Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
PowerPC RoadMap from Motorola
- Thread starter MacRumors
- Start date
- Sort by reaction score
DHM, i really don't know what benchmarks you're talking about, but all the benchmarks I've seen show the DP G4's beating the single 1.6 G4 by large margins. I have already posted one benchmark that shows the 1.6 G4 barely beating a 1.5 powerbook in FCP rendering. Mhz for Mhz, I really don't believe there is that big a difference. Why don't you post links to these supposed benchmarks that show the superiority of the G5. I can provide plenty of benchmarks that will show that the 1.6 is not competitive with most of the faster DP powermacs.
Mr. MacPhisto said:
If Apple bought that load of crap, then they're not half the company they ought to be right now. Motorola took how long to make those last weak jumps? Until Freescale establishes themselves as being past the shortsighted failures of their forebears, they're suspect and complicit in the performance gap that IBM and Apple closed considerably.
If the VX is actually dead, then that's a sad, sad thing to have had happened. IBM is far better than Motorola, in my opinion, and even if there are issues with 90nm, the 750vx would still have been better at 130nm than the G4 was at the same process.
Dont Hurt Me said:
There are no 1.6G5 duals. What are you talking about? Also, as someone just up this thread pointed out, the new PowerBook 1.5s are pretty close to the G5 1.6s in performance. Not a 1:1, but not bad for only being 100mhz behind.
Actually, there is G5-aware and optimized software, but it's almost universally expensive pro-level apps like Photoshop CS.
While I don't expect this amazing G4 revision, the one thing that does make me pause is that they're not using the older 74xx core. The ex00 line is a modified G4 core, and so I might end up eating my words if Motoroloa does create a chip somewhat like the 750vx. Taking the slightly older tech and adding advances since then might yield the power saving features with higher performance, the same way that the Centrino ends up competing with P4s that are double its clock.
If that kind of computing efficiency could be achieved, I'd look at Freescale favorably and try to think of them as separate from their roots. Until it does, though, this is just smoke and mirrors in the form of a press release.
My god, why does everyone think G5 optimized software is going to be so much faster - it wont.
For a great example of this, look at Linux and AMD64. 64 Bit Linux has been out for a hell of a long time now, and it's not twice as fast. Infact, it's basically the same but better at complex mathematical operations.
This isn't like going from 16 bit to 32 bit. 16 bit was hugely flawed and was restraining computing. 32bit will do us just fine - the only reason we need to go to 64 is for extra memory addressing.
Also, I think it would be better if Apple just switched to x86-based arch. However, as we all know that will not happen and instead we will be left with CPUs that seem great at first but seriously start to wane when they have problems bumping up the speed. I would actually be moderately surprised if they hit the target of 3GHz by summer - look at the problems IBM is having with the 970FX and personally I doubt they are going to be able to mass produce hundreds of thousands of these.
For a great example of this, look at Linux and AMD64. 64 Bit Linux has been out for a hell of a long time now, and it's not twice as fast. Infact, it's basically the same but better at complex mathematical operations.
This isn't like going from 16 bit to 32 bit. 16 bit was hugely flawed and was restraining computing. 32bit will do us just fine - the only reason we need to go to 64 is for extra memory addressing.
Also, I think it would be better if Apple just switched to x86-based arch. However, as we all know that will not happen and instead we will be left with CPUs that seem great at first but seriously start to wane when they have problems bumping up the speed. I would actually be moderately surprised if they hit the target of 3GHz by summer - look at the problems IBM is having with the 970FX and personally I doubt they are going to be able to mass produce hundreds of thousands of these.
stockscalper said:
Really?
According to their FCP rendering speeds, the 1.5 G4 Powerbook is only slightly slower than the 1.6 G5.
Which is not surprising. If you go through their benchmarks, clock for cock, the dual G4s aren't *that* much slower than the dual G5s. (A 2ghz G5 is 1.4 times as fast (clock speed wise) than a 1.4ghz G4, and performs around 1.5-1.6 times faster.)
I'm sure if Moto fixed the FSB and scaling issues, the G4 would compete favourably.
The G4 is a good chip with a choking fsb.
Don't Hurt Me, your theory about the G4 going from 1.42->1.5 in one year is flawed.
First, is that it is generally accepted that the 1.42 was an overclocked 1.33 (or something of that nature) when it went into the PowerMacs.
The next... in 10 months IBM will has gone from 2.0ghz to what? OH WAIT, they haven't done a speed bumb since introduction... i almost forgot.
What benchmarks? Game Benchmarks? A videocard on a laptop is far inferior to most modern video cards... even with the same model number.
The 1.5ghz "has" fast RAM, it doesn't "use" fast ram.
The only benchmarks i saw on barefeats were for games, and one MP aware photoshop test, in which the single 1.5 is compared to duals.
and a couple of notes DIRECTLY FROM barefeats.com...
First, is that it is generally accepted that the 1.42 was an overclocked 1.33 (or something of that nature) when it went into the PowerMacs.
The next... in 10 months IBM will has gone from 2.0ghz to what? OH WAIT, they haven't done a speed bumb since introduction... i almost forgot.
What benchmarks? Game Benchmarks? A videocard on a laptop is far inferior to most modern video cards... even with the same model number.
The 1.5ghz "has" fast RAM, it doesn't "use" fast ram.
The only benchmarks i saw on barefeats were for games, and one MP aware photoshop test, in which the single 1.5 is compared to duals.
and a couple of notes DIRECTLY FROM barefeats.com...
valid remarks but IBM has some legs and we will see them the next few years this is why Apple is moving that way. Moto the G4 could be a kick ass chip if the FSB problem could be fixed. i guess it cant Moto and all. though solid cant stand eye for eye with Intel or AMD and G5 still needs two to stand with them. G4 is a hasbeen and G5 is a newborn getting its legs. sort of like os9 and 10 or rather X.
thatwendigo said:
1. Motorola roadmaps are way too optimistic. It's either vaporware or 1-2 years behind. So unless Freescale can actually deliver this e600 in large volume, it doesn't exist, as far as I'm concerned.
2. The 7447A is at the end of the road, no? Unless Apple does something radical to increase the battery power and keep the temp. down, I can't see the G4 going past 1.6 ghz or so. And outside of some wild wild rumors, I haven't heard of any further 74xx revisions.
3. 970fx in a PB? I'm doubtful of that. The 970fx seems pretty close to the 7447A in terms of power consumption at the same clock. Add to that, the potentially hotter-running controller, etc. of the new architecture, the 970fx is too hot. I'm putting my money on something else, a 970-derivative tailored for laptops. Sometime in the spring of 2005.
4. Probably one very modest PB upgrade until then. Minor changes to graphics, RAM, HD, etc. with maybe a 100 mhz increase. Nothing worth writing home about.
Kind of like Intel's "Centrino." Of course don't tell them MHz matters there 
Mr. MacPhisto said:
I have to agree with everyone who talked about the 'putting all your eggs in one basket' thing. Having alternatives is ALWAYS a good thing. It had seemed to me that Motorola had announced in the past that they were working on a 64-bit chip but it fell through, I don't know if they are anywhere near getting that back on track or not, but either way some higher clocked 74xx based 32 bit chips with higher FSB's would be excellent for iBooks, eMacs, iMacs, and possibly PBooks until a G5 is crammed into a PBook..
thatwendigo said:
A large part of the shortsightedness of Motorola was due to their old fab. The Crolles2 facility is able to go down to 32nm and is as advanced, if not moreso, as Fishkill. Moto is also teaming with Phillips and STMicroelectronics in France. There is good reason to believe that Moto has turned the corner. Crolles is already producing chips at 90nm and should be pumping them out in volume this summer.
Apple invested millions of dollars in VX - and they killed it off when it's production had to be pushed back. They had to have had some very good reasons for terminating this project, and I don't consider Apple a stupid company.
The VX (it was not a 750) was slightly better than the current G4, although the 7447A could give it a run for its money clock-to-clock, even with its crippled bus.
I can also tell you that IBM considers Motorola to be a very formidable competitor once more and expects some excellent chips from them this summer. It's helping to drive IBM's own chip designs - and could help them also improve.
Plus, there's been talk that FreeScale will deliver something this summer. I'm sure Apple knows precisely what's coming - and has good reason to believe that it will be delivered. I understand the skepticism - but I'm sure we all hope that FreeScale delivers bigtime and Apple has two solid chipmakers to turn to.
whoa.. hey what happened to the MPC 7457 chip? I thought that was the newest rev 13 micron G4 variant? From what I heard that chip used less power/ran cooler than the 7447 which was a 18 micron chip..
ltgator333 said:
No. The 7447 is a 130nm chip. The 7457 is the G4 with L3 cache capability. The 7447 is that chip with fewer pins - basically the embedded version. The 7455 is the 180nm chip that the 7447/57 replaced. I'm not sure if the 7457 was ever produced because of the switch to the G5. Apple isn't using the L3 versions currently.
nmk:
Dont Hurt Me:
Mr. MacPhisto:
Quite true, but that's not gona stop Apple from rolling out a G5 PB at their first oppurtunity. People think its way faster and Apple will be able to find enough benchmarks to make that impression stick.
Dont Hurt Me:
Yeah that was a 1.42ghz chip in a desktop, under a huge copper heatsink, at the hairy edge of combustion, vs a 1.5 in a laptop.
Mr. MacPhisto:
See thats why I never belived a word about the 750vx. Why on earth would IBM design a fancy new chip to go head to head with Moto in a crowded little market... and then not even have it be significantly superior? People just aren't connecting with the fact that IBM isn't getting rich selling processors to Apple: the 970's are just diluting the R&D of their big PPC's, and the 750's have a huge market other than Apple.
G4 is a good chip, it just suffers from the bottleneck called the front side bus. make it a +400MHz bus and re-introduce the L3 cache, and we'll have pretty good performance.
G5 is a killer because of the fsb throughput. there's nothing magical in the chip itself.
G5 is a killer because of the fsb throughput. there's nothing magical in the chip itself.
What about a Dual Processor G5 Transportable?
What about a Dual Processor G5 Transportable?
- Apple's got the Motherboard built with the XServe
- It's got the 20 display
- It's got the design capability.
Put it all together, and you've got a cool machine.
No need to wait for cooler ( 65nm G5's ) for the portable line.
LaMerVipere said:
What about a Dual Processor G5 Transportable?
- Apple's got the Motherboard built with the XServe
- It's got the 20 display
- It's got the design capability.
Put it all together, and you've got a cool machine.
No need to wait for cooler ( 65nm G5's ) for the portable line.
Some links about Motorola's chip making
That's some good informative insight into what Motorola and IBM are doing.
Here are some links to backup what you are saying about Motorola:
A June 2003 interview with the chief technology officer of Motorola's semiconductor products. In the article Motorola's Claudine Simson states in the last paragraph that a 90-nm PowerPC will be in production in 2004.
http://www.siliconstrategies.com/article/printableArticle.jhtml?articleID=10802061
A shorter article which reiterates some information from the above interview:
http://www.siliconstrategies.com/article/printableArticle.jhtml?articleID=10802058
Statements from the director of device/integration at Motorola's MOS-13 wafer fab:
http://www.siliconstrategies.com/article/printableArticle.jhtml?articleID=10802309
In the fourth paragraph he states that Motorola's goal is "a frequency doubling every 18 months or so, and get into the 2-GHz range for PowerPC, but at very low power consumption of say 20 watts"
If you add about 33% frequency improvement to the 1.5GHz G4 from a process shrink at 90-nm, that takes it to about 2GHz. Moving from 180-nm to 130-nm produced about a 33% increase in frequency.
In another article, Motorola's manager of PowerPC products platform division mentions incorporating a memory controller on the PowerPC and why that would be beneficial (the next G4 is expected to have a onboard memory controller):
http://www.siliconstrategies.com/article/printableArticle.jhtml?articleID=10808952
Mr. MacPhisto said:
That's some good informative insight into what Motorola and IBM are doing.
Here are some links to backup what you are saying about Motorola:
A June 2003 interview with the chief technology officer of Motorola's semiconductor products. In the article Motorola's Claudine Simson states in the last paragraph that a 90-nm PowerPC will be in production in 2004.
http://www.siliconstrategies.com/article/printableArticle.jhtml?articleID=10802061
A shorter article which reiterates some information from the above interview:
http://www.siliconstrategies.com/article/printableArticle.jhtml?articleID=10802058
Statements from the director of device/integration at Motorola's MOS-13 wafer fab:
http://www.siliconstrategies.com/article/printableArticle.jhtml?articleID=10802309
In the fourth paragraph he states that Motorola's goal is "a frequency doubling every 18 months or so, and get into the 2-GHz range for PowerPC, but at very low power consumption of say 20 watts"
If you add about 33% frequency improvement to the 1.5GHz G4 from a process shrink at 90-nm, that takes it to about 2GHz. Moving from 180-nm to 130-nm produced about a 33% increase in frequency.
In another article, Motorola's manager of PowerPC products platform division mentions incorporating a memory controller on the PowerPC and why that would be beneficial (the next G4 is expected to have a onboard memory controller):
http://www.siliconstrategies.com/article/printableArticle.jhtml?articleID=10808952
ddtlm said:
IBM wasn't really designing it for their embedded customers. They could have used it, but the chip was being designed for Apple - and Apple was fronting most of the cost for research and development of it. I'm sure IBM would have sold it to others, but they weren't the ones that killed it. Apple killed it because Apple was paying for it.
Mr. MacPhisto said:
Yeah that's really kinda sad.. when looking at the specs between different procs the one glaring *ouch* on 74xx procs is the bus speed.. 167Mhz is terrible- how you combat terrible bus bandwidth is having a lotta L2/L3 cache... and I don't see why they couldn't have (I didn't look really close so I may be eating my words here...) bumped the FSB frequency up to 200Mhz for the new PBooks, it just don't seem to me like when it comes to the current G4 based systems like anyones trying all that hard.
JFreak:
The PC world contains many examples of higher clocked busses not offering much return due to latency problems, for example there have been countless chipsets that offer poor returns when RAM and FSB are run asyncronously. Numerous times a new chipset arrived on the market supporting a new faster FSB or new faster RAM... and failed to impress anyone until version B showed up that ran RAM and FSB at the same clock speed. And lemme tell ya, the G5's FSB is far more asyncronous with RAM than anything the PC world has ever had. Two one-way 32-bit 1ghz busses with a packet-based protolcol and no separate address lines must talk to 128-bit wide 400mhz RAM and a separate address line. And Apple's chipset has to support two of those FSBs in simultaneous operation. Youch! Apple should be dreaming of the day that they can get on-die memory controllers such as AMD has, that would probably unlock a lot of non-streaming performance in the G5 designs we already have.
Edit:
After thinking about it, I think I should be more clear about what I mean be FSB's and RAM at the "same" clock speed. You see, they often arent really the same, for example Vias old KT133 vs KT133A where going to a 266mhz (DDR) FSB from a 200mhz FSB helped add significant performance, when coupled with PC133 RAM. In the KT133A they even removed the ability to run the FSB and RAM at a different "base clock", and it paid dividends. Really what matters is the "base clock", which is often (or always?) the address clock speed. Intel's quad-pumped FSB has the same address clock as DDR at half the data clock, so you'll see that arrangement a lot.
Mr. MacPhisto:
I remain sceptical of the whole thing.
Even when G4's had L3's and a much for favorable core-to-FSB ratio they got their butts kicked by the competition. I maintain that the FSB is less of a problem for most tasks than people think.
I'd say your putting way to much wieght on FSB clock speeds. Sure a high bandwidth FSB helps, but the other number people should look at is the memory access latency through a FSB. Although infomation has been spotty, everything I've read agrees with the edjucated guess that the G5's FSB and memory system is pretty high latency, compared to what the G4 is running on. High bandwidth helps "steaming" performance (compression, decoding, etc), low latency helps everything else.
The PC world contains many examples of higher clocked busses not offering much return due to latency problems, for example there have been countless chipsets that offer poor returns when RAM and FSB are run asyncronously. Numerous times a new chipset arrived on the market supporting a new faster FSB or new faster RAM... and failed to impress anyone until version B showed up that ran RAM and FSB at the same clock speed. And lemme tell ya, the G5's FSB is far more asyncronous with RAM than anything the PC world has ever had. Two one-way 32-bit 1ghz busses with a packet-based protolcol and no separate address lines must talk to 128-bit wide 400mhz RAM and a separate address line. And Apple's chipset has to support two of those FSBs in simultaneous operation. Youch! Apple should be dreaming of the day that they can get on-die memory controllers such as AMD has, that would probably unlock a lot of non-streaming performance in the G5 designs we already have.
Edit:
After thinking about it, I think I should be more clear about what I mean be FSB's and RAM at the "same" clock speed. You see, they often arent really the same, for example Vias old KT133 vs KT133A where going to a 266mhz (DDR) FSB from a 200mhz FSB helped add significant performance, when coupled with PC133 RAM. In the KT133A they even removed the ability to run the FSB and RAM at a different "base clock", and it paid dividends. Really what matters is the "base clock", which is often (or always?) the address clock speed. Intel's quad-pumped FSB has the same address clock as DDR at half the data clock, so you'll see that arrangement a lot.
Mr. MacPhisto:
I remain sceptical of the whole thing.
While my first response went something like this:
"The true g5.... orgasm"
I quickly realized.. so what? It might be news to us, but I'm quite sure Steve knew this was coming. We've had our g5's for a year or so now... let's not look back. If it makes it into the i-family.. then awesome, but this isn't good for much else.
"The true g5.... orgasm"
I quickly realized.. so what? It might be news to us, but I'm quite sure Steve knew this was coming. We've had our g5's for a year or so now... let's not look back. If it makes it into the i-family.. then awesome, but this isn't good for much else.
ddtlm said:JFreak:
Even when G4's had L3's and a much for favorable core-to-FSB ratio they got their butts kicked by the competition. I maintain that the FSB is less of a problem for most tasks than people think.
Mr. MacPhisto:
I remain sceptical of the whole thing.
I've never seen any evidence of this. I remember some tests of AfterEffects that showed the Dual 1.42 GHZ PowerMacs getting toasted by a 3.2GHZ P4, BUT AfterEffects is not optimized for the Mac at all (G4 or G5) and is well optimized for the Pentium, so it's not a fair comparison at all. I also believe those G4s only had 256K of L2 cache compared to the P4's 512K. Jack up the L2 to equal levels, jack up the bus speed to 400MHZ+ (Moto chips are more efficient with their bus than Intel chips are, btw. MaxBus isn't spectacular anymore, but a 167MHZ FSB on a G4 is far more efficient than the same on a Pentium. Also remember that it's running without clock doubling, quadrupling tech. The Pentiums run on buses that have been quadrupled usually), and put an on-chip memory controller - with the chip running at 2GHZ, and my guess is the G4 will keep up with the new 3.6GHZ Prescott - one of the most inefficient hogs of a chip ever fabbed. Make the G4 a dual-core chip that draws 1/4 the power of the Prescott and you've got yourself a great chip that will run circles around the Pentiums and Centrinos of the world - all the while being more energy efficient and much faster due to SOC design.
Also note that Moto already is working on 65nm, 45nm, and 32nm. Who knows then Intel will get down to 65nm - and thus far there's no indication they can go beyond that. I've talked to enough people so I'm definitely not skeptical. Mark my words, sometime between this summer and fall you will be astonished by what Moto puts out. You also will be floored by the price. Crolles will make their chips much, much cheaper. No longer does Moto have to make due with a dirty fab - they've got one of the world's best fabs, and some of the top engineers in the world working there. It's not just Moto working on these new chips - they're doing it with Phillips and STMicroelectronics. You may not believe it, but with IBM and Moto on board Apple's CPU future looks much brighter than the PC's.