Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
RedTomato said:
Seems I was right on the pipeline thing, at least for the POWER5.
No, you are wrong. POWER5 has the same pipeline as POWER4. POWER5 does more work per cycle compared to POWER4 since it has:
- Lower memory latency due to ondie memory controller
- Higher cache and memory bandwidth
- Much faster in multithreaded code due to the addition of SMT
 
thunng8 said:
No, you are wrong. POWER5 has the same pipeline as POWER4. POWER5 does more work per cycle compared to POWER4 since it has:
- Lower memory latency due to ondie memory controller
- Higher cache and memory bandwidth
- Much faster in multithreaded code due to the addition of SMT

You are right and I made a mistake.

I was actually referring to the POWER6 as compared with the POWER5 as Wiki says the POWER6 has a short 2-stage pipeline.

I'm not 100% sure if that's indeed shorter than the POWER5.
 
Ok hypothetically speaking, could something like that ever end up in a home computer.
 
dmw007 said:
Probably not, Apple is unlikely to go back to PowerPC at this point in time.
Im talking more in general, how feasible is it that these chips could wind up in a personal computer of any kind.
 
thunng8 said:
No, you are wrong. POWER5 has the same pipeline as POWER4. POWER5 does more work per cycle compared to POWER4 since it has:
- Lower memory latency due to ondie memory controller
- Higher cache and memory bandwidth
- Much faster in multithreaded code due to the addition of SMT
Don't forget the advance power & thermal management of the Power5.

Which besides the process improvements, what the G-ified chip needed besides all the other stuff that came along with the next "Power" chip to make it into the notebooks.

MacNut said:
Ok hypothetically speaking, could something like that ever end up in a home computer.
Only if you want to pay IBM 150-250 million for your next Amiga, or Mac clone. This is basically what IBM told Apple, if you want the CPUs -- pay for them.

IBM's stance of the future of the desktop CPUs was clear, that it was a bomb in the marketplace, failed to replace any G4s, so IBM canceled development of future Power-based desktop CPUs.
 
Sun Baked said:
IBMs old big iron, and their workstations.

To be more precise:

- IBM Sytem P servers (entire UNIX range running AIX and which can optionally run Linux).
http://www-03.ibm.com/systems/p/
Currently from 1U 1-4 core servers to large 64 core servers.

- IBM System i servers
http://www-03.ibm.com/systems/i/
Again a huge range from semi small to large servers

- Intellistation POWER workstations (running AIX)
http://www-03.ibm.com/servers/intellistation/power/
Currently a 970MP based workstation and a POWER5+ model

- POWER based blades
http://www-03.ibm.com/systems/bladecenter/power-based.html
Currently has 970MPs running at up to 2.7Ghz, but later will likely change to a Power6Lite processor

- IBM Storage servers
http://www-03.ibm.com/servers/storage/disk/
I think the latest models have POWER5+ processors
 
srf4real said:
yeah, still hoping Apple turns back before it's too late and we're all working on pc s*^#!:rolleyes:


As much as I enjoyed PowerPC based Macs, I think that the switch to Intel has many benefits. But the bottom line is the OS, as long as I have Mac OS X to use, I don't care if it is running on a G5 or a Core 2 Duo. :)
 
Sun Baked said:
IBM's stance of the future of the desktop CPUs was clear, that it was a bomb in the marketplace, failed to replace any G4s, so IBM canceled development of future Power-based desktop CPUs.

True, Apple had some help from IBM in deciding to make the switch to Intel.
 
IBM has determined that there is a niche they can fill that won't be eaten alive by the attack of the low end processors (x86, in this case). That niche is one in which they use exotic chip packaging (and cooling systems) and extremely high bandwidth system interconnects to stay ahead of the mainstream performance. POWER5 has 2313 signalling pins. Desktop CPUs have only recently approached 1000 total pins (signalling + power).

If you detach the POWER5 from it's remarkable cache+memory hierarchy, it suddenly looks a lot less interesting. If you don't detach it, you end up having to pay the rather huge costs of routing and connecting all those high speed busses. POWER6 will continue this with no fewer than 9 distinct bi-directional interconnects totaling 300GB/sec of bandwidth when running at 5GHz. While this is likely to once again bring them to the front of the markets they're targeting, there's no way a desktop motherboard vendor could hope to support it.

As for CELL... wait for the PS3 to be released, and watch the benchmarks of linux apps on it. I predict you'll see a few of them go blazingly fast, and the rest be around the level of the G4. Current programming paradigms are simply unable to handle CELL's low per-thread performance, asymmetric design, and exotic memory management requirements.
 
That would have been a good chip. How good would that chip have been though?? The transition to intel at the same clock speed on the imac made it 2.5 times faster so this wouldnt even be much better than the intel core duo.
 
®îçhå®? said:
That would have been a good chip. How good would that chip have been though?? The transition to intel at the same clock speed on the imac made it 2.5 times faster so this wouldnt even be much better than the intel core duo.

The quoted 2.5x faster is due to the Core Duo having 2 cores and running multithreaded apps.

The 2X faster is for POWER6 vs POWER5, not the G5. The POWER5 is much faster than the G5.
 
dmw007 said:
Probably not, Apple is unlikely to go back to PowerPC at this point in time.

You're right, but since new apps are "universal binary" it means Apple really could switch back at any time, or use a mixture of both in their lineup if they want. ;)
 
Me1000 said:
everyone is talking more speed, but did you ever wonder why intel moved just about everything over to dual core?

2 1GHZ processors, are faster the 1 2GHZ processor, because they can multitask better!

Not really true; 2 1GHz processors are slower, because of SMP overheads. You'll never get 100% performance, usually you'll get the equivilant of a single 1.5 GHz processor on multithreaded apps.

However, from an engineering standpoint it is much easier to get a dual 3 GHz processor than a single 6 GHz one.
 
Catfish_Man said:
IBM has determined that there is a niche they can fill that won't be eaten alive by the attack of the low end processors (x86, in this case). That niche is one in which they use exotic chip packaging (and cooling systems) and extremely high bandwidth system interconnects to stay ahead of the mainstream performance. POWER5 has 2313 signalling pins. Desktop CPUs have only recently approached 1000 total pins (signalling + power).

If you detach the POWER5 from it's remarkable cache+memory hierarchy, it suddenly looks a lot less interesting. If you don't detach it, you end up having to pay the rather huge costs of routing and connecting all those high speed busses. POWER6 will continue this with no fewer than 9 distinct bi-directional interconnects totaling 300GB/sec of bandwidth when running at 5GHz. While this is likely to once again bring them to the front of the markets they're targeting, there's no way a desktop motherboard vendor could hope to support it.

As for CELL... wait for the PS3 to be released, and watch the benchmarks of linux apps on it. I predict you'll see a few of them go blazingly fast, and the rest be around the level of the G4. Current programming paradigms are simply unable to handle CELL's low per-thread performance, asymmetric design, and exotic memory management requirements.

Good post. Someone who knows of what they speak!
 
Catfish_Man said:
[A POST wherein Catfish_Man displays intelligence and informed thinking.]
Just wanted to point out that you win the Internets, sir. I'm tired of people going around saying Cell is some amazing performance panacea and worlds faster than anything else at everything. No, it's quite good at some things, but the requirements to GET it there are heady indeed.
 
MacNut said:
Ok hypothetically speaking, could something like that ever end up in a home computer.
Potentially in 10 years, we could all be running 5GHz laptops, there just need to make transistors small enough
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.