x86 processors and console
- Thread starter Simon82
- Start date
inefficient
Veteran
For the longest time the main advantage of using x86 processors over other CPUs was backwards compatability. x86 backwards compatibily is legendary. And for a long time BC was the thing that was holding the x86 back as well.
More recently the reason to chose a non x86 processor over an x86 one is for lower power requirments and lower heat output. Recent intel chips are doing much better in this regard but typically a desktop computer's requirments in the power/heat dept are going to be more loose than what a console needs.
Because x86 is past it's sell by date?Im no expect in either architecture but if i had to guess i put it down to performace.
Performance may be 1 issue, depending on what sort of performance you are looking at. I am sure there are many, many areas where a chip like Core2Duo mops the floor many times over compared to a chip like Xenon. But Xenon gots the nice marketing flops and from browsing the MS patents it seems one of the ideas behind Xenon was doing a but of graphic work on the CPU, and that is where the beefy VMX units, Dot product output, cache locking and datastreaming from cache to the GPU come to play. Not that MS chose Xenon to begin with, they actually wanted an OOOe CPU.
Anyhow, I think the major reasons are
(1) IP licensing. MS wanted more control over the IP, both for backwards compatibility purposes as well as for future process shrinks and possible system wide integration (e.g. GPU and CPU on a single die). Intel and AMD have little reason to make such concessions to MS. MS also wants control over the IP so it isn't fixated on one fab partner. They can dangle the contract to TSMC or Charter and so forth and get the best deal. If MS went with Intel it would have been locked in with Intel for pricing.
(2) Intel and AMD are very tapped in regards to fab supply/demand. They sell all they can make. It costs Intel and AMD ~$40 to make a CPU. They in turn sell these to mass wholesale retailers anywhere from $150 to $1000. Considering they are binning their chips (best chips are sorted as $1000 CPUs, chips that hit lower frequencies become the $150 chips, and everywhere inbetween) they make a HUGE amount of profit. They have a lot of R&D, fab maintenance, marketing, etc to pay, but that $1000 processor cost them $40 to make. On the console side MS would want to give them razor thin markup -- probably much less than the 30-40% markup GPU makers make. Considering the very high end of Xbox 360 projected sales would be 60M units in 5-6 years they are looking at 10M units at very low margins. Intel and AMD ship about 200M PCs a year at high margins. As a business, do you cut out the 10M high profit PC chips for 10M low profit console chips?
That said, I believe Gates had lobbied for Intel a bit. If MS could have got an X2 or C2D in the 360... boy, we would be having a lot of fun debates right now
Instead MS kind of got screwed. We hear some complaints from PC devs about Cell, but you don't ever hear, "But boy, that Xenon, what a dreamy chip!" hehe If MS had an X2 or C2D in there I am sure we would be hearing a lot of praise, especially from the PC devs.tongue_of_colicab
Veteran
Wouldnt cost and the need to change the design be a bigger factor? You only have AMD and Intel that really matter and both are focussing on the pc market with their cpu's. Couldnt it just be that AMD and Intel dont want to spend their time on designing a custom chip for ms/sony/nintendo? I wouldnt be suprised if that is one of the reasons why MS went to IBM and not to intel for the x360 cpu. Because they wanted this and that at a certain price point but intel just didnt want, or could give them that.
I think this too. Maybe as Nvidia make with Xbox and now with PS3, some company don't want to spend money to customize something that already has millions of dollars spent for PC part. We find RSX as a classic G70 chip and we find the Xbox cpu as a classic Celeron cpu. Maybe sometime is easy to make something at home (see PSP CPU and GPU) or let IBM do it.
RollingBalls
Newcomer
PowerPC, MIPS, ARM are highly scalable.
Can you explain in detail you opinion?
tongue_of_colicab
Veteran
So?
The advantage of IBM is that they turned more into a ''custom'' chip designer. IBM doesnt have the ''burden'' AMD and Intel have. Especially since they arnt making chips for apple anymore.
Probably they need money and custom projects like Xbox360 CPU or Cell let them make good profits.
tongue_of_colicab
Veteran
Probably. But I doubt they mind it because its only making them lots of money and they have the manpower anyway.
RollingBalls
Newcomer
IBM/MIPS/ARM offer configurable cores that can be used in a wide variety of applications from servers/routers/pda's/cellphones/automotive/military/medical/gameconsoles/etc
They're not a one size fits all as opposed to intel's x86 architecture.
ISA has been characterized as a factor of second-order importance with regards to performance. That said, an x86 core will consume some amount of extra power if it is forced to work around a number of difficulties in the ISA.
Console software has a lot less legacy code to worry about, so the big reason why x86 is big--it's already big-- doesn't help.
There are architectures that require less work, (though POWER has a memory addressing model more complicated in some respects than x86), and it can't be assumed that the console CPU can have the luxury of the high-end processes x86 chips rely on for their lead.
There are also far fewer hurdles with regard to licensing and IP.
The x86 license isn't open for everyone, those that have it aren't going to be signing away any portion of their crown-jewel IP, and the manufacturers of high-performance x86 chips are not going to take a bath on the extremely small margins a console CPU would give them.
You can buy a mid-range Core2 chip for almost the price of an entire console, so there is no reason why Intel or AMD would waste fab space or design effort on a chip that will go for the price of a Happy Meal (maybe a little exaggerated, but still peanuts).
Console software has a lot less legacy code to worry about, so the big reason why x86 is big--it's already big-- doesn't help.
There are architectures that require less work, (though POWER has a memory addressing model more complicated in some respects than x86), and it can't be assumed that the console CPU can have the luxury of the high-end processes x86 chips rely on for their lead.
There are also far fewer hurdles with regard to licensing and IP.
The x86 license isn't open for everyone, those that have it aren't going to be signing away any portion of their crown-jewel IP, and the manufacturers of high-performance x86 chips are not going to take a bath on the extremely small margins a console CPU would give them.
You can buy a mid-range Core2 chip for almost the price of an entire console, so there is no reason why Intel or AMD would waste fab space or design effort on a chip that will go for the price of a Happy Meal (maybe a little exaggerated, but still peanuts).
Last edited by a moderator:
tongue_of_colicab
Veteran
Yes but that has nothing to do with the actuall cost. CPU's are sold at higher costs because intel/amd need to make profit + r&d cost on them. Console makers sell their console with a loss because they can make it up with game sales.
Why would Intel or AMD sell to console makers at cost when they can sell CPUs at >50% gross margin?
Cheers
Nobody else is going to be making x86 chips, and the way things are, nobody new is going to get the license to make them.
There's VIA, but that also means giving up on performance and going without the high-end processes Intel and AMD use. Even the market VIA goes for has better markups.
Unless a current x86 manufacturer goes and makes its own console, there's no real way of doing it.
I don't think there is a single reason for this, the reasons are complex, some have been outlined above.
Another is that both Cell and Xenon were both custom designed to do a certain type of computation at high speed.
x86 designs are optimised for high speed scalar integer execution, it takes a lot of complex hardware in order to achieve this. Things like OOO and branch prediction are necessary for this type of computation and hiding the complexities of the x86 ISA.
Cell and Xenon are designed for high speed vector execution. OOO provides little benefit to this and branches should be avoided altogether if possible (both have instructions to do this).
One of the main advantages of OOO is that it allows the hardware to fetch data before it's needed, Xenon and Cell both get around this with cache & DMA instructions respectively.
An x86 processor could be designed for high vector throughput but it'd be a complex job because they'd need to drop OOO and that would cripple it, keeping OOO would produce far too much heat. This is the reason both Intel and AMD are designing vector add-ons which will be added as separate cores rather than internal additions to their x86 cores.
Another reason is IBM's silicon process is pretty much the best around for this sort of processor, it's designed for all out speed above everything else. Intel's is relatively simpler and is designed for mass manufacturing. While this gives Intel a huge advantage in yields it gives them a relative disadvantage in performance, they make up for this by using customised rather than automated design.
On the console processors IBM have done both customised design and used their high speed process. This is why Microsoft's ability to switch manufacturers is somewhat questionable, there's very few companies who have a process that fast, companies like UMC or TSMC may simply not be capable of building Xenon, even Intel would have difficulties - at least at 90nm, it'll get easier at 65nm.
The architecture of all the console processors (including Wii) was built to order, Intel and AMD don't do build to order (though they may for something as big as a console).
PowerPCs have long been tuned for high vector throughput, it's no coincidence Apple always used to pick vector heavy benchmarks when they used PowerPC, they beat x86 designs at that sort of workload. Xenon and Cell are a more extreme version of the same thing so will give x86 an even bigger beating.
x86 will wallop the console chips on SPECInt but it isn't useful for games.
Another is that both Cell and Xenon were both custom designed to do a certain type of computation at high speed.
x86 designs are optimised for high speed scalar integer execution, it takes a lot of complex hardware in order to achieve this. Things like OOO and branch prediction are necessary for this type of computation and hiding the complexities of the x86 ISA.
Cell and Xenon are designed for high speed vector execution. OOO provides little benefit to this and branches should be avoided altogether if possible (both have instructions to do this).
One of the main advantages of OOO is that it allows the hardware to fetch data before it's needed, Xenon and Cell both get around this with cache & DMA instructions respectively.
An x86 processor could be designed for high vector throughput but it'd be a complex job because they'd need to drop OOO and that would cripple it, keeping OOO would produce far too much heat. This is the reason both Intel and AMD are designing vector add-ons which will be added as separate cores rather than internal additions to their x86 cores.
Another reason is IBM's silicon process is pretty much the best around for this sort of processor, it's designed for all out speed above everything else. Intel's is relatively simpler and is designed for mass manufacturing. While this gives Intel a huge advantage in yields it gives them a relative disadvantage in performance, they make up for this by using customised rather than automated design.
On the console processors IBM have done both customised design and used their high speed process. This is why Microsoft's ability to switch manufacturers is somewhat questionable, there's very few companies who have a process that fast, companies like UMC or TSMC may simply not be capable of building Xenon, even Intel would have difficulties - at least at 90nm, it'll get easier at 65nm.
The architecture of all the console processors (including Wii) was built to order, Intel and AMD don't do build to order (though they may for something as big as a console).
PowerPCs have long been tuned for high vector throughput, it's no coincidence Apple always used to pick vector heavy benchmarks when they used PowerPC, they beat x86 designs at that sort of workload. Xenon and Cell are a more extreme version of the same thing so will give x86 an even bigger beating.
x86 will wallop the console chips on SPECInt but it isn't useful for games.
It's also helped that for the longest time, IBM's been really hurting for utilization on its high-end fab.
If IBM's volumes were larger on the high-end, they wouldn't be wasting wafers on low-margin chips.
If IBM's volumes were larger on the high-end, they wouldn't be wasting wafers on low-margin chips.
The absolute most important reason is owning the IP. Others have touched on this point, but failed to emphasize how important this is. Any console maker needs to own the rights to their processor design so that they can cost reduce later in the lifecycle, via die shrinks or integrating chips together. For example, look at the PS2: it has a PSX chipset embedded in it for backwards compatibility, and introduced a smaller (and far cheaper to produce) PSTwo later in its lifecycle. Neither would have been possible were it not for Sony owning the IP. MS learned this lesson the hard way with Xbox; they actually had a great processor in it (733 MHz P-III/Celeron, IIRC), but because they didn't own the IP they couldn't cost reduce it and they were constantly forced into difficult negotiations with Intel over production of the chip.
x86 isn't actually very good as an ISA; it's dated, clunky, and comes with a lot of legacy baggage. The only reason to use x86 is that you can use an off-the-shelf processor with great performance (as well as very mature compiler support). However, the down side of an off-the-shelf processor is as stated above; you don't own the IP rights so you can't control cost reductions and integration later. MS is the only one to have tried it in recent years, but that was because they were in a rush. After their experience, you can bet that no one will try it again for a long time.
x86 isn't actually very good as an ISA; it's dated, clunky, and comes with a lot of legacy baggage. The only reason to use x86 is that you can use an off-the-shelf processor with great performance (as well as very mature compiler support). However, the down side of an off-the-shelf processor is as stated above; you don't own the IP rights so you can't control cost reductions and integration later. MS is the only one to have tried it in recent years, but that was because they were in a rush. After their experience, you can bet that no one will try it again for a long time.
I will make my paragraph a sentance longer next time to add even more emphasis
