IBM's Power5 is Coming

[ktlx]

Re: Re: Re: Re: Re: JtheLemur

Originally posted by jettredmont
Intel has said this? The only estimates I've heard from Intel is that the P4 is set to go up to 3.6 by the end of this year, and that the 90nm P5 will debut at 3.2GHz in the fall (but it's supposed to be more efficient than the P4 so a 3.2GHz P5 *should* meet or beat a 3.6GHz P4). Unless you've seen something that no one else has, Intel is quite tight-lipped about what will happen to the Pentium line after this year.

Intel has published their Pentium 4 roadmap in the past and I have seen it reproduced on places like Ace's Hardware, Tom's Hardware and ArsTechnica so it is hardly like no one else has seen it. They do not give dates beyond this year, but they did give ranges for the 130nm and 90nm processes.

In any case, 3.6GHz P4-class is the end-of-year projection from Intel. Not 5GHz. Not "wGHz" whatever 'w' stands for.

Please bother to at least invest the time in reading my post. I never attached any dates to those changes. Intel has said that their 90nm process version of the Pentium can scale to over 5Ghz but they did not say when that would be. In fact, I said just the opposite that we do not know how quickly each team can ramp up.

And by the way, w is a variable. I thought that would be obvious.

There's more to it than process changes and raw GHz. Processor design and architecture also make a big difference in overall system performance. You are correct that Intel is the only company out there with experience in produce its volume level of production of CPUs, and experience counts for a lot. However, boiling that down to how fast they can shrink their manufacturing process is overly simplifying the matter. IBM has a world-class design team, and a world-class implementation team. If anyone has the goods to challenge Intel, it is IBM.

Quite true but once you leave process changes and raw Ghz, you are just speculating on how some processors sometime in the future will compare with each other. IBM does have a world class design team and I have no doubt in my mind that if the issue was strictly a technical one, IBM would do as well or better than Intel.

The problem is that the issue is really primarily an economic one. We do not know how the market for the PPC 970 will turn out. IBM will not invest the money if they cannot make a profit at it. And it is not clear they can. If Intel executes perfectly on all of their plans and IBM makes any mistake, the market may turn out no better than the G4.

Even if IBM executes on the PPC 970 flawlessly, IBM still has to convince people that AIX/PPC or Linux/PPC is as good or better as Linux/x86-64 for blade servers and workstations. If they are not successful, then the PPC 970 turns into a processor for Apple and the telecom market. Just like the G4.

 

[ffakr]

Originally posted by JtheLemur

The one thing that IBM HAS said is that the 970 tops at about 1.8GHz. HOPEfully, that 1.8GHz is as powerful as a Pentium 4 @ more than 4 GHz, because IF and WHEN Apple uses a 970, Intel will be way over 4GHz. Right now, Avid on a loaded P4 is a heck of a lot smoother than FCP on a loaded G4.

Well, there is NO guarantee that P4 will ship at 4 GHz by the time the 970 is available (to possibly ship in macs). Most industry analysts seem to predict a move to 3.3 GHz this spring and probably around 3.6 by fall.

The fact is, Intel doesn't release faster chips until they need to. They only need to beat AMD, they don't need to release P4s as fast as possible. Making the P4 too fast is bad for margins (they can't sell slower, higher yield chips for more) and it's bad for their enterprise goals (P4 can't be more powerful than Itanic).

Consider these *facts*.
*K7 has little future. AMD has had a LOT of trouble pushing the MHz up any further, even with the latest revision. This relieves pressure on Intel to release faster processors...
*Athlon64 has been pushed back till August or September depending on who you trust....
*Intel has demo'ed 'sample' chips for years that clock way above what they can produce in mass numbers. The fact that they've demo'ed chips running significantly faster than 3GHz doesn't mean they are anywhere near releasing chips at that speed...
*The ultra high speed approach will become self defeating... a 4GHz chip is only 33 percent faster than 3GHz, but it will have much more trouble fetching instructions fast enough. Expect diminishing returns...

And, of course, the other side of the coin is that Apple is likely to put dual 970s into a pro machine if they feel that they need to crush x86. They have plenty of expertise in shipping dual proc boxes.

Personally, I don't see Apple having too much trouble competing with Intel in the 2nd half of this year.

....ffakr.

 

[Telomar]

Originally posted by JtheLemur
The one thing that IBM HAS said is that the 970 tops at about 1.8GHz.

That isn't quite accurate. At release it will top out at 1.8 GHz. IBM hasn't made a public statement regarding scaling after that.

Originally posted by ffakr
*Athlon64 has been pushed back till August or September depending on who you trust....

It's September currently unless AMD moves it again.

 

[Dave Marsh]

Re: Dell Dual Xeon 3.06GHz System

Well, I hopped over to that link and ran the numbers. Yes, a dual 3.06GHz Xeon system with XP Pro, Superdrive, 1GB RAM, 80GB HD, and no monitor can be yours for ~$4400. Don't wait, order one today!

 

[macphoria]

It doesn't matter whether Pentium 4 goes over 4Ghz. Intel just released news that they were coming out with a new mobile chip that actually has lower Ghz rating but better performance, suggesting speed doesn't mean everything. Sounds familiar? I do not have all the information on new 64 bit chips that AMD and Intel are workign on, but I doubt they will start out at 4Ghz. It might be higher than IBM's 970, but still I think if 970 can hit the market first, it will have good chance of catching up and competing.

 

[Death2PCs]

IF, and a big IF... apple uses this processor itll be atleast untill MWSF 04. theres no mid this year or anything.

its too big of an announcement not to do at apples biggest show.

 

[ffakr]

Re: Re: Dell Dual Xeon 3.06GHz System

Originally posted by Dave Marsh
Well, I hopped over to that link and ran the numbers. Yes, a dual 3.06GHz Xeon system with XP Pro, Superdrive, 1GB RAM, 80GB HD, and no monitor can be yours for ~$4400. Don't wait, order one today!

Actually I poped over there and configured it more like a base config for a top end dual G4. Obviously the main difference would be the dual P4s - 3.06GHz, but I put in a 120GB HD, and I selected the DVD authoring software bundle... and other options to make the Dell mimic what you get from Apple (low end speakers...)

My price was $4680.

The prices of the dual 2.0 GHz P4 Dell, and the dual 2.4 GHz P4 Dell come in just over and under the cost of a dual 1.42 GHz Macintosh.

now, consider that The P4 is designed for high clock at any cost... and that an AthlonXP 2000+ only needs to run at 1600MHz to be performance competitive.

I'm not saying that x86 offerings aren't faster. They are.
I'm just pointing out that these comparisons are usually more complex than they appear at first... and that Apple does dual-proc farily reasonably compared to dual-proc X86 boxes.

 

[jettredmont]

Re: Re: Re: Dell Dual Xeon 3.06GHz System

Originally posted by ffakr

The prices of the dual 2.0 GHz P4 Dell, and the dual 2.4 GHz P4 Dell come in just over and under the cost of a dual 1.42 GHz Macintosh.

now, consider that The P4 is designed for high clock at any cost... and that an AthlonXP 2000+ only needs to run at 1600MHz to be performance competitive.

I'm not saying that x86 offerings aren't faster. They are.
I'm just pointing out that these comparisons are usually more complex than they appear at first... and that Apple does dual-proc farily reasonably compared to dual-proc X86 boxes.

Yes, but the P4's you are looking at enjoy 566MHz front side buses (per chip I believe), which keeps them from being memory starved. The dual 1.4sGHz G4s share a single 166MHz FSB, which is just plain slow in comparison and leads to absolute memory starvation. The G4s would definitely blow the pants off the P4s in any benchmark that fits entirely in cache, and have unbelievable MTOPS scores comparatively, but the memory logjam kills those benefits in most if not all consumer-level programs.

 

[ffakr]

Re: Re: Re: Re: Dell Dual Xeon 3.06GHz System

Originally posted by jettredmont
Yes, but the P4's you are looking at enjoy 566MHz front side buses (per chip I believe), which keeps them from being memory starved. The dual 1.4sGHz G4s share a single 166MHz FSB, which is just plain slow in comparison and leads to absolute memory starvation.

Actually, I know that Apple has an individual bus for each processor. They went out of their way to point this out recently. Athlons do the same, but I'm not sure if Intel does or not.

Apple recently made the argument that the lack of DDR support isn't as bad as it looks because the Processors don't share the same pipe.

Upon request.. I'll try and find some on-line references to the above claims. I'll be out fix'in a stinky windows box for a while though.

 

[eric_n_dfw]

Re: Re: Re: Re: Re: Dell Dual Xeon 3.06GHz System

Originally posted by ffakr
Actually, I know that Apple has an individual bus for each processor. They went out of their way to point this out recently. Athlons do the same, but I'm not sure if Intel does or not.

Apple recently made the argument that the lack of DDR support isn't as bad as it looks because the Processors don't share the same pipe.

Please include links to support your claim. I am very interested in reading about it.

As far as I can tell (from Apple.com, ArsTechnica.com and other sites) the dual G4's share a single MaxBus to the System Controler. The System Controler then handles separate busses to the PCI slots, DDR RAM, ATA100 drives and other things - but the 2 processors share the pipe to the system controller.

Please, prove me wong here. (I don't think I am though)

 

[wdlove]

Originally posted by Death2PCs
IF, and a big IF... apple uses this processor itll be atleast untill MWSF 04. theres no mid this year or anything.

its too big of an announcement not to do at apples biggest show.

Has Apple made a decison yet regarding MWNY in July?

 

[eric_n_dfw]

Originally posted by wdlove
Has Apple made a decison yet regarding MWNY in July?

I thought that issue was for the 2004 Summer MacWorld, not this years. Right?

 

[MorganX]

Originally posted by macphoria
It doesn't matter whether Pentium 4 goes over 4Ghz. Intel just released news that they were coming out with a new mobile chip that actually has lower Ghz rating but better performance, suggesting speed doesn't mean everything. Sounds familiar? I do not have all the information on new 64 bit chips that AMD and Intel are workign on, but I doubt they will start out at 4Ghz. It might be higher than IBM's 970, but still I think if 970 can hit the market first, it will have good chance of catching up and competing.

Prescott will be for desktops and have 1MB on die cache and an 800MHz FSB. Oh, and hyperthreading too.

Too bad OS X doesn't run on PC hardware. I think even IBM will have a hard time catching Intel.

 

[ffakr]

Originally posted by MorganX
Prescott will be for desktops and have 1MB on die cache and an 800MHz FSB. Oh, and hyperthreading too.

Too bad OS X doesn't run on PC hardware. I think even IBM will have a hard time catching Intel.

but it will still be the P4 core.
Unfortunately, memory bus is tied to the processor bus... thats why the memory on current P4s is only DDR 266MHz... as in 133MHz x4 = 533MHz bus and 133MHz x2 = DDR266 memory.
The QDR 800 MHz bus on the upcomming P4s will ensure that the memory Bus will be DDR400. It doesn't matter if 333 CAS2 seems to perform as well or better for less, and it doesn't matter that faster memory may become available in the future.

...and the bus on the 970 is half the speed of the processor... so the 1.8GHz part will have a 900MHz processor bus, the 2 Gig part will have a 1Gig Bus... and it can use its choice of memory.

Prescott will be a smoking chip, but so will Opteron and the 970.

 

[PyroTurtle]

i'm only concerned with laptops as of now...
desktops will be powerful enough and this is the first time in my life i can remember saying that most computers are fast enough in general for my needs...however, laptops need a nice cool proc that runs friggin fast...that's all i want these new IBM chips to do...along with run at 10THz of cource as well, and not burn my legs off after 15 minutes of use....

 

[ddtlm]

ffakr:

Unfortunately, memory bus is tied to the processor bus...

This is not the case at all. There are P4's with 533mhz FSBs and DDR-333.

...and the bus on the 970 is half the speed of the processor... so the 1.8GHz part will have a 900MHz processor bus, the 2 Gig part will have a 1Gig Bus... and it can use its choice of memory.

IBM's own PPC-970 pdf claims "up to" 900mhz, and also the FSB can apparently clock at integer fractions of the core speed other than 1/2 (read that in an Arstechnica article).

Actually, I know that Apple has an individual bus for each processor. They went out of their way to point this out recently. Athlons do the same, but I'm not sure if Intel does or not.

You should definately find evidence of this. Intel uses a shared bus, even on Itaniums (although they have multile shared busses for the big machines).

 

[wdlove]

Originally posted by eric_n_dfw
I thought that issue was for the 2004 Summer MacWorld, not this years. Right?

The last that I read on the subject, Boston Globe. IDG and Apple are having intense negotiations about MWNY 2003. Apple has only agreed to MWSF. Has anyone heard anything more recent?

 

[eric_n_dfw]

Originally posted by wdlove
The last that I read on the subject, Boston Globe. IDG and Apple are having intense negotiations about MWNY 2003. Apple has only agreed to MWSF. Has anyone heard anything more recent?

Here we go: http://news.com.com/2100-1040-962417.html

Aparently, IDG wants to move the summer 2004 MacWorld to Boston, and Apple is against it. Because of that, Apple is "reevaluating" participation in MWNY this year.

 

[evoluzione]

Re: Re: Re: Power 5 in Nuclear Simulators

Originally posted by DavPeanut
What, thats all? I have a TI-83+ graphing calculator. Oh, and get this. My school has a bunch of TI-81 graphing calculators. They'll bet you any day at graphing y=x.

Or maybe only the days your computer is Defraging.

damn, you just reminded me of one of my first 'puters. a TI-99-4A lol

 

[wdlove]

Originally posted by eric_n_dfw
Here we go: http://news.com.com/2100-1040-962417.html

Aparently, IDG wants to move the summer 2004 MacWorld to Boston, and Apple is against it. Because of that, Apple is "reevaluating" participation in MWNY this year.

I would think they will need to be making a decison soon. Plans for the exhibition hall, exhibitors, & Mac Fans!

 

[mathiasr]

Re: Reality Check

Originally posted by Frobozz
However, since they've already done the Power4 -> 970 conversion, perhaps a Power5 -> 980 conversion would be shorter. They may learn a lot about the process. However, just because a chip is available, doesn't mean Apple will use it. If they are getting yields of the 970's running at 3.6+ GHz in 2 years (double intro speed?), I would assume they'd milk it until it's dry. After all, once they eek out enough performance from the 970, they can migrate it to lower end systems, cost permitting, and slap in a 980 in the high end.

The problem is that pushing the GHz of the 970 is not an efficient way to get more power out of the CPU, memory latencies will grow and grow...
That's why IBM will add SMT features to the POWER5, it will hide memory latencies. Since the PowerPC 970 is expected to run at even faster clock speeds than the POWER4 line, it would benefit from SMT sooner. I would not by surprised if IBM introduces the 980 only 4 or 6 months after the POWER5.

 

[Cubeboy]

How fast Intel decides to ramp up the clockspeeds of Prescott (Pentium 5) would depend on how much competition their getting from Athlon 64. Due to .09 micron process and strained silicon, the Prescott will be able to scale to 5.20 ghzs without any problems. However unless Athlon 64 completely blows the roof off performance-wise (which is dubious), I doubt you'll see a 5+ ghz Prescott anytime soon. Remember, the longer you keep a processor on the market, the more money you rake in, constantly upgrading processors cuts into profits, Intel will try to ramp up clockspeeds in such a way as to maximize profits while still offering products that compete very well with AMD's offerings.

 

[Cubeboy]

Regarding PPC970's competitiveness, I have no doubt that it will compete very well with Prescott and Hammer, remember, the current clock speeds are for a .13 PPC970, with .09 micron process, the 970 will be able to scale far higher, probably to at least 3 ghzs making it very competitive to the Prescott.

 

[Snowy_River]

Re: chase one dream with another

Originally posted by esome
great. we're drooling over the power 5 now and the toned-down version of the power 4 that we _hope_ will be in Macs sometime in early 2004 (maybe) is still a pipe dream.

Uh... early 2004? Perhaps you've not been paying attention, but it seems not too far fetched that we'd see the first 970s announced in about a month, as in mid-2003. In any event, from what information there is, it seems an easy piece of reasoning that Apple will use the 970, and that IBM is ahead of schedule with it (originally projected as having production shipments around Q3 or Q4 '03). So, I'm optomistic...


50...

 

[Snowy_River]

Re: Power5 is coming

Originally posted by Swift
I have it on good authority that not only will this happen before the end of the year, but it will blow your socks off, and there's a several-year development plan already in operation. Starting very soon, it'll be back to the days when the G3 COULD toast the Pentium II. And that's just the beginning.

Oh boy, I hope that your 'good authority' is dead on...

Originally posted by Mr. MacPhisto
...Knowing IBM and their amazing R&D, I'd say that Apple is in for quite a ride over the next few years and should consistently pound Intel chips in performance by this time next year. OS X will likely also increase in efficiency, especially when the Power5 derivative is introduced.

And this is one ride that I'm looking forward to. (I hope that I'm not disappointed by it...)


49...