II gets 674, compared to scores above 1100 for the 3.06GHz P4. That's a whopping 63% difference! The fastest Itanium II is almost 40% slower than the top of the line non-Xeon Pentium 4!
The Itanium II does fantastically in specfp -- bested, I believe, only by the DEC Alpha, which is sadly being pushed under the carpet for reasons more political than I'd like (Alpha IP is owned by HP and Intel, the companies that created the Itanium's core architecture) -- and many other benchmarks. But you can't simply ascribe a single, simple feature to the performance advantages of the processor. Yeah, the processor can address a 64-bit memory space and, yeah, the processor has 64-bit GP registers. But you're ignoring many other features piled on top. Itanium II is a server processor, so it can afford to have some extra doodads added to it, doodads that would be considered financially unfeasible on mainstream processors.
Hmmm, I did a quick google search, so I apologize if I pulled incorrect info on the following:
The Itanium II has a more than a megabyte and a half of cache memory on the die of the processor. It seems to optionally go up to 3MB on-die L3 cache. In comparison, the Pentium 4 has 512KB cache (there's some more cache, the L1, but that's inclusive), and the Athlon XP has either 384KB or 640KB cache (depending on whether you're counting the older Thoroughbred or the newer Barton). So the Itanium gets about three times as much cache memory on the processor die!
Itanium II has a 400MHz, 128-bit data path to the chipset. Pentium 4 is 533MHz, 64-bit. So the P4 gets a chipset that can send it 4.27 GB per second while the Itanium II gets a chipset sending it data at 6.4GB/s.
The Itanium II has more functional/execution units. The Itanium II gets predication, which is a very expensive (in terms of how much bulk it adds to the die) feature that effectively gets rid of a lot of the penalty associated with branch misprediction (a problem which is rather huge with the trillion-stage netburst microarchitecture of the Pentium 4, though I'm told that the multithreading implementation of the P4 can help alleviate some of that).
The number of bits in the processor don't matter *that* much, not after the 32-bit level. Yeah, it helps, but you have to take the whole package into account. A 64-bit scalar one-stage processor with a ten-byte, off-die cache would get its ass kicked mercilessly by an 80486DX-50.
To take another tack: I'm somewhat interested in possibly purchasing an Athlon 64 late this year or early next year. But if the Athlon 64 was just an Athlon with 64-bit extensions, I wouldn't give it the time of day. I'm interested because the Athlon 64 will have an on-die memory controller. I'm interested because the Athlon 64 will support twice as many registers as a typical x86 chip (which may decrease the need for cache accesses, which could increase performance on my recompiled linux apps). Either of these two advantages promise a far greater advantage for me than the simple increase in register size and memory addressability.
-JC
According to IBM itself the 970 will be cheaper to produce than the G4, due to its small die size.
I especially like your idea that X11 is bolted on to OS X. Damn, dude, X11 is "bolted on" to every Unix!
I cannot even remember the variety of Unixes I've used...PDP 11/33, 680x0 boxes of various kinds, SGI Indigo (is Irix a real Unix?), HP "Snake" (is HP-UX a true Unix to you mister G4dget?), AIX on RS/6000, perhaps a half-dozen to a dozen forgettable and forgotten boxes, and lately the glorious Mac OS X.
Oh hell look at all that and I forgot to mention the Sun pizza box that was on my desk. Did Sun ever ship a "true UNIX" oh mister expert nit-picker G4dget?!
So get over it. Not only is Mac OS X a real, true, complete Unix, but due to its incredible shipping volume, it is now THE UNIX.