Team Xbox speculates on Xbox 2 CPU.

Brimstone

B3D Shockwave Rider
Veteran
Inside the Xbox 2 - Part 2
By: Cesar
Dec. 9th, 2003 07:18 am

Contrary to other sites, we usually don’t report every single rumor that we hear. It is easy to post rumors about fake stories like Halo 1.5 arriving on Nov. 15 or Perfect Dark Zero being shown at X03 than going to the source and finding out if there is any legitimacy to the stories.

When believable stories land in our laps, however, we take things seriously and dig deep to bring you the goods. You might remember a little story that we posted regarding Microsoft teaming up with IBM to manufacture the Xbox 2 CPU. It was a world exclusive and of course, with long-time rivals being the protagonists of the rumor, nobody believed our story. Low and behold the story became official on November 3rd with the announcement of a semiconductor technology agreement between Microsoft and IBM.

So far, Microsoft has announced three partners for its Xbox 2. First ATI, as its prime graphics supplier; then IBM as its semiconductor processor technology provider, and finally SiS as the chipset contractor. What do these companies have in common? A lot. And suspiciously, all roads lead to AMD.


The ATI/AMD64 Connection

ATI replacing nVIDIA as the graphic chip provider for the Xbox successor was a no brainier. The Ontario-based graphic chips maker has proven its R3XX VPU is far superior to nVIDIA’s offer, the GeForce FX family of GPUs (aka NV3X), especially under DirectX 9 Shader tests. The future of graphics is all about Pixel and Vertex Shader programs.

The interesting part of this connection is not the graphic processor that ATI will provide for the Xbox 2 (check our early preview) but the chances ATI might also design the motherboard chipset. It is worth explaining that nowadays the chipset (a set of integrated circuits to support the processor) basically consists of a couple of chips, the North Bridge and the South Bridge; found in the current Xbox in the form of the XGPU and the MCPX chips respectively, both provided by nVIDIA.

Last month, at its Reseller Seminar in Korea, Advanced Micro Devices revealed some future ATI chipsets for the AMD 64-bit platform. It’s particularly interesting that the RS480 (a North Bridge) features an integrated DirectX 9 graphics core using UMA (Unified Memory Architecture); a technology already used in Xbox for its 64 MB of RAM. Accompanying the RS480 is the South Bridge providing I/O capabilities; the IXP400. Both core-logic products will go into mass production in Q2 2004.


The IBM/AMD64 Connection

Although IBM is instantly associated to the PowerPC architecture, several sources within the videogame and chips industries have told us that the Xbox 2 CPU won’t be a PowerPC. I repeat: it WON’T BE a PowerPC processor. And we’re starting to believe them.

So far, no one has dared to confirm the Xbox 2 CPU will be a PowerPC processor. Everyone speculates about its architecture but nobody has been able to confirm it.

Switching to a PowerPC CPU, and therefore abandoning the x86 architecture, would make pieces of one of the flagship advantages to the original Xbox: ease of development. On top of that, you have to consider switching from x86 to PowerPC would make backward compatibility not a viable feature, unless they use a state-of-the-art emulation technology such as VirtualPC from Connectix, now a Microsoft company.

But where’s the IBM/AMD64 connection? Everywhere. Last January, Advanced Micro Devices and IBM said they were co-developing microprocessing technologies for use in future chips. Specifically, their joint venture is intended for develop semiconductor manufacturing technologies for 65-nm and 45-nm chips (a nanometer is one-billionth of a meter). Current microprocessors are built using a 130-nm process implemented on 12-inch (300mm) silicon wafers. Both Intel and AMD plan to switch to a 90-nanometer manufacturing process early next year.

Furthermore, several publications have reported that AMD engineers are being re-located from AMD’s Sunnyvale offices to AMD-IBM joint venture R&D centre in IBM's East Fishkill facility.


The Microsoft/AMD64 Connection

AMD doesn’t have the financial power to prevail against Intel. This quarter would be the first time in nearly three years AMD has made money quarter to quarter. Now it’s embarking on building their 65-nm Fab 36, which will cost $2.4 billion to construct. So AMD needs one thing in the coming years: cash. Who has lot of cash? Microsoft. A few months ago the Redmond-based company was considering paying a special dividend of $10 billion to slim down its cash reserves of more than $46 billion!

It is also worth mentioning that a 64-bit version of Longhorn, the next version of Windows, has been already sent to testers, even before the 64-bit version of Windows XP for AMD64 hit the streets. Depending on the release date of Xbox 2, the Xbox successor might use several (if not all) of the new technologies Microsoft is developing for Longhorn, including DirectX 10 and Palladium security technology.


The SiS/AMD64 Connection

Just one month ago, Microsoft announced that SiS Corp. will be developing “advanced media Input/Output technologies†for the Xbox 2.

Guess who has the best chipset for the Athlon 64 these days? nVIDIA with the nForce3 150? Nope. Via with its K8T800? Wrong again. The correct answer is Silicon Integrated Systems with its SiS 755/SiS 964 chipset. It won the Editor's Choice as the best Athlon64 chipset at Anandtech.

Part of SiS’s lead in performance is thanks to its proprietary MuTIOL 1GB/s Bandwidth bus, which avoids a slower North/South Bridge bus; a common problem found in competitors chipsets. SiS’s offer is the only one that connects the processor at the highest HyperTransport (an AMD spec to interconnect devices) speed while offering an equally speedy bus to connect both the North Bridge and the South Bridges.

Although the most likely scenario is that SiS is building an entire chipset for Xbox 2, it is possible that Microsoft is only licensing their MuTIOL technology. Only time will tell.

It is worth mentioning that the current Athlon64 and Opteron/Athlon64FX processors have a built-in DDR memory controller, resulting in motherboards that have a non-standard North Bridge -- the part of motherboard chipset that holds the memory controller. This implementation also implies an improved memory performance.


Conclusion

To sum things up, it is possible that the Xbox 2 might use an AMD64 processor (Athlon 64FX or the upcoming K9); probably a modified version, highly optimized for gaming purposes (ala Intel PIV Extreme Edition), that will be built by IBM using the most advanced manufacturing process available at the time the Xbox 2 ships; probably a 90-nanometer manufacturing process or, a 65-nm one, if the Xbox 2 arrives late 2005/early 2006.

The chipset could be designed by ATI, using some licensed technologies from SiS. That would have ATI not only designing the graphic processor but also the chipset; just exactly as nVIDIA does nowadays by providing their NV2X GPU and nForce chipset for the existing Xbox.

So there you have it. This is our small conspiracy theory. It is pure speculation but based on rumors, inside info, and technology facts.

Today, our friends from The Inquirer have a couple of interesting stories regarding Microsoft, IBM, AMD and Xbox 2, which help to support our theories. Be sure to check out their stories, available on their website at this URL.


TeamXbox

So an AMD CPU? If I remember correctly, Panjev speculated this might be the case when the IBM news broke a while back.
 
A few people speculated that as well, but with statements like this coming out of IBM:

"A few minutes ago we announced that we have won a deal to design, develop and make the processor for the next generation of Microsoft's Xbox game system."

" The new Xbox technologies will be based on the latest in IBM's family of state-of-the-art processors;"

...I'm not sure how IBM can be so involved in the "design" and "develop" phases of a CPU that would ultimately be based on AMD's Hammer line, nor how Microsoft could possible pursue said angle without bringing in another level of partnership with AMD for the chip. IBM is linked to AMD, but heck they're linked to EVERYONE these days! ATi, nVidia, AMD, Apple, Sony, Nintendo, Microsoft... :p I don't think any x86-64 chips could as "the lastest in IBM's family of state-of-the-art processors."

Speculation is speculation, and there are plenty of concerns about just what/why/how a PowerPC core in Xbox 2 would mean, but by comments so far it's leaning much more that way. Offhand, with other connections between Microsoft and AMD, I assume MS is just wanting to put their fingers in every pie as usual. ;) They want to take full advantage of any marketplace shifts that they can.
 
Switching to a PowerPC CPU, and therefore abandoning the x86 architecture, would make pieces of one of the flagship advantages to the original Xbox: ease of development

OK, sorry, but... LAUGH! :LOL:
 
Isn’t the most important thing that they can make direct-x run, on whatever CPU they decide on?
 
Interesting speculation, and for me as an AMD fan, good news. :) Would ensure AMD's survival over the foreseeable future.

Still, I don't quite believe this story. For one thing, using AMD64 tech would kill an UMA design stone dead. The built-in memory controller in the CPU isn't fast enough to feed a GPU more advanced than a GF4MX, and the hypertransport interface is also too slow and adds (probably too much) latency.

I'm still guessing it'll be a PPC chip in XB2.


*G*
 
I didn't know anything about AMD's Fab 36, so I did a quick search and found this intresting nugget of information from X-Bit Labs.

The 300mm equipment for 65nm chips installed in IBM’s East Fishkill, New York, foundry and to be set up in AMD’s Fab 36 in Dresden, Saxony, will be “pretty much identicalâ€. Since IBM and AMD co-develop 65nm process together, there will be an option for AMD to make chips for IBM’s clients, such as NVIDIA Corporation, to fully engage its manufacturing capacities.

AMD “has discussion underway with several potential partners in one form or anotherâ€. Co-manufacturing with IBM “will be a natural thing to consider,†said Hector Ruiz, the CEO of Advanced Micro Devices.

X-Bit Labs

Could one of AMD's potential partners be Microsoft? AMD fabbing both a CPU and VPU for an X-Box 2 would be similar to the rumored first X-Box deal that almost happened until Intel came in.
 
All three of those connections are not actually connections of course but instead circumstantial facts which have nothing to do with Xbox and are inflated artificially by the author's breathless tone. I fail to see the relevance of the fact that SiS makes the best AMD chipset, or that Microsoft now has a 64 bit WinOS and AMD has a 64 bit CPU to run it. Oh well. :rolleyes:

And how does trading x86 for PPC reduce MSoft's edge in "ease of programming?" :rolleyes: The x86 ISA is the biggest piece of trash this side of 65816, PPC (or ARM or MIPS or anything else) is so much cleaner and more intuitive. Sure, you can't compare two arbitary strings against each other with two instys, but that's a small price to pay for, among other things, a truly orthogonal GPR set or being able to put the results of an operation in a register that wasn't also one of the operands.
 
I've used many an instruction set and fundamentally after learning your 14th asm you come to one conclusion: there's no real difference.

Although there are a few ISA's which are a genuine bugger to work with (56001 DSP and 6502 are the ones I spent longest with), everything else is pretty much the same. They all have their plus and minus points.

Especially given that us asm hackers are a rare breed now anyway, ditching one architecture for another JUST because of the ISA is crazy - assuming the initial ISA is reasonable (which x86 is). In contrast, claiming that changing ISA makes something harder to program is also pretty equally daft.
 
Dio said:
I've used many an instruction set and fundamentally after learning your 14th asm you come to one conclusion: there's no real difference.

Although there are a few ISA's which are a genuine bugger to work with (56001 DSP and 6502 are the ones I spent longest with), everything else is pretty much the same. They all have their plus and minus points.

Especially given that us asm hackers are a rare breed now anyway, ditching one architecture for another JUST because of the ISA is crazy - assuming the initial ISA is reasonable (which x86 is). In contrast, claiming that changing ISA makes something harder to program is also pretty equally daft.

Uh, 6502, the torture... :cry: They all look easy when you learn 6502 first, then x86. :oops: And of course, if the bulk of your code is hand written in ASM these days, you have much, much, much bigger problems than which ISA you use.
 
Part of SiS’s lead in performance is thanks to its proprietary MuTIOL 1GB/s Bandwidth bus, which avoids a slower North/South Bridge bus; a common problem found in competitors chipsets. SiS’s offer is the only one that connects the processor at the highest HyperTransport (an AMD spec to interconnect devices) speed while offering an equally speedy bus to connect both the North Bridge and the South Bridges.

No it's not. That's utter nonsense.

There are a few other things in the speculation that are utter nonsense as well. Like:

But where’s the IBM/AMD64 connection? Everywhere. Last January, Advanced Micro Devices and IBM said they were co-developing microprocessing technologies for use in future chips. Specifically, their joint venture is intended for develop semiconductor manufacturing technologies for 65-nm and 45-nm chips (a nanometer is one-billionth of a meter). Current microprocessors are built using a 130-nm process implemented on 12-inch (300mm) silicon wafers. Both Intel and AMD plan to switch to a 90-nanometer manufacturing process early next year.

Well Intel are doing this too, so is Sony... Sony will therefore use an Intel Processor in PS3. Yep you heard that here first.

AMD doesn’t have the financial power to prevail against Intel. This quarter would be the first time in nearly three years AMD has made money quarter to quarter. Now it’s embarking on building their 65-nm Fab 36, which will cost $2.4 billion to construct. So AMD needs one thing in the coming years: cash. Who has lot of cash? Microsoft. A few months ago the Redmond-based company was considering paying a special dividend of $10 billion to slim down its cash reserves of more than $46 billion!

It is also worth mentioning that a 64-bit version of Longhorn, the next version of Windows, has been already sent to testers, even before the 64-bit version of Windows XP for AMD64 hit the streets. Depending on the release date of Xbox 2, the Xbox successor might use several (if not all) of the new technologies Microsoft is developing for Longhorn, including DirectX 10 and Palladium security technology.

MS has not announced anything about giving AMD any money. Surely they would need to do that soon if not already. You can't just secretly give $2.4 billion (or ever a few million) to a company hoping no one notices.

I do agree MS is looking forward to and actively supporting the AMD64 technology but this is more likely to do with not letting Intel dominate their OS....

Just one month ago, Microsoft announced that SiS Corp. will be developing “advanced media Input/Output technologiesâ€￾ for the Xbox 2.

Although the most likely scenario is that SiS is building an entire chipset for Xbox 2, it is possible that Microsoft is only licensing their MuTIOL technology. Only time will tell.

Sounds like a Southbridge to me rather than a Northbridge which ATI is meant to be handling as well going by this article?? Or ATI are only doing the Integrated GPU like they have no experience in designing integrated graphics until the AMD64 anyway.. (GameCube? IXP250/300 whatever it is called for Pentium 4, the failed integrated Socket A motherboards from Sapphire).

So we have a chipset from ATI which licenses technology but only the MulTiOL and not any other Southbridge functions.. and ATI cant do that since they are busy with designing possibly the first PCI - Express motherboard and gfx card right?

This editorial is a complete mess and makes Xbox2 sound like a mongrel, just like this stupid post of mine...

Heh.. sorry about that - I'm more confused now than I was 13 seconds ago (anyone got any Sands of Time spare?)...
 
Grall said:
Interesting speculation, and for me as an AMD fan, good news. :) Would ensure AMD's survival over the foreseeable future.

Still, I don't quite believe this story. For one thing, using AMD64 tech would kill an UMA design stone dead. The built-in memory controller in the CPU isn't fast enough to feed a GPU more advanced than a GF4MX, and the hypertransport interface is also too slow and adds (probably too much) latency.

I'm still guessing it'll be a PPC chip in XB2.


*G*

I don't think they need to go unified .

But athlon 64 to the main ram

Athlon 64 via hypertansport to the graphics card (hypertransport shoudl be fast enough for that) THen the gpu to ram

Wouldn't that work fine ?
 
jvd said:
I don't think they need to go unified .

But athlon 64 to the main ram

Athlon 64 via hypertansport to the graphics card (hypertransport shoudl be fast enough for that) THen the gpu to ram

Wouldn't that work fine ?

No. Consider this fact: suppose the CPU wishes to write to a location that the GPU has reserved for itself as a render target. We would be writing over our frame buffer with data, and since most modern CPU's dump cache only at block-replace time our problem is only compounded. Or alternatively, let's assume that the CPU allocates some memory space to store elements in a dynamic list, and our GPU, ignorant of the fact that the CPU has reserved this memory, writes over this with a cube environment map. This is why we don't have two memory controllers, to prevent memory hazards such as this from proliferating.

EDIT: I assume you meant a segmented memory structure. Sorry. In that case, I agree with you, I think segmented video-audio-main memory is something that has gotten an unfairly bad reputation for some reason. However, as the IBM press release clearly states IBM as the primary responsible party for the design of XCPU 2, we can safely rule out an AMD K8. My guess is a multicore PPC like the rumored CPU in PS3.
 
Super Grafx said:
good read, even though it is mostly speculation.


AMD64 makes alot of sense, more so than PowerPC.

One thing tho' is cost and availability. IBM has better capacity by far, and would likely undercut AMD64...likely but not necessarily.

I also believe emulation is underrated, why did MS by Connectix?
 
Switching to a PowerPC CPU, and therefore abandoning the x86 architecture, would make pieces of one of the flagship advantages to the original Xbox: ease of development.
These speculations get funnier with time. :LOL:

My guess is a multicore PPC like the rumored CPU in PS3.
PS3 is not gonna use a multicore PPC, but well...
:p
 
Tahir said:
I also believe emulation is underrated, why did MS by Connectix?

to help enterprise customers consolidate archaic windows nt systems on newer hardware running windows server 2003?
 
Tahir said:
Super Grafx said:
good read, even though it is mostly speculation.


AMD64 makes alot of sense, more so than PowerPC.

One thing tho' is cost and availability. IBM has better capacity by far, and would likely undercut AMD64...likely but not necessarily.

I also believe emulation is underrated, why did MS by Connectix?


IBM basicly made amds soi and .09 processes . There may be a deal that allows ibm to liscense the k8 tech for very cheap.



It could make sense though . Ms has the os market all to its self basicly . Intel has a strong hold but its not a full hold on the x86 market. So Ms would want amd around for a long time so that intel can't strong arm ms but ms can strong arm intel. So they buy amd chips from ibm. Ibm makes the bulk of the money . But amd gets some money thrown at them. Ibm could also be making the ati chips too.
 
Fafalada said:
My guess is a multicore PPC like the rumored CPU in PS3.
PS3 is not gonna use a multicore PPC, but well...
:p

:?: Why do you say that? I admit - assuming CELL does end up in PS3 - that each APU being a PPC core is quite ambitious. But there is still very little solid information about PS3. For e.g, a "CELL CPU" may be implemented as an "abstracted"(well, more or less) "processor farm" in the system, with a seperate MIPS/PPC CPU(s) farming instructions to the CELL, controlling main program flow, etc.

Unless you already have some info. 8) At least, enough to know that PPC is not in the design. MIPs maybe?

Is Axel Entertainment 1st/2nd/3rd party to SCEK?
 
akira888 said:
No. Consider this fact: suppose the CPU wishes to write to a location that the GPU has reserved for itself as a render target.

Except, the GPU CAN'T DO THAT. It can't allocate memory, it will put the render target starting at the address it has been instructed to, and if it overwrites something it'll either be catched by the CPU's bus snoop unit and declared invalid in the CPU's cache, or data will simply be corrupted and the machine will crash next time it tries to run cube mapping data as program code. Either way it'll be catched by developers (hopefully :LOL:) before the game's finished.

Naturally, this is a scenario hardware designers are aware of, and would work to prevent happening. AMD has already done a lot of groundwork on this with Opteron and its NUMA memory architecture, where serious risk of memory corruption exists when you have multiple processors with their own pools of RAM.

Anyway, you wouldn't see a GPU rendering a cubemap across the hypertransport interface to main RAM, just as you wouldn't see a CPU write dynamic lists across the hypertransport link into video ram. That would be counterproductive. :)

It IS a cool thought to imagine GPU being joined gluelessly to the CPU via a hypertransport interface though, that would lead to a highly efficient way of communicating between the two, but still, I think M$ wants to go UMA this time as well, considering how expensive fast memory is. Buying two pools of fast (and probably different; possibly DDR2 for CPU and GDDR3 for GPU) RAM when they could buy just one... They'll probably reason the CPU won't need all that much bandwidth, with all the cache they have these days, and the GPU won't miss giving away some since it'll be rendering TV resolutions anyway.

In any case, I bet we're all going to be tickled pink to hear final specs of each of the upcoming consoles. :) Even if PS3 don't reach the magic 1Tflops even if both CPU & GPU are included in the calculations, I bet all of us will be awestruck by the monster Sony will have designed. I think the same can be said about Nintendo and M$'s cubes and boxes as well. :)

I personally hope and pray Nintendo calls their next console "Starcube" like they were gonna do with GC. It's just such a cool name! :)


*G*
 
Back
Top