[olePigeon]

Well, maybe not the Cell specifically, but an architecture very similar. It looks like AMD is moving to a Cell-like philosophy for their CPUs. I remember Intel commenting about Cell being a glimpse into how microprocessors will be organized in the future.

A while back I broke my Apple Prediction streak by saying that they'd be moving to the Cell architecture, but then Apple moved to Intel. I think it threw a lot of people off. However, I might be able to salvage a little bit of that prediction.

IBM, Sony, and Toshiba have the right idea about the future of microprocessors. You can get more out of clusters of specific processor units that can each be ordered to do a special task more efficiently than a single, general processor unit (even with multiple cores.) Cell has the PPE (CPU) with SPE coprocessors, AMD will have a CPU with xPU coprocessors.

Instead of creating an entirely new architecture like IBM, Sony, and Toshiba did, AMD will be bridging existing technologies by placing the CPU, GPU, and APU on a single die, then move to the more Cell-like layout.

I think Intel will move in the same direction, and we'll have Macs using architectures similar in design to the Cell anyway.

[Dark Helmet]

Apple is doing better than ever with the current Intel chips and nobody is complaining.

Screw IBM, they had their chance and they screwed it up worse than Motorola.

[goMac]

AMD's vision is not like the Cell. It's actually the anti-cell. Move processing off the CPU onto other processing unit's built for doing specialized tasks. IBM's Cell is trying to move all processing onto the CPU into processing units built for very general purposes. All 3 ideas, the Cell, multicored processing, and AMD's new idea just share the idea that they will parallelize processing and reduce the amount of heat processors put out.

Apple will never use the Cell. It simply sucks for computers, and is unusable. It has no out of order processing, and only single precision floating point support.

Edit:

As for AMD's assertion that the Core wars were a mistake, it's dead wrong. Multiple cores is the way forward. Increasing mhz exponentially increases the heat output and energy requirements of a chip. Adding more cores only increases those requirements linearly.

And the most interesting part of this all is AMD's plan still involves multiple cores. It is just very different than the Cell, more like the way ATI has seen the idea behind their Crossfire. You install more Cores that have been optimized for specific tasks when you want to add a feature to your machine.

[Big Mac]

Originally Posted by Dark Helmet 

Apple is doing better than ever with the current Intel chips and nobody is complaining.

Screw IBM, they had their chance and they screwed it up worse than Motorola.

I'm complaining, but I also must concede the sales figures. As far as IBM screwing up worse than Motorola, I don't think IBM left us at 500MHz for 3 years like Motorola did.

[andreas_g4]

Originally Posted by Dark Helmet 

Screw IBM, they had their chance and they screwed it up worse than Motorola.

So true, but everyone should stay open for what the future brings. As long as Apple is able do deliver systems which work as reliable as the current intel-driven machines, I don't care what CPU is in them.

[olePigeon]

I wasn't saying that Apple's leaving Intel or anything, I'm just saying that having a central CPU doing everything is going out the window.

You're right about IBM's being more general CPUs, but they're kinda like Transmeta's chips where they can be designated to do a specific task even if the CPU is more general.

AMD is going to be more specific. Instead of having general CPUs targeted for a specific task, they're just going to be specialized CPUs for doing only one specific type of task.

It's hard to say which design would be better. Having the option of making those coprocessors do whatever you want them to do sure seems appealing. A couple of my dad's buddies were saying that the Cell is actually a really cool set to work with for a programmer if you're doing specialized computations (like high-end physics stuff.)

[andreas_g4]

Originally Posted by Big Mac 

I'm complaining, but I also must concede the sales figures. As far as IBM screwing up worse than Motorola, I don't think IBM left us at 500MHz for 3 years like Motorola did.

No, they left us without a mobile and yet capable chip since the G5's introduction…

[Gossamer]

Originally Posted by Big Mac 

I'm complaining, but I also must concede the sales figures. As far as IBM screwing up worse than Motorola, I don't think IBM left us at 500MHz for 3 years like Motorola did.

But they did fail to get us 3GHz.

[Dark Helmet]

Originally Posted by Big Mac 

I'm complaining, but I also must concede the sales figures. As far as IBM screwing up worse than Motorola, I don't think IBM left us at 500MHz for 3 years like Motorola did.

No 3GHz in 12 months... heck even 24 months later.

No portable G5's.

9 fans and liquid cooling.. are they nuts?

[olePigeon]

Originally Posted by goMac 

And the most interesting part of this all is AMD's plan still involves multiple cores. It is just very different than the Cell, more like the way ATI has seen the idea behind their Crossfire. You install more Cores that have been optimized for specific tasks when you want to add a feature to your machine.

They're different approaches to the same conclusion.

The differences in their implementation is kinda like the Mac vs. PC argument:

STI (Mac): Gives you everything you could need, then you decide what you want.
AMD (PC): Lets you choose what you want, then gives you what you need.

AMDs approach will certainly be a lot cheaper and a much better route for home PCs. STI's (the Sony/Toshiba/IBM alliance) will be a lot better for high-end, scalable servers.

We'll see how it pans out.

[goMac]

Originally Posted by olePigeon 

It's hard to say which design would be better. Having the option of making those coprocessors do whatever you want them to do sure seems appealing.

But that's kinda where we are now. You can do games on the CPU, but having a GPU makes things nicer. You can do physics on the CPU, but a PPU makes physics run better. You can run movies off the CPU, but a GPU with an accelerator for movies helps out. It makes it so you don't need specialized components on your machine to do things, but enthusiasts who want better performance can add those things. And honestly, I think some things, like GPU's, are necessary.

However, I think AMD's new "scheme" amounts to nothing more than a greedy company who has suddenly found itself behind in the CPU race and needs to come up with a new idea quickly. Right now, if your GPU isn't satisfactory, you can upgrade it. If your graphics card doesn't support the latest MPEG4, that is fine because usually you can play it off your CPU.

But what AMD is talking about is crippling the CPU's power in favor of pushing that power out to specialized sub processors (which they see as fixing the heat and power problems or a traditional CPU). In other words, what happens when MPEG5 comes out and your MPEG4 processor can't play it? You have to buy an entirely new processor that supports MPEG5 because AMD will cripple the processor and it won't have enough general use power to play an MPEG5. What happens when you need to support DirectX 11 or OpenGL 3? You buy a new processor. And if graphics are the only performance issue you are having? Tough. You need to replace your entire processor anyway.

The entire plan sounds like a way for AMD to grab as much money from you as they can.

[olePigeon]

The Cell is a little more different than general CPUs with multiple cores. With the Cell, anything from an individual thread, to an instruction, to an entire program can be managed and designated to a specific SPE coprocessor. While they're generalized, the data can be delegated a lot more efficiently than a standard multi-core CPU.

How about a mix of the two with some of Sun's ingenuity, maybe Intel will do this:

Keep the high-bandiwth interlinking like the Cell, but use AMD's approach on specialized CPUs, but keep it all socketable and interchangeable.

You'd have an array of, say, 8 sockets. You could buy 2 CPUs, 1 APU, 3 GPUs, and 3 PPUs. If you want more realistic physics but don't care about the graphics, you could do 2 CPUs, 5 PPUs, 1 GPU, and 1 APU. If you don't care about sound at all (you only have stereo speakers or headphones) you can do away with the APU completely and stick in 3 CPUs to speed up load times between levels on a game.

So your end result would be a motherboard or some such with an array of 8 sockets; composed of 3 CPUs for fast load times, 1 GPU cuz you don't care about the graphics, 5 PPUs because you want the physics as realistic as possible, and no APUs cuz you don't have 7.1 surround or whatever and you'll just offload the stereo to that extra CPU (which results in an overall net performance increase for load times for an application.)

I think that'd be killer in customizability. You'd have the specialized CPUs like AMD wants, but you can stick them in any number of combinations in an array like Sun does with their high-end scalable SPARC servers, and have endless options of features like STI's implementation. They'd all be socketable, interchangeable (doesn't matter what chip you stick in what socket, as long as at least 1 of them is a CPU), and expandable.

The race here would be motherboards that offer more sockets for whatever chips you want to stick in.

THAT would be awesome.

[goMac]

Originally Posted by olePigeon 

The Cell is a little more different than general CPUs with multiple cores. With the Cell, anything from an individual thread, to an instruction, to an entire program can be managed and designated to a specific SPE coprocessor. While they're generalized, the data can be delegated a lot more efficiently than a standard multi-core CPU.

Right. My argument about the Cell has never been that it doesn't have it's uses. It's just that I don't think the Cell is a good fit for gaming or general computing.

Originally Posted by olePigeon 

How about a mix of the two with some of Sun's ingenuity, maybe Intel will do this:
Keep the high-bandiwth interlinking like the Cell, but use AMD's approach on specialized CPUs, but keep it all socketable and interchangeable.

Isn't this what we have now with current chipsets? I mean, isn't this basically what a PCI slot is?

Originally Posted by olePigeon 

You'd have an array of, say, 8 sockets. You could buy 2 CPUs, 1 APU, 3 GPUs, and 3 PPUs. If you want more realistic physics but don't care about the graphics, you could do 2 CPUs, 5 PPUs, 1 GPU, and 1 APU. If you don't care about sound at all (you only have stereo speakers or headphones) you can do away with the APU completely and stick in 3 CPUs to speed up load times between levels on a game.

But again, this is basically the same system we have now. Why does it need to be changed? All the sound processors, graphics processors, physics processors, are all connected back to the CPU by a PCI Express Bus. Yeah, it's not as lightning fast as having something more direct, but then again, the GPU's we have now are not exactly fast enough to be pushing past the restrictions of PCI Express.

Originally Posted by olePigeon 

So your end result would be a motherboard or some such with an array of 8 sockets; composed of 3 CPUs for fast load times, 1 GPU cuz you don't care about the graphics, 5 PPUs because you want the physics as realistic as possible, and no APUs cuz you don't have 7.1 surround or whatever and you'll just offload the stereo to that extra CPU (which results in an overall net performance increase for load times for an application.)

I think that'd be killer in customizability. You'd have the specialized CPUs like AMD wants, but you can stick them in any number of combinations in an array like Sun does with their high-end scalable SPARC servers, and have endless options of features like STI's implementation. They'd all be socketable, interchangeable (doesn't matter what chip you stick in what socket, as long as at least 1 of them is a CPU), and expandable.

The race here would be motherboards that offer more sockets for whatever chips you want to stick in.

THAT would be awesome.

Again, it's the same thing we have now. Your graphics card has a specialized processor on it, so does your sound card, physics card, etc, and they are all already linked back to the CPU. This is what I'm saying. The system we have now isn't broken. AMD just wants to move everything onto one part so when one co-processor doesn't meet our needs we have to upgrade the whole processor.

[olePigeon]

Originally Posted by goMac 

Isn't this what we have now with current chipsets? I mean, isn't this basically what a PCI slot is?

...

Again, it's the same thing we have now. Your graphics card has a specialized processor on it, so does your sound card, physics card, etc, and they are all already linked back to the CPU. This is what I'm saying. The system we have now isn't broken. AMD just wants to move everything onto one part so when one co-processor doesn't meet our needs we have to upgrade the whole processor.

I think you're right in the sense that the independence of each CPU is there, but they're completely different in the implementation.

AMD is right in that there are lot better ways of getting the CPU, GPU, PPU, and APU to talk to each other. I agree with you that AMD's approach is faulty if they're going to stick them all on the same die. That'd suck because, as you mentioned, you can't upgrade them (or at least not upgrade individual parts you want to as you can now.)

My idea is that you'd take all those components off the PCI bus and put them on an array similar to a Cell (but socketable) so those components talk directly to each other, instead of through some sort of BUS management system.

Like the Cell, each of the xPUs (CPU, GPU, PPU, APU) would be scalable and talk to each other in a grid-computing fashion. Unlike the Cell, those xPUs really would be CPUs, GPUs, PPUs, and APUs. So instead of having an array of general CPUs like the Cell, you could have any mix of CPUs, GPUs, PPUs, and APUs in that array that would all talk directly to each other and expand if there are any other like-xPUs in the array.

So you'd get the high-bandwith interconnect between the xPUs like AMD wants, and how the Cell already has (but using more generalized coprocessors), but expandable and customizable like we have now.

(I think I need to draw a diagram to get my point across.)

[goMac]

Originally Posted by olePigeon 

My idea is that you'd take all those components off the PCI bus and put them on an array similar to a Cell (but socketable) so those components talk directly to each other, instead of through some sort of BUS management system.

But again, it's not exactly like PCI is broken. PCI still has more than enough speed to do what we need today, and still has tons of room to grow. And generally, to make things just plain work better, you're going to need some sort of bridge in between the PCI devices at the CPU to do a nice job of abstracting things.

Originally Posted by olePigeon 

Like the Cell, each of the xPUs (CPU, GPU, PPU, APU) would be scalable and talk to each other in a grid-computing fashion. Unlike the Cell, those xPUs really would be CPUs, GPUs, PPUs, and APUs. So instead of having an array of general CPUs like the Cell, you could have any mix of CPUs, GPUs, PPUs, and APUs in that array that would all talk directly to each other and expand if there are any other like-xPUs in the array.

Again, I'm not sure why this design would be adventagious. Having a lot of different devices from a lot of different vendors talking directly to each other without any restrictions is usually asking from trouble (See Microsoft Windows 95 as an example of this in software land). And again, I'm not sure why this is necessary. GPU's, CPU's, APU's, and PPU's are already scalable. This is done by the chipset, which I'm sure would still have to be present in such a design like yours, just moved onto the CPU.

Originally Posted by olePigeon 

So you'd get the high-bandwith interconnect between the xPUs like AMD wants, and how the Cell already has (but using more generalized coprocessors), but expandable and customizable like we have now.

But we have no need in the forseeable future for such a high bandwidth system. I don't want to say 640k is enough for anybody, but I don't think a move to such a system would be worth it when bus technology such as PCI still has a lot of room left to grow. A system like yours still have to have a set bus speed (processors have to have a designated speed they will talk to each other at), and will probably still have to have some sort of chipset to manage these added processors, bringing us back to where we are now with PCI.

Don't get me wrong, I'm not saying your idea isn't drool worthy, I completely agree with how you are saying things should be. My problem is that it's similar enough to PCI I just don't see any reason to shift to a system like yours.

This is also why AMD would want to move everything onto one die. By moving everything onto the CPU it allows them to force specifications and do away with the chipset and the bus system, or worry about bus speeds, which is one convenience I don't think computing should go without.

[olePigeon]

This is a thread about the future, not what we have now, but here's my response:

Originally Posted by goMac 

Again, I'm not sure why this design would be adventagious. Having a lot of different devices from a lot of different vendors talking directly to each other without any restrictions is usually asking from trouble (See Microsoft Windows 95 as an example of this in software land).

Even now you choose between AMD or Intel for your motherboard and chipsets. It'd be no different in the future. Regardless if AMD or Intel makes the motherboard, they both talk to the GPU or sound card the same. This wouldn't change. What would change is the option of how many of which kind you want loaded onto it by just purchasing the designated xPU. Each additional xPU would complement and increase processing capabilities of other like-xPUs. You wouldn't have to worry about the bus type or its speed as they'd talk directly with each other.

The PCI slots would still be there, but they'd be used for SCSI cards, network cards, audio/video jacks, etc.

Originally Posted by goMac 

And again, I'm not sure why this is necessary. GPU's, CPU's, APU's, and PPU's are already scalable. This is done by the chipset, which I'm sure would still have to be present in such a design like yours, just moved onto the CPU.

You can't just stick in an additional Audio card and expect it to make encoding sound any faster or upgrade your stereo to 5.1 by adding 3 more jacks. You have to buy a whole new audio card Regardless if you even have 5 ports for audio (or 7), that doesn't mean you can now do 7 uncompressed signals through them, nor does adding another sound card enable you to handle higher quality signals.

GPUs are getting to the point where you can stick in multiple video cards, but you have to sacrifice 4 PCI slots to do it. They're huge and bulky.

PPUs are almost unheard of.

Originally Posted by goMac 

But we have no need in the forseeable future for such a high bandwidth system. I don't want to say 640k is enough for anybody, but I don't think a move to such a system would be worth it when bus technology such as PCI still has a lot of room left to grow. A system like yours still have to have a set bus speed (processors have to have a designated speed they will talk to each other at), and will probably still have to have some sort of chipset to manage these added processors, bringing us back to where we are now with PCI.

This is where I think we're getting confused. This wouldn't be a basic BUS system like PCI, it would be more like Cell. Imagine, instead, that you have a big processor like AMD is proposing, but instead of having those features permanent, they're left open with sockets (obviously, it'd be more spread out.) The end user gets to decide which features will be "on" the CPU, whether it's additional CPUs or if they want to add GPUs, PPUs, or APUs.

Originally Posted by goMac 

Don't get me wrong, I'm not saying your idea isn't drool worthy, I completely agree with how you are saying things should be. My problem is that it's similar enough to PCI I just don't see any reason to shift to a system like yours.

You're right, but this is about the future. I also agree that AMD is jumping the gun. Maybe they'll take the fall for it so that Intel can do it properly later on. People are still adjusting to PCI Express, there's no need to make everyone jump hurdles for a new architecture.

Originally Posted by goMac 

This is also why AMD would want to move everything onto one die. By moving everything onto the CPU it allows them to force specifications and do away with the chipset and the bus system, or worry about bus speeds, which is one convenience I don't think computing should go without.

That would be the cheaper way to do it. That's partly why Cell is so expensive. My idea would probably be the most expensive of the ideas. The highspeed interconnect of the Cell, but the specialization of AMD.

[goMac]

Originally Posted by olePigeon 

Even now you choose between AMD or Intel for your motherboard and chipsets. It'd be no different in the future. Regardless if AMD or Intel makes the motherboard, they both talk to the GPU or sound card the same. This wouldn't change. What would change is the option of how many of which kind you want loaded onto it by just purchasing the designated xPU. Each additional xPU would complement and increase processing capabilities of other like-xPUs. You wouldn't have to worry about the bus type or its speed as they'd talk directly with each other.

But you have other issues. You couldn't mix components from different suppliers. Once you bought an AMD CPU, you are locked into an AMD compliant sound card. In addition, if you're going to allow for the sort of scalability and modularity you're suggesting, you have to reintroduce the chipset in somewhere to manage that.

Originally Posted by olePigeon 

The PCI slots would still be there, but they'd be used for SCSI cards, network cards, audio/video jacks, etc.

Right, and my guess is in order to keep their systems unlocked from one system, people would use the most open standard they can, which would be PCI. My guess is people would continue to buy PCI graphics cards. I just don't see PCI like systems reaching a point where they can no longer handle the graphics cards or physics cards people demand. Sure, the current PCI Express system may run out of steam, but it will be replaced with something like it that works better. And the chipset does add a bit of cost to the board, but it's not an exorbitant amount. And if copper ever becomes too slow for a system bus, we could move to fiber optics (also, if copper became too slow for a system bus, we'd have to come up with a new kind of processor too, but I digress).

Originally Posted by olePigeon 

You can't just stick in an additional Audio card and expect it to make encoding sound any faster or upgrade your stereo to 5.1 by adding 3 more jacks. You have to buy a whole new audio card Regardless if you even have 5 ports for audio (or 7), that doesn't mean you can now do 7 uncompressed signals through them, nor does adding another sound card enable you to handle higher quality signals.

No, but I'm not sure you could do this in parallel like you are suggesting. Otherwise the cards we have these days could work in parallel just fine. The sort of scalability I was referring to though was the ability to feed more audio streams to more outputs.

Originally Posted by olePigeon 

GPUs are getting to the point where you can stick in multiple video cards, but you have to sacrifice 4 PCI slots to do it. They're huge and bulky.

This isn't a problem with PCI per say, it's a problem with the GPU's themselves. You can't simply move the chip off a PCI card and expect it to solve the problem. That chip is still going to require the same amount of power, and put out the same amount of heat, meaning no matter where you move if you're still going to have to strap a giant fan to it. There is no reason a PCI card can't be small and thin. The problem simply is that GPU's circuitry is too complex to go without the sort of cooling a graphics card these days comes with, and that's true for no matter where you move it. I have plenty of non-bulky PCI cards simply because those don't require significant cooling.

Originally Posted by olePigeon 

PPUs are almost unheard of.

They're getting more standard, mostly because in Crossfire a GPU can be repurposed as a PPU. I felt PPU's were an appropriate example for this argument.

Originally Posted by olePigeon 

This is where I think we're getting confused. This wouldn't be a basic BUS system like PCI, it would be more like Cell. Imagine, instead, that you have a big processor like AMD is proposing, but instead of having those features permanent, they're left open with sockets (obviously, it'd be more spread out.) The end user gets to decide which features will be "on" the CPU, whether it's additional CPUs or if they want to add GPUs, PPUs, or APUs.

Even the Cell requires a chipset and a bus for all it's onboard chips. That's what the 3.2 ghz PowerPC is on the Cell. It's the Cell's chipset to manage all the other co-processors. That's my point. Any way you do it, a modular system requires a chipset to manage what is plugged in and what isn't. That's why AMD has to move everything onto one die to make it work without a chipset. That way the user can't change anything, and they can hardwire the components because they know nothing will be changing.

Originally Posted by olePigeon 

You're right, but this is about the future. I also agree that AMD is jumping the gun. Maybe they'll take the fall for it so that Intel can do it properly later on. People are still adjusting to PCI Express, there's no need to make everyone jump hurdles for a new architecture.

Well my point is that AMD is sacrificing flexibility for speed that I can't see being needing. Processors are never going to outrun chipsets, because a chipset in itself is a processor. I see it as a marketing ploy by AMD.

Originally Posted by olePigeon 

That would be the cheaper way to do it. That's partly why Cell is so expensive. My idea would probably be the most expensive of the ideas. The highspeed interconnect of the Cell, but the specialization of AMD.

The Cell is actually cheap compared to other processors in it's class, that's what makes it so compelling. I'd say it's expensive right now because it's in such short supply.

But my point is you're trying to combine the chipset-less design of AMD with a modular design of adding co-processors onto a board, when a modular design requires a chipset, thus negating the chipset-less design of AMD and putting us where we are today.

(Yes, I'm aware there would be a chipset for legacy devices in AMD's plan, but that's not relevant to the point. We aren't talking about their handling of legacy devices).

[mduell]

I think all three philosophies (Intel with lots of "equal" cores, IBM with a large generic core and many smaller stream cores, and AMD with a generic core and specialized cores) have their place. IBM's works great for game consoles. AMD's works well for supercomputers and other dedicated compute clusters that are going to be doing the same or similar tasks. I think Intel's will work the best on the desktop.
Sun's Niagra is a variation on Intel's philosopy with the addition of tons of SMT. Works well for some kinds of servers.

Originally Posted by Big Mac 

I'm complaining, but I also must concede the sales figures. As far as IBM screwing up worse than Motorola, I don't think IBM left us at 500MHz for 3 years like Motorola did.

No, but they did leave Apple at 2.mumble Ghz for 3 years.

[Hash]

Originally Posted by olePigeon 

Well, maybe not the Cell specifically, but an architecture very similar. It looks like AMD is moving to a Cell-like philosophy for their CPUs. I remember Intel commenting about Cell being a glimpse into how microprocessors will be organized in the future.

A while back I broke my Apple Prediction streak by saying that they'd be moving to the Cell architecture, but then Apple moved to Intel. I think it threw a lot of people off. However, I might be able to salvage a little bit of that prediction.

IBM, Sony, and Toshiba have the right idea about the future of microprocessors. You can get more out of clusters of specific processor units that can each be ordered to do a special task more efficiently than a single, general processor unit (even with multiple cores.) Cell has the PPE (CPU) with SPE coprocessors, AMD will have a CPU with xPU coprocessors.

Instead of creating an entirely new architecture like IBM, Sony, and Toshiba did, AMD will be bridging existing technologies by placing the CPU, GPU, and APU on a single die, then move to the more Cell-like layout.

I think Intel will move in the same direction, and we'll have Macs using architectures similar in design to the Cell anyway.

stupid

[olePigeon]

Originally Posted by goMac 

No, but I'm not sure you could do this in parallel like you are suggesting. Otherwise the cards we have these days could work in parallel just fine. The sort of scalability I was referring to though was the ability to feed more audio streams to more outputs.

That's exactly how Cell works. Each chip detects other chips, then shares loads. Or, you can tell a chip to specifically handle only one type of instruction so it's designated for a specific task only. Instead of telling a chip to do a specific task it's not necissarily good at, you're sticking in an already specialized chip. It still acts in a way similar to Cell in which it'll look for other chips just like it, then share the load.

On the PCI side you'd put in another card with more jacks, then you'd have software on the OS that would pass the processed streams from the APUs to the audio jacks.

Same with video. The actual processing would be handled by the GPUs and PPUs on the array, then the resulting information is passed to the display card in the PCI slot. Moving the GPU to a more direct route to the CPU itself would decrease latency.

Originally Posted by goMac 

This isn't a problem with PCI per say, it's a problem with the GPU's themselves. You can't simply move the chip off a PCI card and expect it to solve the problem. That chip is still going to require the same amount of power, and put out the same amount of heat, meaning no matter where you move if you're still going to have to strap a giant fan to it. There is no reason a PCI card can't be small and thin. The problem simply is that GPU's circuitry is too complex to go without the sort of cooling a graphics card these days comes with, and that's true for no matter where you move it. I have plenty of non-bulky PCI cards simply because those don't require significant cooling.

The problem is that they're increasing the MHz and the power required to sustain it, giving off a lot of heat and requiring a lot of power. That's the exact problem the non-multiple core CPUs were having. The Pentium 4 could get up to 4GHz or whatever, but it gave off so much heat and required a lot of power. It's more efficient to stick in 2 2.5GHz cores than a single 4GHz core. If you have an array, you can put in slower MHz GPUs, but in greater quantities. They would generate less heat and require less power.

Originally Posted by goMac 

But my point is you're trying to combine the chipset-less design of AMD with a modular design of adding co-processors onto a board, when a modular design requires a chipset, thus negating the chipset-less design of AMD and putting us where we are today.

OK, I finally understand what you were saying. It sounded cool in my head. Just dropping in whatever xPUs you want to increase computing capacity in the areas you want, with reduced latency.

So you'd still need the specialized chipset in my idea, but I think (in the future) that it's not a bad idea as it would move all the core processing chips closer to each other and reduce latency. You could use fiberoptic interconnects and lasers for the BUS. However, you're right, that without PCI as an intermediary and a common gateway for the components to talk to each other, each company would have to make a chip that can talk to either Intel, AMD, or someone else. That'd cause too much trouble.

Oh well.

[mitchell_pgh]

Apple would be ABSOLUTELY foolish to switch the CPU at this point.

The fact that Windows runs on Macs is a MAJOR advantage to bringing more people to the platform. Also, PPC rarely was faster than x86, and when it was, it was only on the most expensive systems (sure a 4K Mac could beat a 4K PC, but my $1200 Mac was spanked by a $300 PC) . The average Mac users had CPUs that were months/years behind regarding computing power.

[goMac]

I think PowerPC CPUs grew in spurts. At introduction they were frequently faster than x86, but that growth was hard to sustain. x86 grew more consistently, so for most of the life of each PPC processor Intel's were faster.

[goMac]

Originally Posted by olePigeon 

The problem is that they're increasing the MHz and the power required to sustain it, giving off a lot of heat and requiring a lot of power. That's the exact problem the non-multiple core CPUs were having. The Pentium 4 could get up to 4GHz or whatever, but it gave off so much heat and required a lot of power. It's more efficient to stick in 2 2.5GHz cores than a single 4GHz core. If you have an array, you can put in slower MHz GPUs, but in greater quantities. They would generate less heat and require less power.

This issue has already been solved on the PCI bus where possible. GPU's, PPU's, they can be run in parallel as PCI cards reducing power and cooling needs. Some PCI cards even use dual GPU's instead of one massive one to reduce cooling needs.

Audio processors I don't think can stack like that for technical reasons.

[olePigeon]

Originally Posted by mitchell_pgh 

Apple would be ABSOLUTELY foolish to switch the CPU at this point.

The fact that Windows runs on Macs is a MAJOR advantage to bringing more people to the platform. Also, PPC rarely was faster than x86, and when it was, it was only on the most expensive systems (sure a 4K Mac could beat a 4K PC, but my $1200 Mac was spanked by a $300 PC) . The average Mac users had CPUs that were months/years behind regarding computing power.

I wasn't suggesting or saying they were going to switch. I was just saying the it looks like the CPU technology is heading towards the same design philosophy as Cell.

[mitchell_pgh]

Now if only OS X was more threaded... and programs were more threaded.

[MindFad]

Speaking of threading, aren't there reports that developers are having trouble with multithreading on the Cell, at least when it comes to the PS3? Something about it not multithreading nearly as well as the 360. "Reports" is more like me reading that people said there were reports; I don't know for sure, but I am curious about the technology.

[goMac]

Originally Posted by MindFad 

Speaking of threading, aren't there reports that developers are having trouble with multithreading on the Cell, at least when it comes to the PS3? Something about it not multithreading nearly as well as the 360. "Reports" is more like me reading that people said there were reports; I don't know for sure, but I am curious about the technology.

You don't really thread on the Cell. What you do is you create a separate program and load it on to each SPE. The idea is the programs will work in tandem with one each other. Now, I don't know how much this is abstracted in the SDK's, but this alone sounds like a PITA.

Now that sounds pretty cool if you just wanted each SPE to do each independently do some SETI@home action, and it would be pretty easy too. You could have each SPE chewing on a work unit. This is why the Cell is big in super-computing. But get those SPE's to work on something together? That's going to be a bit more difficult.

[Eriamjh]

Intel has done great things for Macs. Macs are faster than ever.

[PB2K]

that's not a miracle, they've always been

[Gossamer]

Originally Posted by PB2K 

that's not a miracle, they've always been

true.
But I'd put money on them being faster now than if we were still with PPC now (G5 laptop anyone?).