Between you and Moki I feel like I'm sitting in the lobby of a theater while my friends poke their heads out of the auditorium every few minutes so they can tell me over and over that I'm missing a really fantastic show and how it's too bad I couldn't get a ticket.
I guess I'll just have to wait until Christmas morning to unwrap my present from Uncle Steve. I'm going to have to start ignoring all the rumors and speculation. My blood pressure can't handle it. <img src="graemlins/hmmm.gif" border="0" alt="[Hmmm]" />
. . . In previous posts here and elsewhere, the time to manufacture cpu's has been mentioned to be about 60 days, from start to finish.
Does any one know if this has been significantly reduced?
</strong><hr></blockquote>
I'll just comment that it is easy to get the wrong idea about the production process. IBM may have a large inventory of 970 chips that are close to completion, but waiting for demand before they finish them. It is not an assembly line, where raw materials goes in, and 60 days later a finished processor pops out. I have no idea whether 60 days is correct, but it could be.
Chips start out on wafers, which contain many chips each. These go through steps of depositing the several layers of an integrated circuit. A photographic process places an image for the first layer on the wafer to start. These wafers then go through a furnace, where gases implant the correct chemical impurities into the silicon for that layer. The photo-resist image keeps the gases from contacting parts of the chips that do not get this impurity. This process is repeated for every layer of the chip, and the final process lays down the surface electrical interconnections. So the chips on the wafers get built up in stages, which take time.
To evaluate a new chip, it is necessary to start several runs of wafers, to take into account the many variation that can occur in the process of building up the chips. Now, at the end of several such runs, IBM could have a whole bunch of chips. Many chips on a wafer, and several wafers per run, and enough runs to feel confident in the yield and performance. Next, they test chips on the wafers, using probes. Bad chips are marked. If the runs all turn out well, there are a lot of good chips on these wafers, and they can be stored until needed. Some chips are cut out of their wafers and packaged for complete testing.
Even though the process takes a long time, IBM could have a lot of chips on wafers, just waiting to be diced up and packaged. They do not take up much space in this form.
As far as shipping systems go, how soon before actual shipping systems would a software company get hold of hardware to test before the systems ship?
I know a person who is working for one software company that has new Apple systems ("G5"). It is very hush, hush and they built a special room dedicated to test on the hardware and to keep prying eyes out.
I unfortunately don't know how long they have been testing, but it sounds like they have for at least a few weeks to a couple of months.
I'll just comment that it is easy to get the wrong idea about the production process. IBM may have a large inventory of 970 chips that are close to completion, but waiting for demand before they finish them. It is not an assembly line, where raw materials goes in, and 60 days later a finished processor pops out. I have no idea whether 60 days is correct, but it could be.
Chips start out on wafers, which contain many chips each. These go through steps of depositing the several layers of an integrated circuit. A photographic process places an image for the first layer on the wafer to start. These wafers then go through a furnace, where gases implant the correct chemical impurities into the silicon for that layer. The photo-resist image keeps the gases from contacting parts of the chips that do not get this impurity. This process is repeated for every layer of the chip, and the final process lays down the surface electrical interconnections. So the chips on the wafers get built up in stages, which take time.
To evaluate a new chip, it is necessary to start several runs of wafers, to take into account the many variation that can occur in the process of building up the chips. Now, at the end of several such runs, IBM could have a whole bunch of chips. Many chips on a wafer, and several wafers per run, and enough runs to feel confident in the yield and performance. Next, they test chips on the wafers, using probes. Bad chips are marked. If the runs all turn out well, there are a lot of good chips on these wafers, and they can be stored until needed. Some chips are cut out of their wafers and packaged for complete testing.
Even though the process takes a long time, IBM could have a lot of chips on wafers, just waiting to be diced up and packaged. They do not take up much space in this form.</strong><hr></blockquote>
<strong>First, QE is just openGL, there's no need to aquire Raycer to implement this clever idea.
Second, if we don't see fruits of the Raycer aquisition by the time the first 970-based powermacs emerge, then it is my feeling we will never see it.
The raycer goodness, from the sound of things, had to do with greatly speeding up 3d functions, not mapping 2d windows on a 3d z-ordered desktop.</strong><hr></blockquote>
The HD on the photo was manufactured (or, more precisely, time-stamped) on July, 25. So the photo is relatively new. Can we say the same about the motherboard? Does anybody recognize the chip set?
The more I look at the MB and the more I read Ars Forums, the more I suspect that this home-brewed bottom-end mazaboard doesn't have a PPC970 on it. This, by the way, may be the reason for IBM to pull this photo from any website, in order to save geeks' lives in the peril of laughing death.
i admit, there´s something fishy about that muthaboard. i looked and flipped and looked for several times now... but i´m not sure if i can see what *certain notorious teasers* on these boards are coughing at. i mean, it´s pretty obvious that the harddrive somehow doesn´t belong there. it could even be simply doctored, but for what purpose? to demonstrate size proportions? <img src="confused.gif" border="0">
[quote]Originally posted by Transcendental Octothorpe:
<strong>
I'm not so sure.
</strong><hr></blockquote>
Given how Quartz is currently handled (and lack of Quartz snappiness in comparison with MS Windows), it would make a lot of sense to offload the drawing load to a custom external hardware coprocessor.
<strong>i admit, there´s something fishy about that muthaboard. i looked and flipped and looked for several times now... but i´m not sure if i can see what *certain notorious teasers* on these boards are coughing at. i mean, it´s pretty obvious that the harddrive somehow doesn´t belong there. it could even be simply doctored, but for what purpose? to demonstrate size proportions? </strong><hr></blockquote>
...no, the Travelstar is supposed to be there. There is a connector that rises off the edge of the mobo. I'm not sure if thats how IBM does it with all there blades, but blades are pretty damn thin and being that I would think they do most of their work in memory, HD speed isn't that crucial, or if they do need data they get it on a rack mounted RAID or sumpin.
No, the only thing that stumps me is the PCI slots. An Xserve has two PCI slots. Is it possible they might share mobos or something.
<strong>i admit, there´s something fishy about that muthaboard. i looked and flipped and looked for several times now... but i´m not sure if i can see what *certain notorious teasers* on these boards are coughing at. i mean, it´s pretty obvious that the harddrive somehow doesn´t belong there. it could even be simply doctored, but for what purpose? to demonstrate size proportions? </strong><hr></blockquote>
I spent some time nosing around the IBM site and found an image of their dual Xeon blade board. The drive is actually legit - there's a connector to the board above it - like the one to its right. The right edge is identical, with twin Gb ethernet connectors, so I'm convinced this is a legitimate blade board. The PCI slots are different, as is the entire center-to-left end of the board. The Xeon board has four banks of DDR RAM, oriented at right angles to the one SDR bank shown on the prototype. The heat-sinks are smooth and gold-colored, and there are several chips present on the Xeon board this proto lacks. I wish I knew enough to be able to analyze all this.
The only things I could figure out that looked a bit odd (only because I haven't seen them on "ordinary" MBs - they may be standard on these blades) were a couple of what appeared to be connectors on the lower center on the board: one labeled HD(maybe an 8? a B?) Program Header and another labeled VPO Header. Don't know if they're significant or not.
Yeah, IBM pulled the press release... seems someone wasn't happy with that information getting out... nor the fairly hires screenshot of the blade server. Hint: if you still have it, try looking at it very closely.</strong><hr></blockquote>
moki
Thank you for the response. Fortunately, I did save the picture of the blade server. Regretably, I could stare at it until the next millenium and not know what I'm looking at. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
I'm truely a technological dunderhead and stay in a constant state of <img src="confused.gif" border="0">
I'll just comment that it is easy to get the wrong idea about the production process. IBM may have a large inventory of 970 chips that are close to completion, but waiting for demand before they finish them....
Even though the process takes a long time, IBM could have a lot of chips on wafers, just waiting to be diced up and packaged. They do not take up much space in this form.</strong><hr></blockquote>
snoopy
Thank you for the information, I'm always interested in learning.
1. The blades are thin, right? The PCI slots are vertical. I haven't seen a PCI card that would fit vertically into the case. Look at the RAM slot. It's skewed because it would not fit in the upright position (if I'm missing the real reason, please, correct me). So they want us to believe there are PCI cards smaller than a PC133.
2. A rackmount server with the only DIMM slot is crap.
Yeah, IBM pulled the press release... seems someone wasn't happy with that information getting out... nor the fairly hires screenshot of the blade server. Hint: if you still have it, try looking at it very closely.</strong><hr></blockquote>
I'VE GOT IT!!!!
It's not a blade server at all!!!
It's actually a controller for a Fusion Pulse Cannon off of an Auroran Cruiser! Those fiends at IBM are in league with the Aurorans!
Come now people. Has nobody ever seen a IIsi or Quadra 610, PowerMac 6100 on the inside? It would not be hard to imagine an "L" shaped adapter so you can mount a PCI card parallel to the board.
<strong>Come now people. Has nobody ever seen a IIsi or Quadra 610, PowerMac 6100 on the inside? It would not be hard to imagine an "L" shaped adapter so you can mount a PCI card parallel to the board.</strong><hr></blockquote>
When you talk L-shaped PCI adapters, you certainly don't mean $30k rackmounts, right? As far as I can remember, PowerMac 6100 was not used in server configurations, either.
What I wanted to say is this thing was not supposed to appear in a 1U block in this form. I bet it existed only for testing purposes and, I'm afraid, it's 3 years old.
Comments
<strong>
I'm not so sure.
[ 03-02-2003: Message edited by: Transcendental Octothorpe ]</strong><hr></blockquote>
Sigh...
Between you and Moki I feel like I'm sitting in the lobby of a theater while my friends poke their heads out of the auditorium every few minutes so they can tell me over and over that I'm missing a really fantastic show and how it's too bad I couldn't get a ticket.
I guess I'll just have to wait until Christmas morning to unwrap my present from Uncle Steve. I'm going to have to start ignoring all the rumors and speculation. My blood pressure can't handle it. <img src="graemlins/hmmm.gif" border="0" alt="[Hmmm]" />
Second, if we don't see fruits of the Raycer aquisition by the time the first 970-based powermacs emerge, then it is my feeling we will never see it.
The raycer goodness, from the sound of things, had to do with greatly speeding up 3d functions, not mapping 2d windows on a 3d z-ordered desktop.
<strong>
. . . In previous posts here and elsewhere, the time to manufacture cpu's has been mentioned to be about 60 days, from start to finish.
Does any one know if this has been significantly reduced?
</strong><hr></blockquote>
I'll just comment that it is easy to get the wrong idea about the production process. IBM may have a large inventory of 970 chips that are close to completion, but waiting for demand before they finish them. It is not an assembly line, where raw materials goes in, and 60 days later a finished processor pops out. I have no idea whether 60 days is correct, but it could be.
Chips start out on wafers, which contain many chips each. These go through steps of depositing the several layers of an integrated circuit. A photographic process places an image for the first layer on the wafer to start. These wafers then go through a furnace, where gases implant the correct chemical impurities into the silicon for that layer. The photo-resist image keeps the gases from contacting parts of the chips that do not get this impurity. This process is repeated for every layer of the chip, and the final process lays down the surface electrical interconnections. So the chips on the wafers get built up in stages, which take time.
To evaluate a new chip, it is necessary to start several runs of wafers, to take into account the many variation that can occur in the process of building up the chips. Now, at the end of several such runs, IBM could have a whole bunch of chips. Many chips on a wafer, and several wafers per run, and enough runs to feel confident in the yield and performance. Next, they test chips on the wafers, using probes. Bad chips are marked. If the runs all turn out well, there are a lot of good chips on these wafers, and they can be stored until needed. Some chips are cut out of their wafers and packaged for complete testing.
Even though the process takes a long time, IBM could have a lot of chips on wafers, just waiting to be diced up and packaged. They do not take up much space in this form.
I know a person who is working for one software company that has new Apple systems ("G5"). It is very hush, hush and they built a special room dedicated to test on the hardware and to keep prying eyes out.
I unfortunately don't know how long they have been testing, but it sounds like they have for at least a few weeks to a couple of months.
-tink
[ 03-03-2003: Message edited by: tink ]</p>
<strong>
I'll just comment that it is easy to get the wrong idea about the production process. IBM may have a large inventory of 970 chips that are close to completion, but waiting for demand before they finish them. It is not an assembly line, where raw materials goes in, and 60 days later a finished processor pops out. I have no idea whether 60 days is correct, but it could be.
Chips start out on wafers, which contain many chips each. These go through steps of depositing the several layers of an integrated circuit. A photographic process places an image for the first layer on the wafer to start. These wafers then go through a furnace, where gases implant the correct chemical impurities into the silicon for that layer. The photo-resist image keeps the gases from contacting parts of the chips that do not get this impurity. This process is repeated for every layer of the chip, and the final process lays down the surface electrical interconnections. So the chips on the wafers get built up in stages, which take time.
To evaluate a new chip, it is necessary to start several runs of wafers, to take into account the many variation that can occur in the process of building up the chips. Now, at the end of several such runs, IBM could have a whole bunch of chips. Many chips on a wafer, and several wafers per run, and enough runs to feel confident in the yield and performance. Next, they test chips on the wafers, using probes. Bad chips are marked. If the runs all turn out well, there are a lot of good chips on these wafers, and they can be stored until needed. Some chips are cut out of their wafers and packaged for complete testing.
Even though the process takes a long time, IBM could have a lot of chips on wafers, just waiting to be diced up and packaged. They do not take up much space in this form.</strong><hr></blockquote>
Thank you, snoopy.
<strong>First, QE is just openGL, there's no need to aquire Raycer to implement this clever idea.
Second, if we don't see fruits of the Raycer aquisition by the time the first 970-based powermacs emerge, then it is my feeling we will never see it.
The raycer goodness, from the sound of things, had to do with greatly speeding up 3d functions, not mapping 2d windows on a 3d z-ordered desktop.</strong><hr></blockquote>
QE is not just OpenGL
All modern graphics cards support hardware acceleration of Opengl, thus the 2d desktop is hardware accelerated.
If you know otherwise, please post.
<strong>
I'm not so sure.
</strong><hr></blockquote>
Given how Quartz is currently handled (and lack of Quartz snappiness in comparison with MS Windows), it would make a lot of sense to offload the drawing load to a custom external hardware coprocessor.
What if it came out as
[CPU]-->[CUSTOM-QUARTZ-GPU](drawing)->AGP->[VIDEOCARD](compositing)?
[ 03-03-2003: Message edited by: ag ]</p>
<strong>i admit, there´s something fishy about that muthaboard. i looked and flipped and looked for several times now... but i´m not sure if i can see what *certain notorious teasers* on these boards are coughing at. i mean, it´s pretty obvious that the harddrive somehow doesn´t belong there. it could even be simply doctored, but for what purpose? to demonstrate size proportions?
...no, the Travelstar is supposed to be there. There is a connector that rises off the edge of the mobo. I'm not sure if thats how IBM does it with all there blades, but blades are pretty damn thin and being that I would think they do most of their work in memory, HD speed isn't that crucial, or if they do need data they get it on a rack mounted RAID or sumpin.
No, the only thing that stumps me is the PCI slots. An Xserve has two PCI slots. Is it possible they might share mobos or something.
<strong>i admit, there´s something fishy about that muthaboard. i looked and flipped and looked for several times now... but i´m not sure if i can see what *certain notorious teasers* on these boards are coughing at. i mean, it´s pretty obvious that the harddrive somehow doesn´t belong there. it could even be simply doctored, but for what purpose? to demonstrate size proportions?
I spent some time nosing around the IBM site and found an image of their dual Xeon blade board. The drive is actually legit - there's a connector to the board above it - like the one to its right. The right edge is identical, with twin Gb ethernet connectors, so I'm convinced this is a legitimate blade board. The PCI slots are different, as is the entire center-to-left end of the board. The Xeon board has four banks of DDR RAM, oriented at right angles to the one SDR bank shown on the prototype. The heat-sinks are smooth and gold-colored, and there are several chips present on the Xeon board this proto lacks. I wish I knew enough to be able to analyze all this.
The only things I could figure out that looked a bit odd (only because I haven't seen them on "ordinary" MBs - they may be standard on these blades) were a couple of what appeared to be connectors on the lower center on the board: one labeled HD(maybe an 8? a B?) Program Header and another labeled VPO Header. Don't know if they're significant or not.
That's all I can contribute, unfortunately.
[ 03-03-2003: Message edited by: TJM ]</p>
<strong>
Yeah, IBM pulled the press release... seems someone wasn't happy with that information getting out... nor the fairly hires screenshot of the blade server. Hint: if you still have it, try looking at it very closely.</strong><hr></blockquote>
moki
Thank you for the response. Fortunately, I did save the picture of the blade server. Regretably, I could stare at it until the next millenium and not know what I'm looking at. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
I'm truely a technological dunderhead and stay in a constant state of <img src="confused.gif" border="0">
<strong>
I'll just comment that it is easy to get the wrong idea about the production process. IBM may have a large inventory of 970 chips that are close to completion, but waiting for demand before they finish them....
Even though the process takes a long time, IBM could have a lot of chips on wafers, just waiting to be diced up and packaged. They do not take up much space in this form.</strong><hr></blockquote>
snoopy
Thank you for the information, I'm always interested in learning.
1. The blades are thin, right? The PCI slots are vertical. I haven't seen a PCI card that would fit vertically into the case. Look at the RAM slot. It's skewed because it would not fit in the upright position (if I'm missing the real reason, please, correct me). So they want us to believe there are PCI cards smaller than a PC133.
2. A rackmount server with the only DIMM slot is crap.
<strong>
Yeah, IBM pulled the press release... seems someone wasn't happy with that information getting out... nor the fairly hires screenshot of the blade server. Hint: if you still have it, try looking at it very closely.</strong><hr></blockquote>
I'VE GOT IT!!!!
It's not a blade server at all!!!
It's actually a controller for a Fusion Pulse Cannon off of an Auroran Cruiser! Those fiends at IBM are in league with the Aurorans!
[ 03-03-2003: Message edited by: TJM ]</p>
<strong>Come now people. Has nobody ever seen a IIsi or Quadra 610, PowerMac 6100 on the inside? It would not be hard to imagine an "L" shaped adapter so you can mount a PCI card parallel to the board.</strong><hr></blockquote>
When you talk L-shaped PCI adapters, you certainly don't mean $30k rackmounts, right? As far as I can remember, PowerMac 6100 was not used in server configurations, either.
What I wanted to say is this thing was not supposed to appear in a 1U block in this form. I bet it existed only for testing purposes and, I'm afraid, it's 3 years old.