Quad Processor Macs???

[wrldwzrd89]

wrldwzrd89:


That's certainly possible, but its still really just a 4 CPU machine. SMT does not make one CPU perform like two.

Yes, you are correct - however, the advantages of such a design over a traditional 4-CPU design are numerous:
1. Because there are still only 2 CPU "units" (each containing 2 CPUs), Apple doesn't need to make significant design changes to the G5 to accomodate the new chips.
2. Only two system buses would be required, further simplifying the design.
3. Even though SMT isn't the same as having true dual processors, Apple would surely market the fact that SMT is supported in these Macs (perhaps even going so far as to call it an "effective-8 CPU PowerMac").
4. Assuming that SMT gives you 130% of the performance you'd get without it, 30*4=120, so this quad-CPU design will perform like it has over 5 CPUs.

 

[Sun Baked]

I was just thinking - why not swing for the fences with an 8 CPU design? Here's how this one would work: the Mac would have two dual-core CPUs (that's 4 right there), plus SMT to create 8 effective CPUs. This way, Apple wouldn't need 4 or 8 system buses, and would still get most of the performance benefits from the additional processors. (BTW, if Mac OS X is dual-CPU aware, that automatically makes it n-CPU aware.)

If you have a handy-dandy copy of Tiger sitting around, try this command in the terminal...

sysctl -a

Which "turns up two interesting values, specifically the number of physical and logical processors." (possible SMT hint from Catfish_Man42).

The only thing that strikes me odd about MOSR true quad is the "maximum" power dissipation at 3.0 GHz Power5-UL which would probably be 100-120W per CPU -- times four -- on the consevative side.

The 970 @ 2GHz is near the top of that and the 2.5GHz 970FX near the low end of the 100-120W, why would a 3.0GHz chip that's 20% larger than the FX and clocked higher be much lower power consumer on the same process (it probably wouldn't).

I can see a 2.5GHz Power5-UL kicking in with a 35-40% performance boost at the same power consuming fury of the 970 -- for a psuedoquad, but not 4 of them.

 

[Sun Baked]

[Delete me]

 

[ddtlm]

wrldwzrd89:

Because there are still only 2 CPU "units" (each containing 2 CPUs), Apple doesn't need to make significant design changes to the G5 to accomodate the new chips.

This is true, and its not a new idea. Intel has been talking about dual core hyperthreadeed P4's (and presumbly Xeons) already.

Only two system buses would be required, further simplifying the design.

System busses? IBM needs to follow AMD's lead and go on-die memory controllers.

Assuming that SMT gives you 130% of the performance you'd get without it, 30*4=120, so this quad-CPU design will perform like it has over 5 CPUs.

SMT is probably not worth that much, and even if it were in some cases, it would get progressively harder to benefit from as the number of real CPUs increases. I don't know of anyone still making a quad CPU workstation, let alone quad SMT processors. Even "power users" will have a hard time keeping so many CPUs busy (unless they are doing distributed computing, but then I would wonder why they don't save money and just get a dedicated machine for that).

Sun Baked:

The 970 @ 2GHz is near the top of that and the 2.5GHz 970FX near the low end of the 100-120W, why would a 3.0GHz chip that's 20% larger than the FX and clocked higher be much lower power consumer on the same process (it probably wouldn't).

Hold on there, those power specs have never been released. It's almost certain that no shipping PPC970 have ever breached 90W of dissipation, based on the "typical" ratings. A more reasonable estimate on the 2.5ghz 970fx is probably 70W peak.

It's true that IBM's PowerX chips dissipate a lot, but thats two cores per die and they've never been fabbed at 90nm, yet.

 

[Freg3000]

Fuel to the fire:

The Quad Power are going to be announced at Apple Expo Paris, according to multitude of sources found deep in the grape vine( 67.7% reliability). Rumors previously stated that the Quads featured 4 3.0 GHz processors but now sources say that 2.8GHz Quad PowerMacs also exist.

This is a breaking story, please stay tuned for details and sketches.

67.7% I love this site.

 

[Duff-Man]

Fuel to the fire:



67.7% I love this site.

Duff-Man says....check out todays postings...he's got a "prototype" G5 Powerbook...uh huh...Duff-Man wants some of what he's smokin'....oh yeah!

 

[Chip NoVaMac]

Fuel to the fire:



67.7% I love this site.

From some of what i have been reading over there, I would not want to hang out with these guys. I like my life and limbs in one piece

They seem too accident prone.

 

[Sun Baked]

Hold on there, those power specs have never been released. It's almost certain that no shipping PPC970 have ever breached 90W of dissipation, based on the "typical" ratings. A more reasonable estimate on the 2.5ghz 970fx is probably 70W peak.

Are you basing the reasonable on Motorola's typical being about 75% of Maximum or Apple's marketing releases which put the maximum at twice the typical?

The 2.5ghz 970fx is right about where the 1.8ghz 970 was about 100W -- and the 2.0ghz 970 was about 10-15W more on the typical side.

Doubling typical brings the numbers close to the middle of the range IBM/Apple are stating. IBM says a little less than double and Apple a little bit more than double.