That's an interesting statement. A lot of what I read in these forums slams the Itanium pretty severly.
It only clocks to 800MHz
Performs at the equivalent of a PIII
etc.
Am I being lied to?</strong>
Itanium has Intel behind it. That's pretty much my sole reasoning for saying it. Intel will provide nearly everything for SGI. They'll sell the chip at about the same cost as a MIPS R12/14/16K. They'll sell core logic chipsets. They'll sell the motherboard. All SGI has to do is develop the graphics capabilities for which they should be known for.
SGI has no other options. Power4 seems to be an IBM only product. Alpha was dying. PowerPC is pretty iffy. SPARC is the processor of the enemy (Sun). Itanium has an 800 lb gorilla behind it. Mind that I wouldn't have moved away from MIPS, but if you have to reduce costs and abandon MIPS, go with the gorilla and keep your core competency, graphics.
As far as performance is concerned, this is Intel we're talking about. They are on their way to fixing some of the Itanium's problems with the upcoming McKinley version. The Merced version was decent in FP, but average in integer. It still would have competed against POWER3, SPARC and PA-RISC though. The key was more IRIX/Itanium, not NT/Itanium nor Linux/Itanium though. The double switch is hard, and SGI made that mistake by trying to sell NT/x86 machines. That custom core logic ASIC SGI built for its Pentium machines must have been costly.
It depends. $60 to 80 million annually is a lot if the processor is only going to end up in your own products. R&D plus third-party fab costs would be more, I suspect, than Apple currently pays out for supplies of G3s and G4s.</strong>
The gamble is that by designing one's own processor, one will be sure to design a faster processor than the competition and thereby sell more product. By selling more product, one gets more revenue.
<strong>Apple would need to know that the performance of a homegrown chip would be enough to attract new customers to Macintosh.</strong>
Considering the line of processors from the 750 to 7450, I think it wouldn't have been that hard of task.
Early on, if Apple decided to put backside cache onto the 604e, it should have outperformed the 750. If they extended the pipeline and added AltiVec, even better. (The 604e was a quad-issue chip and should have benifited more from memory improvements and better compilers.) It could have also given Apple the chance to keep clones around if they had exclusive rights to the fastest PPC chips.
<strong>I think it would also have to look at selling the chip - or some derivative - for purposes other than computers.
Sony's Emotion Engine is a good example - it was developed for the Playstation 2 (and apparently scaling for 3 and 4), but was designed (I assume from the start) to be a huge source of income for Sony through sales to third parties.</strong>
I think this is more due to the fact that Sony/Hitachi built a brand new 0.18 micron fab to manufacture the EE and they needed to recoup that investment. For Apple, they could have just piggybacked onto Intel's ample fab capability or an East Asian company's fabs.
<strong>Erm, afaik graphics chips are at least as, if not more complex than CPUs now. Sure, they used to be simpler, whitout 2D and 3D accelleration, back in the days of the first color monitors.
G-News</strong><hr></blockquote>
While modern GPUs have a higher transistor count than modern CPUs they are not more complex to design. They still for the most part use standard cell logic rather than the custom design logic that CPU designers use. That is why it only takes Nvidia 18 months to design a new GPU and a company like Intel or AMD 4-5 years to design a CPU.
No one's saying Apple should enter the fab business. They design the CPU. Intel, IBM, UMC, TSMC, TI, AMD, and Moto can fab it.</strong><hr></blockquote>
Chip design and fab process are intricate : that's why the yield of the G4 first version where so poor. In others words you need a strong cooperation between , design engineer , and fab engineer.
It's like building a skycrepers , the architecture depends on what kind of material you use (steel, stone).
Comments
That's an interesting statement. A lot of what I read in these forums slams the Itanium pretty severly.
It only clocks to 800MHz
Performs at the equivalent of a PIII
etc.
Am I being lied to?</strong>
Itanium has Intel behind it. That's pretty much my sole reasoning for saying it. Intel will provide nearly everything for SGI. They'll sell the chip at about the same cost as a MIPS R12/14/16K. They'll sell core logic chipsets. They'll sell the motherboard. All SGI has to do is develop the graphics capabilities for which they should be known for.
SGI has no other options. Power4 seems to be an IBM only product. Alpha was dying. PowerPC is pretty iffy. SPARC is the processor of the enemy (Sun). Itanium has an 800 lb gorilla behind it. Mind that I wouldn't have moved away from MIPS, but if you have to reduce costs and abandon MIPS, go with the gorilla and keep your core competency, graphics.
As far as performance is concerned, this is Intel we're talking about. They are on their way to fixing some of the Itanium's problems with the upcoming McKinley version. The Merced version was decent in FP, but average in integer. It still would have competed against POWER3, SPARC and PA-RISC though. The key was more IRIX/Itanium, not NT/Itanium nor Linux/Itanium though. The double switch is hard, and SGI made that mistake by trying to sell NT/x86 machines. That custom core logic ASIC SGI built for its Pentium machines must have been costly.
It depends. $60 to 80 million annually is a lot if the processor is only going to end up in your own products. R&D plus third-party fab costs would be more, I suspect, than Apple currently pays out for supplies of G3s and G4s.</strong>
The gamble is that by designing one's own processor, one will be sure to design a faster processor than the competition and thereby sell more product. By selling more product, one gets more revenue.
<strong>Apple would need to know that the performance of a homegrown chip would be enough to attract new customers to Macintosh.</strong>
Considering the line of processors from the 750 to 7450, I think it wouldn't have been that hard of task.
Early on, if Apple decided to put backside cache onto the 604e, it should have outperformed the 750. If they extended the pipeline and added AltiVec, even better. (The 604e was a quad-issue chip and should have benifited more from memory improvements and better compilers.) It could have also given Apple the chance to keep clones around if they had exclusive rights to the fastest PPC chips.
<strong>I think it would also have to look at selling the chip - or some derivative - for purposes other than computers.
Sony's Emotion Engine is a good example - it was developed for the Playstation 2 (and apparently scaling for 3 and 4), but was designed (I assume from the start) to be a huge source of income for Sony through sales to third parties.</strong>
I think this is more due to the fact that Sony/Hitachi built a brand new 0.18 micron fab to manufacture the EE and they needed to recoup that investment. For Apple, they could have just piggybacked onto Intel's ample fab capability or an East Asian company's fabs.
[ 02-06-2002: Message edited by: THT ]</p>
<strong>Erm, afaik graphics chips are at least as, if not more complex than CPUs now. Sure, they used to be simpler, whitout 2D and 3D accelleration, back in the days of the first color monitors.
G-News</strong><hr></blockquote>
While modern GPUs have a higher transistor count than modern CPUs they are not more complex to design. They still for the most part use standard cell logic rather than the custom design logic that CPU designers use. That is why it only takes Nvidia 18 months to design a new GPU and a company like Intel or AMD 4-5 years to design a CPU.
<strong>[qb
No one's saying Apple should enter the fab business. They design the CPU. Intel, IBM, UMC, TSMC, TI, AMD, and Moto can fab it.</strong><hr></blockquote>
Chip design and fab process are intricate : that's why the yield of the G4 first version where so poor. In others words you need a strong cooperation between , design engineer , and fab engineer.
It's like building a skycrepers , the architecture depends on what kind of material you use (steel, stone).