Getting back to the topic of cells, it is likely too early for outsiders to know much. It may not be something that replaces todays processors, but is a means of taking processors to the next level of performance. There may be nothing to prevent cell technology from implementing PPC and AltiVec instructions. With IBMs heavy investment in the PPC architecture, I am sure they would be working on ways to extend the usefulness of the PPC instruction set, not to replace it completely with something totally new.
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.
PS3 chip = 1 teraflop
- Thread starter chibianh
- Start date
- Sort by reaction score
sure, we made the mistake of the 9.90GFLOPS calculation, but you haven't proven wrong the main point of our posts. All you've stated is that a G4 is 51.5% faster than a P4, not that a P4 is 50% slower than a G4.
15*0.5<9.90
(15-X)/15 = 0.34
where X, the GFLOPS of a P4 = 9.90
therefore the P4 is only 34% slower than a G4
umm, the P4's cache is running at full processor speed (ie 2.54Ghz), while the current G4, the PPC 7455 has only 256k of L2 cache running at a max of 1Ghz. The L3 cache is only 4ns, meaning that it can only run at a rated speed of 250Mhz (500Mhz effective. ie 1000/4 = 250). More cache means squat if there's no speed to back it up
I'll try another angle here.
It's completely obvious from the discussion so far that there is no real way to compare a G4 and a P4 that isn't inherently biassed. In terms of there being no available mobo for the G4 to match the P4's available set.... that just re-enforces my point. You're not actually comparing chips so don't say you are.
You may as well be argueing about comparing a full-size cargo diesel to a Japanese Motorcycle like a Honda CBR-900. There is just too much difference in what they're designed to do and how power is delivered.
As to the main topic: IBM has been siting on multi-core chip technology and core cross-connect for quite a while and it IS already on the PPC roadmap.
It's completely obvious from the discussion so far that there is no real way to compare a G4 and a P4 that isn't inherently biassed. In terms of there being no available mobo for the G4 to match the P4's available set.... that just re-enforces my point. You're not actually comparing chips so don't say you are.
You may as well be argueing about comparing a full-size cargo diesel to a Japanese Motorcycle like a Honda CBR-900. There is just too much difference in what they're designed to do and how power is delivered.
As to the main topic: IBM has been siting on multi-core chip technology and core cross-connect for quite a while and it IS already on the PPC roadmap.
Re: I'll try another angle here.
You can compare them based on their theoretical maximum performance or on their actual real-world performance. You're apparently trying to compare based on the former, and I'm trying to compare based on the latter. What good is theoretical maximum performance when it's only theoretical and it will realistically never be achieved? Sure a 486 could whip the pants off a P4 if it were overclocked to 5GHz with 500MHz DDR RAM, but it's impossible to do that, so who cares? Same with the G4: Yes, it would probably be as fast or faster than the P4 if it were able to utilize some faster memory and if it were clocked higher, but until it is, who cares.
Alex
Originally posted by mischief
It's completely obvious from the discussion so far that there is no real way to compare a G4 and a P4 that isn't inherently biassed. In terms of there being no available mobo for the G4 to match the P4's available set.... that just re-enforces my point. You're not actually comparing chips so don't say you are.
You can compare them based on their theoretical maximum performance or on their actual real-world performance. You're apparently trying to compare based on the former, and I'm trying to compare based on the latter. What good is theoretical maximum performance when it's only theoretical and it will realistically never be achieved? Sure a 486 could whip the pants off a P4 if it were overclocked to 5GHz with 500MHz DDR RAM, but it's impossible to do that, so who cares? Same with the G4: Yes, it would probably be as fast or faster than the P4 if it were able to utilize some faster memory and if it were clocked higher, but until it is, who cares.
Alex
We all forget one thing here...
A few years ago Apple predicted a market moving towards 64 bit processing and eventually to 128 bit processing much like our beloved N65's and PS2's.
Unfortunately for Apple, computers are sold based on clock speed, backwards compadibility is a must, and windows dominates the market. Therefore the market stayed at the current 32bit processing.
If Apple correctly predicted the market then the G4 processor would be a faster 32 bit processor or perhaps what the G5 is going to be, a 64 bit processor.
Everyone is talking about how the G5 is going to be super fast, but the truth of the story is that Apple is just going to recess its current technology to gain a marketable quality, clockspeed.
This 128 bit architechture is also the reason why Apple lags in bus speed. It is much harder to create a 400 mhz DDR system bus for a 128 bit architecture than a 32 bit architecture.
A few years ago Apple predicted a market moving towards 64 bit processing and eventually to 128 bit processing much like our beloved N65's and PS2's.
Unfortunately for Apple, computers are sold based on clock speed, backwards compadibility is a must, and windows dominates the market. Therefore the market stayed at the current 32bit processing.
If Apple correctly predicted the market then the G4 processor would be a faster 32 bit processor or perhaps what the G5 is going to be, a 64 bit processor.
Everyone is talking about how the G5 is going to be super fast, but the truth of the story is that Apple is just going to recess its current technology to gain a marketable quality, clockspeed.
This 128 bit architechture is also the reason why Apple lags in bus speed. It is much harder to create a 400 mhz DDR system bus for a 128 bit architecture than a 32 bit architecture.
Re: We all forget one thing here...
This is kind of misleading, because you're using different definitions of bitness. 128 bit here refers to the width of the on-chip vector unit. This vector unit still deals with the same 32-bit or 64-bit words as any other 32- or 64-bit CPU; it's just that it processes more than one simultaneously. The Dreamcast, for example, has a CPU that is 32-bit (according to its maximum word length), but it has an on-board 128-bit vector unit which can process four 32-bit words simultaneously (not two 64-bit words or one 128-bit word). Everybody calls the SH3 in the Dreamcast a 128-bit CPU, but by the common definition of the word, it's 32-bit.
It is possible to extend most 32-bit architectures to 64-bit with no loss of backwards compatibility. I know MIPS did this in the early '90s - it is possible to run a 32-bit MIPS binary from 1991 on a modern-day MIPS CPU with no loss of hardware compatibility (software compatibility is another story).
In the short term this is what they'll do. Nobody knows what they'll do in the coming years. If the G5 were to arrive today with the specs it is said to have, it would be super-fast, but it has been vaporware basically forever, so...
Alex
This is kind of misleading, because you're using different definitions of bitness. 128 bit here refers to the width of the on-chip vector unit. This vector unit still deals with the same 32-bit or 64-bit words as any other 32- or 64-bit CPU; it's just that it processes more than one simultaneously. The Dreamcast, for example, has a CPU that is 32-bit (according to its maximum word length), but it has an on-board 128-bit vector unit which can process four 32-bit words simultaneously (not two 64-bit words or one 128-bit word). Everybody calls the SH3 in the Dreamcast a 128-bit CPU, but by the common definition of the word, it's 32-bit.
It is possible to extend most 32-bit architectures to 64-bit with no loss of backwards compatibility. I know MIPS did this in the early '90s - it is possible to run a 32-bit MIPS binary from 1991 on a modern-day MIPS CPU with no loss of hardware compatibility (software compatibility is another story).
In the short term this is what they'll do. Nobody knows what they'll do in the coming years. If the G5 were to arrive today with the specs it is said to have, it would be super-fast, but it has been vaporware basically forever, so...
This is not really true - the problem is that the G4 wasn't designed to support DDR. Presumably Apple has spent all this time grafting a DDR hack onto the new Power Mac motherboards.
Alex
Getting back in the flame war
A lot people who look at the gigaflop numbers and compare to Pentium 4's forget that 15 gigaflops is the theoretical maximum for *TWO* PowerPC G4's. And you're comparing against a single processor Pentium 4. How unfair. Compare to a Dual Xeon, it's no contest; or hell, a dual Athlon MP. Ever wondered why Lucasarts doesn't use G4's for render farms? They use Athlon MPs for render farms; Macs for drawing up and producing stuff. G4 is a great processor, especially for embedded products; but it sure sucks in the desktop market.
Message to Apple: Screw Motorola. Go IBM. Or hell, work with AAI (Apple, AMD, IBM) and you've got a kick ass combination. *sigh* If that would only ever exist. Or just take Power4, add water [actually, Altivec] and other accelerators, produce it in SOI .7micrometers, reduce power intake to around same level as Athlon, and you've got a kick ass PowerPC architecture CPU that can be used in a Mac [to my understanding, since Power4 is just an extremely powerful PowerPC, if I'm wrong, slap me in the face and bitch at me].
A lot people who look at the gigaflop numbers and compare to Pentium 4's forget that 15 gigaflops is the theoretical maximum for *TWO* PowerPC G4's. And you're comparing against a single processor Pentium 4. How unfair. Compare to a Dual Xeon, it's no contest; or hell, a dual Athlon MP. Ever wondered why Lucasarts doesn't use G4's for render farms? They use Athlon MPs for render farms; Macs for drawing up and producing stuff. G4 is a great processor, especially for embedded products; but it sure sucks in the desktop market.
Message to Apple: Screw Motorola. Go IBM. Or hell, work with AAI (Apple, AMD, IBM) and you've got a kick ass combination. *sigh* If that would only ever exist. Or just take Power4, add water [actually, Altivec] and other accelerators, produce it in SOI .7micrometers, reduce power intake to around same level as Athlon, and you've got a kick ass PowerPC architecture CPU that can be used in a Mac [to my understanding, since Power4 is just an extremely powerful PowerPC, if I'm wrong, slap me in the face and bitch at me].
alex_ant
Your wrong... You say your comparing the real world speeds of these computers.
How many fools go and buy a PC for graphic design.... If I recall correctly not many they prefere Macs... word processing doesn't matter as it doesn't require anything... internet doesn't matter as it doesn't require anything... 3d rendering.... okay so this is a function where your arguement may stand, but as far as I know Apple don't go for the 3d rendering market.... that's why PIXAR don't use G4's and everybody knows that!
Okay audio... well processor wise that isn't a huge ask it's just the bits and pieces to allow you to set it up thats the draw back.
and then there's games... well all the games I've played on my mac.. have run just as good if not better on my mac... so that doesn't matter much either.
So when your comparing real world speeds... well you sort of have to think about real world operations... and I find it hard to believe that everybody wants to do full scale rendering on there mac!
So please SHUT UP your posts are sooooo boring!
(as said by DannyZR2: "I feel a flame post coming on")
Your wrong... You say your comparing the real world speeds of these computers.
How many fools go and buy a PC for graphic design.... If I recall correctly not many they prefere Macs... word processing doesn't matter as it doesn't require anything... internet doesn't matter as it doesn't require anything... 3d rendering.... okay so this is a function where your arguement may stand, but as far as I know Apple don't go for the 3d rendering market.... that's why PIXAR don't use G4's and everybody knows that!
Okay audio... well processor wise that isn't a huge ask it's just the bits and pieces to allow you to set it up thats the draw back.
and then there's games... well all the games I've played on my mac.. have run just as good if not better on my mac... so that doesn't matter much either.
So when your comparing real world speeds... well you sort of have to think about real world operations... and I find it hard to believe that everybody wants to do full scale rendering on there mac!
So please SHUT UP your posts are sooooo boring!
(as said by DannyZR2: "I feel a flame post coming on")
Re: Re: Re: Re: PS3 chip = 1 teraflop
MMX went out of fashion in 1998, I suggest you read up on SSE, 3dnow! and SSE2
all of which DO SIMD on FP data, SSE2 even works with double precision FP
(MMX OTOH was a cheap 'n' nasty hack that didn't help much for pretty much anything.)
ARS Technica writeup on SIMD on the Pentium 4
MMX went out of fashion in 1998, I suggest you read up on SSE, 3dnow! and SSE2
all of which DO SIMD on FP data, SSE2 even works with double precision FP
(MMX OTOH was a cheap 'n' nasty hack that didn't help much for pretty much anything.)
ARS Technica writeup on SIMD on the Pentium 4
DDR Over-rated???
Read this from BareFeats.com
http://www.barefeats.com/xserve2.html
"I was curious how the Xserve with its DDR RAM and 2.1GB/s memory throughput would fair running non-server applications. Like, what if I wanted to use the Xserve as a number cruncher or several rack mounted Xserve's as a rendering farm? Is it worth the extra money?
I tested both the dual and single CPU Xserve models. Check it out...
After all the hype touting the Xserve's "phenomenal processing power," I was disappoined in the performance. The more expensive Double Data Rate Memory gives it no measureable advantage. "
As I said before.. I wonder if everyone is putting too much hope in the DDR, when we already have a great architecture with the L3 cache... and if they up that to 4MB per chip.. oh yeah.. that'd be nice...
Read this from BareFeats.com
http://www.barefeats.com/xserve2.html
"I was curious how the Xserve with its DDR RAM and 2.1GB/s memory throughput would fair running non-server applications. Like, what if I wanted to use the Xserve as a number cruncher or several rack mounted Xserve's as a rendering farm? Is it worth the extra money?
I tested both the dual and single CPU Xserve models. Check it out...
After all the hype touting the Xserve's "phenomenal processing power," I was disappoined in the performance. The more expensive Double Data Rate Memory gives it no measureable advantage. "
As I said before.. I wonder if everyone is putting too much hope in the DDR, when we already have a great architecture with the L3 cache... and if they up that to 4MB per chip.. oh yeah.. that'd be nice...