IBM PPC 970MP A Reality?

[melgross]

That is why we have the 970GX( single core version of MP) and 970MP. They are based off the same core of the FX but is able to reach faster speeds like 3 Ghz. While wizard is right about the noise thing, also the 90nm process the heat is in a more condensed area where air cooling will have a hard time cooling it. Some people say the 2.5 Ghz G5 is overclocked. There is no evidence to prove it.

Unfortunately, we don't know anything about the 970GX. We don't even know if it will exist. We also don't know about the dual core 970MP. There haven't been any detailed specs released on the 970MP as yet. The most we have is info about the thermal diodes, and interfacing with them. That tells us nothing about the speed of the chip itself, only how to monitor its temps.

Where did you read that they will be faster. This sounds like the Rumors Department, Wishful Thinking Section.

The other info we have is tentitive at best. 1MByte shared cache, 1 shared memory buss, etc.

 

[daveL]

Unfortunately, we don't know anything about the 970GX. We don't even know if it will exist. We also don't know about the dual core 970MP. There haven't been any detailed specs released on the 970MP as yet. The most we have is info about the thermal diodes, and interfacing with them. That tells us nothing about the speed of the chip itself, only how to monitor its temps.

Where did you read that they will be faster. This sounds like the Rumors Department, Wishful Thinking Section.

The other info we have is tentitive at best. 1MByte shared cache, 1 shared memory buss, etc.

Get your rumors straight, will you ! It's 1 MB cache *per* core.

 

[melgross]

Get your rumors straight, will you ! It's 1 MB cache *per* core.

That's not what I read.

But a kiss is just a kiss. A smile is... Oh, sorry, I was dreaming. I meant a rumor is just a rumor.

 

[daveL]

That's not what I read.

But a kiss is just a kiss. A smile is... Oh, sorry, I was dreaming. I meant a rumor is just a rumor.

Well you obviously read a different, inaccurate, rumor Anyway, it will be interesting when the facts come to light.

 

[melgross]

Well you obviously read a different, inaccurate, rumor Anyway, it will be interesting when the facts come to light.

I'd love to read an accurate rumor. MOSR just said that the intro was pushed back to the beginning of May. That correlates to my info. But it IS MOSR so.

Maybe you can make up one here. I'll contribute, and we can spread it.

At least we'll know the sources are reliable.

 

[melgross]

I'd love to read an accurate rumor. MOSR just said that the intro was pushed back to the beginning of May. That correlates to my info. But it IS MOSR so.

Maybe you can make up one here. I'll contribute, and we can spread it.

At least we'll know the sources are reliable.

Hey!!! I think I just got my rumor threads mixed up!

See how confusing this all can get? That was the OS rumor thread.

Oh well, I guess it's good here as well.

so, how about it?

 

[daveL]

I'd love to read an accurate rumor. MOSR just said that the intro was pushed back to the beginning of May. That correlates to my info. But it IS MOSR so.

Maybe you can make up one here. I'll contribute, and we can spread it.

At least we'll know the sources are reliable.

So, they announce it at NAB with availability in early May. That would be normal for a PM release. Then, the actual shipments will be delayed another 4 to 6 weeks, due to problems with the chip supplier's inability to deliver. Then they'll be endless MR threads about how Apple's supply chain sucks. And then people will actually get the new PMs, and we'll have countless MR threads about how disappointed everyone is, because by that time DDR3, SATA3 and quad-core CPUs will be the next great thing and, of course, Apple didn't come through, again. Come on Apple, can't you get your shxt together?

 

[daveL]

Hey!!! I think I just got my rumor threads mixed up!

See how confusing this all can get? That was the OS rumor thread.

Oh well, I guess it's good here as well.

so, how about it?

I thought you did, but I went with the HW anyway

 

[ksz]

You're kidding, right. This is a fun post?

The 970x series is going to top out at about 3.0GHz. I hope we get there. Maybe .2GHz less or more.

If IBM could get to, say, 3.5 or so in the next few months, there wouldn't be a rush from them to get to dual cores. They could wait until early next year.

5GHz in 3 years. Well, 3 years is a looong time. By then they might be using 45nm. If the problems with the waste heat from numerous areas in the chip design can be reduced sufficiently, then maybe. But it hardly pays to be talking about mid 2008. We don't even know what's going to happen by January 2006!

Remember that these cooling problems are not just about the total amount of watts. The 2.5GHz chips dissipate less excess wattage, not more. It's the fact that the wattage is coming from an ever decreasing area. If you use half the wattage, but the area that is utilizing it is one quarter the size, then the temp produced is going to be higher. Less overall "heat", but concentrated over a smaller area, the temp rises.

The problem with this is that the temp diffuses through the chip and case at the same thermal speed as it did before. That causes a temp rise to be more serious that it was before. It becomes more difficult to cool because there is less area for the cooling system to interface with. The thermal resistence of the overall chip/cooling system is higher, slowing heat transfer from the INSIDE of the chip to the cooling interface.

While liquid cooling systems have a lower thermal resistence than do air cooled systems (because liquid has a greater thermal mass, as molecules in a liquid are closer together, eg. more of them per cubic cm), the problem remains as to how to get the "heat" out of the chip itself.

There are schemes as to how to do that, but so far they are only in the experimental stages. They present drastically different solutions to these problems, compared with even the cooling systems being used by Apple today, though ultimately some of them do use liquid cooling.

After all of this, if you have actually read it all, we can see that While Apple's cooling solution is good for todays chips, it may not be sufficient for those of 2 or 3 years from now.

Basically I'm sayng that you should not bet on anything remaining the same. IBM's new chips down the road might require radically new methods of cooling. If these methods were available, and the chips were designed to take advantage of them, we would be seeing 3.5GHz, or even higher speeds today.

This is a good summary. Power dissipation is the most significant obstacle for maintaining the historical rate of speed increases. From 1970 to about 1990 transistors were manufactured using the bipolar process. In 1970 average power dissipation from these bipolar devices was about 1 watt per square-cm. A steam iron dissipates 5 W/cm2. By about 1985 the IBM 3090 was dissipating approximately 4 W/cm2. By the late 1980s the IBM 3090s had reached 7 W/cm2. It was becoming clear that the bipolar curve could not be sustained and a newer, lower-power technology was necessary. By the end of the 1980s, the IBM ES9000 was hitting 13 W/cm2.

Fortunately, a savior was waiting in the wings. It was the CMOS process that brought power consumption all the way back to the 1970 levels and allowed provided a new growth platform. To squeeze more and more performance, "scaling" was the key. Feature sizes simply scaled down from quarter-micron design rules to 180nm, 130nm, 90nm, and are at 65nm today with 45nm and 30nm process technologies in R&D. However, power consumption has been increasing significantly. The Pentium 4 exceeds 5 W/cm2 while other devices are once again hitting 13 W/cm2. In effect, the CMOS curve is reaching the limits of practicality. As a result:

* More innovative cooling and on-chip power management features are being developed.

* Scalability is no longer the buzzword. More performance will not come from the same torrid rise in clock speeds. In fact, microprocessor operating frequencies will increase at half the historical rate.

* Product differentiation will be based increasingly on function/features rather than clock speed.

Whereas CMOS was the savior waiting in the wings to take over from Bipolar, there is *no new savior* as of yet. The semiconductory industry jumped from the Bipolar track to the CMOS track in the early 1990s, and it needs to jump tracks again, but there is no new track. New process technologies will emerge eventually, but as of now the CMOS track is going to slow down considerably and progress will be measured by added functionality such as dual or multiple cores (simultaneous multithreading), sub-processor partitioning, dynamic firmware updates, and enhanced memory subsystems.

 

[melgross]

I thought you did, but I went with the HW anyway

It's too bad that I have to eat dinner now. This is all finally starting to make sense.

 

[daveL]

ksz

I agree, and you did a great job of presenting the problem.

 

[quagmire]

Unfortunately, we don't know anything about the 970GX. We don't even know if it will exist. We also don't know about the dual core 970MP. There haven't been any detailed specs released on the 970MP as yet. The most we have is info about the thermal diodes, and interfacing with them. That tells us nothing about the speed of the chip itself, only how to monitor its temps.

Where did you read that they will be faster. This sounds like the Rumors Department, Wishful Thinking Section.

The other info we have is tentitive at best. 1MByte shared cache, 1 shared memory buss, etc.

I give you all the details of the 970MP and 970GX from the most reliable Apple news site( it has sort of evolved from a rumor site to news) Thinksecret.

http://www.thinksecret.com/news/0411ppc.html

http://www.thinksecret.com/news/antares.html

Give or take a few Mhz, this chip sounds impressive.

 

[melgross]

I agree, and you did a great job of presenting the problem.

Thanks.

Ok, I'm back for a while.

What I'm wondering is when Apple will be putting out their first Quantum computer. There must be some rumors about that we can start.

 

[daveL]

Thanks.

Ok, I'm back for a while.

What I'm wondering is when Apple will be putting out their first Quantum computer. There must be some rumors about that we can start.

G5 PowerBook, 'nuf said.

 

[melgross]

G5 PowerBook, 'nuf said.

You know, this may not be as strange as it sounds. They actually have Einstein-Bose Condensates functioning in microchips now at room temperature. Just a couple of years ago it was thought to be impossable.

 

[blitzkrieg79]

I am beginning to think that those 970GX and 970MP "slip ups" were done on purpose to throw people off... I remember that it was said that the current Power4 derivative is supposed to be for 18 months and June 2003 is when the G5 powerMacs were released so if my math skills are still in good shape that means its almost 2 years now = 24 months... After the 18 month period Apple was supposed to release he Power 5 derivative processor which was developed simultaneously with Power 5 (and since they had trouble to clock 970fx higher than 2.5GHZ it does make sense to basically stop wasting money on something that already didnt live up to expectations and invest it in something new especially that it seems Apple had more involvement in the Power 5 development) and Power 5 has been out for a while now so I am beginning to think that if Apples next PM release might be a rather significant one cuz definately offers a lot more clock for clock than Power 4... And the fact that Apple has been developing Power5 derivative simultaneously means that it wasnt the rush job as the current 970 was and they actually took time to refine it to suit Apples needs (not that the current 970 is bad but if IBM and Apple was able to develop 970 in a rather rushed atmosphere than I know they can really come up with something that will reall be a fastest personal computer in the world bars none)...

 

[melgross]

I am beginning to think that those 970GX and 970MP "slip ups" were done on purpose to throw people off... I remember that it was said that the current Power4 derivative is supposed to be for 18 months and June 2003 is when the G5 powerMacs were released so if my math skills are still in good shape that means its almost 2 years now = 24 months... After the 18 month period Apple was supposed to release he Power 5 derivative processor which was developed simultaneously with Power 5 (and since they had trouble to clock 970fx higher than 2.5GHZ it does make sense to basically stop wasting money on something that already didnt live up to expectations and invest it in something new especially that it seems Apple had more involvement in the Power 5 development) and Power 5 has been out for a while now so I am beginning to think that if Apples next PM release might be a rather significant one cuz definately offers a lot more clock for clock than Power 4... And the fact that Apple has been developing Power5 derivative simultaneously means that it wasnt the rush job as the current 970 was and they actually took time to refine it to suit Apples needs (not that the current 970 is bad but if IBM and Apple was able to develop 970 in a rather rushed atmosphere than I know they can really come up with something that will reall be a fastest personal computer in the world bars none)...

I would imagine (hope) that one of those upgrades would be the Express bus. That's what I'm waiting for. I wouldn't buy the first version though. A whole new architecture will need some shakedown cruising first.

PCI-X is just an in-between bus. It will be difficult to find cards for it once Express is established in a year or so.

Also some of these new cpu technologies will require the bandwidth that only Express offers.