I was even more stoked when I saw Intel's future roadmap, and integer 'performance per watt'
It is of course a completely meaningless metric.
IBM 440 PowerPC chips yield extremely good performance per watt for example but they aren't going in Macs. It depends which end of that ratio you attach importance. 'Performance per watt' lets Apple weasel out of delivering any performance.
I can hear Jobs now - 'Hey, it may not be faster than a G5 but you can now use your Powerbook for 16 hours straight and ok, it takes 3 times longer to encode video now, but it's whisper quiet'.
The current Intel roadmap isn't very exciting at all although at least they have one, whereas IBM's isn't forthcoming. Dual Yonah is coming in at 2.16Ghz, 667 FSB and using 31W so they aren't really useful for desktops and neither for laptops.
If you watch the WWDC 2005 keynote in Quicktime, Steve Jobs says, "I think a lot of you would like a G5 in your PowerBook, and we haven't been able to deliver that to you yet."
Key word there is "yet."
Jobs continues, "But these aren't even the most important reasons. The most important reasons are that as we look ahead, though we have great products right now...and we have some *great* PowerPC products still yet to come..." and Jobs then proceeds to talk about the future of the PowerPC roadmap vs. Intel's.
So the way I parse the whole thing is that the near term is OK for the G5s, but long-term it's not as good as Intel's will be. That leaves the door open for the PowerBook G5, even it's just for a year or so.
Watch the keynote, and it's about 21 minutes in. See for yourself.
Yes, this was all in my memory as well. We never know what they plan, and it might be that the Powerbooks are not going to be the first to get Intel processors, which would make a year or so with a G5 even more sensible. I'm really looking forward to seeing what direction they go this fall.
Some of you guys are talking like you think Apple hasn't known about these IBM chips since their inception. Of course Apple has known about them, and they know about the next ones too... and it is because of these ones and the next ones (compared to Intel's roadmap) that Apple decided to switch to Intel.
This "low power" 970 is a yawner, if you ask me. No additional cache, no lower power bus, and no other performance improvements. They tweaked the process and downclocked, from the looks of it. At 1.6 GHz w/ 512K L2 this 970 will get stomped by the current Pentium-Ms, nevermind what Intel has in the pipeline. I doubt these 970s will even do that well on typical Mac software compared to the 7448 at the same power levels. Better on media intensive stuff, but lower clock rate really hurts the 970 -- its designed for high frequencies and lots of bandwidth.
The 970MP is the interesting announcement -- two cores and double the cache per core. That is probably why the PowerMac is switching to Intel last.
Apple should play up the qualities of their machines and software, and downplay the cpu's. Make people concentrate on what's important.
Tell that to the people buying powermacs. The CPU matters. You can't just woo people with software. You need something to back it up, and if this no-hardware-improvement trend continues, and software development keeps up, then eventually we won't be able to run that software on our pretty, underpowered machines. Actually, this is already happening, alebit minor (Dashboard ripple effect in MacMini).
The hardware has to keep up with the software, and right now that's not happening, so that's where Intel comes in.
I can tell you that this statement is 100% wrong. Yonah is already done and is sampling at Intel (has been that wasy since March). Jobs said that they would have their first line of Intel based Macs BY the spring 06' conference. To say that it will be July at the earliest flies right in the face of all of the facts.
Bottom line is Intel already ramping up production and could launch Yonah fall of 05' but are waiting for Q1 06' because that is when they said it would launch. They gave that timeframe expecting the same problems from 130 nm to 90 nm. bottom line is 90 nm to 65 nm was been one of the easiest transitions in recent Intel history.
OEMs are going to be getting early Yonah generation centrinos (dual core and single core Pentium Ms, late beta chipsets, and the a/b/g chip minus the WiMax pretty soon). Since it takes anywhere from 2 to 6 months for companies to make products from the samples we can expect the market to be flooded with Yonah Centrino right from the launch date. They just need to make a few more tweeks with the chipsets and get the last of WiMax intigrated with their wireless chip.
It would be shocking if Apple couldnt produce a Yonah product in Q1 06'. If they couldnt that would meen they are having a problem with software or that they are just so new with Centrino they are needing a lot of outside help.
Well, again it's the Dev conf, not July, that was a mistake.
Don't forget that when Apple said that 10.4 would be out in the first half of 2005, people were also predicting that it would be out in Jan. Even the Powerpage said January (but he's NEVER right!).
Look when it did come out though. Almost at the very end. I suspect that it would have come out during the conf. but Jobs didn't want to detract from the Intel announcement. That's why it's so buggy. They rushed it at the last instant.
My feeling is that we will see the same schedule here. Shortly before the conf.
There is really no reason for Apple to rush this. He has to give time for his own team to get the Intel version of the OX finished. He also has to give the developers time to get their work done.
This transfer MUST go smoothly. No excuses. Apple will also, no doubt, be working on Rosetta, to make it work better, possibly to let work with more programs. Does it use both processors? If not, it could be a problem.
There are a lot of issues that aren't being considered here. It's not only the chips. Does Apple have a new case design? Will they go to Express? That bus is also for mobile use. Just because Apple has been considering going to x86 doesn't mean that these, and other, problems have been worked out yet. It's very possible that with Apple's concerns for secrecy, they haven't let any plans out to manufacturing.
We just don't know. It's not worth arguing that they will, or won't have a machine out by January.
Why don't we try to figure out what they will contain instead? That's really more interesting anyway.
Tell that to the people buying powermacs. The CPU matters. You can't just woo people with software. You need something to back it up, and if this no-hardware-improvement trend continues, and software development keeps up, then eventually we won't be able to run that software on our pretty, underpowered machines. Actually, this is already happening, alebit minor (Dashboard ripple effect in MacMini).
The hardware has to keep up with the software, and right now that's not happening, so that's where Intel comes in.
When the word "consumer" is used, it doesn't mean "pro". We have four Powermacs at home, and in my business we had eighteen.
Mom and dad going to the Apple store aren't looking to compare a P4 (or M) to the G5 or G4. They just want what seems to be snappy, does what they want, looks good, and isn't subject to internet garbage.
At least that's what the surveys of people buying these machines say.
And again i'll state, from someone who has bought a lot of Powermacs over the years; we buy our Macs based on the premise that we WANT Macs as opposed to Windows machines. We compare a new Mac to the older ones we have.
If x86 machines are MUCH faster, maybe we would consider them instead. But they have to be MUCH faster.
I can tell you from inside the media industry that the reason that Powermac sales have fallen is mostly because people are sitting on their current models, not because they are rushing out and buying XP. I'm sure that this has happened in a few cases, but not many. And those are usually those who were not comfortable with Macs anyway.
In the future Apple sees the x86 line getting MUCH better than the PPC. That's a problem So they are switching.
If you watch the WWDC 2005 keynote in Quicktime, Steve Jobs says, "I think a lot of you would like a G5 in your PowerBook, and we haven't been able to deliver that to you yet."
Key word there is "yet."
Jobs continues, "But these aren't even the most important reasons. The most important reasons are that as we look ahead, though we have great products right now...and we have some *great* PowerPC products still yet to come..." and Jobs then proceeds to talk about the future of the PowerPC roadmap vs. Intel's.
So the way I parse the whole thing is that the near term is OK for the G5s, but long-term it's not as good as Intel's will be. That leaves the door open for the PowerBook G5, even it's just for a year or so.
Watch the keynote, and it's about 21 minutes in. See for yourself.
I don't think the one word "yet" outweighs the fact that Apple would have to engineer a PowerBook first for the G5 and then for an Intel so quickly. It just doesn't make sense, especially given that there's good reason to believe that the PowerBooks will be among the first Macs to go Intel.
I don't think the one word "yet" outweighs the fact that Apple would have to engineer a PowerBook first for the G5 and then for an Intel so quickly. It just doesn't make sense, especially given that there's good reason to believe that the PowerBooks will be among the first Macs to go Intel.
This is really hard to say. There is so much at stake here. Apple will lose a lot of sales over the next year or two.
We don't know how close they are to having a G5 plan ready. It might not be as bad as is thought. It might be worth it to Apple to have a G5 machine just to slow down the deterioration in portable sales. If they spend $50 million or even $100 million but save $500 million to a billion in sales, then it's worth it. Apple also has to show that they are doing something. If the performance is even 25% better than it would otherwise have been, then it might be worthwile.
These costs have to be thought of as being advertising, or publicity. If the 7448 isn't much better than the 7447a, they might have to take steps. Just getting a G5 in might be enough. G5 performance is about 30-50% better than a G4 clock for clock. And with no L3 on the 744x series, it could be more. Then bus speed becomes even more important.
G5 performance is about 30-50% better than a G4 clock for clock. And with no L3 on the 744x series, it could be more. Then bus speed becomes even more important.
The G5 doesn't have L3 cache either.
And the 7448 has twice the L2 cache of the G5 970fx.
7448 has a 200Mhz FSB and you can probably bet on a portable G5 being throttled back for power concerns.
IMHO, everything points to a 7448 update. I'd not be surprised if we see a 1.8 to 2.0Ghz 7448 appearing next in Powerbooks and iBooks and that they'll easily out perform a G5 at everything apart from manipulating large files and video - something you don't expect to be fast on a laptop anyway with their slow drives and limited GPUs.
Of course, I'm prepared to be wrong. I was totally off with switching to Intel.
And the 7448 has twice the L2 cache of the G5 970fx.
7448 has a 200Mhz FSB and you can probably bet on a portable G5 being throttled back for power concerns.
IMHO, everything points to a 7448 update. I'd not be surprised if we see a 1.8 to 2.0Ghz 7448 appearing next in Powerbooks and iBooks and that they'll easily out perform a G5 at everything apart from manipulating large files and video - something you don't expect to be fast on a laptop anyway with their slow drives and limited GPUs.
Of course, I'm prepared to be wrong. I was totally off with switching to Intel.
Yes, of course the G5 has no L3 cache, what's your point? It never had L3 cache. That's why the comparison is so interesting. It performs as well as it does vs a vs the G4 WITH cache, despite that. so they will slow the bus down to 1/3rd instead of 1/2. That hasn't taken a real bite out of its performance. Even if they dropped to 1/4 it wouldn't matter that much. It's power requirements closely match the 7448's. I don't agree with your performance expectations, at least not from the specs. that we know so far.
If Apple clocks the 7448 up, the power requirements will go up with it as well, and not neccessarly in a linear way either. It seems that when chips are clocked out of their factory specs they don't behave the way they do otherwise. so a clocked up 7448 would use more power then a 1.6 G5 would. If that's the case, and Apple is willing to live with it, then they could clock the G5 up as well.
What I'm saying is that we don't know what Apple is doing, or what their thinking on this is. You're being conservative. Nothing wrong with that. But these are "interesting times". I wouldn't be surprised if Apple stretches out as far as they can on this one.
In a previous life I was a partner and designer in a Pro and hi-end audio firm, Magnum Opus. We recieved advanced info on many components from manufacturers. Oftentimes more than a year in advance. We would start preliminary work on products based on those specs. Sometimes the specs would change, for better or worse. So did our plans for them.
I'm sure that Apple has been working on a G5 plan for quite some time. The only question is how far advanced it is, and whether these chips will make the cut. Depending on where both Apple and IBM are on this, we could see an announcement as early as Janurary, to be available in Feb or March. It wouldn't be the first time they had a month (or three) lead time.
I don't think the one word "yet" outweighs the fact that Apple would have to engineer a PowerBook first for the G5 and then for an Intel so quickly.
Your supposing that Apple is just finding out about the low-power G5s through IBM's recent announcement in Japan. More than likely Apple has long been playing around with samples and engineering them into prototypes. We knew of last summer the "PowerBook 7,1" and the "PowerBook 7,2." They appeared in a plist file of the 10.3.5 update as you may recall, around the same time that we found out the next iMac was based around the G5. Now that info is encrypted in the plist file.
A lot of people assumed that the PowerBooks represented were the 15-inch and 17-inch PowerBooks. But it is noteable that the 12-inch PowerBook and the 12-inch iBook G4 are in the same family, relative to model numbers. So the PowerBook 7,1 and PowerBook 7,2 fits nicely into the Digitimes rumor of PowerBook G5s and iBook G5s.
We are still in the quarter referenced in the Digitimes article that Apple would be shipping these so-called PowerBook G5s and iBook G5s, we are the heels of IBM's official announcement of the 970MP and the low power 970FX, Apple announced a weak PowerBook update in January, and iBooks haven't been updated since last October. Something's gotta be coming. Something better be coming.
I'm just creating some hype here. Hell, we have had a dearth of exciting product announcements from Apple lately. We could use anything to lift up our spirits!
"The first PowerBook to sport an Intel processor is not expected until July 2006 at the earliest."
This should read:
"The first PowerBook to sport an Intel processor is expected between January and July 2006."
Close, but I disagree. It should read
"The first PowerBook to sport an Intel processor is expected by July 2006 at the latest."
That matches what Steve said.
He also said "We want the transition to be mostly complete by WWDC 2007", and fully complete by the end of 2007 - and described that as "a 2 year transition" - which implies some Intel based Macs early next year.
He may be counting from his announcement, but that makes the transition a 2.5 year transition. He may have meant "18 month transition) but was being general. I guess it won't be this year because he says the 2 year transition is 2006-2007.
I don't think the intel PB will appear till MWSF 2007. The PBs are too important to pro users. If there is some problem making an intel chip perform in a Mac like a Mac should then you want to see that problem in the Mac Mini or the iBook first.
Suppose the Mac Mini with intel comes out May of 2006. Then an iBook comes out in September of 2006. Then the PB could come out in January, shipping same day as the announcement. This would give Apple six months experience with the intel chips in the field both in the Mac Minis and in a portable. Six months to see if there are problems with Rosetta, Quark, Adobe CS, Photoshop, Applescript, printing, sleep, wakeup, CPU usage, WiFi connections, Firewire, etc.
If there is some goofy bug in the compiler that causes an unexpected problem I want to see that in a Mac Mini that I might use as a kiosk or for light surfing rather than on the PB that I use to earn a living.
Presumably we'll see Leopard arrive then too. There may be some benefit in delaying the new PB to coincide with that as well.
If we get a 7448 update this fall and possibly a dual core G4 next spring with a speed bump later in the year that wouldn't be too bad. Remember, in addition to the CPU changes we'll be seeing bigger, faster HDs, possibly more FW800 ports, a better display, better battery life, nicer GPU. What's not to like?
"The first PowerBook to sport an Intel processor is expected by July 2006 at the latest."
That matches what Steve said.
He also said "We want the transition to be mostly complete by WWDC 2007", and fully complete by the end of 2007 - and described that as "a 2 year transition" - which implies some Intel based Macs early next year.
He may be counting from his announcement, but that makes the transition a 2.5 year transition. He may have meant "18 month transition) but was being general. I guess it won't be this year because he says the 2 year transition is 2006-2007.
While it might be nice to see an Intel machine in early 2006, I don't believe it for all of the reasons I've given before, and because it's too soon for an entirely new machine to come out. There are still issues. The dev machines use a BIOS chip. It's believed that Aple won't use that, but will instead use Intel's equivalent of Open Firmware. Too many issues for six or seven months. Apple doesn't do things quickly. It can take Apple two or more years to come out with a new mobo, for example. It takes the far eastern manufacturers six months. Even if the board will be from Intel, and not an Apple design, it still has to be done and tested, sent to manufacturing etc.
I think that March would be pushing it. Again, my other reasons above.
Yes, of course the G5 has no L3 cache, what's your point? It never had L3 cache.
That's exactly my point. Arguing about L3 cache is pointless as neither the 744x or G5 have one. And neither will have, especially in a laptop.
The 7448 has a 1MB L2 - twice as big as the G5 and previous G4 7447a. That is way more significant than L3. IBM really needed to up the L2 on the low power G5 but then it would consume more power. Cache is expensive power wise.
When the 750fx went to 1MB L2 cache, it rocked and in some ways was faster than a similarly clocked G4 in the iBook. If it wasn't for Altivec, it quite possibly would have been a quicker chip all told in general use.
That kind of worries me about going Intel too - Having used 500Mhz G3 and 400Mhz G4 at the same time, you notice how much faster Altivec makes things. It may not matter up at 2Ghz I guess but it's niggling.
Quote:
Originally posted by melgross
That's why the comparison is so interesting. It performs as well as it does vs a vs the G4 WITH cache, despite that.
Compare the G5 2.0Ghz SP tests with G4 upgrade speeds. Neither have L3. That's an overclocked 1.3Ghz 7447A running at 2.0Ghz on a 133Mhz FSB and slow AGP slot compared to a 2.0Ghz G5 PowerMac running a 1Ghz FSB and faster AGP slot. It's never more than 25% behind the G5 in the CPU bound tests and even draws with it on Photoshop SP actions. Your assertion that it's 30-50% behind is plain wrong. Tests do not bear that out.
Now, with the 7448, double the L2 and raise the FSB to 200Mhz. Remember we're sticking this in a laptop so drop the G5 to say 533Mhz like an iMac and run the machine in auto power mode instead of Highest all the time which knocks off about 30% CPU score anyway. IMHO, the G4 may beat the G5 if not be very close. So close it really doesn't matter if you go G4 or G5. Look at the 1.4Ghz results there with 2MB L3 - it comes second to a G5 in the multi-processor aware Photoshop tests. 1MB of L2 is real nice - better than 2MB L3 in most uses as you don't have to go off chip.
The G5 will win in memory intensive applications and shoving data at the GPU but in a laptop the GPU is usually throttled back with slower RAM.
The G4 will win on disk intensive tasks because Apple's G5 disk controller sucks. The G4 systems are about twice as fast at disk access.
As an other example, how about Powerbook G4 1.5Ghz v 1.8G5 iMac...
If that PB was 1.8Ghz, it'd be pretty closely ran. The Motion test is an oddity because Motion uses the GPU. iMac loses because it's GPU is very slow, by comparison.
I've suggested to Rob at Barefeats that he runs a 1.67 G4 Powerbook v 1.6Ghz G5 PowerMac v 1.6/1.7 Pentium M test. Might be interesting to see cycle-for-cycle who wins then add in the 7448 if that ships and faster Pentium Ms and G5s. Since Intel is heading down the path of using the Pentium M core for most of it's roadmap it may be nice to see where we're heading also. They're even using Yonah on low end Xeons next year.
Quote:
Originally posted by melgross
so they will slow the bus down to 1/3rd instead of 1/2. That hasn't taken a real bite out of its performance. Even if they dropped to 1/4 it wouldn't matter that much. It's power requirements closely match the 7448's. I don't agree with your performance expectations, at least not from the specs. that we know so far.
I was basing my guesses on actual benchmarks not the specs. There's a bit of extrapolation and guestimation but I think they'll be broadly comparable. Each CPU as pluses and minuses.
Quote:
Originally posted by melgross
If Apple clocks the 7448 up, the power requirements will go up with it as well, and not neccessarly in a linear way either. It seems that when chips are clocked out of their factory specs they don't behave the way they do otherwise. so a clocked up 7448 would use more power then a 1.6 G5 would. If that's the case, and Apple is willing to live with it, then they could clock the G5 up as well.
The low power 970fx is 16w at 1.6Ghz - no faster speeds have been announced and we know already that the 970fx at 2.0ghz is too hot for a laptop. We've also no idea what they've done to get it low power - is it low voltage? 200Mhz FSB? No L2? So why no low power 2.0Ghz?
Freescale said the 7448 at 1.7Ghz consumes 15w. It's already cooler than the low power 970fx. The 7447a is listed as consuming 20W at 1.42Ghz (they don't list the 1.67). There's obviously room there to grow.
For comparison the Pentium M at 1.4 is 18W typical.
Quote:
Originally posted by melgross
What I'm saying is that we don't know what Apple is doing, or what their thinking on this is. You're being conservative. Nothing wrong with that. But these are "interesting times". I wouldn't be surprised if Apple stretches out as far as they can on this one.
Perhaps, but I don't see any technical reason, or financial reason for Apple to come out with a whole new architecture just before they scrap it and go Intel. It just doesn't make any sense. Better to just apply minor tweaks to the current one. It'd be majorly expensive to support a G5 powerbook.
Quote:
Originally posted by melgross
In a previous life I was a partner and designer in a Pro and hi-end audio firm, Magnum Opus. We recieved advanced info on many components from manufacturers. Oftentimes more than a year in advance. We would start preliminary work on products based on those specs. Sometimes the specs would change, for better or worse. So did our plans for them.
In a previous life I wrote compilers, run-time systems and native code generators for various, often prototype CPUs. I worked on a Pentium system before they were released. Worked on OSs before they were even beta release, even debugged the source of one particular OSs filesystem at one point. I'm sure I can say that now but at the time I was NDAd up to my eyeballs by Intel, Sun, DG, Digital, Microsoft and even ARM at one point.
So I'm aware how changing architectures is a major thing you don't do readily. Doing it twice in two years - suicide.
Quote:
Originally posted by melgross
I'm sure that Apple has been working on a G5 plan for quite some time. The only question is how far advanced it is, and whether these chips will make the cut. Depending on where both Apple and IBM are on this, we could see an announcement as early as Janurary, to be available in Feb or March. It wouldn't be the first time they had a month (or three) lead time.
Or you could be right, and it could be a 7448.
If they announce for March next year, then I'd expect it'll be Intel based.
But in the meantime a 7448 based Powerbook or iBook can't be far off. iBook hasn't been updated since October last year - now almost 4 months overdue an upgrade. Powerbooks are a month away from their average update cycle. Even if they only stick in the 1.7Ghz quoted by Freescale, with the faster bus and 1MB L2, it's an ok upgrade. Dropping say 8-9W off the power requirement is nice too. You could see 1.7Ghz 12" powerbooks and those would rock. I'd buy one in a flash. If I was to guess, we're a month or so away from 7448 updates. I'd also guess, as usual, Freescale has faster than the announced to the embedded market versions for Apple. 2Ghz perhaps? One can hope. I'm sure people will still moan it's a G4 even though it may actually be faster and cooler than a G5.
But anyway, I'll wait for Apple to prove me entirely wrong again.
Job's said the first machines would be out by Sping (WWDC) 2006.
My guess for the 1st machines would either be the iBook and/or Mac mini. The the biggest bet on the Mac mini. My main reason is performance. The iBook & Mac mini have the least performance-critical customers. Thus, "Rosetta" might work for them. Especially since these customers can be new (no previous comparison or a VERY old computer to compare against). Could be a mini at MW SF, an iBook in May, and so on?
I don't think a PB customer will buy until the Intel box runs their sw faster than their currrent version (and probably not quicker than the 1.67 current high end). So the PB is contingent on it being faster, even with Rosetta, AND having the major power apps (Final Cut Pro, Photoshop, MS Office, etc) Intel native (fat/univeral binary). I think the same is true for PowerMacs, but the hurdle is higher (likely 2 dual 970 MP's vs a single 1.7 G4), so they'll be out the latest. iMacs would, be in the middle (less performance sensitive, but still can come out THAT much slower than the current line).
But what if the Intel-based Macs run at such better performance with native apps that the slowest of the Rosetta emulated apps run better on Intel than they do on native PowerPC hardware?
But what if the Intel-based Macs run at such better performance with native apps that the slowest of the Rosetta emulated apps run better on Intel than they do on native PowerPC hardware?
That's exactly my point. Arguing about L3 cache is pointless as neither the 744x or G5 have one. And neither will have, especially in a laptop.
The 7448 has a 1MB L2 - twice as big as the G5 and previous G4 7447a. That is way more significant than L3. IBM really needed to up the L2 on the low power G5 but then it would consume more power. Cache is expensive power wise.
When the 750fx went to 1MB L2 cache, it rocked and in some ways was faster than a similarly clocked G4 in the iBook. If it wasn't for Altivec, it quite possibly would have been a quicker chip all told in general use.
That kind of worries me about going Intel too - Having used 500Mhz G3 and 400Mhz G4 at the same time, you notice how much faster Altivec makes things. It may not matter up at 2Ghz I guess but it's niggling.
Compare the G5 2.0Ghz SP tests with G4 upgrade speeds. Neither have L3. That's an overclocked 1.3Ghz 7447A running at 2.0Ghz on a 133Mhz FSB and slow AGP slot compared to a 2.0Ghz G5 PowerMac running a 1Ghz FSB and faster AGP slot. It's never more than 25% behind the G5 in the CPU bound tests and even draws with it on Photoshop SP actions. Your assertion that it's 30-50% behind is plain wrong. Tests do not bear that out.
Now, with the 7448, double the L2 and raise the FSB to 200Mhz. Remember we're sticking this in a laptop so drop the G5 to say 533Mhz like an iMac and run the machine in auto power mode instead of Highest all the time which knocks off about 30% CPU score anyway. IMHO, the G4 may beat the G5 if not be very close. So close it really doesn't matter if you go G4 or G5. Look at the 1.4Ghz results there with 2MB L3 - it comes second to a G5 in the multi-processor aware Photoshop tests. 1MB of L2 is real nice - better than 2MB L3 in most uses as you don't have to go off chip.
The G5 will win in memory intensive applications and shoving data at the GPU but in a laptop the GPU is usually throttled back with slower RAM.
The G4 will win on disk intensive tasks because Apple's G5 disk controller sucks. The G4 systems are about twice as fast at disk access.
As an other example, how about Powerbook G4 1.5Ghz v 1.8G5 iMac...
If that PB was 1.8Ghz, it'd be pretty closely ran. The Motion test is an oddity because Motion uses the GPU. iMac loses because it's GPU is very slow, by comparison.
I've suggested to Rob at Barefeats that he runs a 1.67 G4 Powerbook v 1.6Ghz G5 PowerMac v 1.6/1.7 Pentium M test. Might be interesting to see cycle-for-cycle who wins then add in the 7448 if that ships and faster Pentium Ms and G5s. Since Intel is heading down the path of using the Pentium M core for most of it's roadmap it may be nice to see where we're heading also. They're even using Yonah on low end Xeons next year.
I was basing my guesses on actual benchmarks not the specs. There's a bit of extrapolation and guestimation but I think they'll be broadly comparable. Each CPU as pluses and minuses.
The low power 970fx is 16w at 1.6Ghz - no faster speeds have been announced and we know already that the 970fx at 2.0ghz is too hot for a laptop. We've also no idea what they've done to get it low power - is it low voltage? 200Mhz FSB? No L2? So why no low power 2.0Ghz?
Freescale said the 7448 at 1.7Ghz consumes 15w. It's already cooler than the low power 970fx. The 7447a is listed as consuming 20W at 1.42Ghz (they don't list the 1.67). There's obviously room there to grow.
For comparison the Pentium M at 1.4 is 18W typical.
Perhaps, but I don't see any technical reason, or financial reason for Apple to come out with a whole new architecture just before they scrap it and go Intel. It just doesn't make any sense. Better to just apply minor tweaks to the current one. It'd be majorly expensive to support a G5 powerbook.
In a previous life I wrote compilers, run-time systems and native code generators for various, often prototype CPUs. I worked on a Pentium system before they were released. Worked on OSs before they were even beta release, even debugged the source of one particular OSs filesystem at one point. I'm sure I can say that now but at the time I was NDAd up to my eyeballs by Intel, Sun, DG, Digital, Microsoft and even ARM at one point.
So I'm aware how changing architectures is a major thing you don't do readily. Doing it twice in two years - suicide.
If they announce for March next year, then I'd expect it'll be Intel based.
But in the meantime a 7448 based Powerbook or iBook can't be far off. iBook hasn't been updated since October last year - now almost 4 months overdue an upgrade. Powerbooks are a month away from their average update cycle. Even if they only stick in the 1.7Ghz quoted by Freescale, with the faster bus and 1MB L2, it's an ok upgrade. Dropping say 8-9W off the power requirement is nice too. You could see 1.7Ghz 12" powerbooks and those would rock. I'd buy one in a flash. If I was to guess, we're a month or so away from 7448 updates. I'd also guess, as usual, Freescale has faster than the announced to the embedded market versions for Apple. 2Ghz perhaps? One can hope. I'm sure people will still moan it's a G4 even though it may actually be faster and cooler than a G5.
But anyway, I'll wait for Apple to prove me entirely wrong again.
I'm familliar with the Bare Feats tests. Frankly I've never found then to be all that accurate. They don't agree with what I've found on my own machines. A number of tests that he's done aren't processor bound anyway, or depend upon the GPU more than the cpu.
I also don't agree that going to 1MB L2 is way more significant than 2MB of L3. It certainly helps, but L3 isn't used because of power requirements, not because it isn't of as much use. I agree that 1MB of L2 on the G5 would be better. a larger cache is always better. Intel has shown that with the Xenon's. But again it's probably power requirements. I felt that IBM shouls have gone to 1MB with the 2.7's, but again, it might have been too hot.
Altivec is a stopper. No doubt about that. That's one thing that bothers me about the switch. But rendering tests using two 3.6 Xenons vs two 2.5 G5's show the Xenons to be slightly faster, so maybe we're worrying about nothing. Of course an on board mem controller ala Opteron would help even more.
As I said, I've never gotten results like that so I can't speak for them. MacWorld's tests, for example don't show as close a result as his either. Neither do others that I've seen.
We do know that the low power FX has 512KB of L2 because the press release said it did. FSB? low voltage? Though low voltage seems most likely. It could be a combination, but we'll have to wait.
This isn't a new architecture for Apple after all. It would be a new mobo, but they have two years of experience with G5's, and have no doubt been working on this for a while now. It just might need to go to production.
I really doubt very much that a 7448 machine would be faster than the 1.6 G5 unless they clocked it. something Apple has done a number of times in the past.
I'm not saying that Apple would do this. But there are reasons why they might. It would depend on when the chips come out as well as some political decisions. I don't know the schedule for the 7448's.
Comments
Originally posted by 1337_5L4Xx0R
I was even more stoked when I saw Intel's future roadmap, and integer 'performance per watt'
It is of course a completely meaningless metric.
IBM 440 PowerPC chips yield extremely good performance per watt for example but they aren't going in Macs. It depends which end of that ratio you attach importance. 'Performance per watt' lets Apple weasel out of delivering any performance.
I can hear Jobs now - 'Hey, it may not be faster than a G5 but you can now use your Powerbook for 16 hours straight and ok, it takes 3 times longer to encode video now, but it's whisper quiet'.
The current Intel roadmap isn't very exciting at all although at least they have one, whereas IBM's isn't forthcoming. Dual Yonah is coming in at 2.16Ghz, 667 FSB and using 31W so they aren't really useful for desktops and neither for laptops.
Originally posted by DHagan4755
If you watch the WWDC 2005 keynote in Quicktime, Steve Jobs says, "I think a lot of you would like a G5 in your PowerBook, and we haven't been able to deliver that to you yet."
Key word there is "yet."
Jobs continues, "But these aren't even the most important reasons. The most important reasons are that as we look ahead, though we have great products right now...and we have some *great* PowerPC products still yet to come..." and Jobs then proceeds to talk about the future of the PowerPC roadmap vs. Intel's.
So the way I parse the whole thing is that the near term is OK for the G5s, but long-term it's not as good as Intel's will be. That leaves the door open for the PowerBook G5, even it's just for a year or so.
Watch the keynote, and it's about 21 minutes in. See for yourself.
Yes, this was all in my memory as well. We never know what they plan, and it might be that the Powerbooks are not going to be the first to get Intel processors, which would make a year or so with a G5 even more sensible. I'm really looking forward to seeing what direction they go this fall.
This "low power" 970 is a yawner, if you ask me. No additional cache, no lower power bus, and no other performance improvements. They tweaked the process and downclocked, from the looks of it. At 1.6 GHz w/ 512K L2 this 970 will get stomped by the current Pentium-Ms, nevermind what Intel has in the pipeline. I doubt these 970s will even do that well on typical Mac software compared to the 7448 at the same power levels. Better on media intensive stuff, but lower clock rate really hurts the 970 -- its designed for high frequencies and lots of bandwidth.
The 970MP is the interesting announcement -- two cores and double the cache per core. That is probably why the PowerMac is switching to Intel last.
Originally posted by melgross
Apple should play up the qualities of their machines and software, and downplay the cpu's. Make people concentrate on what's important.
Tell that to the people buying powermacs. The CPU matters. You can't just woo people with software. You need something to back it up, and if this no-hardware-improvement trend continues, and software development keeps up, then eventually we won't be able to run that software on our pretty, underpowered machines. Actually, this is already happening, alebit minor (Dashboard ripple effect in MacMini).
The hardware has to keep up with the software, and right now that's not happening, so that's where Intel comes in.
Originally posted by cwestpha
I can tell you that this statement is 100% wrong. Yonah is already done and is sampling at Intel (has been that wasy since March). Jobs said that they would have their first line of Intel based Macs BY the spring 06' conference. To say that it will be July at the earliest flies right in the face of all of the facts.
Bottom line is Intel already ramping up production and could launch Yonah fall of 05' but are waiting for Q1 06' because that is when they said it would launch. They gave that timeframe expecting the same problems from 130 nm to 90 nm. bottom line is 90 nm to 65 nm was been one of the easiest transitions in recent Intel history.
OEMs are going to be getting early Yonah generation centrinos (dual core and single core Pentium Ms, late beta chipsets, and the a/b/g chip minus the WiMax pretty soon). Since it takes anywhere from 2 to 6 months for companies to make products from the samples we can expect the market to be flooded with Yonah Centrino right from the launch date. They just need to make a few more tweeks with the chipsets and get the last of WiMax intigrated with their wireless chip.
It would be shocking if Apple couldnt produce a Yonah product in Q1 06'. If they couldnt that would meen they are having a problem with software or that they are just so new with Centrino they are needing a lot of outside help.
Well, again it's the Dev conf, not July, that was a mistake.
Don't forget that when Apple said that 10.4 would be out in the first half of 2005, people were also predicting that it would be out in Jan. Even the Powerpage said January (but he's NEVER right!).
Look when it did come out though. Almost at the very end. I suspect that it would have come out during the conf. but Jobs didn't want to detract from the Intel announcement. That's why it's so buggy. They rushed it at the last instant.
My feeling is that we will see the same schedule here. Shortly before the conf.
There is really no reason for Apple to rush this. He has to give time for his own team to get the Intel version of the OX finished. He also has to give the developers time to get their work done.
This transfer MUST go smoothly. No excuses. Apple will also, no doubt, be working on Rosetta, to make it work better, possibly to let work with more programs. Does it use both processors? If not, it could be a problem.
There are a lot of issues that aren't being considered here. It's not only the chips. Does Apple have a new case design? Will they go to Express? That bus is also for mobile use. Just because Apple has been considering going to x86 doesn't mean that these, and other, problems have been worked out yet. It's very possible that with Apple's concerns for secrecy, they haven't let any plans out to manufacturing.
We just don't know. It's not worth arguing that they will, or won't have a machine out by January.
Why don't we try to figure out what they will contain instead? That's really more interesting anyway.
Originally posted by mynamehere
Tell that to the people buying powermacs. The CPU matters. You can't just woo people with software. You need something to back it up, and if this no-hardware-improvement trend continues, and software development keeps up, then eventually we won't be able to run that software on our pretty, underpowered machines. Actually, this is already happening, alebit minor (Dashboard ripple effect in MacMini).
The hardware has to keep up with the software, and right now that's not happening, so that's where Intel comes in.
When the word "consumer" is used, it doesn't mean "pro". We have four Powermacs at home, and in my business we had eighteen.
Mom and dad going to the Apple store aren't looking to compare a P4 (or M) to the G5 or G4. They just want what seems to be snappy, does what they want, looks good, and isn't subject to internet garbage.
At least that's what the surveys of people buying these machines say.
And again i'll state, from someone who has bought a lot of Powermacs over the years; we buy our Macs based on the premise that we WANT Macs as opposed to Windows machines. We compare a new Mac to the older ones we have.
If x86 machines are MUCH faster, maybe we would consider them instead. But they have to be MUCH faster.
I can tell you from inside the media industry that the reason that Powermac sales have fallen is mostly because people are sitting on their current models, not because they are rushing out and buying XP. I'm sure that this has happened in a few cases, but not many. And those are usually those who were not comfortable with Macs anyway.
In the future Apple sees the x86 line getting MUCH better than the PPC. That's a problem So they are switching.
Originally posted by DHagan4755
If you watch the WWDC 2005 keynote in Quicktime, Steve Jobs says, "I think a lot of you would like a G5 in your PowerBook, and we haven't been able to deliver that to you yet."
Key word there is "yet."
Jobs continues, "But these aren't even the most important reasons. The most important reasons are that as we look ahead, though we have great products right now...and we have some *great* PowerPC products still yet to come..." and Jobs then proceeds to talk about the future of the PowerPC roadmap vs. Intel's.
So the way I parse the whole thing is that the near term is OK for the G5s, but long-term it's not as good as Intel's will be. That leaves the door open for the PowerBook G5, even it's just for a year or so.
Watch the keynote, and it's about 21 minutes in. See for yourself.
I don't think the one word "yet" outweighs the fact that Apple would have to engineer a PowerBook first for the G5 and then for an Intel so quickly. It just doesn't make sense, especially given that there's good reason to believe that the PowerBooks will be among the first Macs to go Intel.
Originally posted by BRussell
I don't think the one word "yet" outweighs the fact that Apple would have to engineer a PowerBook first for the G5 and then for an Intel so quickly. It just doesn't make sense, especially given that there's good reason to believe that the PowerBooks will be among the first Macs to go Intel.
This is really hard to say. There is so much at stake here. Apple will lose a lot of sales over the next year or two.
We don't know how close they are to having a G5 plan ready. It might not be as bad as is thought. It might be worth it to Apple to have a G5 machine just to slow down the deterioration in portable sales. If they spend $50 million or even $100 million but save $500 million to a billion in sales, then it's worth it. Apple also has to show that they are doing something. If the performance is even 25% better than it would otherwise have been, then it might be worthwile.
These costs have to be thought of as being advertising, or publicity. If the 7448 isn't much better than the 7447a, they might have to take steps. Just getting a G5 in might be enough. G5 performance is about 30-50% better than a G4 clock for clock. And with no L3 on the 744x series, it could be more. Then bus speed becomes even more important.
Originally posted by melgross
G5 performance is about 30-50% better than a G4 clock for clock. And with no L3 on the 744x series, it could be more. Then bus speed becomes even more important.
The G5 doesn't have L3 cache either.
And the 7448 has twice the L2 cache of the G5 970fx.
7448 has a 200Mhz FSB and you can probably bet on a portable G5 being throttled back for power concerns.
IMHO, everything points to a 7448 update. I'd not be surprised if we see a 1.8 to 2.0Ghz 7448 appearing next in Powerbooks and iBooks and that they'll easily out perform a G5 at everything apart from manipulating large files and video - something you don't expect to be fast on a laptop anyway with their slow drives and limited GPUs.
Of course, I'm prepared to be wrong. I was totally off with switching to Intel.
Originally posted by aegisdesign
The G5 doesn't have L3 cache either.
And the 7448 has twice the L2 cache of the G5 970fx.
7448 has a 200Mhz FSB and you can probably bet on a portable G5 being throttled back for power concerns.
IMHO, everything points to a 7448 update. I'd not be surprised if we see a 1.8 to 2.0Ghz 7448 appearing next in Powerbooks and iBooks and that they'll easily out perform a G5 at everything apart from manipulating large files and video - something you don't expect to be fast on a laptop anyway with their slow drives and limited GPUs.
Of course, I'm prepared to be wrong. I was totally off with switching to Intel.
Yes, of course the G5 has no L3 cache, what's your point? It never had L3 cache. That's why the comparison is so interesting. It performs as well as it does vs a vs the G4 WITH cache, despite that. so they will slow the bus down to 1/3rd instead of 1/2. That hasn't taken a real bite out of its performance. Even if they dropped to 1/4 it wouldn't matter that much. It's power requirements closely match the 7448's. I don't agree with your performance expectations, at least not from the specs. that we know so far.
If Apple clocks the 7448 up, the power requirements will go up with it as well, and not neccessarly in a linear way either. It seems that when chips are clocked out of their factory specs they don't behave the way they do otherwise. so a clocked up 7448 would use more power then a 1.6 G5 would. If that's the case, and Apple is willing to live with it, then they could clock the G5 up as well.
What I'm saying is that we don't know what Apple is doing, or what their thinking on this is. You're being conservative. Nothing wrong with that. But these are "interesting times". I wouldn't be surprised if Apple stretches out as far as they can on this one.
In a previous life I was a partner and designer in a Pro and hi-end audio firm, Magnum Opus. We recieved advanced info on many components from manufacturers. Oftentimes more than a year in advance. We would start preliminary work on products based on those specs. Sometimes the specs would change, for better or worse. So did our plans for them.
I'm sure that Apple has been working on a G5 plan for quite some time. The only question is how far advanced it is, and whether these chips will make the cut. Depending on where both Apple and IBM are on this, we could see an announcement as early as Janurary, to be available in Feb or March. It wouldn't be the first time they had a month (or three) lead time.
Or you could be right, and it could be a 7448.
Originally posted by BRussell
I don't think the one word "yet" outweighs the fact that Apple would have to engineer a PowerBook first for the G5 and then for an Intel so quickly.
Your supposing that Apple is just finding out about the low-power G5s through IBM's recent announcement in Japan. More than likely Apple has long been playing around with samples and engineering them into prototypes. We knew of last summer the "PowerBook 7,1" and the "PowerBook 7,2." They appeared in a plist file of the 10.3.5 update as you may recall, around the same time that we found out the next iMac was based around the G5. Now that info is encrypted in the plist file.
A lot of people assumed that the PowerBooks represented were the 15-inch and 17-inch PowerBooks. But it is noteable that the 12-inch PowerBook and the 12-inch iBook G4 are in the same family, relative to model numbers. So the PowerBook 7,1 and PowerBook 7,2 fits nicely into the Digitimes rumor of PowerBook G5s and iBook G5s.
We are still in the quarter referenced in the Digitimes article that Apple would be shipping these so-called PowerBook G5s and iBook G5s, we are the heels of IBM's official announcement of the 970MP and the low power 970FX, Apple announced a weak PowerBook update in January, and iBooks haven't been updated since last October. Something's gotta be coming. Something better be coming.
I'm just creating some hype here. Hell, we have had a dearth of exciting product announcements from Apple lately. We could use anything to lift up our spirits!
Originally posted by sdfisher
"The first PowerBook to sport an Intel processor is not expected until July 2006 at the earliest."
This should read:
"The first PowerBook to sport an Intel processor is expected between January and July 2006."
Close, but I disagree. It should read
"The first PowerBook to sport an Intel processor is expected by July 2006 at the latest."
That matches what Steve said.
He also said "We want the transition to be mostly complete by WWDC 2007", and fully complete by the end of 2007 - and described that as "a 2 year transition" - which implies some Intel based Macs early next year.
He may be counting from his announcement, but that makes the transition a 2.5 year transition. He may have meant "18 month transition) but was being general. I guess it won't be this year because he says the 2 year transition is 2006-2007.
I don't think the intel PB will appear till MWSF 2007. The PBs are too important to pro users. If there is some problem making an intel chip perform in a Mac like a Mac should then you want to see that problem in the Mac Mini or the iBook first.
Suppose the Mac Mini with intel comes out May of 2006. Then an iBook comes out in September of 2006. Then the PB could come out in January, shipping same day as the announcement. This would give Apple six months experience with the intel chips in the field both in the Mac Minis and in a portable. Six months to see if there are problems with Rosetta, Quark, Adobe CS, Photoshop, Applescript, printing, sleep, wakeup, CPU usage, WiFi connections, Firewire, etc.
If there is some goofy bug in the compiler that causes an unexpected problem I want to see that in a Mac Mini that I might use as a kiosk or for light surfing rather than on the PB that I use to earn a living.
Presumably we'll see Leopard arrive then too. There may be some benefit in delaying the new PB to coincide with that as well.
If we get a 7448 update this fall and possibly a dual core G4 next spring with a speed bump later in the year that wouldn't be too bad. Remember, in addition to the CPU changes we'll be seeing bigger, faster HDs, possibly more FW800 ports, a better display, better battery life, nicer GPU. What's not to like?
Originally posted by GregAlexander
Close, but I disagree. It should read
"The first PowerBook to sport an Intel processor is expected by July 2006 at the latest."
That matches what Steve said.
He also said "We want the transition to be mostly complete by WWDC 2007", and fully complete by the end of 2007 - and described that as "a 2 year transition" - which implies some Intel based Macs early next year.
He may be counting from his announcement, but that makes the transition a 2.5 year transition. He may have meant "18 month transition) but was being general. I guess it won't be this year because he says the 2 year transition is 2006-2007.
While it might be nice to see an Intel machine in early 2006, I don't believe it for all of the reasons I've given before, and because it's too soon for an entirely new machine to come out. There are still issues. The dev machines use a BIOS chip. It's believed that Aple won't use that, but will instead use Intel's equivalent of Open Firmware. Too many issues for six or seven months. Apple doesn't do things quickly. It can take Apple two or more years to come out with a new mobo, for example. It takes the far eastern manufacturers six months. Even if the board will be from Intel, and not an Apple design, it still has to be done and tested, sent to manufacturing etc.
I think that March would be pushing it. Again, my other reasons above.
Originally posted by melgross
Yes, of course the G5 has no L3 cache, what's your point? It never had L3 cache.
That's exactly my point. Arguing about L3 cache is pointless as neither the 744x or G5 have one. And neither will have, especially in a laptop.
The 7448 has a 1MB L2 - twice as big as the G5 and previous G4 7447a. That is way more significant than L3. IBM really needed to up the L2 on the low power G5 but then it would consume more power. Cache is expensive power wise.
When the 750fx went to 1MB L2 cache, it rocked and in some ways was faster than a similarly clocked G4 in the iBook. If it wasn't for Altivec, it quite possibly would have been a quicker chip all told in general use.
That kind of worries me about going Intel too - Having used 500Mhz G3 and 400Mhz G4 at the same time, you notice how much faster Altivec makes things. It may not matter up at 2Ghz I guess but it's niggling.
Originally posted by melgross
That's why the comparison is so interesting. It performs as well as it does vs a vs the G4 WITH cache, despite that.
That's really not the case.
Please read http://www.barefeats.com/g4up.html
Compare the G5 2.0Ghz SP tests with G4 upgrade speeds. Neither have L3. That's an overclocked 1.3Ghz 7447A running at 2.0Ghz on a 133Mhz FSB and slow AGP slot compared to a 2.0Ghz G5 PowerMac running a 1Ghz FSB and faster AGP slot. It's never more than 25% behind the G5 in the CPU bound tests and even draws with it on Photoshop SP actions. Your assertion that it's 30-50% behind is plain wrong. Tests do not bear that out.
Now, with the 7448, double the L2 and raise the FSB to 200Mhz. Remember we're sticking this in a laptop so drop the G5 to say 533Mhz like an iMac and run the machine in auto power mode instead of Highest all the time which knocks off about 30% CPU score anyway. IMHO, the G4 may beat the G5 if not be very close. So close it really doesn't matter if you go G4 or G5. Look at the 1.4Ghz results there with 2MB L3 - it comes second to a G5 in the multi-processor aware Photoshop tests. 1MB of L2 is real nice - better than 2MB L3 in most uses as you don't have to go off chip.
The G5 will win in memory intensive applications and shoving data at the GPU but in a laptop the GPU is usually throttled back with slower RAM.
The G4 will win on disk intensive tasks because Apple's G5 disk controller sucks. The G4 systems are about twice as fast at disk access.
As an other example, how about Powerbook G4 1.5Ghz v 1.8G5 iMac...
http://www.barefeats.com/imacg5.html
If that PB was 1.8Ghz, it'd be pretty closely ran. The Motion test is an oddity because Motion uses the GPU. iMac loses because it's GPU is very slow, by comparison.
I've suggested to Rob at Barefeats that he runs a 1.67 G4 Powerbook v 1.6Ghz G5 PowerMac v 1.6/1.7 Pentium M test. Might be interesting to see cycle-for-cycle who wins then add in the 7448 if that ships and faster Pentium Ms and G5s. Since Intel is heading down the path of using the Pentium M core for most of it's roadmap it may be nice to see where we're heading also. They're even using Yonah on low end Xeons next year.
Originally posted by melgross
so they will slow the bus down to 1/3rd instead of 1/2. That hasn't taken a real bite out of its performance. Even if they dropped to 1/4 it wouldn't matter that much. It's power requirements closely match the 7448's. I don't agree with your performance expectations, at least not from the specs. that we know so far.
I was basing my guesses on actual benchmarks not the specs. There's a bit of extrapolation and guestimation but I think they'll be broadly comparable. Each CPU as pluses and minuses.
Originally posted by melgross
If Apple clocks the 7448 up, the power requirements will go up with it as well, and not neccessarly in a linear way either. It seems that when chips are clocked out of their factory specs they don't behave the way they do otherwise. so a clocked up 7448 would use more power then a 1.6 G5 would. If that's the case, and Apple is willing to live with it, then they could clock the G5 up as well.
The low power 970fx is 16w at 1.6Ghz - no faster speeds have been announced and we know already that the 970fx at 2.0ghz is too hot for a laptop. We've also no idea what they've done to get it low power - is it low voltage? 200Mhz FSB? No L2? So why no low power 2.0Ghz?
Freescale said the 7448 at 1.7Ghz consumes 15w. It's already cooler than the low power 970fx. The 7447a is listed as consuming 20W at 1.42Ghz (they don't list the 1.67). There's obviously room there to grow.
For comparison the Pentium M at 1.4 is 18W typical.
Originally posted by melgross
What I'm saying is that we don't know what Apple is doing, or what their thinking on this is. You're being conservative. Nothing wrong with that. But these are "interesting times". I wouldn't be surprised if Apple stretches out as far as they can on this one.
Perhaps, but I don't see any technical reason, or financial reason for Apple to come out with a whole new architecture just before they scrap it and go Intel. It just doesn't make any sense. Better to just apply minor tweaks to the current one. It'd be majorly expensive to support a G5 powerbook.
Originally posted by melgross
In a previous life I was a partner and designer in a Pro and hi-end audio firm, Magnum Opus. We recieved advanced info on many components from manufacturers. Oftentimes more than a year in advance. We would start preliminary work on products based on those specs. Sometimes the specs would change, for better or worse. So did our plans for them.
In a previous life I wrote compilers, run-time systems and native code generators for various, often prototype CPUs. I worked on a Pentium system before they were released. Worked on OSs before they were even beta release, even debugged the source of one particular OSs filesystem at one point. I'm sure I can say that now but at the time I was NDAd up to my eyeballs by Intel, Sun, DG, Digital, Microsoft and even ARM at one point.
So I'm aware how changing architectures is a major thing you don't do readily. Doing it twice in two years - suicide.
Originally posted by melgross
I'm sure that Apple has been working on a G5 plan for quite some time. The only question is how far advanced it is, and whether these chips will make the cut. Depending on where both Apple and IBM are on this, we could see an announcement as early as Janurary, to be available in Feb or March. It wouldn't be the first time they had a month (or three) lead time.
Or you could be right, and it could be a 7448.
If they announce for March next year, then I'd expect it'll be Intel based.
But in the meantime a 7448 based Powerbook or iBook can't be far off. iBook hasn't been updated since October last year - now almost 4 months overdue an upgrade. Powerbooks are a month away from their average update cycle. Even if they only stick in the 1.7Ghz quoted by Freescale, with the faster bus and 1MB L2, it's an ok upgrade. Dropping say 8-9W off the power requirement is nice too. You could see 1.7Ghz 12" powerbooks and those would rock. I'd buy one in a flash. If I was to guess, we're a month or so away from 7448 updates. I'd also guess, as usual, Freescale has faster than the announced to the embedded market versions for Apple. 2Ghz perhaps? One can hope. I'm sure people will still moan it's a G4 even though it may actually be faster and cooler than a G5.
But anyway, I'll wait for Apple to prove me entirely wrong again.
My guess for the 1st machines would either be the iBook and/or Mac mini. The the biggest bet on the Mac mini. My main reason is performance. The iBook & Mac mini have the least performance-critical customers. Thus, "Rosetta" might work for them. Especially since these customers can be new (no previous comparison or a VERY old computer to compare against). Could be a mini at MW SF, an iBook in May, and so on?
I don't think a PB customer will buy until the Intel box runs their sw faster than their currrent version (and probably not quicker than the 1.67 current high end). So the PB is contingent on it being faster, even with Rosetta, AND having the major power apps (Final Cut Pro, Photoshop, MS Office, etc) Intel native (fat/univeral binary). I think the same is true for PowerMacs, but the hurdle is higher (likely 2 dual 970 MP's vs a single 1.7 G4), so they'll be out the latest. iMacs would, be in the middle (less performance sensitive, but still can come out THAT much slower than the current line).
Originally posted by DHagan4755
But what if the Intel-based Macs run at such better performance with native apps that the slowest of the Rosetta emulated apps run better on Intel than they do on native PowerPC hardware?
And monkeys will fly out of people's arses.
Originally posted by aegisdesign
That's exactly my point. Arguing about L3 cache is pointless as neither the 744x or G5 have one. And neither will have, especially in a laptop.
The 7448 has a 1MB L2 - twice as big as the G5 and previous G4 7447a. That is way more significant than L3. IBM really needed to up the L2 on the low power G5 but then it would consume more power. Cache is expensive power wise.
When the 750fx went to 1MB L2 cache, it rocked and in some ways was faster than a similarly clocked G4 in the iBook. If it wasn't for Altivec, it quite possibly would have been a quicker chip all told in general use.
That kind of worries me about going Intel too - Having used 500Mhz G3 and 400Mhz G4 at the same time, you notice how much faster Altivec makes things. It may not matter up at 2Ghz I guess but it's niggling.
That's really not the case.
Please read http://www.barefeats.com/g4up.html
Compare the G5 2.0Ghz SP tests with G4 upgrade speeds. Neither have L3. That's an overclocked 1.3Ghz 7447A running at 2.0Ghz on a 133Mhz FSB and slow AGP slot compared to a 2.0Ghz G5 PowerMac running a 1Ghz FSB and faster AGP slot. It's never more than 25% behind the G5 in the CPU bound tests and even draws with it on Photoshop SP actions. Your assertion that it's 30-50% behind is plain wrong. Tests do not bear that out.
Now, with the 7448, double the L2 and raise the FSB to 200Mhz. Remember we're sticking this in a laptop so drop the G5 to say 533Mhz like an iMac and run the machine in auto power mode instead of Highest all the time which knocks off about 30% CPU score anyway. IMHO, the G4 may beat the G5 if not be very close. So close it really doesn't matter if you go G4 or G5. Look at the 1.4Ghz results there with 2MB L3 - it comes second to a G5 in the multi-processor aware Photoshop tests. 1MB of L2 is real nice - better than 2MB L3 in most uses as you don't have to go off chip.
The G5 will win in memory intensive applications and shoving data at the GPU but in a laptop the GPU is usually throttled back with slower RAM.
The G4 will win on disk intensive tasks because Apple's G5 disk controller sucks. The G4 systems are about twice as fast at disk access.
As an other example, how about Powerbook G4 1.5Ghz v 1.8G5 iMac...
http://www.barefeats.com/imacg5.html
If that PB was 1.8Ghz, it'd be pretty closely ran. The Motion test is an oddity because Motion uses the GPU. iMac loses because it's GPU is very slow, by comparison.
I've suggested to Rob at Barefeats that he runs a 1.67 G4 Powerbook v 1.6Ghz G5 PowerMac v 1.6/1.7 Pentium M test. Might be interesting to see cycle-for-cycle who wins then add in the 7448 if that ships and faster Pentium Ms and G5s. Since Intel is heading down the path of using the Pentium M core for most of it's roadmap it may be nice to see where we're heading also. They're even using Yonah on low end Xeons next year.
I was basing my guesses on actual benchmarks not the specs. There's a bit of extrapolation and guestimation but I think they'll be broadly comparable. Each CPU as pluses and minuses.
The low power 970fx is 16w at 1.6Ghz - no faster speeds have been announced and we know already that the 970fx at 2.0ghz is too hot for a laptop. We've also no idea what they've done to get it low power - is it low voltage? 200Mhz FSB? No L2? So why no low power 2.0Ghz?
Freescale said the 7448 at 1.7Ghz consumes 15w. It's already cooler than the low power 970fx. The 7447a is listed as consuming 20W at 1.42Ghz (they don't list the 1.67). There's obviously room there to grow.
For comparison the Pentium M at 1.4 is 18W typical.
Perhaps, but I don't see any technical reason, or financial reason for Apple to come out with a whole new architecture just before they scrap it and go Intel. It just doesn't make any sense. Better to just apply minor tweaks to the current one. It'd be majorly expensive to support a G5 powerbook.
In a previous life I wrote compilers, run-time systems and native code generators for various, often prototype CPUs. I worked on a Pentium system before they were released. Worked on OSs before they were even beta release, even debugged the source of one particular OSs filesystem at one point. I'm sure I can say that now but at the time I was NDAd up to my eyeballs by Intel, Sun, DG, Digital, Microsoft and even ARM at one point.
So I'm aware how changing architectures is a major thing you don't do readily. Doing it twice in two years - suicide.
If they announce for March next year, then I'd expect it'll be Intel based.
But in the meantime a 7448 based Powerbook or iBook can't be far off. iBook hasn't been updated since October last year - now almost 4 months overdue an upgrade. Powerbooks are a month away from their average update cycle. Even if they only stick in the 1.7Ghz quoted by Freescale, with the faster bus and 1MB L2, it's an ok upgrade. Dropping say 8-9W off the power requirement is nice too. You could see 1.7Ghz 12" powerbooks and those would rock. I'd buy one in a flash. If I was to guess, we're a month or so away from 7448 updates. I'd also guess, as usual, Freescale has faster than the announced to the embedded market versions for Apple. 2Ghz perhaps? One can hope. I'm sure people will still moan it's a G4 even though it may actually be faster and cooler than a G5.
But anyway, I'll wait for Apple to prove me entirely wrong again.
I'm familliar with the Bare Feats tests. Frankly I've never found then to be all that accurate. They don't agree with what I've found on my own machines. A number of tests that he's done aren't processor bound anyway, or depend upon the GPU more than the cpu.
I also don't agree that going to 1MB L2 is way more significant than 2MB of L3. It certainly helps, but L3 isn't used because of power requirements, not because it isn't of as much use. I agree that 1MB of L2 on the G5 would be better. a larger cache is always better. Intel has shown that with the Xenon's. But again it's probably power requirements. I felt that IBM shouls have gone to 1MB with the 2.7's, but again, it might have been too hot.
Altivec is a stopper. No doubt about that. That's one thing that bothers me about the switch. But rendering tests using two 3.6 Xenons vs two 2.5 G5's show the Xenons to be slightly faster, so maybe we're worrying about nothing. Of course an on board mem controller ala Opteron would help even more.
As I said, I've never gotten results like that so I can't speak for them. MacWorld's tests, for example don't show as close a result as his either. Neither do others that I've seen.
We do know that the low power FX has 512KB of L2 because the press release said it did. FSB? low voltage? Though low voltage seems most likely. It could be a combination, but we'll have to wait.
This isn't a new architecture for Apple after all. It would be a new mobo, but they have two years of experience with G5's, and have no doubt been working on this for a while now. It just might need to go to production.
I really doubt very much that a 7448 machine would be faster than the 1.6 G5 unless they clocked it. something Apple has done a number of times in the past.
I'm not saying that Apple would do this. But there are reasons why they might. It would depend on when the chips come out as well as some political decisions. I don't know the schedule for the 7448's.