FWIW, I was poking around on Ars Battlefront yesterday to watch some fur fly after this announcement and I found this quote interesting:
[quote]
As a Macintosh user, I live in interesting times. And I believe that Apple will see those those kinds of speeds, and more, from their as yet unannouced "switcher" program of their own.
<hr></blockquote>
At first I thought...huh? But now it's making a bit more sense.
<strong>everyone sure has gotten their expectations up. There has been no announcement that apple will even use the 970 or anything else for that matter. THey could do something completely arbitruary. Lets get back to earth. Mac users this is Houston... do you read me?</strong><hr></blockquote>
Maybe you're right...no wait, I cannot even imagine that a Apple will use anything BUT an IBM processor in the Powermacs upcoming. All things point to Apple. Of course there is no announcement, this is Apple. Have they ever pre-announced something like this? Hell no. Houston, ignore kraig911, he's lost all his senses
Looks like an interesting future indeed...</strong><hr></blockquote>
I agree the future looks interesting but here you are comparing *actual* P4 SPEC numbers to a linear extrapolation of *estimated* 970 SPEC numbers. And the P4 3.067 is shipping now. SPEC numbers never see a linear scaling with Hz without other improvements as well - from 1.8 to 2.5 GHz you'd probably get at least a 5% drop from linear scaling. I think the match up would look more like this on release of the 970:
[code]
PPC 970 @ 2.5 GHz SPECint2k = 1236
SPECfp2k = 1386
P4 @ 3.4 GHz SPECint2k = 1278 (up from 1130(*))
SPECfp2k = 1249 (up from 1103(*)) </pre><hr></blockquote>
For the P4 I assumed 10% super-linear scaling of the measured Dell #s since Intel are moving from a 533 MHz FSB to an 800 MHz FSB which will be very beneficial. The hardware sites are expecting at least a 10% increase based on this move.
Of course, since the 970 #s are estimates they may be conservative. Also, the FSB scaling with the internal clock may prevent sub-linear scaling of performance. Thirdly, the VMX/AltiVec unit is superior to SSE and is finally getting bandwidth.
My point is, the 970 will definitely be competetive with the Wintel world but has yet to prove it's trample-all-over-ness.
<strong>everyone sure has gotten their expectations up. There has been no announcement that apple will even use the 970 or anything else for that matter. THey could do something completely arbitruary. Lets get back to earth. Mac users this is Houston... do you read me?</strong><hr></blockquote>
I'm not sure I follow you. Where would Apple go if they don't use the 970. They surely aren't going to hit 8 Billion of revenue with the G4...PERIOD. I realize you're probably the type of person that doesn't believe stuff until it's actually announced but it would really take someone really dense to discount the "mounds" of information available. This really is an unguarded secret. The only people that will be purchasing G4 Powermacs are the ones that can't wait.
That assumes you're correct that it takes 60 days to fab the PPC970 and I'm not convinced that's the case. It may take 60 days (or so) to ramp up to full production level yields, but that doesn't mean there won't be chips available (although constrained) to the market before then.</strong><hr></blockquote>
Added to this is Apple's penchant for anouncing new systems before they actually ship. They might have everything in place and make the anouncement when they get the first batch of production chips and have tested a few for quality & compatibility.
How about a May announcement and 1st July shipping?
<strong>Andrew for president!!! Now THIS is how you lift the spirit of a people!</strong>
Originally posted by moki:
<strong>
Wish I had something to do with it, but I don't. Thank the researchers at IBM, assuming things hold up as well in the real world as they do "on paper".</strong><hr></blockquote>
Anybody who doesn't take credit for something they didn't do could never be president nowadays.
That assumes you're correct that it takes 60 days to fab the PPC970 and I'm not convinced that's the case. It may take 60 days (or so) to ramp up to full production level yields, but that doesn't mean there won't be chips available (although constrained) to the market before then.</strong><hr></blockquote>
Sorry, my mistake. I really was under the impression, from other sources, that the actual process from beginning to end in manufacturing a cpu took about 60 days. I mean that's a lot of itty bitty lines to etch and fill with Cu and to do it on what 6-9 layers of silicon.
<strong>Better: Why didn't Apple update the 15" PB, and why hasn't it now that the channel has been bone dry for some time?
How many 17" PB's have shipped? Over at MacNN, there are several increasingly furious threads about shipping delays.<hr></blockquote>
I think this is likely to have more to do with contractual obligations than with any CPU plans. As I recall, Apple recently switched (or added) a new LCD screen supplier and/or has a new laptop manufacturer, so existing component shortages due to the transition is the most likely culprit here.
[quote]As an idle thought, how many 1.42 GHz PowerMacs have actually shipped?</strong><hr></blockquote>
Not that many, they've only been shipping for a few months now and demand has been slack. If Apple is purposely reducing channel inventory they probably started weeks ago.
<strong>I mean that's a lot of itty bitty lines to etch and fill with Cu and to do it on what 6-9 layers of silicon.</strong><hr></blockquote>
Well yeah, but they don't do them one line at a time!
Seriously though, I think the big problem is not so much the actual time it takes to do deposition and etch the traces, etc. but more in determining what the optimal time and technique to use at each stage is. Initially, it's an ongoing process that's constantly being tweaked, but once you hit a sweet spot, you pretty much nail it down and crank them out. And judging by the way they went from 1.8Ghz at the top end of the range (in October) to 1.8GHz at the bottom end (now), I'd say they've come pretty far in determining what that "sweet spot" is.
<strong>If a PPC970 @ 2.5GHz already exists (even if there are 2 or 3 of them) and things are really going well at IBM, what chance is there that PPC970 hits mass production earlier than in autumn? In other words, could one of moki's two surprises be an earlier-to-unexpected release date of our brand new PowerMac?
Could, indeed, the second surprise be Marklar? Consider that Mac OS X on Intel is not like VirtualPC for Mac. It would be times faster. And it would begin a next round of Apple vs. Microsoft box match.
What do you think?</strong><hr></blockquote>
As to the 970's production timetable, I believe that they stated that these will ship i the secnd half of the year. So the earliest we will see something from them is July 1st (the 1st day of the7th month) and the latest would be December 31st of 03.
Now pure speculation, IBM is now touting a tope end at introduction of 2.5 Ghz, since the top end speed of a chip is determined by testing the chips then I would think that they have some initial test units off of the production line to back up this statement, and are preparing to go into full production very soon. They probably need to build up an inventory before they start shipping them, and some final assembly time at Apple....say 90 days earliest till a shipping 970 hits the market. I think that it is very possable that Apple might have shipping units at Macworld this summer, and that would be a great time to introduce and demo a dual 970 PM running OS X 10.3 in a Maya bake-off.
<strong>Reducing inventory of NEW machines shortly after announcing them...it's a sad time, currently. Oh the difference a few months will make!</strong><hr></blockquote>
I said IF. It's just as likely (no, more likely) it's a blip in the supply chain and nothing to do with inventory reduction.
Oh I know; it's just the way I read and interpretted it the first time that struck me as funny that Apple would be clearing inventory when they just announced the 1.42s. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
<strong>There's been some talk about Apple incorporating the Bochs emulation (open-source Windows emulator for Unix) system into OS X, possibly as soon as 10.3. <a href="http://bochs.sourceforge.net/"; target="_blank">Bochs homepage</a>
That would be the REAL butt-kicker - if the 970+Bochs gives Apple native IA-32 emulation that is nearly as fast as the best x86 hardware, built right into the OS, who needs Marklar?</strong><hr></blockquote>
Unfortunately, Bochs is so amazingly slow that I think you may as well give up on that idea.
<strong>Well... You have the possibility to buy a cheap peecee and run OSX on it, or you will have a PIV crushing PPC 970 with OSX on it.. Consumers buy peecee's AND OSX, top-end d00d's buy PPC 970's with OSX... Apple's marketshare KABOOOMS. ==> Apple sells more macs ==> More marketshare ==> More Apple sales... -Get my point? </strong><hr></blockquote>
oh i think i get it now <img src="graemlins/surprised.gif" border="0" alt="[surprised]" /> ... "get this copy of the great Mac OS X Demo for PC for just $100 and look how well it runs on bad hardware!!! after that take a look at the new OS X on a PowerMac970 and be blown away!!!" hmmm ... ok if the windows-people finished playing windows they can try Mac OS X Demo... i don't care... <img src="graemlins/cancer.gif" border="0" alt="[cancer]" />
All the power in the world is not going to make Bochs a champion.
I find it ironic that VPC costs more than a PC now. Hell if you need a PC $300 will take care of that easily. I don't want Apple to spend one Iota of time ingrating that horrible codebase into OSX. I prefer OSX to ship "virus" free
Unfortunately, Bochs is so amazingly slow that I think you may as well give up on that idea. </strong><hr></blockquote>
From what I gather, Bochs has been basically a one-man operation. How much of its sluggishness is just due to inefficient coding (no offense intended to Kevin Lawton, if you're out there)? I would imagine Apple engineers could massage his code to a fare-thee-well...
I'm mostly interested in SpecFP, so I dropped the others for simplicity.
[quote]Originally posted by MartianMatt:
<strong>My point is, the 970 will definitely be competetive with the Wintel world but has yet to prove it's trample-all-over-ness.
</strong><hr></blockquote>
SpecFP estimates:
[code]
1x G4+ 1.0GHz 187
1x G4+ 1.4GHz 261
2x G4+ 1.4GHz 523
1x P4 3.4GHz 1249
1x 970 1.8GHz 1051
1x 970 2.5GHz 1459
2x 970 2.5GHz 2918
4x 970 2.5GHz 5839
</pre><hr></blockquote>
The multi-CPU P4's & Xeons have heat issues & cost through the nose. (Yes, there are things more expensive than Macs.)
That's not the main thing though.
All you need to do is read the line "2xG4" and the line "2x 970"
5.57x as fast. Trample-all-overness.
And the thing that's been the big hurdle historically at Apple to Quads is the _bus_. With the bus fixed, a Quad-ppc can easily undercut a Quad-Xeon in price... Which means the '4x 970' line might be interesting.
Having the bus fixed should also be a HUGE benefit to AltiVec, which isn't reflected (at all) in the Spec numbers. No Mac-types were too concerned when 'pc Hz' was around 1.3-1.5x Mac Hz - AV quite effective for what it does.
Comments
[quote]
As a Macintosh user, I live in interesting times. And I believe that Apple will see those those kinds of speeds, and more, from their as yet unannouced "switcher" program of their own.
<hr></blockquote>
At first I thought...huh? But now it's making a bit more sense.
<strong>everyone sure has gotten their expectations up. There has been no announcement that apple will even use the 970 or anything else for that matter. THey could do something completely arbitruary. Lets get back to earth. Mac users this is Houston... do you read me?</strong><hr></blockquote>
Maybe you're right...no wait, I cannot even imagine that a Apple will use anything BUT an IBM processor in the Powermacs upcoming. All things point to Apple. Of course there is no announcement, this is Apple. Have they ever pre-announced something like this? Hell no. Houston, ignore kraig911, he's lost all his senses
<strong>IBM PPC 970:
SPECint2000
? 937 @ 1.8 GHz
SPECfp2000
? 1051 @ 1.8 GHz
Dhrystone MIPS
? 5220 @ 1.8 GHz
....
So let's extrapolate this to 2.5ghz...
SPECint2000
? 1301 @ 2.5 GHz
SPECfp2000
? 1459 @ 2.5 GHz
Dhrystone MIPS
? 7250 @ 2.5 GHz
.....
[snip]
.....
Oh, and here's a Pentium IV @ 3.06 GHz:
SPECint2000
? 1032 @ 3.06 GHz
SPECfp2000
? 1092 @ 3.06 GHz
.....
Looks like an interesting future indeed...</strong><hr></blockquote>
I agree the future looks interesting but here you are comparing *actual* P4 SPEC numbers to a linear extrapolation of *estimated* 970 SPEC numbers. And the P4 3.067 is shipping now. SPEC numbers never see a linear scaling with Hz without other improvements as well - from 1.8 to 2.5 GHz you'd probably get at least a 5% drop from linear scaling. I think the match up would look more like this on release of the 970:
[code]
PPC 970 @ 2.5 GHz SPECint2k = 1236
SPECfp2k = 1386
P4 @ 3.4 GHz SPECint2k = 1278 (up from 1130(*))
SPECfp2k = 1249 (up from 1103(*)) </pre><hr></blockquote>
* Dell Precision WorkStation 350 @ 3.067 GHz - see <a href="http://www.spec.org"; target="_blank">www.spec.org</a>
For the P4 I assumed 10% super-linear scaling of the measured Dell #s since Intel are moving from a 533 MHz FSB to an 800 MHz FSB which will be very beneficial. The hardware sites are expecting at least a 10% increase based on this move.
Of course, since the 970 #s are estimates they may be conservative.
My point is, the 970 will definitely be competetive with the Wintel world but has yet to prove it's trample-all-over-ness.
MM
[ 02-28-2003: Message edited by: MartianMatt ]</p>
<strong>everyone sure has gotten their expectations up. There has been no announcement that apple will even use the 970 or anything else for that matter. THey could do something completely arbitruary. Lets get back to earth. Mac users this is Houston... do you read me?</strong><hr></blockquote>
I'm not sure I follow you. Where would Apple go if they don't use the 970. They surely aren't going to hit 8 Billion of revenue with the G4...PERIOD. I realize you're probably the type of person that doesn't believe stuff until it's actually announced but it would really take someone really dense to discount the "mounds" of information available. This really is an unguarded secret. The only people that will be purchasing G4 Powermacs are the ones that can't wait.
<strong>
That assumes you're correct that it takes 60 days to fab the PPC970 and I'm not convinced that's the case. It may take 60 days (or so) to ramp up to full production level yields, but that doesn't mean there won't be chips available (although constrained) to the market before then.</strong><hr></blockquote>
Added to this is Apple's penchant for anouncing new systems before they actually ship. They might have everything in place and make the anouncement when they get the first batch of production chips and have tested a few for quality & compatibility.
How about a May announcement and 1st July shipping?
MM
<strong>Andrew for president!!! Now THIS is how you lift the spirit of a people!</strong>
Originally posted by moki:
<strong>
Wish I had something to do with it, but I don't. Thank the researchers at IBM, assuming things hold up as well in the real world as they do "on paper".</strong><hr></blockquote>
Anybody who doesn't take credit for something they didn't do could never be president nowadays.
<strong>
That assumes you're correct that it takes 60 days to fab the PPC970 and I'm not convinced that's the case. It may take 60 days (or so) to ramp up to full production level yields, but that doesn't mean there won't be chips available (although constrained) to the market before then.</strong><hr></blockquote>
Sorry, my mistake. I really was under the impression, from other sources, that the actual process from beginning to end in manufacturing a cpu took about 60 days. I mean that's a lot of itty bitty lines to etch and fill with Cu and to do it on what 6-9 layers of silicon.
<strong>Better: Why didn't Apple update the 15" PB, and why hasn't it now that the channel has been bone dry for some time?
How many 17" PB's have shipped? Over at MacNN, there are several increasingly furious threads about shipping delays.<hr></blockquote>
I think this is likely to have more to do with contractual obligations than with any CPU plans. As I recall, Apple recently switched (or added) a new LCD screen supplier and/or has a new laptop manufacturer, so existing component shortages due to the transition is the most likely culprit here.
[quote]As an idle thought, how many 1.42 GHz PowerMacs have actually shipped?</strong><hr></blockquote>
Not that many, they've only been shipping for a few months now and demand has been slack. If Apple is purposely reducing channel inventory they probably started weeks ago.
If.
<strong>If Apple is purposely reducing channel inventory they probably started weeks ago.</strong><hr></blockquote>
Reducing inventory of NEW machines shortly after announcing them...it's a sad time, currently. Oh the difference a few months will make!
<strong>I mean that's a lot of itty bitty lines to etch and fill with Cu and to do it on what 6-9 layers of silicon.</strong><hr></blockquote>
Well yeah, but they don't do them one line at a time!
Seriously though, I think the big problem is not so much the actual time it takes to do deposition and etch the traces, etc. but more in determining what the optimal time and technique to use at each stage is. Initially, it's an ongoing process that's constantly being tweaked, but once you hit a sweet spot, you pretty much nail it down and crank them out. And judging by the way they went from 1.8Ghz at the top end of the range (in October) to 1.8GHz at the bottom end (now), I'd say they've come pretty far in determining what that "sweet spot" is.
<strong>If a PPC970 @ 2.5GHz already exists (even if there are 2 or 3 of them) and things are really going well at IBM, what chance is there that PPC970 hits mass production earlier than in autumn? In other words, could one of moki's two surprises be an earlier-to-unexpected release date of our brand new PowerMac?
Could, indeed, the second surprise be Marklar? Consider that Mac OS X on Intel is not like VirtualPC for Mac. It would be times faster. And it would begin a next round of Apple vs. Microsoft box match.
What do you think?</strong><hr></blockquote>
As to the 970's production timetable, I believe that they stated that these will ship i the secnd half of the year. So the earliest we will see something from them is July 1st (the 1st day of the7th month) and the latest would be December 31st of 03.
Now pure speculation, IBM is now touting a tope end at introduction of 2.5 Ghz, since the top end speed of a chip is determined by testing the chips then I would think that they have some initial test units off of the production line to back up this statement, and are preparing to go into full production very soon. They probably need to build up an inventory before they start shipping them, and some final assembly time at Apple....say 90 days earliest till a shipping 970 hits the market. I think that it is very possable that Apple might have shipping units at Macworld this summer, and that would be a great time to introduce and demo a dual 970 PM running OS X 10.3 in a Maya bake-off.
<strong>Reducing inventory of NEW machines shortly after announcing them...it's a sad time, currently. Oh the difference a few months will make!</strong><hr></blockquote>
I said IF. It's just as likely (no, more likely) it's a blip in the supply chain and nothing to do with inventory reduction.
<strong>I said IF.</strong><hr></blockquote>
Oh I know; it's just the way I read and interpretted it the first time that struck me as funny that Apple would be clearing inventory when they just announced the 1.42s. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
<strong>There's been some talk about Apple incorporating the Bochs emulation (open-source Windows emulator for Unix) system into OS X, possibly as soon as 10.3. <a href="http://bochs.sourceforge.net/"; target="_blank">Bochs homepage</a>
That would be the REAL butt-kicker - if the 970+Bochs gives Apple native IA-32 emulation that is nearly as fast as the best x86 hardware, built right into the OS, who needs Marklar?</strong><hr></blockquote>
Unfortunately, Bochs is so amazingly slow that I think you may as well give up on that idea.
<strong>Well... You have the possibility to buy a cheap peecee and run OSX on it, or you will have a PIV crushing PPC 970 with OSX on it.. Consumers buy peecee's AND OSX, top-end d00d's buy PPC 970's with OSX... Apple's marketshare KABOOOMS. ==> Apple sells more macs ==> More marketshare ==> More Apple sales... -Get my point? </strong><hr></blockquote>
oh i think i get it now <img src="graemlins/surprised.gif" border="0" alt="[surprised]" /> ... "get this copy of the great Mac OS X Demo for PC for just $100 and look how well it runs on bad hardware!!! after that take a look at the new OS X on a PowerMac970 and be blown away!!!" hmmm ... ok if the windows-people finished playing windows they can try Mac OS X Demo... i don't care... <img src="graemlins/cancer.gif" border="0" alt="[cancer]" />
<strong>
Unfortunately, Bochs is so amazingly slow that I think you may as well give up on that idea.
The question that begs to be answered Mokis what and how long would it take to make it as robust or, preferably, more robust than VPC?
I find it ironic that VPC costs more than a PC now. Hell if you need a PC $300 will take care of that easily. I don't want Apple to spend one Iota of time ingrating that horrible codebase into OSX. I prefer OSX to ship "virus" free
<strong>
Unfortunately, Bochs is so amazingly slow that I think you may as well give up on that idea.
From what I gather, Bochs has been basically a one-man operation. How much of its sluggishness is just due to inefficient coding (no offense intended to Kevin Lawton, if you're out there)? I would imagine Apple engineers could massage his code to a fare-thee-well...
[ 02-28-2003: Message edited by: TJM ]</p>
[quote]Originally posted by MartianMatt:
<strong>My point is, the 970 will definitely be competetive with the Wintel world but has yet to prove it's trample-all-over-ness.
</strong><hr></blockquote>
SpecFP estimates:
[code]
1x G4+ 1.0GHz 187
1x G4+ 1.4GHz 261
2x G4+ 1.4GHz 523
1x P4 3.4GHz 1249
1x 970 1.8GHz 1051
1x 970 2.5GHz 1459
2x 970 2.5GHz 2918
4x 970 2.5GHz 5839
</pre><hr></blockquote>
The multi-CPU P4's & Xeons have heat issues & cost through the nose. (Yes, there are things more expensive than Macs.)
That's not the main thing though.
All you need to do is read the line "2xG4" and the line "2x 970"
5.57x as fast. Trample-all-overness.
And the thing that's been the big hurdle historically at Apple to Quads is the _bus_. With the bus fixed, a Quad-ppc can easily undercut a Quad-Xeon in price... Which means the '4x 970' line might be interesting.
Having the bus fixed should also be a HUGE benefit to AltiVec, which isn't reflected (at all) in the Spec numbers. No Mac-types were too concerned when 'pc Hz' was around 1.3-1.5x Mac Hz - AV quite effective for what it does.
<strong>
Unfortunately, Bochs is so amazingly slow that I think you may as well give up on that idea.
A separate idea: Would it be possible that Apple has their own IA-32 emulation environment in the works (i.e. not Bochs OR VPC)?