1.6 GHz, anyone? :D

[MrMacMan]

Originally posted by cb911
cool. would people be likely to OC these though? could be good, this might start a Mac OC'ing cult.

I would LOVE that!

What I would love more is people acually MAKING there own Mac, like all expect PC users, make their own comp, but us we are stuck so far, with the parts and no really walk throughs or guides, i want to see this, maybe I can give this to the world, I dunno.

My project forever.

 

[WannabeSQ]

IMO the only reason this upgrade market for g4s is flourishing, is that they are compatible with many Powermacs, from the first AGPs and on. Large user base, because the basic design has not changed significantly (same CPU slot, form factor etc)

When they switch to a whole new Mobo chipset (which they will eventually) it will hurt the upgrade market. I don't think there will be 970 upgrades for current G4 powermacs, unless they put practically a whole new mobo on it, with its own memory bus, or a lot of really fast cache.

 

[MacBandit]

Re: Re: Re: Re: Re: 1.6 GHz, anyone?

Originally posted by ZeeOwl
OK. I see your point. Of course, real world max speed will never attain theoretical max speed. But my point was that the slow FSB is the G4's Achilles Heel. And here's some real world numbers (ran this test myself) to illustrate what I mean.

CineBench 2003 R8.1 results:
G3 350MHz - 901.0 seconds
G4 700MHz - 632.1 seconds

I had disabled the L2 cache of both processors for this test. Both machines have 1,024 MiB of CL2 RAM, on a 100MHz FSB. Now, "in theory", the G4 should have produced the render in 450.5 seconds. There are only two explanations that I can think of to explain these results...
a) The G3 is a much more efficient design than the G4. That I doubt, because from the documention that I've seen, while each processor has it's strengths and weaknesses, they are about evenly matched (not counting AltiVec). Which reminds me that C4D R6 and up make mild use of AltiVec. So the G4, still "in theory", should have produced the render in less than 450.5 seconds. Not much less, but maybe 2 or 3 seconds.
b) The FSB can't supply the G4 running at 7x the FSB frequency with data quickly enough to saturate it's processing power. That explanation I find much more likely.

So on your dual 1GHz G4, with a 133MHz FSB, you're in an even worst position, running at 15x the FSB frequency. In the MacBidouille test, they were running at 16x, and with only 1 processor. So it's not surprising that their fractal score was lower than your's. I'll be able to compare apples with apples in about two weeks, when I get my replacement G3 800MHz upgrade. I'll compare it to the original G3 scores at 7x, 8x, and 9x if I can get the upgrade to run reliably at that speed.

Now that's why I'm very much looking forward to the new 970 PowerMacs. Because, assuming Apple implements true DDR memory into these machines, I think they'll run circles around the current line of PowerMacs. Even if they only have 200MHz DDR RAM, that's still only a factor of 4.5x for a 1.8GHz single processor, 9x for a dual. About the same as my more "efficient" G3. A much more reasonable ratio than the current line of PowerMacs, which run at 7.5x to 17x. And if the rumours coming from MacBidouille are correct, Apple has equipped these new beasts with Twin Bank memory, which means that the 970s will actually be operating at 2.25x and 4.5x respectively. Which according to IBM's literature, is close to their designed saturation point of 2x. I can't wait to see how their real-world performance measures up to theory.

Okay just some simple stuff. A doubling in MHz will never ever render a doubling in processing performance with the same chip or fairly equal chips. It just doesn't happen. Once again it's a real world thing. As the clock speed goes up so does the overhead. Second of all you hit on but didn't guess it. The G3 has a much shorter pipeline then the G4 you are testing I would say at least half to 3 times as short. This is also why doubling did not create a doubling of performance in this situation or a lot of situations. As manufacturers try to increase clock rates they run into problems with different tasks being done faster then others in the pipeline of the processor so they add empty segments to just give the next action time to catch up. So lets try something. Lets say you have a 500MHz processor with 5 segments to it's pipeline and a 1GHz processor with 10 segments to it's pipeline. Well just mathematically the 500MHz processor will perform equally with the 1GHz but once again in real world that isn't the case. I would say the 1GHz would probably perform about 50% better just as a hunch.


On another note my dual/GHz Powermac has a 167MHz bus. Also there are other places that the DDR helps without the double data rate FSB. Just moving data back and forth from RAM to Hard drive to PCI bus to AGP bus etc.. The overall increase in overall system speed between the non DDR macs and the DDR macs is approximately 30% so it definitely made a difference.

 

[cb911]

Originally posted by MrMacman
I would LOVE that!

What I would love more is people acually MAKING there own Mac, like all expect PC users, make their own comp, but us we are stuck so far, with the parts and no really walk throughs or guides, i want to see this, maybe I can give this to the world, I dunno.

My project forever.

that would be good to see. like in PC mags now they have about 4 pages full of PC cases (and sometimes just internals- caseless) that have been modded and OC'ed.

i also just remembered reading something about Intel having some sort of hardware lock on OC'ing. i forget the details, but it was supposed to stop people from OC'ing the new Intel chips... i think mainly the P4 3GHz+.

and what will the OC'ing potential of the 970's be like? that would be so good if you could just put on a massive fan and add like 5 or 6 case fans and get a few hundred MHz increase. there may be hope for the Mac OC'ing scene.

 

[ZeeOwl]

Re: Re: Re: Re: Re: Re: 1.6 GHz, anyone?

Originally posted by MacBandit
A doubling in MHz will never ever render a doubling in processing performance with the same chip or fairly equal chips.

Yes, I know that. It would only happen if the speed of all other the elements involved also doubled. That's why I put "in theory" between quotes.

As the clock speed goes up so does the overhead.

What do you mean by "overhead"? The only overhead I see is the (relatively) slower FSB.

As manufacturers try to increase clock rates they run into problems with different tasks being done faster then others in the pipeline of the processor so they add empty segments to just give the next action time to catch up.

Don't get that part. If the clock speed is increased, shouldn't all tasks run proportionally faster? I'm talking only about tasks already in the processor's pipeline, of course.

On another note my dual/GHz Powermac has a 167MHz bus.

I was under the impression that only the latest round of DP MDD towers had a 167MHz bus. That means you're running at 12x the FSB (@ 2x1GHz), not 15x. And the MacBidouille overclock test was at 16x (@ 1.6GHz). You got 8 Gflops, they got 6 Gflops. That just confirms a hunch that I had; that past a certain FSB multiplier more GHz doesn't make much difference. It's the FSB bandwidth that has the most influence on performance. Ya, I know, I have a thick skull. OK, well let's settle this in proper scientific fashion. Test the hypothesis with empirical data. As soon as I get my replacement G3 back from Powerlogix, I'll try running it with my FSB set to 100MHz(8x) and 67MHz(12x). If my hypothesis is correct, performance with the 67MHz FSB should be around 75% of 100MHz. I'll keep you posted.

Also there are other places that the DDR helps without the double data rate FSB. Just moving data back and forth from RAM to Hard drive to PCI bus to AGP bus etc.. The overall increase in overall system speed between the non DDR macs and the DDR macs is approximately 30% so it definitely made a difference.

Impressive. I didn't think that so much non-processor stuff was going on in a Mac MB. Shows that Apple really did a good job of working around the G4's limitations.

 

[ZeeOwl]

BYOM (Build Your Own Mac)

Originally posted by MrMacman
I would LOVE that!

What I would love more is people acually MAKING there own Mac, like all expect PC users, make their own comp, but us we are stuck so far, with the parts and no really walk throughs or guides, i want to see this, maybe I can give this to the world, I dunno.

My project forever.

Actually, my 13-year-old son is presently building his own Mac. He started out with a mildly damaged B&W motherboard I gave him (one of the ATA channels is dead, and one of the RAM slots). It's actually quite easy. With a B&W anyways. He's got it nearly complete. All that's missing is the processor (he's waiting like a vulture for my new processor to get back so he can have my current one), and some plexiglass for a Kool Klear Kase.

It's going to be a fairly decent machine. Here are the specs:
350 MHz G3 (overclocked 300)
512 MiB RAM
30 GB 5,400 RPM HD
Rage 128 16MiB video card
14" 640x480 VGA monitor
1 x 32-bit 67MHz and 3 x 64-bit 33MHz PCI slots
10/100 Mbps Ethernet
2 x USB 1.1 ports
16-bit stereo audio in and out
Global Village 56K modem
Apple Compact USB Keyboard and Puck Mouse
Baby-AT 200W power supply (modified)
OS X 10.2.6 and OS 9.2.2

No CD-ROM for now, partly for budgetary reasons, and partly because, since only 1 ATA bus is working, he can't have both a HD and a CD-ROM. He'll need a PCI ATA card to be able to add that later. Software can be installed by hooking up his machine to my B&W via Ethernet. Total cost: about 260$CA (~185$US), less ATA card, CD-ROM and case. I gave him the motherboard and the RAM modules. He already had the monitor. Factor in about an extra 390$US for those items.

 

[MacBandit]

Re: Re: Re: Re: Re: Re: Re: 1.6 GHz, anyone?

Originally posted by ZeeOwl
Yes, I know that. It would only happen if the speed of all other the elements involved also doubled. That's why I put "in theory" between quotes.



What do you mean by "overhead"? The only overhead I see is the (relatively) slower FSB.



Don't get that part. If the clock speed is increased, shouldn't all tasks run proportionally faster? I'm talking only about tasks already in the processor's pipeline, of course.



I was under the impression that only the latest round of DP MDD towers had a 167MHz bus. That means you're running at 12x the FSB (@ 2x1GHz), not 15x. And the MacBidouille overclock test was at 16x (@ 1.6GHz). You got 8 Gflops, they got 6 Gflops. That just confirms a hunch that I had; that past a certain FSB multiplier more GHz doesn't make much difference. It's the FSB bandwidth that has the most influence on performance. Ya, I know, I have a thick skull. OK, well let's settle this in proper scientific fashion. Test the hypothesis with empirical data. As soon as I get my replacement G3 back from Powerlogix, I'll try running it with my FSB set to 100MHz(8x) and 67MHz(12x). If my hypothesis is correct, performance with the 67MHz FSB should be around 75% of 100MHz. I'll keep you posted.



Impressive. I didn't think that so much non-processor stuff was going on in a Mac MB. Shows that Apple really did a good job of working around the G4's limitations.

Okay let me try to explain the CPU pipe length thing. A cpu has to perform several actions a piece of data from the time it enters the cpu and then exits it. This is why you sometimes hear people talking about the length of the pipe or channel. The early G3s had 4 segments to this path in which each segment did a processing task. Each segment requires a cpu clock cycle. As you increase the speed of a processor some of those segments are capable of processing faster then others and this causes a bottleneck within the CPU and can lead to crashes. So what they do to increase the speed of the processor and maintain stability is to have empty segments that do nothing.

So if you have two cpus that runs at 100MHz and one has a 5 stage pipeline and another has a 10 stage pipeline the one with the 5 stage will be able to process the same amount of data in half the time.

This is the big part of what makes a faster say a 1GHz processor slower per clock cycle then a slower cpu. The faster cpu will almost always have a longer pipeline.

So even if you double the speed of the system bus you will not see a doubling of overall speed of the machine. A big part of this has to do with how fast the data can get to the system bus not all data is always cached in ram so in this situation the bus would be far from the slowest peice of puzzle and in fact the hard drive would be the slowdown. Another thing to consider is that the bus only becomes a bottleneck when the cpu is able to process the data that is being given it faster then it needs the data. Often data is put in the cpu chache and processed more then once this would leave the cpu without the need for new data from the bus at all. All in all the system bus as it is is not the biggest slow down and this was proven when the PCs went to double and quad pumped buses and saw no more then 10-15% gain. Macs rely on the bus even less then PCs as they don't stream all data through the CPU like on PC hardware as I have said processing devices such as video cards are capable of bypassing the cpu and gathering their data directly from the hard drive or ram.