Xbox moving away from Intel

GrandMasterTerwynn · Post by **GrandMasterTerwynn** » 2003-11-04 01:08am

Thar she blows!

It seems that the next-generation Xbox will likely be using a PowerPC CPU mated with an ATI graphics chipset.

Hamel · Post by **Hamel** » 2003-11-04 01:14am

If Sony isn't bullshitting on Cell specs then nothing MS can pump out will compete

Durandal · Post by **Durandal** » 2003-11-04 01:18am

Which is precisely why IBM was contracted. If anyone can shove specs down Sony's throat, it's IBM.

The Kernel · Post by **The Kernel** » 2003-11-04 01:34am

The CPU is much less relevent to the Xbox architecture then the Playstation. The PS3 "Cell" chip will handle T&L, geometry and all the graphics effects like textures, bump maps and shading. It is then sent to a very simple renderer (GS2) where it is rendered. The Xbox (and Xbox2) uses the CPU for little more than setup and AI while the GPU handles the graphics exculsively (this is much like a modern PC)

The CPU of the Xbox didn't matter that much which is exactly why it was a 733 MHz Celeron with the P III cache tags activated. Even at 733 MHz it is no match for the Emotion Engine, which has two very powerful FP vector units that can cream the P III on FP code. Yet the Xbox is a more powerful machine because it doesn't need the CPU for graphics work and its total graphics power exceeds that of the PS2.

That being said, IBM PowerPC chips are great for low power, high performance computing and since Microsoft probably wants to make the Xbox2 smaller then the Xbox, this would be a step towards that.

EDIT: In case you didn't figure this out Hamel, the Xbox2 will indeed be able to compete with the PS3. The Cell chips have potential but they are much harder to code for because of their massive number of independent vector units. Although it is admirable for programmers to add scalablity and multiple threads to their code, very few actually do (just look at the PC SMP scene and that's only creating threads for TWO processors). Unless Sony releases some killer tools for PS3 along with actual shader code (which I doubt since they really don't have enough time) then the PS3 won't see NEAR its potential for several years after launch. Xbox2 on the other hand already has Microsoft's HLSL and all its tools to work from.

phongn · Post by **phongn** » 2003-11-04 10:54am

The Kernel wrote:That being said, IBM PowerPC chips are great for low power, high performance computing and since Microsoft probably wants to make the Xbox2 smaller then the Xbox, this would be a step towards that.

Bah, I'm not that impressed with the PPC750 series. It's a nice chip, but Motorola SPS (now spun off, IIRC) does better with the PPC7400 series.

The Kernel · Post by **The Kernel** » 2003-11-04 10:57am

phongn wrote:
The Kernel wrote:That being said, IBM PowerPC chips are great for low power, high performance computing and since Microsoft probably wants to make the Xbox2 smaller then the Xbox, this would be a step towards that.
Bah, I'm not that impressed with the PPC750 series. It's a nice chip, but Motorola SPS (now spun off, IIRC) does better with the PPC7400 series.

It will likely be a POWER5 based chips similar to the G5 sucessor, not an older G3 chip.

phongn · Post by **phongn** » 2003-11-04 10:59am

The Kernel wrote:It will likely be a POWER5 based chips similar to the G5 sucessor, not an older G3 chip.

Since when was the POWER series (or their derivatives a'la PPC970) low-power?

The Kernel · Post by **The Kernel** » 2003-11-04 11:05am

phongn wrote:
The Kernel wrote:It will likely be a POWER5 based chips similar to the G5 sucessor, not an older G3 chip.
Since when was the POWER series (or their derivatives a'la PPC970) low-power?

I should modify my statement slightly. They are low power in the sense that they meet the Xbox2 specifications (passive cooling) at low enough clock speeds (probably ~2GHz at 90 nano). The thermal budget is really the only thing they have to work around here seeing as this is a game console, not a portable.

Actually, these chips draw much less power then their Intel counterparts due to a number of factors (lower clockspeed, SOI, RISC-design) IIRC, around 45 watts for the top end G5 compared to 100+ for Intel's fastest. Of course increases in heat on higher end POWER parts are mostly due to cache issues, which are irrelevent to Xbox2 that needs only the bare minimum in this implementation (512KB-1MB). Also, the Xbox2 chip might very well have the "Velocity Engine" stripped out, much like the POWER series does for simple lack of need and cost savings.

phongn · Post by **phongn** » 2003-11-04 11:21am

2GHz @ 90nm is a bit ambitious for passive cooling, I think, but we'll see.

I also know that Intel and AMD processors typically have higher thermal dissipation than their Motorola and IBM counterparts, but the G5 is still a rather warm processor all told. Apple needed those enormous heatsinks to get 'quiet' cooling and it's still not really passive.

As for AltiVec, will POWER5 even have it? An Xbox2 part probably won't, but that vector stuff has its uses.

The Kernel · Post by **The Kernel** » 2003-11-04 11:29am

phongn wrote:2GHz @ 90nm is a bit ambitious for passive cooling, I think, but we'll see.

It is fairly agressive but if there is one thing IBM is good at it is their process engineering.

I also know that Intel and AMD processors typically have higher thermal dissipation than their Motorola and IBM counterparts, but the G5 is still a rather warm processor all told. Apple needed those enormous heatsinks to get 'quiet' cooling and it's still not really passive.

True, but Xbox2 probably won't be totally passive either. If you look inside the current Xbox (which has a passive cooling solution according to MSFT) the exaust fan is directly behind the heatsinks for the CPU and GPU, acting much like the Apple cooling solution (although smaller).

As for AltiVec, will POWER5 even have it? An Xbox2 part probably won't, but that vector stuff has its uses.

Although IBM has whispered about it, there is no indication that they will include vector processing in POWER5. However, the PPC970 successor (based on POWER5) will have it, so inclusion in the Xbox2 CPU is still a possibility. I'm not sure that vector units would be all that helpful for the amount of die space they take up (which isn't insignificant and may limit clockspeeds) considering the Xbox2's GPU will do most of the heavy lifting.

Durandal · Post by **Durandal** » 2003-11-04 02:22pm

It really depends on when Microsoft decide to lock in the hardware specs. They could go with a 970, but they could use a 980 with SMT, as well, but there's really no telling how far off that is. It's pretty clear that the 970's AltiVec unit was built specifically for Apple, so Microsoft may decide to just forego it entirely.

Of course, this means that Microsoft must be working on a set of PowerPC DirectX drivers as well. Drivers can benefit heavily from vectorization and the use of SIMD. So Microsoft may opt to use AltiVec if it can give enough of a performance boost. Does anyone know how extensively the current X-Box utilizes SSE? Or if it even has it? Really, they have a ton of different options, which is why I suspect they went with IBM.

And by the way, Hamel, the Cell is an IBM processor, as well.

Hamel · Post by **Hamel** » 2003-11-04 05:25pm

And by the way, Hamel, the Cell is an IBM processor, as well.

Haha

IBM is a double agent

phongn · Post by **phongn** » 2003-11-04 06:34pm

If Nintendo stays with PPC for their next-generation console (should one appear), this means that all three companies will be using IBM processors

Is Cell MIPS-like, though? PS1 and PS2 both used varients on the MIPS design.

The Kernel · Post by **The Kernel** » 2003-11-04 06:56pm

phongn wrote:If Nintendo stays with PPC for their next-generation console (should one appear), this means that all three companies will be using IBM processors

Is Cell MIPS-like, though? PS1 and PS2 both used varients on the MIPS design.

Cell is indeed a MIPS design. More of a multi-core version of the Emotion Engine in the PS2 with shared resources then anything else.