[BLAZE_MkIV]

Does any one think that apple may do away with the daughtercard technique with the 970? I would think that with its high FSB bandwith the large L3 cache wouldn't be needed if they could keep the memory latency low enough. That and it would lower the cost of manufacturing and make upgrades easier, though that isn't exactly their pattern.

[OreoCookie]

The PPC970 doesn't have a L3 cache at all (no controller for that).

[:XI:]

If they did away with the cpu daughter cards they'd have to mount the cpu directly to the mothorboard. How would this make upgrades easier?
I can't really comment on manufacturing costs because I don't know enough about that. But I'm going to throw my thoughts in anyway.
Which would be cheaper a motherboard with a faulty cpu on a daughtercard or a faulty cpu soldered to the board?
Which would be cheaper a faulty motherboard with a cpu on a daughtercard or a cpu soldered to the board?

[OreoCookie]

Judging from the pix that I have seen, the PPC970 has a usual CPU socket like all PC cpus sold today. No daughtercards should be needed here -- plug it in, set the jumpers if you have to, and viol�.

[Mac Zealot]

XI why even bother with that? They could just stick it onto a ZIF socket

[RealMac]

In addition to this, it would make dealing with different models more difficult for apple. They'd essentially have to have three different motherboards for the cpus at the three different price points. No one in the industry is going back to such a design. It's all about easily swapping of parts.

Technically, it might not even be feasible to solder the 970s to a motherboard. Dual 970s are going to generate a lot of heat and take up a lot of space on the motherboard (vs a connector). Just my 2 cents.

[:XI:]

Originally posted by Mac Zealot:
XI why even bother with that? They could just stick it onto a ZIF socket

Ok, good point. I read the first post as "get rid of the daughter cards and build the cpus into the motherboard, the same way as the iMacs, iBooks and Powerbooks."

My bad.

[BLAZE_MkIV]

I probably should have been more specific. I was thinking of going back to a zip style socket but without the daughter card pcb. I seems to me that apple has to 1 buy the cpus then 2 get someone to paste them onto the PCB. I would think that the incrase in MB space would be more than offset in the loss of the daughtercard.

I know they have test equipment in the factory were they assemble pcb's to test them prior to shipping, and they probably test cpu's during manufacturing now as it is, otherwise their 50% or so initial failure rate would reach into the product lines.

[D'Espice]

Originally posted by BLAZE_MkIV:
Does any one think that apple may do away with the daughtercard technique with the 970? I would think that with its high FSB bandwith the large L3 cache wouldn't be needed if they could keep the memory latency low enough. That and it would lower the cost of manufacturing and make upgrades easier, though that isn't exactly their pattern.

The question is not what they're about to do, the question is: who cares? Why can't you people wait until these Macs are going to be available? Discussing issues of absolutely no importance with absolutely no background information is pretty silly, don't you think?

[Scotttheking]

I could tell you, but then I'd have to kill you.

[Eug]

Originally posted by D'Espice:
The question is not what they're about to do, the question is: who cares? Why can't you people wait until these Macs are going to be available? Discussing issues of absolutely no importance with absolutely no background information is pretty silly, don't you think?

Ummm... This is a speculation thread. Nothing wrong with that.

[BLAZE_MkIV]

I expressely labeled it as speculation so go troll elswere.

[Scotttheking]

careful there, mods don't troll

[BLAZE_MkIV]

So moderator status automatically confers them with a cloak of troll protection? I don't think so. But thats way off topic just like his comments.

[voodoo]

Originally posted by Scotttheking:
careful there, mods don't troll

Steve does.

[aaanorton]

Originally posted by Scotttheking:
careful there, mods don't troll

Somebody should tell all the mods that.

[Athens]

Originally posted by :XI::
If they did away with the cpu daughter cards they'd have to mount the cpu directly to the mothorboard. How would this make upgrades easier?
I can't really comment on manufacturing costs because I don't know enough about that. But I'm going to throw my thoughts in anyway.
Which would be cheaper a motherboard with a faulty cpu on a daughtercard or a faulty cpu soldered to the board?
Which would be cheaper a faulty motherboard with a cpu on a daughtercard or a cpu soldered to the board?

You havent installed a Pentium 4 have you? Lift up arm, pop CPU in, justs glides on in, lock the arm. Clamp the Heatsink and fan done. Its painless. No screw drivers are anything, AMD could use some lessions on Intels design. Apple should adopbt that too. Standard CPU socket, so ppl can upgrade there Macs later on.

[Mac Zealot]

thats what i was talking about and apple DID use it before.

But anyway, CPU testing is thrown into the product testing process, I'm rather sure when apple runs these machines off the production line they do a quick check on the boards, plug the machine in, see if they get an apple logo and a test screen, and then kill the power.

Not sure raelly, what I *CAN* tell you is that it isn't very in depth.. and if anything each machine gets no more then what.. 10 seconds of testing?

This is apple, not alienware lol.

[veryniceguy2002]

Originally posted by Athens:
You havent installed a Pentium 4 have you? Lift up arm, pop CPU in, justs glides on in, lock the arm. Clamp the Heatsink and fan done. Its painless. No screw drivers are anything, AMD could use some lessions on Intels design. Apple should adopbt that too. Standard CPU socket, so ppl can upgrade there Macs later on.

Well, Apple doesn't want people to upgrade their Macs... they just want people to buy a new one from the Apple store when their existing one can run the latest version Mac OS X, or haven't got the latest hardware feature

So in Apple perspective, it only add cost and no clear advantage to make CPU upgrade easy.

[-Q-]

Originally posted by veryniceguy2002:
So in Apple perspective, it only add cost and no clear advantage to make CPU upgrade easy.

Yes, but that's counteracted by the need for apple to make the mobo's cost effective. And it's much easier to to introduce new processor speeds if they just have to install the new processor via an easily insertable slot rather than having to solder things to the mobo.

[veryniceguy2002]

Originally posted by -Q-:
Yes, but that's counteracted by the need for apple to make the mobo's cost effective. And it's much easier to to introduce new processor speeds if they just have to install the new processor via an easily insertable slot rather than having to solder things to the mobo.

In terms of manufacturing it is cheaper to soldier the CPU then make an insertable slot! It is because for those insertable CPU slot (like those Intel ones) would involve a human operator to push the lock arm, and that cost $$$$$. Whereas if the CPU soldered on, the whoel process can be done by machines in the factory, and this saves $$$$$!

[:XI:]

Originally posted by Athens:
You havent installed a Pentium 4 have you? Lift up arm, pop CPU in, justs glides on in, lock the arm. Clamp the Heatsink and fan done. Its painless. No screw drivers are anything, AMD could use some lessions on Intels design. Apple should adopbt that too. Standard CPU socket, so ppl can upgrade there Macs later on.

No, but I have installed Pentiums, same thing.
I missread the topic starters post and completely forgot about ZIF sockets.

[Scotttheking]

Originally posted by Athens:
You havent installed a Pentium 4 have you? Lift up arm, pop CPU in, justs glides on in, lock the arm. Clamp the Heatsink and fan done. Its painless. No screw drivers are anything, AMD could use some lessions on Intels design. Apple should adopbt that too. Standard CPU socket, so ppl can upgrade there Macs later on.

That is how AMD processors are installed...
Of course, AMD has been using the same socket for a while, while intel keeps changing it so you can't upgrade.

[CIA]

My Beige G3 was like that. Switching it to a 500Mhz G3 Upgrade was a matter of popping it in the zif slot and setting jumpers....

[OreoCookie]

Originally posted by RealMac:
In addition to this, it would make dealing with different models more difficult for apple. They'd essentially have to have three different motherboards for the cpus at the three different price points. No one in the industry is going back to such a design. It's all about easily swapping of parts.

Technically, it might not even be feasible to solder the 970s to a motherboard. Dual 970s are going to generate a lot of heat and take up a lot of space on the motherboard (vs a connector). Just my 2 cents.

Heat would be ok. A Pentium 4 or Athlon XP consumes about as much as two 970s (@ half the SPECmarks). Cooling would be feasible.

[-Q-]

Looks like the speculation is getting more detail-oriented:

AppleInsider

[Eug]

Originally posted by OreoCookie:
Heat would be ok. A Pentium 4 or Athlon XP consumes about as much as two 970s (@ half the SPECmarks). Cooling would be feasible.

A 3 GHz P4 has significantly higher SPECmarks than a 1.8 GHz PPC 970. The 1.8 would come in at around 2.3 GHz in today's P4 terms.

[osxisfun]

unless IBM is, as moki has hinted, is "sandbagging"...


June23rd be there or be square...

[MindFad]

Originally posted by -Q-:
Looks like the speculation is getting more detail-oriented:

AppleInsider

Man, I so hope that's the case. What an awesome surprise that would be. And the developers wouldn't have to lynch poor Steve.

[Scotttheking]

There's nothing technologically impressive about those specs. If they had come out a year ago, it would be ahead of the PC crowd. Now, it's just on par. I just hope there is a decent, DEAD SILENT case to go with the system, assuming it's released.

[Eug]

Originally posted by osxisfun:
unless IBM is, as moki has hinted, is "sandbagging"...

There could be increases in performance from more (compiler, etc.) optimization since then, and also from the possibility 1.8 GHz won't be the initial top end, but I doubt they're really sandbagging in terms of their reported SPEC indicators. At best they could have somewhat conservative "estimates", but most companies tend to overestimate performance if anything.

Why is moki saying they're sandbagging, and where did you see that?

[MindFad]

Originally posted by Scotttheking:
There's nothing technologically impressive about those specs. If they had come out a year ago, it would be ahead of the PC crowd. Now, it's just on par. I just hope there is a decent, DEAD SILENT case to go with the system, assuming it's released.

What about the 64-bit potential for apps that can take advantage of the larger chunks of data (rendering, database stuff, etc.), or that these are *estimated* numbers *without* the AltiVec extensions, or the huge bus? I imagine we're going to see some really cool stuff done with these chips. I'm already impressed, though I'd really like to get my hands on one.

[kupan787]

Originally posted by Eug:
A 3 GHz P4 has significantly higher SPECmarks than a 1.8 GHz PPC 970. The 1.8 would come in at around 2.3 GHz in today's P4 terms.

Umm....

PowerPC 970 @ 1.8 ghz
SpecInt base: 937
SpecFP base: 1051

Pentium4 @ 3GHz:
SPECint - 1099
SPECfp - 1077

Pentium4 @ 2.4GHz
SPECint score of 819
SPECfp score of 806

So the 970 and the 3.0GHz P4 are much close than you make it sound. And the 970 beats up a 2.4GHz P4. Looks like you were a bit off Eug...

And that is single 970 vs single P4 at twice the MHz!!! Apple will ship duals, so expect more ass kicking.

And just to throw in:
MOT G4 does:
SPECint2000 418 @ 1.4 GHz
SPECfp2000 248 @ 1.4 GHz

Note that the G4 can hang with the P4 in some tasks (small few), but mainly gets beat up by the 3.0GHz P4 in a majority of tasks. Now the G4's "spec" numbers suck. So if the 970 has very similar marks to a P4, why should it not just blow out the P4 for most things?

[Eug]

Originally posted by kupan787:
Umm....

PowerPC 970 @ 1.8 ghz
SpecInt base: 937
SpecFP base: 1051

Pentium4 @ 3GHz:
SPECint - 1099
SPECfp - 1077

Pentium4 @ 2.4GHz
SPECint score of 819
SPECfp score of 806

So the 970 and the 3.0GHz P4 are much close than you make it sound. And the 970 beats up a 2.4GHz P4. Looks like you were a bit off Eug...

You've forgotten that Intel has upgraded their chips since then.

The current 2.40C has both Hyper-Threading and an 800 MHz bus. The old 2.4 GHz didn't do so well because it was on a 400 MHz (or maybe 533 MHz) bus, and the chip didn't have Hyper-Threading. (Maybe the compiler or whatever has been updated too?)

The (base) SPEC numbers for the new Intel chips:

P4 2.40C GHz - SPECint 984, SPECfp 1081
P4 3.0 GHz - SPECint 1164, SPECfp 1213.

See my thread here.

If you use the 2.40C integer bench as a gauge (and if the SPEC numbers are at all reliable), the 1.8 GHz PPC 970 would be equivalent to about a P4 2.3 GHz in current terms.

What should be noted however, is that a P4 2.2 GHz would be significantly slower, since there is no such thing as a Hyper-Threaded 800 MHz bus P4 2.2 GHz. It's pretty big jump to go from a P4 2.2 GHz to a P4 2.4 GHz these days.

[Commodus]

Of course, it remains to be seen whether IBM's initial benchmarks hold true. Remember that IBM actually said that those SPEC scores were conservative figures. I'm not expecting a massive gain - but they could well be improved upon.

The other question mark concerns dual processors. I don't expect a 90% efficiency gain through a second CPU, but even 50% would push a dual-1.8 past the P4 (probably equalling or beating a 3.2 GHz P4).

Even if they were a bit slower, I don't think most people here would care. It would be a lot faster than what we have now.

[osxisfun]

re: sandbagging

it was in one of the many 970 threads at appleinsider.com

(don't know which one)

[Eug]

Forget Jaguar! Forget Panther! PPC 970 is gonna run on Smeagol!!!

[kupan787]

Originally posted by Commodus:
Of course, it remains to be seen whether IBM's initial benchmarks hold true. Remember that IBM actually said that those SPEC scores were conservative figures. I'm not expecting a massive gain - but they could well be improved upon.

Actually this is a good point. When AMD released figures for the Opteron (sp?) the initial spec numbers were lower than once chips were relased.

Preliminary Marks:
Opteron @ 1.8 ghz
SpecInt base: 1048
SpecFP base: 998

"Released chips" Marks (for lack of better term):
Opteron @ 1.8 ghz
SPECint: 1170
SPECfp: 1219

That is a pretty good increase. 970 might see a increase in spec results as well.

What what does spec really compare? I have seen mention over at appleinsider that spec is pretty useless as a "benchmark" as the P4 numbers are very optimized (really good compilers, specifically tuned, etc). So I guess this might all be moot. Lets wait for some "realworld" results before we let judgment pass.

[Catfish_Man]

Originally posted by kupan787:
And just to throw in:
MOT G4 does:
SPECint2000 418 @ 1.4 GHz
SPECfp2000 248 @ 1.4 GHz

Actually, Moto managed to get a SPECint of 500 from a 1.25GHz PowerMac using some altiveced libraries. It was in a PDF recently, can't remember where.

In response to the post with the P4 SPEC scores: I hadn't seen that, but ~1200 SPECfp is extremely impressive from the P4. I imagine it's mostly due to the bus, since afaik SPEC doesn't use duals (so it wouldn't use SMT).

[Eug]

Originally posted by Catfish_Man:
Actually, Moto managed to get a SPECint of 500 from a 1.25GHz PowerMac using some altiveced libraries. It was in a PDF recently, can't remember where.

In response to the post with the P4 SPEC scores: I hadn't seen that, but ~1200 SPECfp is extremely impressive from the P4. I imagine it's mostly due to the bus, since afaik SPEC doesn't use duals (so it wouldn't use SMT).

Here is the Intel benchmark bragging page.

And here is SPECfp. Just look for the ones with the Intel 875 mobo.

But enough about Intel. I wanna see a PPC 970 dammit!

[kupan787]

Here is a good analogy I saw posted over at AppleInsder about the 970:

Originally posted by Programmer
Imagine a 32-bit machine as a 2 door car. Imagine a 64-bit machine as a 4 door car. Now imagine that you have a driver's license that restricts you to carrying only yourself and 1 passenger. That's what running Jaguar on a 970 is like... you've only ever had 2 door cars so while now you've got one that has these extra doors, it doesn't do anything for you until you get a new license. It just so happens that the new car (in addition to having more doors) is roomier, faster, more fuel efficient and cheaper so its well worth having even though you can't use all the doors (yet). In a couple of months you'll get a new license and then you can unlock the doors and bring more people on board (a lot more -- while the two doors only have 2 seats, the rear doors have many many many seats).

and he also added a few posts later:

Glad you like it. Just remember that it is only an analogy and that while useful for grasping a concept it is dangerous to look too deep and try to draw conclusions based on the analogy and not the original concept

[OreoCookie]

Originally posted by Eug:
You've forgotten that Intel has upgraded their chips since then.

The current 2.40C has both Hyper-Threading and an 800 MHz bus. The old 2.4 GHz didn't do so well because it was on a 400 MHz (or maybe 533 MHz) bus, and the chip didn't have Hyper-Threading. (Maybe the compiler or whatever has been updated too?)

The (base) SPEC numbers for the new Intel chips:

P4 2.40C GHz - SPECint 984, SPECfp 1081
P4 3.0 GHz - SPECint 1164, SPECfp 1213.

See my thread here.

If you use the 2.40C integer bench as a gauge (and if the SPEC numbers are at all reliable), the 1.8 GHz PPC 970 would be equivalent to about a P4 2.3 GHz in current terms.

What should be noted however, is that a P4 2.2 GHz would be significantly slower, since there is no such thing as a Hyper-Threaded 800 MHz bus P4 2.2 GHz. It's pretty big jump to go from a P4 2.2 GHz to a P4 2.4 GHz these days.

A lot of this is due to new versions of Intel's own compiler. As the figures posted above were from a similar point in time, I'd take them as a direct comparison instead of the latest ones. New versions of gcc have steadily improved performance, too, so the 970's figures could improve somewhat. Then, additionally, the SPECrates are of interest, too. Traditionally, the P4/Xeon line doesn't do too well compared to the Opteron or (of our interest) the Power4. I assume there will be dualies (at least the top model) right from the start -- and these will perform quite well compared to an Intel workstation.

Also of interest should be AltiVec. It doesn't show in the SPECmarks (GNU compiler suite doesn't use AltiVec AFAIK), but it will speed up OS X tremendously.

I'd suggest that we'd postpone that discussion till the release of the new systems (probably in two weeks' time). I mean, the SPECmarks given by IBM were estimates. Plus according to IBM (and common sense), the PPC970's clockspeeds will rise up to at least 2.5 GHz.

[veryniceguy2002]

Originally posted by kupan787:

What what does spec really compare? I have seen mention over at appleinsider that spec is pretty useless as a "benchmark" as the P4 numbers are very optimized (really good compilers, specifically tuned, etc). So I guess this might all be moot. Lets wait for some "realworld" results before we let judgment pass.

According to this PDF file from The University of Texas at Dallas lecture notes by D. Hollenbeck:

SPEC benchmarks are small "kernels" of code extracted from real programs

. May run unrealistically fast

. Programs with large arrays that are accessed using a large stride in memory are likely to run much more slowly than the same program with smaller arrays and a smaller stride

so depends how the chip designer design the chip, it is possible for a chip have a higher SPEC mark number than another, yet in real world application the result could go other way round.

I think the comparison is independent of compiler performance etc. The test could be developed at a very low level.

[Mac Zealot]

*shrug* No matter what the x86 chip will be faster FOR THE MOMENT but it's had the past 30 years to get there.

The PPC chip itself is 10 years old, give or take, and has been maintained by crap manufacturers not keeping to their promises. I would say it's surprising that a processor maintained by a phone company is about as powerful (or was) as THE PROCESSOR TO HAVE.

The 970 is better, yes, but remember that unlike motorola IBM has PLANS for this processor, and so the tables could turn fast, and the processor could become much better.

[veryniceguy2002]

Originally posted by Mac Zealot:
*shrug* No matter what the x86 chip will be faster FOR THE MOMENT but it's had the past 30 years to get there.

The PPC chip itself is 10 years old, give or take, and has been maintained by crap manufacturers not keeping to their promises. I would say it's surprising that a processor maintained by a phone company is about as powerful (or was) as THE PROCESSOR TO HAVE.

You cannot really say Motorola (unless I misread the meaning of your post) is just a phone company. This is a wallstreet analyst's preception of Motorola.

MOT has been doing silicon chips longer than making mobile phones! (MOT had chips back in early 70's, and only start doing mobiel phone development in late 70's)

The 970 is better, yes, but remember that unlike motorola IBM has PLANS for this processor, and so the tables could turn fast, and the processor could become much better.

No. Motorola also have plans, but their plans does not "align" with exactly what Apple wants. That's all.

[Hozie]

Guys, there's more and more indication that the 970's will be announced at WWDC. As some of you might have noticed, a story run by Arn at MacRumors.com was pulled at the request of Apple Legal. It pretty much summarized the AI article.


That, to me, is the biggest piece of evidence that we really are on the verge of seeing the new chips. Just FYI...

[Xaositect]

Back to original topic -

I would guess a daughter card to make dual (and - dare I dream - quad!) processor designs easier (same board as single processors).

Now on the secondary topic -

I will wait and compare shipping models. We just don't know how Apple's logic board is going to affect the performance until we have one.

[proux]

It seems that all the rumors point to single CPU at WWDDC only... big bummer. They might announce the duals but not ship them. August/September for the dual Powermacs and maybe Powerbooks.

[jcadam]

Originally posted by veryniceguy2002:
According to this PDF file from The University of Texas at Dallas lecture notes by D. Hollenbeck:



so depends how the chip designer design the chip, it is possible for a chip have a higher SPEC mark number than another, yet in real world application the result could go other way round.

I think the comparison is independent of compiler performance etc. The test could be developed at a very low level.

Since SPEC is considered THE benchmark, some (most) CPU manufacturers optimize their processors for it. Just like video card manufacturers were heavily optimizing for Quake 3 performance (was the most important 3D benchmark for a while).

Real world performance is bound to be different.

[chrisutley]

Originally posted by proux:
It seems that all the rumors point to single CPU at WWDDC only... big bummer. They might announce the duals but not ship them. August/September for the dual Powermacs and maybe Powerbooks.

I don't think they will announce anything too far in advance, otherwise it's harder for them to sell the existing lines.