The future of Powermac is closer than you think
Firewire 2 is making news.
<a href="http://www.oxsemi.com/press/may02/index.html"; target="_blank">http://www.oxsemi.com/press/may02/index.html</a>;
But with that we need an actual computer to take advantage of this new speed. and with Apple opening up licensing on the Firewire brand, they seem to be preparing for something soon... most likely at MWNY. Now to intergrate Firewire 2 into systems they will have to redesign their controller chips. This is not a small update. But it also means something else. The next Powermac must be getting a major overhaul, and not just a switch to the Xserve architecture. Why do i say this? Well for one, the Xserve does not have Firewire 2. Secondly, look at the Xserve arcitecture. It needs cool chips like the G4. My take is that the powermac introduced this summer will have a new processor. i have no idea if it's the G5 but it will be a hot processor because they would have put this new processor in the Xserve if it was possible. But i think that this processor will need some more cooling that a 1U case can provide.
What makes a processor hot? Running at high MHz. What lets you run at high MHz? A small process of munufacturing but mostly an extended pipeline. I would estimate the pipeline to reach 10-11 stages on Hip7. What else makes processors run hot? More execution units? Sure.
Here are my features for the ideal workstation PowerPC:
A 10 stage pipeline to help in the scaling of processor speed.
On die memory controller for fast thruput. But then you get into a problem with multiple processor.... but here's a solution:
get rid of multiple processors. Hyper threading. Create a processor with twice the integer units, twice the fp units, twice the cache. Multiple threads will just be processed on different execution units. This is one step short of dual core chips, which is also entirely possible.
This leaves PCI/ethernet/firewire/usb/etc. with it's own controller, in which MPX running at 133MHz can more than handle. Or to conserve pins (and this means cheaper chip packaging and that means cheaper processors) use Rapid IO. Using RapidIO will make smaller motherboard designs and this can leave more room for more PCI slots or just smaller cases. With no need for L3 cache, the processor can once again be fitted into a ZIF type socket with normal socket heatsinks and fans. The DIMM slots can be placed much closer to the processor and it may even be possible to have 4 slots if designed correctly.
Something is cooking. There is too much activity in Cupertino leading up to MWNY. Not much left to update. When the PowerMacs stop using the 1GHz G4's then the iMac and PowerBook will inherite them. But big things are comming. And they have PowerMac written all over it.
<a href="http://www.oxsemi.com/press/may02/index.html"; target="_blank">http://www.oxsemi.com/press/may02/index.html</a>;
But with that we need an actual computer to take advantage of this new speed. and with Apple opening up licensing on the Firewire brand, they seem to be preparing for something soon... most likely at MWNY. Now to intergrate Firewire 2 into systems they will have to redesign their controller chips. This is not a small update. But it also means something else. The next Powermac must be getting a major overhaul, and not just a switch to the Xserve architecture. Why do i say this? Well for one, the Xserve does not have Firewire 2. Secondly, look at the Xserve arcitecture. It needs cool chips like the G4. My take is that the powermac introduced this summer will have a new processor. i have no idea if it's the G5 but it will be a hot processor because they would have put this new processor in the Xserve if it was possible. But i think that this processor will need some more cooling that a 1U case can provide.
What makes a processor hot? Running at high MHz. What lets you run at high MHz? A small process of munufacturing but mostly an extended pipeline. I would estimate the pipeline to reach 10-11 stages on Hip7. What else makes processors run hot? More execution units? Sure.
Here are my features for the ideal workstation PowerPC:
A 10 stage pipeline to help in the scaling of processor speed.
On die memory controller for fast thruput. But then you get into a problem with multiple processor.... but here's a solution:
get rid of multiple processors. Hyper threading. Create a processor with twice the integer units, twice the fp units, twice the cache. Multiple threads will just be processed on different execution units. This is one step short of dual core chips, which is also entirely possible.
This leaves PCI/ethernet/firewire/usb/etc. with it's own controller, in which MPX running at 133MHz can more than handle. Or to conserve pins (and this means cheaper chip packaging and that means cheaper processors) use Rapid IO. Using RapidIO will make smaller motherboard designs and this can leave more room for more PCI slots or just smaller cases. With no need for L3 cache, the processor can once again be fitted into a ZIF type socket with normal socket heatsinks and fans. The DIMM slots can be placed much closer to the processor and it may even be possible to have 4 slots if designed correctly.
Something is cooking. There is too much activity in Cupertino leading up to MWNY. Not much left to update. When the PowerMacs stop using the 1GHz G4's then the iMac and PowerBook will inherite them. But big things are comming. And they have PowerMac written all over it.
Comments
<strong>.... but here's a solution:
get rid of multiple processors. Hyper threading. Create a processor with twice the integer units, twice the fp units, twice the cache. Multiple threads will just be processed on different execution units. This is one step short of dual core chips, which is also entirely possible.</strong><hr></blockquote>
It would be a bad move for Apple to get rid of MP's. MP's are adventagious for High en Graphics, 3-D, Animation, Aidio and Video, and Servers. Having software that is optamised for them will show an increased
performance on MP systems. The profesional market needs this advantage, and I can see the home market taking advantage of this in the future as home entertainment and computers merge.
Yeah. But the current 'high' end 'power'Mac G4s are way out of date. Painfully so compared to x86 land.
Outsider:
I like your speculation. Interesting.
We've only got a month to go until we find out for sure. I hope Apple have got some secret killer punch performance option for the 'power'Macs. Let's hope they DO put the mhz issue to bed this Summer.
Lemon Bon Bon :cool:
[ 06-04-2002: Message edited by: Lemon Bon Bon ]</p>
[quote]
"If I were to venture a guess, I would guess that Apple will include 1394b as soon as silicon is available. Apple's history as an innovator, and its commitment to FireWire would support this supposition," James Snider, executive director of the 1394 Trade Association, told NewsFactor.
Rumor has it that Apple will introduce new professional desktops at July's Macworld Expo in New York, paving the way for a FireWire revision. But Apple may not be able to finalize 1394b silicon in time to ship the systems by July.
Snider told NewsFactor that the necessary silicon will not be available until the "end of the summer at the earliest."
<hr></blockquote>
<a href="http://www.newsfactor.com/perl/story/17899.html"; target="_blank">Faster Firewire may come to Macs</a>
So the $64 question is will Apple do an interim update at MWNY without FW2 or perhaps announce machines shipping "at the end of the summer"? If they do ship "end of summer" does it make sense to do a substantial rev of the machines for a 3-4 month run to MWSF or might Apple make the leap to G5 for the "end of summer" machines. Of course this all assumes that Loch Ness....er the G5 is indeed going to appear within the next 7 months
<strong>I'm Still not sure Firewire 2 is going to make it in time for MWNY
<a href="http://www.newsfactor.com/perl/story/17899.html"; target="_blank">Faster Firewire may come to Macs</a>
So the $64 question is will Apple do an interim update at MWNY without FW2 or perhaps announce machines shipping "at the end of the summer"? If they do ship "end of summer" does it make sense to do a substantial rev of the machines for a 3-4 month run to MWSF or might Apple make the leap to G5 for the "end of summer" machines. Of course this all assumes that Loch Ness....er the G5 is indeed going to appear within the next 7 months
I doubt we'll see an interim revision. Either it's announced at MWNY and ships late, or it's announced at Seybold. If the latter, I wouldn't be too surprised to hear a pre-announcement such as happened with the Xserve.
If my comments make no sense, ignore them. I'm pretty ignorant about chip specs.
So the next PowerMac case could ditch the fan - to replace it with A/C, wouldn't that be cool?
ZoSo
<strong>The next request would be for an MP Powermac with these super-processors.</strong><hr></blockquote>
...in clusters.
And then it all on one chip
<strong>The next Powermac must be getting a major overhaul, and not just a switch to the Xserve architecture. Why do i say this? Well for one, the Xserve does not have Firewire 2. Secondly, look at the Xserve arcitecture. It needs cool chips like the G4. My take is that the powermac introduced this summer will have a new processor. i have no idea if it's the G5 but it will be a hot processor because they would have put this new processor in the Xserve if it was possible. But i think that this processor will need some more cooling that a 1U case can provide. </strong><hr></blockquote>
You may be right, but your logic is faulty. Have you considered the idea that new chips weren't used in the Xserve because they weren't ready? And rumor has it, as somebody else posted, that FireWire 2isn't, either.
New Powermacs:
Probably new form factor, faster bus, DDR RAM, *possibly* FW2/USB2, though unlikely.
G4, no breathtaking MHz breakthrough.
That's all.
-s*
<strong>Just a thought - would those nasty multicore POWER2 processors IBM uses in the Regatta servers be binary compatible with current PowerPC code? They're probably hot as hell, but their performance is screaming - even without AltiVec units...
</strong><hr></blockquote>
The Power4 processors in the Regatta aka P690 servers are Power processors. The standard PowerPC instruction set is a subset of the Power instruction set. This means that standard PowerPC instructions run on Power machines, but there are some Power instructions not supported by PowerPC. (You should be able to run G3 binaries on a Power4, but not all Power4 binaries on a G3/G4.) Then, there's the Altivec/Velocity superset on top of the PowerPC instruction set -- this means not all G4 binaries will run on a Power4. Typical G4 binaries have branches when they detect a G3 so they can run non-altivec code, so it should run on a Power4.
The end of all this is, the Power4 is a throughput monster. You think the G4 is constrained by memory throughput, you have no idea how much the Power4 would choke in that machine. I don't think a consumer could afford a mainboard with enough throughput to main memory. I'd rather see a pared down, single core Power4 -- cheaper, cooler, still faster than Intel/AMD. Who knows if we'll ever see that, though. It's undoubtedly a very expensive proposition.
<strong>Alti-Vec is great where it is used, but not everything uses it. If Apple can use Quartz Extream to accelerate graphics performance in the high-end computers, and get a faster chip from IBM without Alti-Vec than what Moto can produce then they might be able to ditch Alti-Vec on high end computers.</strong><hr></blockquote>
Producing a fast chip is one thing. Producing a fast chip that is affordable is another. Yes, the Power4 is a fast chip. Yes, it wipes the floor with Intel and AMD offerings. No, they do not compete for the same market. The Power4 is an example of a "cost is no object" type of chip. It has a MUCH larger die size it is much more expensive to produce. Sure trade-offs could be made to reduce the cache size and add a SIMD unit, etc. AIM is quite capabable of creating a chip that would blow away Intel and AMD. The problem is that there is little incentive for either IBM or Motorola to do so. Apple's market just isn't big enough to justify it. PPC was originally designed to be an alternative reference platform, not just new chips for Macs. After the realization that a common reference platform based on PPC wasn't going anywhere, both IBM and MOT focused on the embedded CPU market. As it turns out, the PPC makes a great embedded CPU and is a "capable" desktop processor.
The bottom line is that unless Apple takes over the design of an Apple specific PPC, things aren't likely to change in terms of Mac performance relative to Wintel. The BookE architecture seems to be very modular by nature, so there is a possibility of a G5 based PPC to suit Apple's needs better. At the same time, things aren't as bad as many of the whiners in this forum make it out to be. The G4 is capable at the low end and the dual CPU strategy is keeping them afloat at the high end. No, Apple is not the the performance leader it once was. We'd all like them to reclaim that glory, but Apple isn't exactly going out of business due to performance issues as some people here seem to suggest.
Steve
Would it be possible to have an external/backside Altivec co-processor? This way, IF IBM or someone else has a PowerPC compatible chip that smokes anything coming out of Moto, could Apple use these chips and add a Mot Altivec-only coprocessor to keep the advantage of Altivec for optimized code? This would obviously need some type of controller interpretting instructions and diverting Altivec code to the coprocessor. Would this be too much overhead, thereby slowing things down too much?
Is this even possible? Cheers!
<strong>Would it be possible to have an external/backside Altivec co-processor?</strong><hr></blockquote>
Well, let's say a POWER4 (or a hypothetical low cost version of it) + AltiVec external co-processor - I think this would pose the problem: "what kind of bus would then Apple use to connect the two units?"
No, I think a solution like this (apart from the fact that the POWER4 is unsuitable to low-cost desktop processing, as has been pointed out) would throw in only more hurdles for Apple to overcome...
I only suggested the POWER4 as an ironical hypothesis - but I'd love to see a Regatta running Jag-wire!
ZoSo
<strong>
No, Apple is not the the performance leader it once was. We'd all like them to reclaim that glory, but Apple isn't exactly going out of business due to performance issues as some people here seem to suggest.
Steve</strong><hr></blockquote>
No, but they're slowly bleeding market-share through a combination of pricing and performance issues. Don't just look at Steve's tenure, think in terms of the last 18 years. The trend is a very slow but definite decline in market-share. They may be beginning a slow upswing, but without better prices (on better performance) it won't last.
I disagree. It is sharply and cruelly exposed compared to a 2.1 gig Athlon Xp. 1 fpu vs 3 fpu. Do the math. (AMD's high end vs Apple's low end. AMD's top end? Priced cheaper. And you get more DDR ram, bigger hard drive and a flaming monitor...)
"and the dual CPU strategy is keeping them afloat at the high end."
I don't think so. Any 3D x86 workstation beats the snot out of the dual 1 gig G4. Show me one bench mark where a dual G4 takes a dual Athlon 2.1 Xp multiprocessor or single (for that matter) pro' system? Then there's price. Heh. I know. It's not funny.
"No, Apple is not the the performance leader it once was."
No. It certainly isn't. No DDR, ancient bus, same G3 with bells on processor and 'power'Mac prices that have remained static and way over priced when most other PC makers have expandable and powerful towers in most price brackets.
If Steve Jobs wants to liken Apple to Mercedes or BMW, then he better drop in the appropriate 'engine' into a 'power'Mac to keep spouting those kinda analogies.
"We'd all like them to reclaim that glory,"
The only thing we agree on. I'm waiting Apple. My money is here when you catch up. Until then, I can limp along with a low end Athlon.
I'll pay the premium all things being relatively equal.
"but Apple isn't exactly going out of business due to performance issues as some people here seem to suggest."
Not anytime soon with $4 billion or so in the bank. Yeah, yeah. It buys them time. Great OS, nice iApps. Right now, Apple make a compelling software startup. Hardware company? Great plastics. Poor out of date specs. If the 'new' imac was beige apple would be in trouble. The spec list would look overpriced for what it is.
But if they want to convince many of those 95% Wintel owners to switch...'X' and free software and posey plastic swivelling gimmicks aint going to do it alone. It may stem the market share bleed...but for how long? I know most of the people I talk to about the Mac's advantages suddenly 'glaze' over when they look at the clock speed. Real or perceived (and anybody who thinks a 900ish G4 can take a 2.1 XP Athlon needs help...) the current 'power'Macs are an out of date and pricey joke.
With Matsu on this.
"No, but they're slowly bleeding market-share through a combination of pricing and performance issues. Don't just look at Steve's tenure, think in terms of the last 18 years. The trend is a very slow but definite decline in market-share. They may be beginning a slow upswing, but without better prices (on better performance) it won't last."
Agreed.
Lemon Bon Bon <img src="graemlins/bugeye.gif" border="0" alt="[Skeptical]" />
PS. Apple will have to do something about it soon given the 'Eighth generation' AMD chip is due to ship this later this year/early next year. On hypertransport. Having been to AMD's website...and seen how compelling, open and positive they are regarding their future processor strategies...I hope Apple have something better than a p*ss poor dual 1.2 G4 on p*ss poor DDR 266 memory.
7500. Soon. That's my bit for future hardware.
[ 06-05-2002: Message edited by: Lemon Bon Bon ]
[ 06-05-2002: Message edited by: Lemon Bon Bon ]</p>