Buying A G4: Ddr Sdram Vs. Quicksilver?, Performance issues in design of new G4sb Options

Here's the situation: Me: bright young computer musician just beginning a career in digital signal processing. Apple: company, which, up until now, made relatively nice computers for creative applications. The Quandry: Apple releases a new round of G4s, equipped with newer ddr sdram. However I have seen much information indicating these new computers are NOT as capable of handling the type of real-time processing I would like to do as the previous quicksilver G4s which used non-ddr sdram. The crux lies in the design: the new machines have a 1mb/proc L3 cache, while the older machines have a 2mb/proc cache. Also Moto's system bus doesnt have the bandwidth to make use of the DDR sdram properly, and may cause more bottleneck problems than it was introduced to solve.
(benchmark info and ddr sdram kludge?).

Does anyone here think the ddr sdram G4s will pose any restraints to heavy real-time number crunching for sound-synthesis and dsp stuff? This is only an issue because Apple has completely pulled the line of older quicksilver G4s and my chances of getting one are rapidly scaling to zero. I emailed a guy at Apple, who does music, about this, but am still waiting for his reply. Thanks for any suggestions the mac music community can provide me.


M.

Here's a response I wrote on gearslutz.com about the SDR vs DDR situation.

Bottom line, really, is that DDR in the new Macs is all about marketing, none about performance. Why? Because on the PC side, systems using DDR also have the equivalent of a DDR based Front Side Bus. It's xfering data at double (or for the P4 and Rambus x4) of the actual FSB clock speed.

I wouldn't call it all "marketing" DDR does have some benefits and I'll explain why.

Yes the Processor to Memory bus is limited to 1.3GBps. That is going to affect the speed in some areas. However Apple was right to add DDR. Here's why



Note the Dual G4's at the top of the pic. The connection you see there from the Procs to the System Controller is 1.3GBps Maximum. However AFTER the System Controller it's 2.6GBS(DDR). You can clearly see that AGP, PCI,RAM, ATA-100 Controller, and the ATA-66 Controller all connect to this System Controller at 2.6GBps.

The next logical question would be "Doesn't ALL data have to flow through the FSB?" and Apple neatly answers that on the Powermac page here



Memory Enhancements

The resulting throughput between main memory and the system controller is up to 2.7GBps, more than double the throughput from the previous dual 1GHz Power Mac G4. DDR SDRAM also increases systemwide memory bandwidth to the processors and all other elements of the system. At the same time, direct memory access allows system elements, such as a hard drive controller or a graphics processing unit, to send and receive data directly from main memory, without going through the processors. The added bandwidth allows system elements to function independently at high data rates, boosting total system performance.

So how will this manifest itself in benchmarks? Most likely it won't. Barefeats tested throughput and speed of ONE application but a design like you see above is Apple getting things like the Gigabit Ethernet, Firewire and ATA-100(which supports 48bit addressing so you CAN use Drives over 137GB BTW) off the slower memory bus of previous Macs and utilizing the 2.6GBps of DDR.

The only real way you could test for this is to run multiple apps/benchmarks simultaneously and look for which computer handles itself best under full load. I think you would find the DDR Macs to be superior in this regard. And quite honestly when we're asking our Macs to run multiple Audio apps simultaneously we're doing just that.

Concerning DDR and FSB. There is a misconception that having a full DDR bus will "unlock" the speed of the G4. While it will undoubtedly help we should not expect huge increases in speed with the addition of "just" a DDR FSB simply because with only 7 Pipline stages the current G4 just doesn't suffer enough of a "hit" by having a branch misprediction. Intel P4's have 20 Pipelines so if there is a misprediction it has an additional 13cycles to obtain the correct data. Hence the need to have FAST memory to retrieve the data as quickly as possible while the processor stalls.

On a related topic, it does appear that IBM and their new "GigaProcessor Ultralite (GPUL)" is the frontrunner for the next generation Macs. It's a 64 bit CPU. No idea of a timetable yet. It appears the G5 may not happen, or in any event, happen soon enough.

Heheheh Brian you're on top of things as usual! jkthtyrt

Here's a link for all interested. How's this for drool worthiness. A GPUL 64bit proc should be twice as fast clock for clock over todays G4. Here's the kicker. It's likely to be a dual core. Yes that means Dually Performance in one chip and that means that we're talking a 4X improvement for Dual Core 2Ghz!!!

http://www.eweek.com/article2/0,3959,543317,00.asp