The future of the G3

[mr evil brkfast]

It has been said that the G3 still has life left.

seeing that the only continuing apple model running a G3 is the ibook, how long do you think apple will leave it in the ibook before a G4 is dropped in.

Do you think apple would consider adding a velocity type engine to the processor- if possible?

 

[ddtlm]

Apple niether designs nor manufactures the G3. IBM does. Apple does not add things to it. I expect IBM will keep making it as-is for some time, and I expect Apple to keep using it in iBooks because it is tiny, cheap, efficient, and overall pretty powerful.

 

[DakotaGuy]

Well technically speaking I believe the current gen G3 is a 750fx "Sahara" from IBM and correct me if I am wrong, but it is actually manufactured better then the G4 is. I also heard that it has some features over the G4 as far as the main processor is concerned. Yes I know it does not have Alti-Vec, but honestly if you stripped out alti-vec which would be a superior processor the 7455(G4) or the 750fx (G3), I think without the Alti-vec or as apple calles it Velocity Engine the G3 would kick the G4's butt.

If you consider the heat output of the current G3 and also the power intake, I think the G3 750fx is actually a better choice for the current design iBooks IMHO. What do we need. Give us another 200 or 300 MHz speed bump and the 133MHz bus speed. Those two things would help this little machine perform close to PowerBooks for less money, heat, and power consumption.

 

[MrMacMan]

Originally posted by Abercrombieboy
Well technically speaking I believe the current gen G3 is a 750fx "Sahara" from IBM and correct me if I am wrong, but it is actually manufactured better then the G4 is. I also heard that it has some features over the G4 as far as the main processor is concerned. Yes I know it does not have Alti-Vec, but honestly if you stripped out alti-vec which would be a superior processor the 7455(G4) or the 750fx (G3), I think without the Alti-vec or as apple calles it Velocity Engine the G3 would kick the G4's butt.

If you consider the heat output of the current G3 and also the power intake, I think the G3 750fx is actually a better choice for the current design iBooks IMHO. What do we need. Give us another 200 or 300 MHz speed bump and the 133MHz bus speed. Those two things would help this little machine perform close to PowerBooks for less money, heat, and power consumption.

Techically speaking you are correct, the reason is that the G3 has less 'steps' for data to go through. The G4 runs faster and has the veolict engine which makes it the better processor.

 

[MacCoaster]

Will continue to scale to 2GHz and beyond.

For some reason I tend to believe the new PowerPC introduced is really just a vamped up G3. POWER4 is in fact just a ultra-super-G3.

 

[rice_web]

From what I've read, the Power4 has many differences from the PowerPC line, though not significant enough to create a new platform. Though, I wouldn't be surprised if it took much more optimization of code to take advantage of the Power4.

The G3 is a great processor, and if it were paired with Altivec would easily beat the G4 (considering the Sahara currently supports 200MHz system buses and DDR memory)

 

[ddtlm]

MacCoaster:

The Power4 is to the G3 what the Athlon is to the K6-2.

In other words, they are very very different but happen to both run PPC code.

 

[Sherman]

Exactly.

that's why they call it a friggin i686 on bootup, that's what the system sees a pentagram as.


386
486
586
686

Why DOS programs will still run on a P4. VERY different processors running the same instruction set.

 

[MacBandit]

Originally posted by rice_web
The G3 is a great processor, and if it were paired with Altivec would easily beat the G4 (considering the Sahara currently supports 200MHz system buses and DDR memory)

The original G4 was nothing more then a G3 with Altivec coprocessors on board. The current G3 has had many improvement including smaller dies and on chip cache. I do not believe that the current G3 would beat the current G4 if it had altivec by much if any. The reason being that you would have to make the chip larger overall to add the Altivec units and this does and would slow down the chip. Small is good. I have read some where that IBM has produced a hand full of G3 chips that test at or near 2Ghz. I believe it was a release from IBM and I read this back around July so who knows what's going on now. Now if they could only add an Altivec unit to a 2Ghz G3 chip then we might be talking but once again I understand that a big part of why the G4 chip has not increased in speed substantially is because of the Altivec unit. Apparently it's very difficult to print at smaller and smaller sizes.

 

[nuckinfutz]

Sorry but the G3 is a poor processor by

Today's standards. The modern processor has a SIMD unit and Capable Integer and FPU performance.

The G3 has no SIMD(Altivec) unit.

It's FPU is slower than a 604 based proc and a G4 at Double Precision work(pretty much MOST 3d apps amongst many other genres)

It does not clock appreciably higher than a G4. It has a 4stage pipleline versus the G4+ 7 stage.


The processor is not the hangup as far as DDR memory. That weakness is due to the MPX bus.

I like the G3 but it's ran it's course. If Sonnet can offer a G4 400mhz upgrade card for $199 then you know that it makes no sense to continue using the G3 as the G4 is just as cheap to manufacture.

 

[mr evil brkfast]

a 2 ghz G3 chip with 200Mhz bus and DDR- who would complain?

 

[Santiago]

A G4 is not a just a G3 with an AltiVec unit. The G3 doesn't fully implement the MESI cache-tagging protocol, which is necessary for multiprocessing systems. (The G4 actually implements a more advanced model known as either MERSI or MOESI.) (One company did come out with some kind of a multiprocessor G3 machine (non-Mac), but it was an ugly hack.) Also, the G4 has higher floating-point performance because I think it has twice as many floating-point subunits.

 

[ddtlm]

Rampant inaccuracies...

MacBandit

Making a chip die larger does not decrease it's clock speed (if so, perhaps you could explain the P4 and Itanium II?).

I have read some where that IBM has produced a hand full of G3 chips that test at or near 2Ghz.

Yeah, and you believed it? Sheesh. No way in heck a 4-stage G3 can get to 2ghz on current tech, it is simply not possible. A Power4 might be able to make it... but that's a 13 or 14-stage pipline. AMD made it with 10-stages when they went to 130nm, and Intel made it on 180nm with the 20-stages of the P4.

Now if they could only add an Altivec unit to a 2Ghz G3 chip then we might be talking but once again I understand that a big part of why the G4 chip has not increased in speed substantially is because of the Altivec unit. Apparently it's very difficult to print at smaller and smaller sizes.

You apparently believe things without enough critical thought. Notice how AMD and Intel each have SIMD units clocking at well over 2ghz. In the case of SSE2, it can even do things AltiVec can't.

A far more reasonable explaination for the G4's clock problems is that it is fabbed on a rather low-tech 180nm process and only has a 7-stage pipeline (short compared to Power4, Athlon, P4).

nuckinfutz:

Your definition of a "modern" processor is wholly arbitrary and holds no water. Besides SIMD units, other things are also very important, such as:

1) Cost: 750fx G3 is cheap. Fabbed on a 130nm SOI processes, the die size is a tiny 34.6 sq mm. Great for low-end Macs.

2) Power dissipation: 750fx dissipates 3.6W typical @ 800mhz, 1.4v. Great for laptops.

3) Size: 750fx is pakaged in 21mm x 21mm... something like .75" x .75". That is tiny, great for latops.

You also mention a 4-stage pipe vs 7-stage in a way that suggests that the 7-stage is better, plainly not correct. The sole advantage of longer pipes is that they allow higher clockspeeds, but a G4 with 7-stages at 700mhz would certainly loose in general to a 4-stage G3 at 700mhz. You also seem to think that the G3 clocks higher than a G4... also not true when speaking of the 7-stage G4. The G3 does clock "appreciably" higher than the old 4-stage G4 which topped out about 533-500mhz, whereas people have easily gotten their 750fx chips to 800mhz, in a laptop.

The 750fx is a perfectly modern processor for laptops, supporting the latest and greatest fab tech, a large L2 cache, 256bit internal data paths, and minimalistic yet sufficient execution resources. Even so it can often beat G4's when AltiVec is not useful. It does it's job perfectly.

 

[P-Worm]

Originally posted by Abercrombieboy
Well technically speaking I believe the current gen G3 is a 750fx "Sahara" from IBM and correct me if I am wrong, but it is actually manufactured better then the G4 is. I also heard that it has some features over the G4 as far as the main processor is concerned. Yes I know it does not have Alti-Vec, but honestly if you stripped out alti-vec which would be a superior processor the 7455(G4) or the 750fx (G3), I think without the Alti-vec or as apple calles it Velocity Engine the G3 would kick the G4's butt.

If you consider the heat output of the current G3 and also the power intake, I think the G3 750fx is actually a better choice for the current design iBooks IMHO. What do we need. Give us another 200 or 300 MHz speed bump and the 133MHz bus speed. Those two things would help this little machine perform close to PowerBooks for less money, heat, and power consumption.

Correct me if I'm wrong, but the G4 has another one up on the G3, Dual Processor ability. From what I understand, you can't do dual G3s. Is this true?

P-Worm

 

[nuckinfutz]

Your definition of a "modern" processor is wholly arbitrary and holds no water. Besides SIMD units, other things are also very important, such as:

Whether it is "Aribitrary" or not. The fact is Intel/AMD both have SIMD units in their "consumer" procs. Apple is a premium Tier 1 Computer Manufacturer. They at the very least should contain what the low end compettion offers.

1) Cost: 750fx G3 is cheap. Fabbed on a 130nm SOI processes, the die size is a tiny 34.6 sq mm. Great for low-end Macs.

I contend that a G4 fabbed at 130nm would be cheap enough to obviate the need for a G3. Moto should have had 130nm G4 by now. Their HiP7 process has be functional for over a year.

2) Power dissipation: 750fx dissipates 3.6W typical @ 800mhz, 1.4v. Great for laptops.

The heat dissipation requirements of a 130nm G4 would be negligle versus a G3 on the same process.

3) Size: 750fx is pakaged in 21mm x 21mm... something like .75" x .75". That is tiny, great for latops.

I think most people would prefer Altivec ehancements.

The 750fx is a perfectly modern processor for laptops, supporting the latest and greatest fab tech, a large L2 cache, 256bit internal data paths, and minimalistic yet sufficient execution resources. Even so it can often beat G4's when AltiVec is not useful. It does it's job perfectly.

I agree to disagree here.

Any apps that support Double Precision math run slower on a G3

4 Pipes IMO are preferrable to 7 but it's a trade off. The G4 gained a %40+ speed advantage from adding the 3 stages ..

The lack of MERSI Cache Coherency is a downer although I wouldn't expect dual proc iBook.

The G3 has done well for Apple but it really needs to be put out to pasture.

Look for Apple to be Minimim G4 by next summer at the latest.

 

[Telomar]

Originally posted by nuckinfutz
I contend that a G4 fabbed at 130nm would be cheap enough to obviate the need for a G3. Moto should have had 130nm G4 by now. Their HiP7 process has be functional for over a year.

Even when the G4 and G3 were manufactured using the same fab technology (or at least as similar as you get at two different fabs) the G4 was the more expensive chip. It's a larger and more complex chip and has lower production yields.

Originally posted by nuckinfutz
The heat dissipation requirements of a 130nm G4 would be negligle versus a G3 on the same process.

No it wouldn't and I have no idea what basis you even have for assuming that. The G3 currently features a design far more condusive to low-power operation.

To put it in perspective when Intel went from 0.18µm to 0.13 the PIV lost 14W at an equal clock speed (2GHz if my memory serves). The G4 wouldn't lose that much and even if it did it would still have a much larger heat dissipation.

Originally posted by nuckinfutz
I think most people would prefer Altivec ehancements.

I think most people would like a SIMD unit, that preferably is as good as Altivec. As long as people program for it I doubt consumers care what it is. A SIMD unit is also already planned for the next evolution of the G3.

Originally posted by nuckinfutz
Any apps that support Double Precision math run slower on a G3

4 Pipes IMO are preferrable to 7 but it's a trade off. The G4 gained a %40+ speed advantage from adding the 3 stages ..

The lack of MERSI Cache Coherency is a downer although I wouldn't expect dual proc iBook.

The G3 has done well for Apple but it really needs to be put out to pasture.

Look for Apple to be Minimim G4 by next summer at the latest.

Not to say Apple won't switch their systems over from the G3 but the G3 has a very long and good roadmap ahead of it. Many of its current shortcomings are to be improved in its next iteration.

The design of the G3 for what it is intended for is excellent and it can easily be evolved into newer and better chips.

 

[Chryx]

Originally posted by Telomar
A SIMD unit is also already planned for the next evolution of the G3.

a G3 with a SIMD unit would pretty much be a G4... so we've had those "next evolution of" G3s since 1999

(yes, I'm aware of the bus and fpu differences also)

 

[MacBandit]

Re: Rampant inaccuracies...

Originally posted by ddtlm
MacBandit

Making a chip die larger does not decrease it's clock speed (if so, perhaps you could explain the P4 and Itanium II?).


Yeah, and you believed it? Sheesh. No way in heck a 4-stage G3 can get to 2ghz on current tech, it is simply not possible. A Power4 might be able to make it... but that's a 13 or 14-stage pipline. AMD made it with 10-stages when they went to 130nm, and Intel made it on 180nm with the 20-stages of the P4.


You apparently believe things without enough critical thought. Notice how AMD and Intel each have SIMD units clocking at well over 2ghz. In the case of SSE2, it can even do things AltiVec can't.

A far more reasonable explaination for the G4's clock problems is that it is fabbed on a rather low-tech 180nm process and only has a 7-stage pipeline (short compared to Power4, Athlon, P4).

You misunderstood what I said about larger sized chips. By shrinking a chip size you also shrink the length of the connections between transistors thus speeding up the chip. This has nothing to do with the clock rate. This is why chip manufacturors are constantly going to smaller and smaller etching.

Also I know that the G4 uses a slightly different architecture that allows simmetric-multiproccessing this doesn't change the fact that the G4 was a very small step from the G3 and in whole was based upon the G3.

Yes I do believe that IBM was able to achieve 2ghz with a G3. I am sure now that it was a release by IBM. Also I realize that it was probably only one chips out of tens of thousands and they probably used some special cooling on it but it did achieve 2ghz. Also the number of stages only affects the end speed in that it's easier to achieve higher clock rates with the chip because if there is an error the chip has more stages to correct it instead of sending off the bad data to the system causing a lock up. This is why overclocking a chip often ends up with system errors because your pushing data through the chip faster then it can handle.

 

[ddtlm]

nuckinfutz:

Whether it is "Aribitrary" or not. The fact is Intel/AMD both have SIMD units in their "consumer" procs. Apple is a premium Tier 1 Computer Manufacturer. They at the very least should contain what the low end compettion offers.

My point is that an arbitrary definition of what makes a processor modern means nothing. I can pull "important" specs out of the air too, and declare them to be what makes a processor "modern" or "useful." I presented a 3-part list that establishes easy-to-see advantages that the G3 holds over any G4 currently available. You have presented "I think" and thin-air psuedo-reasoning which really doesn't refute what I said.

Any apps that support Double Precision math run slower on a G3

Incorrect. Since double-precision math cannot be done by AltiVec, it comes down to FP unit vs FP unit, and while the G4 (I think) has a slightly better one, that hardly means that the G4 will always be faster. The G4 has the disadvantage of a higher branch-mispredict penalty as well as a smaller L2 cache. (Remember that L3 caches are not iBook equiptment.)

4 Pipes IMO are preferrable to 7 but it's a trade off. The G4 gained a %40+ speed advantage from adding the 3 stages ..

It gained less performance than it did clockspeed advantage. A 700mhz G4 with PC100 memory and no L3 would certainly fare poorly against a 750fx with those specs (except in certain AltiVec apps).

 

[ddtlm]

MacBandit:

You misunderstood what I said about larger sized chips. By shrinking a chip size you also shrink the length of the connections between transistors thus speeding up the chip. This has nothing to do with the clock rate. This is why chip manufacturors are constantly going to smaller and smaller etching.

You seem to have some serious misunderstandings about how that works. It is the fundamental nature of synchronous digital devices (such as all current processors) not to get faster or slower without a change in design or a change in clock speed.

Chip makers go to smaller manufacturing processes for several reasons, including the fact that die sizes are smaller which makes the chips cheaper to manufacture, that the chips can clock higher, and that the chips tend to produce less heat at a given clockspeed.

Die size itself does not dictate clock speed. It can influence it because errors are more common during the manufacture of larger dies, and it can influence it if the die is so large that signals cannot propigate as far as they need to during a single clock cycle. However simply adding something like an AltiVec unit to a G3 would not cause either of these to happen to any large degree since the die size is so darn small right now (20% or something the size of a P4).

Also I know that the G4 uses a slightly different architecture that allows simmetric-multiproccessing this doesn't change the fact that the G4 was a very small step from the G3 and in whole was based upon the G3.

Times change. The 750fx has little in common with the MPC7455.

Yes I do believe that IBM was able to achieve 2ghz with a G3.

You are wrong. IBM surely posseses chips that can make it to 2.0ghz, but the 750fx is not one of them. No two ways about it. It is not possible to take a 750fx chip to 2ghz, no matter how lucky you are. There is a limit which clock speeds of a design will approach and never exceed, no matter how many samples are tried. You can rest assured that the limit is less than 2ghz; it's probably in the low 1.x ghz range.

 

[MacBandit]

Originally posted by ddtlm
MacBandit:


You seem to have some serious misunderstandings about how that works. It is the fundamental nature of synchronous digital devices (such as all current processors) not to get faster or slower without a change in design or a change in clock speed.

Chip makers go to smaller manufacturing processes for several reasons, including the fact that die sizes are smaller which makes the chips cheaper to manufacture, that the chips can clock higher, and that the chips tend to produce less heat at a given clockspeed.

Die size itself does not dictate clock speed. It can influence it because errors are more common during the manufacture of larger dies, and it can influence it if the die is so large that signals cannot propigate as far as they need to during a single clock cycle. However simply adding something like an AltiVec unit to a G3 would not cause either of these to happen to any large degree since the die size is so darn small right now (20% or something the size of a P4).


Times change. The 750fx has little in common with the MPC7455.


You are wrong. IBM surely posseses chips that can make it to 2.0ghz, but the 750fx is not one of them. No two ways about it. It is not possible to take a 750fx chip to 2ghz, no matter how lucky you are. There is a limit which clock speeds of a design will approach and never exceed, no matter how many samples are tried. You can rest assured that the limit is less than 2ghz; it's probably in the low 1.x ghz range.

You may be right about the overall clock speed you seem to know a little bit about processor fabrication though you haven't indicated how. What I do know is that by shortening the distance a signal has to travel you have effectively shortened the time of travel which is the measurement of speed. Now when it comes to cpus and your taking about Billions of signals/second sortening the distance those signals have to travel also effectively increases the perceived speed of the processor.

 

[ddtlm]

MacBandit:

I'm a programmer but have taken many system architecture courses and I pay very close attention to quite a number of technology websites. Sometimes at those sites I dare not speak because the issues are over my head.

You are correct that smaller distances allow the signals to get where they are going faster, but the idea with a clocked processor is that everything happens at the clock, so even if the data is waiting to go, the execution units don't operate until their signal to do so. To take advantage of the shorter distances, the chip makers increase the clock speed, which is to say shorten the time between clocks, so that data is showing up just in time again.

The reason why some chips cannot clock as high as others is that errors in the manuafacturing can cause signals to propagate a bit more slowly, signals to be less clear, switches to take longer, and so on. These errors don't make operation impossible, it just takes longer for the parts to work correctly, and so clock speed is lowered until everything is in proper lockstep.

A processor is a bit like a well trained army doing a march. It would be a truely facinating visual if someone could make a screensaver or whatever that showed a simple device at work, signals propagating, states changing. Ooooo.

 

[MacBandit]

Originally posted by ddtlm
MacBandit:

I'm a programmer but have taken many system architecture courses and I pay very close attention to quite a number of technology websites. Sometimes at those sites I dare not speak because the issues are over my head.

You are correct that smaller distances allow the signals to get where they are going faster, but the idea with a clocked processor is that everything happens at the clock, so even if the data is waiting to go, the execution units don't operate until their signal to do so. To take advantage of the shorter distances, the chip makers increase the clock speed, which is to say shorten the time between clocks, so that data is showing up just in time again.

The reason why some chips cannot clock as high as others is that errors in the manuafacturing can cause signals to propagate a bit more slowly, signals to be less clear, switches to take longer, and so on. These errors don't make operation impossible, it just takes longer for the parts to work correctly, and so clock speed is lowered until everything is in proper lockstep.

A processor is a bit like a well trained army doing a march. It would be a truely facinating visual if someone could make a screensaver or whatever that showed a simple device at work, signals propagating, states changing. Ooooo.

Hey, thanks for going through this with me. I was trying to be stubborn I was just stating what I thought I knew. If you know better that may be. It does certainly sound like you have the training.

In the future to help you with your responses to shorten the number of times you have to go back in forth it helps to explain in detail what you mean as you did in this last response.

Again thanks.

 

[nuckinfutz]

Great Thread guys

Nice to have a little "pleasant" disagreements.


As for Double Precision

http://www.xlr8yourmac.com/G3CARDS/XLR8G4/G4vsG3.html

FPU
On the MPC750, single-precision operations involving multiplication have a 3-cycle latency, while their double-precision equivalents take an additional cycle. Because MPC7400 has a full double-precision FPU, double-precision multiplies have the same latency as single-precision multiplies: 3 cycles. Floating-point divides, on the other hand, still have the same latency for the two designs (17 cycles for single-precision, 31 cycles for double-precision).

MPC750 Double-precision floating-point multiply:
All other floating-point add and multiply: 4 cycles
3 cycles

MPC7400 All floating-point add and multiply: 3 cycles

That will produce a noticeable difference in speed.

It'll be interesting to see if Apple stays with the G3 or moves to the G4. Your guess is as good as mine if not better since i'm obviously biased towards the G4.

 

[rice_web]

I'm guessing that we'll see the G4 in the iBook, but still an IBM G4 with their new Altivec-like co-processor. There are so many rumors abound, though, that there isn't a single person on this board that could even make a reasonable guess.