[bbt]

the power mac front side bus speed is not the same as the system memory

does this mean that effectively the system only runs at the speed of the front side bus?

does the DDR memory help in performance?

[clarkgoble]

For all intensive purposes you won't be getting much out of the DDR on most Macs. It is more marketing.

[Jansar]

Originally posted by clarkgoble:
For all intensive purposes you won't be getting much out of the DDR on most Macs. It is more marketing.

Does that mean the whole DDR thing is a lie (or partially)? Maybe it's true with real intensive apps like Final Cut Pro.

[michaelb]

Originally posted by clarkgoble:
For all intensive purposes

I've seen this so much lately I'm starting to doubt my sanity, but I think you'll find the expression is:

"For all intents and purposes"

[keston]

So much for trying to sound intelligent

[Amorph]

Your various I/O devices, ports and the PCI and AGP busses can all access the RAM without going through the frontside bus, so what the DDR RAM allows the memory controller to be able to do is to saturate not only the processor bus (not the processor itself!), but the hard drive, optical drive, AGP bus, FireWire, Ethernet, all at once. So you can have maximum feasible CPU usage without choking any other part of the board. This is especially useful for e.g. Quartz Extreme, which will make heavy use of the AGP bus to access RAM.

This is hard to benchmark, but if you test for it it's there, and it's a real advantage. It will only show up under fairly heavy load, but then that's really when you'd want it to show up.

[bbt]

so what you guys are saying is for me to wait until macs have a faster front side bus

[OptimusG4]

Originally posted by bbt:
so what you guys are saying is for me to wait until macs have a faster front side bus

Basically, yes.

[Catfish_Man]

Originally posted by bbt:
so what you guys are saying is for me to wait until macs have a faster front side bus

Even better. Wait for the PPC970. 800MHz effective FSB speed. The highest the G4's going to go before the 7457-RM is AT MOST 200MHz.

[bradoesch]

Originally posted by Catfish_Man:
Even better. Wait for the PPC970. 800MHz effective FSB speed. The highest the G4's going to go before the 7457-RM is AT MOST 200MHz.

I thought it was supposed to be a 900 MHz FSB?

[bbt]

i know that companies need cash flow for them to operate but all of this waiting can't be good for apple

maybe if people will buy more often then the macs will be cheaper?

[Amorph]

The 970's bus - called GigaBus by IBM - is clocked at 1/4 the speed of the CPU, and double-pumped. So a 1.6GHz 970 has a 400MHz bus that operates at an "effective" 800MHz. The 900MHz number comes from the top speed that IBM has claimed for the 970's initial release: 1.8GHz. That CPU's bus will run at 1.8/4 = 450MHz, double-pumped to an effective 900MHz.

[Catfish_Man]

Originally posted by Amorph:
The 970's bus - called GigaBus by IBM - is clocked at 1/4 the speed of the CPU, and double-pumped. So a 1.6GHz 970 has a 400MHz bus that operates at an "effective" 800MHz. The 900MHz number comes from the top speed that IBM has claimed for the 970's initial release: 1.8GHz. That CPU's bus will run at 1.8/4 = 450MHz, double-pumped to an effective 900MHz.

Almost. The 900MHz bus has an effective speed of 800MHz due to a bit of overhead in the bus protocol (it's packet based).

[Amorph]

Originally posted by Catfish_Man:
Almost. The 900MHz bus has an effective speed of 800MHz due to a bit of overhead in the bus protocol (it's packet based).

Umm, I've never heard it stated that way. For one thing, all busses have less actual than theoretical bandwidth to varying degrees and for various reasons, but I've never heard anyone say, for example, that a 167MHz MaxBus is really a 133MHz bus because of that.

The 970's clock speed is what it is in absolute terms, and its bus speed is tied to an invariant multiplier of the CPU clock. Clock speed is not directly related to bandwidth.

More precisely, at 900MHz (effective), the GigaBus is theoretically capable of 7.2GB/s (3.6GB/s either way), and IBM claims a real-world bandwidth of 6.4GB/s (3.2GB/s either way). That would scale down with the bus' clock speed, which scales down with the CPU's clock speed. So on an 800MHz GigaBus (attached to a 1.6GHz 970) you'd have a theoretical cap of 6.4GB/s and a (claimed) real-world bandwidth of about 5.7GB/s (2.8ish GB/s either way).

All this does add up to saying that a 900MHz GigaBus' actual bandwidth is equal to an 800MHz GigaBus' theoretical bandwidth, but I don't really see how that's helpful in determining what the actual clock speeds of the busses are, or how they relate to the clock speed of the CPU. Those were the points I was clarifying.

[Catfish_Man]

Originally posted by Amorph:
Umm, I've never heard it stated that way. For one thing, all busses have less actual than theoretical bandwidth to varying degrees and for various reasons, but I've never heard anyone say, for example, that a 167MHz MaxBus is really a 133MHz bus because of that.

The 970's clock speed is what it is in absolute terms, and its bus speed is tied to an invariant multiplier of the CPU clock. Clock speed is not directly related to bandwidth.

More precisely, at 900MHz (effective), the GigaBus is theoretically capable of 7.2GB/s (3.6GB/s either way), and IBM claims a real-world bandwidth of 6.4GB/s (3.2GB/s either way). That would scale down with the bus' clock speed, which scales down with the CPU's clock speed. So on an 800MHz GigaBus (attached to a 1.6GHz 970) you'd have a theoretical cap of 6.4GB/s and a (claimed) real-world bandwidth of about 5.7GB/s (2.8ish GB/s either way).

All this does add up to saying that a 900MHz GigaBus' actual bandwidth is equal to an 800MHz GigaBus' theoretical bandwidth, but I don't really see how that's helpful in determining what the actual clock speeds of the busses are, or how they relate to the clock speed of the CPU. Those were the points I was clarifying.

You're correct of course, I was just noting that, from what I've heard, "GigaBus" seems to have slightly more overhead than most (or, perhaps is just being advertised more honestly). Also, I was unclear on whether they can actually get that 6.4GB/sec, or whether it's more like Apple's claim of 1.3GB/sec for the MaxBus (iirc, from Apples optimization page, a 133MHz MaxBus has an actual max of 800MB/sec). It seems that I hadn't read it closely enough, and had missed the "real-world".