Future of PowerMacG5

[dongmin]

Re: Merry Christmas, every two (point four?) of us...

Originally posted by Brother Mugga
This is somewhat speculative, but...

It seems likely to me that Apple will move to the following line up around or just after Xmas (maybe MWSF?):

single 1.8
dual 2.0
dual 2.0+ (possibly around 2.4, still fabbed at 130nm?).

...

Of course if I'm right, waiting till around Xmas (ish) to get a G5 means you not only miss out on any 'teething' troubles with the new machines (I'm intrigued as to how the 'dust' issue will work out) but also cash in on the 2.0 dropping to 'mid-range' prices etc.

As far as the timing goes, I don't think Apple has done a major update right around Xmas. (For reference, the 15" PowerBooks were last updated in early November.) I think early Jan./Feb. 2004 is more likely, especially if these first G5s don't ship until early September.

So yeah, expect the G5 to hit 2.4 ghz by Feb. and to 3 ghz+ by next July/Aug, a year from now. The big question, to me, is whether they'll be able to squeeze 3 ghz out of the current .13-process 970s or will have to go to .09 process.

 

[DharvaBinky]

970 memory control & stuff

Originally posted by Flynnstone
I would like to see IBM/Apple move the memory controller onto the processor like the Opteron. Throughput up, and latency down.

I disagree.

Sure, it would lower memory latencies, but it does, essentially, lock them into a single memory technology (DDR2-500 maybe) until IBM redesigns that controller to support something faster. If the 980 or 990 or whatever came with a memory controller, Apple would be "topped out" at that memory technology until the next iteration of chips. For apple, that seems like a bad deal since this is there general purpose CPU.

Notice that although the AMD Opteron uses this technique, the upcoming Athlon64 does not. It uses the traditional NorthBridge memory controller configuration found in PCs and Macs today.

What I think the 980 should contain, though, is a memory controller for L3 Cache. This would help alleviate the latencies (especially at these new high clockspeeds) associated with northbridge controlled memory. Wtih L3 cache and a nice flexible northbridge memory subsystem, Apple would be able to quickly introduce the next big thing:

!!!!MRAM!!!!

IBM is cooking up MRAM chips in their labs right now. I've read they have prototypes running and want to have a production line for 2005. Gee, doesn't this IBM/Apple partnership seem more and more fruitful every day?



With MRAM and its non-volitile state (doesn't empty the ram when you shut the machine off), Macs would be able to boot almost instantly back to the last state you left off in. Even from being unplugged. Fun for a desktop, sure, but imagine the possibilities for home entertainment. Consumers can't/won't wait for their TV to take 5 minutes to boot, but if it was nearly instant.... hmmmmmm....

Dharvabinky

 

[noverflow]

my tv takes 10sec to boot up.
well to warm up befor it turnes on the tube... then an other 5sec to get a picture.

Why are our tvs going to boot again?

 

[Waluigi]

Re: 970 memory control & stuff

IBM is cooking up MRAM chips in their labs right now.

Besides the fast boot up features of MRAM, what exactly is it? Is it really fast in performance, or just really fast to boot up?

--Waluigi

 

[noverflow]

Re: Re: 970 memory control & stuff

Originally posted by Waluigi
Besides the fast boot up features of MRAM, what exactly is it? Is it really fast in performance, or just really fast to boot up?

--Waluigi

its not fast to boot, it can just hold info in its banks with no power. this ability makes a "save state" ability possible for a theoretical zero boot time. so while it is not the ram that would be booting, it makes it possible

 

[Chryx]

Re: 970 memory control & stuff

Originally posted by DharvaBinky
Notice that although the AMD Opteron uses this technique, the upcoming Athlon64 does not. It uses the traditional NorthBridge memory controller configuration found in PCs and Macs today.

Bzzt, the Athlon64 has an ondie 64bit PC3200 memory controller (compared to the Opterons ondie 128bit PC2700 ECC/Registered memory controller)

 

[wizard]

Re: Future of PowerMacG5

Here is my take on macbidouille. It goes without question that we will have to see a rather large jump in machine performance around the end of the year. Since SJ has publicly stated 3 GHz with in a year, the rational thing would be to get half way there by the start of 2004, which would mean a speed bump to atleast 2.5 GHz.

While PCI-X may only be a begining I can't see the current bus slots going away in a year or twos time. I woudn't accept it and neither would the market place in my opinion. What we could very well see is PCI-Expreess replacing the AGP ports or otherwise acting as a supplemental I/O port. Apple and its customers need PCI-X slots for their unerversal acceptance, PCI-Express will take awhile to gain that sort of market approval.


It is obvious that much effort was put into Apples 970 implementation to grow systems with much higher clock rate. I would not be surprised at all to hear that 3.4 GHz processors make it to market in a years time. Frankly Apple and IBM have no choice whatsoever if they want to remian alternative CPU venodors. The supposed speed leadership that Apple has with these machine will only last a couple of months. AMD and Intel have been competeing for years on speed, a little startup like Apple and IBM will only feed that competition it will not slow it down by any means. This is especially the case when they see all of the free marketing help Apple gets with its claims.

Dave

Originally posted by Macrumors
MacBidouille reports on some rumors of upcoming technologies coming in the PowerMac G5 line.

According to the site, PCI-X is only the beginning, and to expect technologies such as PCI-X 2.0, a 1.7GHz Bus "in the beginning of 2004", 3GIO/PCI-Express in 2004/2005. Macbidouille also claims that plans for 2GHz and 2.5GHz speeds are for the end of the year.

Few details of these rumors are given -- for example, the actual PowerMac G5's bus runs at 1/2 of the 970's Processor Speed. A 1.7GHz bus would imply a 3.4GHz 970 Processor - based on current PPC970 designs. But this is NOT what's implied in the article.

 

[wizard]

Re: Merry Christmas, every two (point four?) of us...

Originally posted by Brother Mugga
This is somewhat speculative, but...

It seems likely to me that Apple will move to the following line up around or just after Xmas (maybe MWSF?):

single 1.8
dual 2.0
dual 2.0+ (possibly around 2.4, still fabbed at 130nm?).

I believe this is right on the money though I think 2.5 GHz or higher would be a better number for the high end machine. The reason being is that there are indications that this level has alread been reached and its just a matter of getting acceptable yields. Also considering Apples OS and the excellent SMP results they get it would be advantageous to have two duals in the line up, the middle of the road unit might be clock slightly slower though to keep the price down.

My reasoning is founded on the existence of the 1.6 and Jobs' chip roadmap.

The 1.6 is just a big old anomaly. I don't get it. It's like a leftover from the development cycle that - having developed a motherboard with (then) acceptable but not bleeding edge technology - they then decided to sell with a load of 1.6 chips they now had knocking around the place (having been pleasantly surprised by the 2.0 for their top end, if you see what I mean). Having a product with different memory/PCI etc. just doesn't make sense if you're trying to migrate to a new form-structure and achieve economies of scale in production and support. So perhaps the 1.6 (machine, not chip...I would have thought the rumoured 1.4 is the chip that died before it got out of East Fishkill) is, in fact, effectively a vestigial structure in the evolution of the G4 to the G5?

The 1.6 unit has its place in the market, that would be people that need a compatable PCI bus. This should not be underestimated, but I believe that Apple will find that market to be to small.

With regards to a 2.0+ around Xmas - well, this is a no-brainer if Jobs' roadmap is to be achieved. I would have also thought that something around 2.4 would be the top end of the 130nm process, after which heat/power issues become prohibitive. This would mean that the '3.0 within 12 months' relies upon migration to the 90nm process. That this is on the cards would appear evident from the G5 promotional video on Apple's website, in which Mr. IBM commented that they have already started development on 'the next generation of chips' (if memory serves, he might have actually said they've already begun sampling, but I'm not sure about that).

Anyway, just a thought.

Of course if I'm right, waiting till around Xmas (ish) to get a G5 means you not only miss out on any 'teething' troubles with the new machines (I'm intrigued as to how the 'dust' issue will work out) but also cash in on the 2.0 dropping to 'mid-range' prices etc.

Which means waiting to switch ISN'T so dumb, Mr. Moltz (you cheeky monkey), and is *in fact* a sign of great fortitude, patience, and, no doubt, rampant virility.

Probably.

Well okay, maybe not the virility part...


Brother Mugga

I'm still of the opinion that the case is a bigger issue than the percieved short comings of the G5. They will need a rev that quickly addresses expandability with in the case. Either that or they need a true workstation Mac housing.

Dave

 

[jettredmont]

Re: Future of PowerMacG5

Originally posted by Macrumors
Few details of these rumors are given -- for example, the actual PowerMac G5's bus runs at 1/2 of the 970's Processor Speed. A 1.7GHz bus would imply a 3.4GHz 970 Processor - based on current PPC970 designs. But this is NOT what's implied in the article.

I read this as "1.7GHz-capable", especially as it notes that the current bus is running at its max frequency right now (1GHz). I find this rumor to be highly believable, and certainly not self-contradictory.

The bus design should be made fairly independant of the processor. If 1/2 proc frequency continues to be the rule, then that means that Apple will either have to develop faster bus controllers (system controller chipsets) with each speed bump, or that Apple will design said controllers in fairly large bumps and throttle them down for the next couple speed bumps' worth of chips. The latter strategy is obviously preferable.

In other words, if this is correct, then the next speed bumps will have a redesigned SC, and we won't necessarily have to see another SC redesign until/unless the 970s surpass 3.4GHz.

 

[DharvaBinky]

Re: Re: 970 memory control & stuff

Originally posted by Waluigi
Besides the fast boot up features of MRAM, what exactly is it? Is it really fast in performance, or just really fast to boot up?

--Waluigi

A little about MRAM and its production

Also... regarding my mistake about the Athlon 64 memory controller...

You were right.

dharvabinky

 

[Dros]

Re: Re: Future of PowerMacG5

Originally posted by wizard
Here is my take on macbidouille. It goes without question that we will have to see a rather large jump in machine performance around the end of the year. Since SJ has publicly stated 3 GHz with in a year, the rational thing would be to get half way there by the start of 2004, which would mean a speed bump to atleast 2.5 GHz.

Dave

Since computer speed increases are perhaps based on a 'doubling' such as suggested by Moore's law, produces an exponential curve, not a line, the point in time half way (6 months) wouldn't produce a speed half way between. It would be less. Just being nitpicky .

On the plus side, it would mean two years out the projected speed would not be 4 GHz, but something more!

Of course, other factors are going to weigh more heavily than some theoretical exponential curve of speed, as Motorola was happy to demonstrate with the G4.

 

[Santiago]

Originally posted by WM.
I do hope that Apple soon brings all three slots to 133 MHz across the line.

That would require adding another PCI-X bus. The way the spec works is that you can have one 133 MHz slot or two 100 MHz slots. The PowerMac G5 has two busses, one of each type.

 

[wizard]

Re: Re: Re: Future of PowerMacG5

Ok; just who was talking about Moore's law?

What we are talking about is that Apple and IBM have stated that they will be at 3GHz in 12 months, December/January would put them close to 6 months, which would be a good time ot release a processor that is half way to 3GHz from where they are currently maxed out. Or in other words 2.5GHz, Moore's law does not come inot the equation. At this point it appear that going beyond 2GHz is a marketing question not a technicla one. Now they could pull a fast one on us and go directly to 3GHz come the end of the year.

By the way guys, 3GHz will probally be required by the end of the year if Apple expects the G5 to remain anywhere near competitive with the AMD/Intel world. As the bench marks start to roll in it will become more obvious that there are performance limitations with respect to the current G5 hardware, so Apple will be under a great deal of pressure to push forward.

Two years out the hope would be that Apple & IBM would have a processor that corrects the short comings of the G5. They are already working on it. don't be surprised if the operational frequency of G5's replacement is not dramtically faster on a per Hz basis.

Just trying to bring a little balance to the excitement.

Dave

Originally posted by Dros
Since computer speed increases are perhaps based on a 'doubling' such as suggested by Moore's law, produces an exponential curve, not a line, the point in time half way (6 months) wouldn't produce a speed half way between. It would be less. Just being nitpicky .

On the plus side, it would mean two years out the projected speed would not be 4 GHz, but something more!

Of course, other factors are going to weigh more heavily than some theoretical exponential curve of speed, as Motorola was happy to demonstrate with the G4.

 

[Dros]

Re: Re: Re: Re: Future of PowerMacG5

Originally posted by wizard
Ok; just who was talking about Moore's law?

What we are talking about is that Apple and IBM have stated that they will be at 3GHz in 12 months, December/January would put them close to 6 months, which would be a good time ot release a processor that is half way to 3GHz from where they are currently maxed out. Or in other words 2.5GHz, Moore's law does not come inot the equation.
Dave

I was just pointing out that while it would be a good thing to release a 2.5 GHz processor, since speed increases get larger over time (hence Moore's law), you wouldn't necessarily expect a speed half way between 2 and 3 at the half way mark. Just a small point, and as I said, other factors will override it. But you are making predictions based on linear speed increases over time, and processors haven't tended to increase like that.

Dros