Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
PowerMac DDR Clarification
- Thread starter arn
- Start date
- Sort by reaction score
ARRRR read my posts .............
no the G4 processor would not allow apple to use 2 different busses for each processor. I mean, they COULD, but... that would be horribley inefficient and would just be stupid. IT would be like a BeoWolf Cluster in a case.
no the G4 processor would not allow apple to use 2 different busses for each processor. I mean, they COULD, but... that would be horribley inefficient and would just be stupid. IT would be like a BeoWolf Cluster in a case.
It was your many posts saying that the powermac was fully DDR while I kept saying the opposite. I hate to say I told you so...
I told you so!
Oh well, I'm not all that into the desktop powermacs anyway. I don't really need speed as much as portability. I'm gonna get another Powerbook! (my duo 2300 is only good for word processing and playing gameboy games ) (besides, my pc already has 512mb PC2700DDR on a true 133DDR bus (overclocked to 166Mhz /333DDR effective)
) (besides, my pc already has 512mb PC2700DDR on a true 133DDR bus (overclocked to 166Mhz /333DDR effective)I'm not trying to start something but...
No one seems all that happy the the improvement made by Apple, just how fast is fast enough? I assumed (ya i know..ass..) that most Apple users were better (smarter) than PC weenies. I'm currious as to what can't you realistically accomplish in a reasonable time with any of the new macs. ( except Photoshop )
I just wish I could afford one.....
Regards,
No one seems all that happy the the improvement made by Apple, just how fast is fast enough? I assumed (ya i know..ass..) that most Apple users were better (smarter) than PC weenies. I'm currious as to what can't you realistically accomplish in a reasonable time with any of the new macs. ( except Photoshop )
I just wish I could afford one.....
Regards,
The difference between Intel, AMD, and Apple is this:
Intel: 100 mhz QDR (transfers data 4 times per clock cycle, making it an effective 400 mhz clock)
AMD: 133 mhz DDR (transfers data 2 times per clock cycle, making it an effective 266 mhz clock)
Apple: 166 mhz SDR (only transfers data once per clock cycle)
Without an architecture change, you can't make a processor that uses a SDR bus go to a DDR bus. You could make it a 333 mhz SDR bus, but it's highly doubtful that the processor would be able to handle that.
Now, the new P4's do use a 133 QDR bus, however, their own chipset still only has a memory bus of 100 mhz QDR, which does still help, but not as much as it could. Remember, even though you have 1.3 GB/s to give to your I/O devices, every I/O device still uses the CPU quite a bit, and generates a lot of traffic on that FSB. Additionally SMP systems generate a lot of traffic on the FSB communicating to each other, and so you end up with only 900 megs of bandwidth left over.
There is no such thing as an I/O device that is independent of a CPU. It will use the CPU in some way, at the bare minimum raising an interrupt everytime data is transferred. You can offload some of the processing, but it's all but impossible to offload everything.
Re: Re: 167 to each or both?
Apple seems to feel that since there are separate busses for memory, PCI, AGP, ATA, and main bus that the main bus can be 133-166 and the computer is still double the practical speed as a wintel P4 PC.
This may be true indeed.
However why not upgrade the syatem bus substantially anyway? I cannot believe it is not practical since 266-400 mhz system busses have neen on wintel boxes for some time now, and even if you assume Apple's adoption should be one year behind the curve it should at least have 266-300 mhz.
I for one hope this lack of emphasis on system bus which IS a major point of customer concern and complaint, is because they have some form of hypertransport on the horizon.
The current system can have 4 ATA drives (2x100, 2x66) plus two optical drives but optimum performance is achieved if you only have one drive on each bus, one on a 66 and one on a 100 ATA bus.
So there are limitations.
On the positive side Apple has as expected delivered a SUBSTANTIAL improvement in overall speed for NO INCREASE in pricing.
Say what you will about what Apple "shoulda done", but it is leading the industry in technology, speed, stability, new application development and profitability. You gotta respect that.
Are you a cheapskate? Buy an iBook G3 or a CRT iMac G3. $1100 and $749 respectively.
Rocketman
Apple seems to feel that since there are separate busses for memory, PCI, AGP, ATA, and main bus that the main bus can be 133-166 and the computer is still double the practical speed as a wintel P4 PC.
This may be true indeed.
However why not upgrade the syatem bus substantially anyway? I cannot believe it is not practical since 266-400 mhz system busses have neen on wintel boxes for some time now, and even if you assume Apple's adoption should be one year behind the curve it should at least have 266-300 mhz.
I for one hope this lack of emphasis on system bus which IS a major point of customer concern and complaint, is because they have some form of hypertransport on the horizon.
The current system can have 4 ATA drives (2x100, 2x66) plus two optical drives but optimum performance is achieved if you only have one drive on each bus, one on a 66 and one on a 100 ATA bus.
So there are limitations.
On the positive side Apple has as expected delivered a SUBSTANTIAL improvement in overall speed for NO INCREASE in pricing.
Say what you will about what Apple "shoulda done", but it is leading the industry in technology, speed, stability, new application development and profitability. You gotta respect that.
Are you a cheapskate? Buy an iBook G3 or a CRT iMac G3. $1100 and $749 respectively.
Rocketman
Could you PLEASE include the width of each companies data path with your example, otherwise you're just proving which of the three operates at the highest effective frequency.
It would be nice to know which expample you're providing actually MOVES the most data.
---
If you convert your table to this
Truck A goes 65 mph
Truck B goes 50 mph
Truck C goes 35 mph
You're concluding Truck A is the best performer based on speed alone.
But if you a little more data.
Truck A (F-150) goes 65 mph and carries .75 tons
Truck B (UPS truck) goes 50 mph and carries 3 tons
Truck C (a loaded semi) goes 35 mph and carries 24 tons
The actual results may not be tied to speed alone.
----
Fun thing is once you get over realizing it's not MHz alone that wins the race, they'll add another layer of fuzz to the equation.
Well hm, while we're on the subject of "I told you so"'s, I'd just like to deliver the following " " to AmbitiousLemon (no hard feelings, of course! ) We all smeg up once in awhile.
" to AmbitiousLemon (no hard feelings, of course! ) We all smeg up once in awhile.
The one area that will shine with the new PowerMacs is games, despite not having a full DDR implementation. With the AGP bus now independent of the CPU, framerates will skyrocket.
Question
We know that the new Power Mac has a crippled DDR implementation. Could the 7455 CPU module in one of these systems be replaced in the future with a different CPU that DOES support DDR, thus un-crippling the memory architecture?
In other words, was DDR hacked onto the motherboard in order to achieve a real performance increase, or was it hacked onto the motherboard in order to facilitate future CPUs and their respective modules that would be able to truly exploit it, providing an upgrade path to a certain future DDR-compatible PowerPC chip?
Alex
We know that the new Power Mac has a crippled DDR implementation. Could the 7455 CPU module in one of these systems be replaced in the future with a different CPU that DOES support DDR, thus un-crippling the memory architecture?
In other words, was DDR hacked onto the motherboard in order to achieve a real performance increase, or was it hacked onto the motherboard in order to facilitate future CPUs and their respective modules that would be able to truly exploit it, providing an upgrade path to a certain future DDR-compatible PowerPC chip?
Alex
Sadly, I'm certain that this means that we won't be seeing anything resembling a "G5" until MWNY 2003. The next rev will be speed bumped processors on this new board. Apple wouldn't eat all that R&D. It makes no sense. That means we're stuck with this BUS bottleneck.
IMHO, I don't think this is what the "pro" users and video effects and compositing people were waiting for.
IMHO, I don't think this is what the "pro" users and video effects and compositing people were waiting for.
a little help from my friends...
Guys, I'm sorry to post this message because it's a little off-topic, but I kind of need your expertise:
My problem is: should I get the new DP 1Ghz or the old DP 1Ghz? do you think that there's a real performance difference?
Thanks for your comments!
Guys, I'm sorry to post this message because it's a little off-topic, but I kind of need your expertise:
My problem is: should I get the new DP 1Ghz or the old DP 1Ghz? do you think that there's a real performance difference?
Thanks for your comments!
Re: a little help from my friends...
HOWEVER (and the following is only speculation), the new case of the new Power Macs may be a hint at some sort of possible upgrade to this Power4-derived PPC we're all waiting for. It would make sense - why give all this cooling and ventilation to a G4 system that doesn't need it? Still though, I would buy the old G4, because there's always eBay.
Alex
If your only consideration is performance, then if you could find the old DP 1GHz for substantially less money than the new one, I would go with that. I would be very surprised to see more than a 5% performance difference between new and old. The new system has a hacked DDR implementation the same as the Xserve, with the same CPU (MPC7455) as both the Xserve and the old DP machine. Various performance benchmarks between the old DP Power Mac and the DP Xserve revealed virtually no performance difference, indicating that this crippled DDR implementation does little to nothing to enhance performance.
HOWEVER (and the following is only speculation), the new case of the new Power Macs may be a hint at some sort of possible upgrade to this Power4-derived PPC we're all waiting for. It would make sense - why give all this cooling and ventilation to a G4 system that doesn't need it? Still though, I would buy the old G4, because there's always eBay.
Alex
Re: a little help from my friends...
Obviously, independent performance tests will indicate to you whether the price differential is worth the performance differential.
There will be a significant performance differential for two main reasons:
First, the data throughput from the main memory to the CPU's will be 325 MB/s faster on the new dual 1 gig than on the old dual 1 gig.
Second, because the AGP bus/connection to the System Controller is independent of the separate bus/connector from the System Controller to the CPU's, your video can speak directly to the main memory and supposedly can speak fluent DDR. That is, you should realize twice the data throughput from data memory to video card than from DDR to the CPU's. That's the simple way to look at it. The video card and CPU's to have to coordinate their actions with one another so that takes away some of the video card b/w. Also, the so-called 'direct' communication from video card to main memory does in some way share b/w with the CPU in sofar as the DDR can crank out data at 2.7 GB/s. If its cranking out 1.3 GB/s to the CPU's then there's that much that it cannot send to the video card or 1.4 GB/s.
On the old dualie, you'd have significantly less data throughput to your video card as well as less to your CPU (both directly and indicerctly in that the bus is shared by both the CPU and video card I believe).
So, with Quartz Extreme, there should be a very significant difference in performance, especially in terms of snappy response.
Conclusion: we should expect at least 20% improved performance for main memory intensive applications for the new 1 gig dualies over that of the old 1 gig dualies.
I'm stretching my understanding of these things here so please be kind oh EE-enlightened folk.
Obviously, independent performance tests will indicate to you whether the price differential is worth the performance differential.
There will be a significant performance differential for two main reasons:
First, the data throughput from the main memory to the CPU's will be 325 MB/s faster on the new dual 1 gig than on the old dual 1 gig.
Second, because the AGP bus/connection to the System Controller is independent of the separate bus/connector from the System Controller to the CPU's, your video can speak directly to the main memory and supposedly can speak fluent DDR. That is, you should realize twice the data throughput from data memory to video card than from DDR to the CPU's. That's the simple way to look at it. The video card and CPU's to have to coordinate their actions with one another so that takes away some of the video card b/w. Also, the so-called 'direct' communication from video card to main memory does in some way share b/w with the CPU in sofar as the DDR can crank out data at 2.7 GB/s. If its cranking out 1.3 GB/s to the CPU's then there's that much that it cannot send to the video card or 1.4 GB/s.
On the old dualie, you'd have significantly less data throughput to your video card as well as less to your CPU (both directly and indicerctly in that the bus is shared by both the CPU and video card I believe).
So, with Quartz Extreme, there should be a very significant difference in performance, especially in terms of snappy response.
Conclusion: we should expect at least 20% improved performance for main memory intensive applications for the new 1 gig dualies over that of the old 1 gig dualies.
I'm stretching my understanding of these things here so please be kind oh EE-enlightened folk.
Re: Re: a little help from my friends...
Alex, it is my understanding that the CPU's in the Xserve speak to the System Controller at 133 MHz or 1.05GB/s of data throughput. The new 1 Gig dualies speaks at 167 MHz or 1.3 GB/s. So, that alone should yield close to a 25% performance increase for MEMORY INTENSIVE operations.
Also, I don't know (don't recall, too lazy to look right now) if those benchmarks were based on 10.1 or a 10.2 beta. It would make for some difference wouldn't it, given Quartz Extreme would off-load some data throughput between the System Controller and CPU?
I'm not CERTAIN about this but I am pretty confident.
Alex, it is my understanding that the CPU's in the Xserve speak to the System Controller at 133 MHz or 1.05GB/s of data throughput. The new 1 Gig dualies speaks at 167 MHz or 1.3 GB/s. So, that alone should yield close to a 25% performance increase for MEMORY INTENSIVE operations.
Also, I don't know (don't recall, too lazy to look right now) if those benchmarks were based on 10.1 or a 10.2 beta. It would make for some difference wouldn't it, given Quartz Extreme would off-load some data throughput between the System Controller and CPU?
I'm not CERTAIN about this but I am pretty confident.