Motorola MPC7457 PDF

[Phil Of Mac]

Re: Re: Re: 7457-RM probably on smaller process size in 2nd half of 2004

Originally posted by Phinius
Well with a G4 topping out at about 2.4 GHz on a 90-nm process and since Motorola will be putting two cpus on one chip, that would make a single G4 chip computer about twice as fast as a dual 1.42 GHz G4 Power Mac. Since Apple would be able to put two of these dual-cpu G4 chips in a computer, then that would be about 4X faster than a 1.42 GHz G4 Power Mac.

Yal, who would want to buy that? 4X faster? Nay it doesn't say G5 though and we all know 5 is a bigger number than 4 now don't we. Geez...

In the second half of 2004, we're going to have PPC 980's at 90 nm, at frequencies well past 3 GHz, quite possibly 4 if not 5 GHz. Apple will sell these in a dual if not quad processor form. In the meantime, Motorola will slip behind schedule again, so your mythical dual-core 2.4 GHz G4 will arrive in 2005, along with the IBM G6

Stick a fork in Motorola. It's done.

 

[Analog Kid]

Re: Re: Re: The watt usage you list is for the lower volt 7447

Originally posted by Sun Baked
On the MPC 7457 Product Page there is a document updated on 7/24/2003 called MPC7457 Part Number Specification for the MPC74x7RXnnnnNx Series

Which must be really hard to find since it's right after the 7457EC pdf.

Ok that's not fair-- going to primary sources... Jeesh!

Thanks for posting this though-- my world just got a lot rounder. This is much more in line with the power numbers I'd have expected from the process shrink.

Also gives a lot more headroom.

Sorry to make you dig that up again, I should have spent more time at the Mot site.

 

[Analog Kid]

Re: Re: Re: Re: 7457-RM probably on smaller process size in 2nd half of 2004

Originally posted by Phil Of Mac
In the second half of 2004, we're going to have PPC 980's at 90 nm, at frequencies well past 3 GHz, quite possibly 4 if not 5 GHz. Apple will sell these in a dual if not quad processor form.

Great, now when Apple releases a dual 3.2GHz G5 next year everybody is going to complain that they remember hearing Apple was supposed to be up over 5GHz by that time in a quad format...

 

[Phinius]

Re: Re: Re: Re: 7457-RM probably on smaller process size in 2nd half of 2004

Originally posted by Phil Of Mac
In the second half of 2004, we're going to have PPC 980's at 90 nm, at frequencies well past 3 GHz, quite possibly 4 if not 5 GHz.

Your dreaming. Both Apple and IBM have announced that the G5 will hit 3 GHz a year from now. The next version of the Pentium will peak at 5 GHz sometime in 2005. So the likelihood of a 970 or 980 processor hitting 4 or 5 GHz by the end of 2004 is extremely unlikely.

In the meantime, Motorola will slip behind schedule again, so your mythical dual-core 2.4 GHz G4 will arrive in 2005, along with the IBM G6

Stick a fork in Motorola. It's done.

Motorola would not have announced the upcoming production of a dual-cpu G4 unless Apple was commited to using it extensively. There would simply not be much demand for it outside of Apple since a dual-cpu G4 chip would have too much power use for the embedded market.

Motorola has joined forces with Phillips and STMicroelectronics to improve chip manufacturing. The head of Motorola's microelectronics division has already announced that the company is using test chips on a 90-nm process in STMicroelectonics Crolles facility in France and PowerPC processors are scheduled to begin production in 2004.

 

[Phil Of Mac]

Phinius, do you work for Motorola, or do you just not get it?

Motorola is done as a supplier of processors to Apple. Next summer, IBM will release the 980. Motorola will still be on the 7457. IBM will be at 90nm. 970's and 980's will quickly migrate to Apple's entire product line. Once IBM reaches 90 nm (probably January or so), the G5 is going to go into PowerBooks and iMacs. iBooks and eMacs won't be far behind. By the second half of 2004, when Motorola is scheduled to have their dual-core G4 (which is a few YEARS late already), Apple might not even have any G4 products! By 2005, when Motorola will finally have that G4 (has Motorola ever been on schedule?), 2.4 GHz will be old news.

OK, Motorola's fixing their production problems. Next they need to fix their design problems. 2.4 GHz and 90 nm by the end of 2004? Psshh. IBM'll have that beat months ahead.

 

[Phinius]

Originally posted by Phil Of Mac
Phinius, do you work for Motorola, or do you just not get it?

Motorola is done as a supplier of processors to Apple.

Maybe you should tell Motorola and Apple since you seem to be much more informed about it than those two companies.

You don't seem to grasp the reality that if Apple eliminates Motorola as a supplier for PowerPC processors, then Apple will be completely dependent and at the mercy of IBM. That makes it very unlikely that Apple would go with one suppler for processors.

Next summer, IBM will release the 980.

There are strong indications that IBM might still be calling it the 970 in mid 2004.

Motorola will still be on the 7457. IBM will be at 90nm. 970's and 980's will quickly migrate to Apple's entire product line.

Motorola will still be on the 7457 in 2004 and it will be made on a 90-nm process. Whether Apple moves the G5 over to the consumer products remains to be seen. Apple's recent history indicates that they like to have different processors on the consumer and pro lines to differentiate them.

Once IBM reaches 90 nm (probably January or so), the G5 is going to go into PowerBooks and iMacs.

IBM is just getting started on a 130-nm process for the 970 in August and you already have the 90-nm process pegged for January? Geez, you expect IBM to move faster than Intel? The 970 should still have some speed improvements to made it on the 130-nm process.

It could be that Apple will move the G5 to the PowerBooks in January. Which would give them a lot more room to improve the speed of the consumer line with the G4.

By the second half of 2004, when Motorola is scheduled to have their dual-core G4 (which is a few YEARS late already), Apple might not even have any G4 products!

What do you mean years late? It's doubtful that Motorola or any other mass production chip manufacturer for consumer products would find it financially feasible have a dual-core chip in production years ago.

By 2005, when Motorola will finally have that G4 (has Motorola ever been on schedule?), 2.4 GHz will be old news.

You seem to be using Intel's chip frequencies as the yardstick by which to measure PowerPC performances. Having two 2.4 GHz processors on one chip has much more performance than a single 2.4 GHz processor of the same design. Motorola is simply boosting performance by adding another G4 processor onto the chip. It will have the equivilant performance of a single G4 processor running at a much higher frequency.

OK, Motorola's fixing their production problems. Next they need to fix their design problems. 2.4 GHz and 90 nm by the end of 2004? Psshh. IBM'll have that beat months ahead.

IBM is unlikely to have a dual-core G5 chip by mid year 2004, whereas Motorola is very likely to have a dual-core G4 chip. That's two processors on one chip. That alone will boost the performance of a G4 chip by 50% or more.

 

[Phil Of Mac]

Originally posted by Phinius
Maybe you should tell Motorola and Apple since you seem to be much more informed about it than those two companies.

Apple already knows this.

Originally posted by Phinius
IBM is just getting started on a 130-nm process for the 970 in August and you already have the 90-nm process pegged for January? Geez, you expect IBM to move faster than Intel? The 970 should still have some speed improvements to made it on the 130-nm process.

Yes, but the 90nm will come quite quickly. Yes, possibly before Intel. January is a date I've seen reported.

Originally posted by Phinius
What do you mean years late? It's doubtful that Motorola or any other mass production chip manufacturer would find it financially feasible have a dual-core chip in production years ago.

The G4 was originally going to be a dual core processor from the outset. Which was 1999. They settled for single core. If they're finally going to reach dual core in 2004, well, that's great. I hear Copland's almost finished too.

Originally posted by Phinius
You seem to be using Intel's chip frequencies as the yardstick by which to measure PowerPC performances. Having two 2.4 GHz processors on one chip has much more performance than a single 2.4 GHz processor of the same design. Motorola is simply boosting performance by adding another G4 processor onto the chip. It will have the equivilant performance of a single G4 processor running at a much higher frequency.

That's nice. By the end of 2004, two G4's should be good enough for half a computer.

Originally posted by Phinius
IBM is unlikely to have a dual-core G5 chip by mid year 2004, whereas Motorola is very likely to have a dual-core G4 chip. That's two processors on one chip. That alone will boost the performance of a G4 chip by 50% or more.

Wow, that'll make it almost half as powerful as the G5.

How likely do you think it is that Motorola will have a dual-core G4 in 2004? Really! Do you have any clue how behind schedule Motorola usually is? Do you remember when they couldn't get the G4 above 500 MHz for about a year? If Motorola promises something for the end of 2004, I'm going to expect it in the summer of 2005. And Apple will too. They've dealt with Motorola enough that they know this far better than I do.

 

[Sun Baked]

Motorola SG1001 Product Selector Guide pdf

also updated 7/2003...

Just a listing of some of the semi sections products, including 7457 and current revisions. Seems 7455A has been bumped.

 

[ryan]

Originally posted by Analog Kid

[snip/]
I don't get the sense that the G5 is going to scale down as well. There's no L3 cache support, so as soon as you narrow the pipe to system memory you're going to starve the CPU (the G5 has more bandwidth to system RAM than the G4 has to it's L3 cache!).
[snip/]

Except that even if a G5 PB is running at *only* 1GHz it presumably would have a 500MHz bus, which is 3x faster than the bus in the current 1GHz PowerBooks, so CPU starvation probably wouldn't be an issue.

On a somewhat related note, I'm not convinced that the G4 is that bad of a chip, but instead it has a starvation problem. I think that is what makes this new info on the 7457 so disappointing, in that it appears it will still have a 167MHz bus so having it run at 1.3GHz won't do anyone much good.

 

[Phinius]

Originally posted by Phil Of Mac
Apple already knows this.

Apple is not going to put all their eggs in one basket and only using IBM processors would limit them to only one basket to choose from. That would not be a wise strategy.

Yes, but the 90nm will come quite quickly. Yes, possibly before Intel. January is a date I've seen reported.

You seem to be getting your information from rumor sites. Intel is the leader in getting a new process technology to market. Intel will have the Prescott version of the Pentium on a 90-nm process in the fourth quarter of 2003. IBM will likely have the 970 moved to a 90-nm process months into 2004 and not in January. Apple could have a 130-nm 970 processor in a PowerBook by the end of January however.

The G4 was originally going to be a dual core processor from the outset. Which was 1999. They settled for single core. If they're finally going to reach dual core in 2004, well, that's great. I hear Copland's almost finished too.

If what you say is true and judging from your previous responses you probably got that information from rumors, then Motorola might have decided to not have a dual-core earlier due to the die size being too cost prohibitive.

That's nice. By the end of 2004, two G4's should be good enough for half a computer.

If there are two 2+ GHz G4 processors on one chip, then Apple could use the same PowerMac G4 motherboard and use 2 of these chips, which would be 4 G4 processors. I'd hardly call 4 G4 processors running at 2+ GHz 'half a computer'.

Wow, that'll make it almost half as powerful as the G5.

Your wildly overestimating the speed advantage of the G5 over a G4 processor. It's very unlikely that a year from now a 3 GHz 970 processor would have double the performance of a dual-core 2 GHz G4 chip. In fact IBM estimated that a 970 running at 1.8 GHz would have a SPECint score of 937 and Motorola states that a single PowerMac G4 running at 1.25 GHz has a SPECint score of 500. Now your probably thinking that aha!, the G5 has almost double the speed of a G4 processor. But the G4 that Motorola mentions is made on the bigger 180-nm process. A 130-nm G4 can have up to a 50% higher frequency than that 1.25 GHz G4. So that would place it at about 750 on SPECint. Which means the G5 would have to score about 1500 on SPECint at 2.8 GHz on the 130-nm level to have double the G4s performance. Since Motorola intends to put two G4s on a chip, then that reduces the odds that a topend 970 processor will have double the performance of a topend G4 chip.

There are advantages and disadvantages to both designs. Because of the simpler design of the G4 processor compared to the 970, Motorola can shrink the size of the processor enough to put two of them on one chip at the 90-nm process level and still keep the die size competitive to the Intel chips. Whereas IBM would have a very large die size with a dual-core G5 chip made on a 90-nm process.

How likely do you think it is that Motorola will have a dual-core G4 in 2004? Really! Do you have any clue how behind schedule Motorola usually is? Do you remember when they couldn't get the G4 above 500 MHz for about a year?

What year is this, 1997 or 2003? Times change. Do you realize that IBM announced in October of 2000 that the 750FX would debut in January 2002 and it would reach speeds as high as 1 GHz? Well it's July of 2003 and IBM now states that the 750FX will only go up to 900 MHz. So you have more faith in IBM?

If Motorola promises something for the end of 2004, I'm going to expect it in the summer of 2005. And Apple will too. They've dealt with Motorola enough that they know this far better than I do.

Well Motorola had previously stated that the 7457 will be in production in the fourth quarter of 2004. I'd expect Apple to be announcing products that use this chip sometime in the next two months.

Motorola has not stated when the dual-core PowerPC processors will be in production. But it's obvious that it would occur no sooner than mid year 2004. Since a Motorola executive also stated that at 1.5 GHz the dual-core G4 would use 25 watts. That would have to be on a 90-nm process size since a 130-nm G4 will use about 17 watts at 1.3 GHz.

You seem to pick out only those points that support your beliefs rather than looking at all the pluses and minuses. The G5 is meant to be the topend chip for Apple that competes against the desktop Pentium processors. But Apple still has to compete against the Celerons and Pentium M processors and a smaller, cheaper to produce processor like the G3 or G4 would be much better suited in that market than the bigger and costlier G5 processors.

Why do you insist on believing that the G5 will replace all the G4 processors when the G4 did not eliminate the G3 in Macs? The G5 is the new topend chip for Apple and the topend G4 processors will now move downstream to the consumer models. With IBM announcing a doubling of L2 cache for the 750 G3 processor scheduled for December, it's very likely that Apple will even continue to use that lowly chip. Or do you honestly believe that a 750GX processor running at 1.1 GHz will replace the upcoming 7457 processors that will run up to 1.8 GHz?

 

[Analog Kid]

Originally posted by ryan
Except that even if a G5 PB is running at *only* 1GHz it presumably would have a 500MHz bus, which is 3x faster than the bus in the current 1GHz PowerBooks, so CPU starvation probably wouldn't be an issue.

Well, yes and no... Remember that you've got two busses to think about with the G4-- the cache bus and the main memory bus.

At first blush (off-the-top-of-my-head), the G4 gets 64bits*200MHz*DDR out of the L3, while the G5 gets 32bits*500MHz*DDR out of system memory-- so the G5 bus might still be a bit faster...

But then there's the fact that system memory is DRAM and cache is SRAM. DRAM usually has a pretty nasty (~50ns) precharge penalty for the first access (and then bursts after that). As your bus speed slows, that first access will start to dominate.

When I get more time, it would be interesting to look at the bus bandwidths at these speeds...

Even if the G5 does win on bandwidth, it's still clocking 30% slower than the G4 here. At these frequencies it might become a wash for most applications.

Originally posted by ryan
On a somewhat related note, I'm not convinced that the G4 is that bad of a chip, but instead it has a starvation problem. I think that is what makes this new info on the 7457 so disappointing, in that it appears it will still have a 167MHz bus so having it run at 1.3GHz won't do anyone much good.

Yeah, that bugged me too until I actually looked at the numbers.

Turns out that the processor speed is jumping 30% and the 133-167MHz bus is a 26% jump. Not that bad given the extra L2 cache size.

If Apple gets a 200MHz bus, then thats about a 50% improvement, which exceeds the core clock boost.


I'm not really interested in doing a full blown analysis of these things knowing that I'll find out eventually anyway, but my gut tells me the G4 will be the better laptop processor for a while...

 

[ryan]

Turns out that the processor speed is jumping 30% and the 133-167MHz bus is a 26% jump.

The current 1GHz PowerBooks already have a 167MHz bus.

 

[herocero]

bad news

http://www.hardmac.com/niouzcontenu.php?date=2003-07-27#263

now if french-mac has a source who even knows stevie's slang for crappy motorola production, well done!

off to buy a 1GHz ti . . .

-cero

 

[panphage]

Originally posted by Analog Kid Even if the G5 does win on bandwidth, it's still clocking 30% slower than the G4 here. At these frequencies it might become a wash for most applications.

I think his point with the 1ghz G5 is that even a G5 that "slow" would waste a similarly clocked G4 because the G5 has fatter pipes. No one has put forward 1Ghz as a serious number for the low end of mobile G5s. I keep reading 1.4. Of course these numbers are coming from uninformed mac zealots who are reading rumor sites in a frenzy of upgrade fever (oops! That's me!) but still, the G5 for the desktop starts at 1.6. If the die shrink is what's needed for mobile G5s, why would we think clock speed would drop by about 1/2?

 

[avus]

"Scum-a-rola"

Originally posted by herocero
http://www.hardmac.com/niouzcontenu.php?date=2003-07-27#263

now if french-mac has a source who even knows stevie's slang for crappy motorola production, well done!

This rumor by MacBidouille is quite funny but depressing at the same time. The bit about Steve giving Motorola a nickname "Scum-a-rola" is very plausible, and I don't understand why some people still have a faith in whatever Motorola executives said about their wonderful "plan" for the G4.

 

[Sun Baked]

Re: "Scum-a-rola"

Originally posted by avus
This rumor by MacBidouille is quite funny but depressing at the same time. The bit about Steve giving Motorola a nickname "Scum-a-rola" is very plausible, and I don't understand why some people still have a faith in whatever Motorola executives said about their wonderful "plan" for the G4.

Could you imagine the wonderful new aluminum machine we would have if Motorola rolled the 7457-RM out on time?

From an Apple pdf (I did not change it)...

 

[Cubeboy]

Originally posted by Analog Kid
Well, yes and no... Remember that you've got two busses to think about with the G4-- the cache bus and the main memory bus.

At first blush (off-the-top-of-my-head), the G4 gets 64bits*200MHz*DDR out of the L3, while the G5 gets 32bits*500MHz*DDR out of system memory-- so the G5 bus might still be a bit faster...

But then there's the fact that system memory is DRAM and cache is SRAM. DRAM usually has a pretty nasty (~50ns) precharge penalty for the first access (and then bursts after that). As your bus speed slows, that first access will start to dominate.

When I get more time, it would be interesting to look at the bus bandwidths at these speeds...

Even if the G5 does win on bandwidth, it's still clocking 30% slower than the G4 here. At these frequencies it might become a wash for most applications.

I haven't read all the technical documentation for the G5 yet but doesn't it utilize dual 32 bit buses? That would be the only way to explain how a 2 GHz G5 can have a 8 GB/s peak bandwidth. If so, than the 1 GHz G5's dual 32 bit buses would be the equivalent of a single 64 bit 500 MHz bus which would make it over twice as fast as the G4's "cache bus". As you've mentioned, the G4's cache bus has much lower latency but I doubt that it can match or even come close to the throughput of a G5, even a 1 GHz model.

Remember that the G5 also has a much stronger FPU than the G4, and will probably win in programs heavy in fp code even if it is clocked 30% slower. This is quite evident in Scalar Jet3d (which is nearly all floating point math) in which the G5, running unoptimized code, performed significantly better clock to clock than the G4.

 

[Ludio]

Re: bad news

Originally posted by herocero
http://www.hardmac.com/niouzcontenu.php?date=2003-07-27#263

now if french-mac has a source who even knows stevie's slang for crappy motorola production, well done!

off to buy a 1GHz ti . . .

-cero

I must say this truly sucks as this is my frist response/post to the forums on reading this news. Im now im a dilemma if I should at the end of the month put in a pre-order for a desktop G5 to arrive in the fist half of sept or should I just get the 17in PB since it will arrive on time if I order it later in august. The school im going to uses nothing but macs on campus so im to worried about capalibity problems but don't know if i'll need the portability or just like as an added convience and the discount with the payment plan is another plus from apple. I just wish the PB was more up to date becuase I know I will be getting rid of it next about this time becuase the tech offered will be severely outdated to desktop and other laptop competitors either way.

 

[Phil Of Mac]

Originally posted by Phinius
Apple is not going to put all their eggs in one basket and only using IBM processors would limit them to only one basket to choose from. That would not be a wise strategy.

As we've just seen, it isn't a wise strategy, because you mind end up with that one supplier utterly stagnating. Like Motorola.

Apple may choose, say, AMD as another supplier. But Motorola only wants the embedded sector. They haven't been doing well with computers and they don't want to bother with it. So a second supplier is a feasible strategy. Motorola is not.

Originally posted by Phinius
You seem to be getting your information from rumor sites. Intel is the leader in getting a new process technology to market. Intel will have the Prescott version of the Pentium on a 90-nm process in the fourth quarter of 2003. IBM will likely have the 970 moved to a 90-nm process months into 2004 and not in January. Apple could have a 130-nm 970 processor in a PowerBook by the end of January however.

Doubtful, unless Apple chooses not to release a new PowerBook G4. The 130nm 970 is already, at low frequencies, comparable with the heat and power of the G4. However, at 90 nm, heat and power necessities will be reduced even more, making it more feasible to build a truly high-speed PowerBook that befits the name.

The 970 at this point requires a very large heat sink that would ruin the sleek design of the PowerBook. 90nm is, essentially, a necessity.

Originally posted by Phinius
If what you say is true and judging from your previous responses you probably got that information from rumors, then Motorola might have decided to not have a dual-core earlier due to the die size being too cost prohibitive.

I suppose that's the same reason it stayed at 500 MHz for a year, it would be too cost prohibitive to design a faster chip.

Originally posted by Phinius
If there are two 2+ GHz G4 processors on one chip, then Apple could use the same PowerMac G4 motherboard and use 2 of these chips, which would be 4 G4 processors. I'd hardly call 4 G4 processors running at 2+ GHz 'half a computer'.

By today's standards, it would be impressive. By the standards of the end of 2004, it might contend with anything, except a dual G5.

Originally posted by Phinius
Your wildly overestimating the speed advantage of the G5 over a G4 processor. It's very unlikely that a year from now a 3 GHz 970 processor would have double the performance of a dual-core 2 GHz G4 chip. In fact IBM estimated that a 970 running at 1.8 GHz would have a SPECint score of 937 and Motorola states that a single PowerMac G4 running at 1.25 GHz has a SPECint score of 500.

IBM *estimated* that? Well! How convienient! Now my suspicion that you don't keep up with the news is confirmed.

Convieniently, there have been two recent benchmark tests of the G5. One of them, an independent test done for Apple, gives the G5 a SPECint of 800, and a SPECfp of 840.

NASA benchmarks rate the 2 GHz G5 at 254 MFLOPS and the 1.25 GHz G4 at 129 MFLOPS, scalar. Vector benchmarks are 2755 and 1612, respectively.

Originally posted by Phinius
Since Motorola intends to put two G4s on a chip, then that reduces the odds that a topend 970 processor will have double the performance of a topend G4 chip.

I suggest you look up "hyperbole". The G5 will still outperform your pathetic 2.4 GHz dual-G4 significantly. Especially if Motorola's past history is any indication. A dual G4 can't keep up with a single Pentium. A single G5 is near parity with a single Pentium and will quickly overtake it. So why do you think a dual G4 after another year of IBM innovation and Motorola's lack thereof will be even close to the G5?

Originally posted by Phinius
What year is this, 1997 or 2003? Times change.

Motorola hasn't changed. If you read the news, you'd know that Motorola is the one responsible for Apple falling behind since the glory days of the G3.

Originally posted by Phinius
The G5 is meant to be the topend chip for Apple that competes against the desktop Pentium processors. But Apple still has to compete against the Celerons and Pentium M processors and a smaller, cheaper to produce processor like the G3 or G4 would be much better suited in that market than the bigger and costlier G5 processors.

Why do you insist on believing that the G5 will replace all the G4 processors when the G4 did not eliminate the G3 in Macs?

The G4's only real advantage over the G3 was AltiVec. There was no cost effectiveness in putting it in all products at once. The G3, however, did immediately replace the 604e and 603e. Why? It was so much faster that it was cost-effective to!

Come 90nm, the G5 will, or should, replace the G4 and G3 entirely. Motorola's unreliability and the advantage of the G5 design necessitate it.

 

[daveL]

Phil Of Mac

Right on (that's a technical term from the 60's)!

 

[sacrilicious]

Re: Phil Of Mac

Originally posted by daveL
Right on (that's a technical term from the 60's)!

Word. I'm inspired to love Apple again and forget that they're never going to release a new 15".

 

[Analog Kid]

Originally posted by panphage
I think his point with the 1ghz G5 is that even a G5 that "slow" would waste a similarly clocked G4 because the G5 has fatter pipes. No one has put forward 1Ghz as a serious number for the low end of mobile G5s. I keep reading 1.4. Of course these numbers are coming from uninformed mac zealots who are reading rumor sites in a frenzy of upgrade fever (oops! That's me!) but still, the G5 for the desktop starts at 1.6. If the die shrink is what's needed for mobile G5s, why would we think clock speed would drop by about 1/2?

Actually, I think he was continuing on a point I made about starving the G5 if you slow down the buses (his point, I think, is that if you slow the CPU down too everything stays happy and the G5 still has a faster bus).

I don't think a 1GHz G5 would "waste" a 1.3GHz G4. They may be quite comprable at those speeds.

Why would the speed drop? More than just a die shrink on the G5 is required to cram it into a laptop. In fact, the more I think about it, that's probably the least important change.

I think the biggest challenge will be keeping that beast fed out of memory without the aluminum case glowing orange.

I've gone over all the system components a dozen times... In order to discuss it any more intelligently, we'd need better benchmarks of the G5 and G4 with and without cache...

It would also be interesting to see how the G5 runs with the front side bus clocking at 1/6 of the system clock, for example.

 

[Analog Kid]

Originally posted by Cubeboy
I haven't read all the technical documentation for the G5 yet but doesn't it utilize dual 32 bit buses? That would be the only way to explain how a 2 GHz G5 can have a 8 GB/s peak bandwidth. If so, than the 1 GHz G5's dual 32 bit buses would be the equivalent of a single 64 bit 500 MHz bus which would make it over twice as fast as the G4's "cache bus". As you've mentioned, the G4's cache bus has much lower latency but I doubt that it can match or even come close to the throughput of a G5, even a 1 GHz model.

Yeah, but they're dual unidirectional 32bit buses, so they don't actually act as a 64bit bus. It was a trick (in the good sense of trick) to get the bus speed up-- they don't have to wait to turn the bus around between cycles.

I don't know how the system will interleave writebacks with reads. There's two busses coming out of the part, but only one to the memory-- so the system controller has to handle all that.

All this makes direct comparison kinda tough, which is why I just took a rough cut at it.

I don't know how quickly I'd write off the DRAM latency... A 50ns precharge latency is 25 beats of a 500MHz system clock, or 50 missed data transfers at DDR. No small potatoes.

At this point I'm only raising questions, I don't pretend to have the answers to most of them. It's just proving to myself that it's not all as cut and dry as some seem to think.

Originally posted by Cubeboy
Remember that the G5 also has a much stronger FPU than the G4, and will probably win in programs heavy in fp code even if it is clocked 30% slower. This is quite evident in Scalar Jet3d (which is nearly all floating point math) in which the G5, running unoptimized code, performed significantly better clock to clock than the G4.

Looks like it's 23% faster clock to clock... It's unoptimized, but I don't know enough about the benchmark to know if the whole thing ran in G4 cache or if the memory bandwidth played any role.

Again, I don't know either way, but it's not as cut and dry as some may think...


I'm really excited about the G5 as a desktop processor where you don't need to compromise. I'm not sure when it'll make sense to stick it in a portable. Heat, battery life, and form factor are important to me-- so I'm more likely to look at power/performance rather than just performance.

 

[Cubeboy]

Analog Kid:

The benchmark is basically all floating point math, it has a pretty small memory footprint (around 1 MB I hear) so bandwidth limitations isn't a problem for either processor.

The entire program should be able to fit inside the G4's L3 cache and might fit in the G5's on-die cache through hardware prefetching provided by the extra bandwidth from the motherboard channel.

I wouldn't doubt a 20%+ performance boost with optimized code, considering the Absoft compiler used was geared towards producing optimized code for the G4.

 

[iPC]

clock speed jump, and more L2 cache. Guess we'll see 1GHz with 1MB L3 cache and 1.3GHz with 2MB L3 cache 15.4" PB sometime September...???

Bus speed of 167MHz instead of 200MHz doesn't mean much.