Beyond3D - Future Console Revisions

Future Console Revisions - 360, Wii and PS3 - Page 1

Published on 5th Jan 2007, written by Stefan Salzl for Consumer Graphics - Last updated: 21st Mar 2007

Introduction

In the latest article of his weekly column, Hiroshige Goto talks about the subject of console costs and the strategies which the different manufacturers try to apply in order to minimize their expenses. We take a look at his statements, strategies and analysis.

For Goto, both Microsoft's Xbox 360 and Sony's Playstation are designed for future cost reductions by semiconductor scaling. According to him there will be 2 process changes during the current console cycle -- 65nm in 2007 and 45nm in 2009 -- which would theoretically cut current dice sizes to a fourth of their current size. Coinciding with process shrinks, the manufacturers would observe "dramatic" improvements in the yield rate. Also, a lower power consumption and heat generation would lower costs for the cooling and the EMI shielding.

Goto sees two results from the shrinks: on the one hand the reduction of the die size would reduce the silicon costs, on the other hand the shrinks should ease the reduction of the number of chips, the substrate costs, etc. -- as such, chip-related costs should scale down faster than the actual die sizes.

The current die sizes at 90nm are about 228mm² for the Cell CPU and ~240-260mm² for RSX in Sony's PlayStation 3. In Microsoft's Xbox360 the Xenon CPU is 168mm²; Xenos is comprised of a 170mm² GPU core and 70mm² EDRAM on the same package. On a 45nm process Goto expects Cell to be about 60mm². Consequently "the chip costs would decrease to a third or a fourth." (Goto).

Sony PlayStation 3

One of the biggest potentials for cost reductions is the number of DRAM chips used in the PlayStation 3 (PS3). Currently PS3 uses 4 512Mib GDDR3 and 4 512Mib XDR DRAMs. "The difficulty in decreasing the number of DRAM chips is maintaining the memory access times and memory bandwidth while decreasing their number" (Goto). Goto estimates that all DRAM chips will be replaced by 1Gib chips in the future, which we would expect in the 2008-09 timeframe.

As for the XDR chips, they're currently using a 64-bit interface to Cell which could be replaced by 1Gib XDR2 chips with a 32-bit bus while the bandwidth would remain the same. "Reducing the Rambus memory chips from 4 to 2 is easy. The memory bandwidth doesn’t change. Since we develop the XDRAM together with Rambus, we keep an eye on all such thing during the development" (Ken Kutaragi).

On the RSX side Goto assumes that the GPU and the 4 512Mib GDDR3 chips, both on the same substrate, could be connected via a custom interface. Thus, the 4 GDDR3 chips could be replaced by 2 1Gib custom DRAMs while still retaining the speed and bandwidth. Due to the high production volume of the PS3, the additional costs would be more than just offset by the cost savings. "If the product volume is small, it's difficult. But in the case of 20 million units per year, the options are manifold. For example, you can think about SIP (System-In-Package)." (Ken Kutaragi). From our point of view, it is perfectly possible that these chips would run at 1.4GHz and use a 64-bit memory bus instead of a 128-bit one, however. This would be more akin to a migration to GDDR4, then.

Another likely direction for cost reduction is the hardware backwards compatibility of the current PS3. In case of a reliable soft emulation solution, the PS2's EE/GS and 2 128Mib RDRAM chips could be removed.

Click here for Goto's proposed PS3 schematics

Microsoft Xbox 360

In contrast to the previous generation Microsoft bought the design from the chip vendors and has the chips produced in foundries. According to Goto this makes responding to advances in the manufacturing processes easier and allows for chip integration, which is an advantage only Sony had in the last generation.

For Goto the cost reductions lie mainly with Xbox 360's GPU (Xenos): The current multi-die setup of a GPU core currently manufactured by TSMC and the eDRAM currently produced by NEC, contributes to higher costs. A one-chip solution (eDRAM & GPU core) would reduce the number of dice produced (and at different foundries) and thus costs.

Goto expects that by the time the 65nm process becomes available, the eDRAM will be integrated in the GPU core. That move would also replace the wiring between the core and the daughter die, which according to Goto will become more difficult at smaller processes. Moreover, it would reduce the power consumption of the GPU. We believe, however, that plenty of problems exist with such a strategy, because EDRAM processes are quite different from traditional ones and not quite as tuned for logic. In the near future, Xenos will be produced on 65nm while the EDRAM will be produced on 55nm; in the longer term, Z-RAM may or may not become a viable option to replace EDRAM in the 45nm timeframe, possibly using SOI at TSMC.

The 8 512Mib GDDR3 DRAM chips connecting to Xenos through a 128-bit interface also offer the potential for a further cost cut: Goto reckons they will be replaced by 4 1Gib RAM chips, as soon they become cheaper through DRAM manufacturing shrinks.

Click here for Goto's proposed Xbox 360 schematics

Nintendo Wii

For Nintendo's Wii, Goto see little room for further cost reduction. After changing their approach toward technology Nintendo already focused on low-costs: Broadway's die size (CPU) is already down to 19 mm², which implies truly negligible costs. The number of DRAM chips has been reduced to 2: 1 GDDR3 chip and a 1T-SRAM chip. The only possibility for reduction is the integration of the 1T-SRAM die, "but the integration of the 1T-SRAM seems difficult because of the size of its capacity" (Goto).

It is furthermore questionable from our point of view whether such integration would actually reduce costs, given the resulting yield differences. We believe straight shrinks of AMD's Hollywood GPU are inevitable, however, which might help a fair bit already given the likely proportion of costs attributed to the chip.

Thus, Goto expects to the costs difference between Wii and Xbox360 & PS3 to shrink, as well as the price difference.

Click here for Goto's proposed Wii schematics

Conclusion

Historically, console cost reductions are met by engaging new process developments and advances in manufacturing technology, shrinking and combining components in order to make the same for less. It seems that the Wii's hardware is already small enough, and therefore cheap enough, for them to likely seek possible cost reductions via process scaling alone, rather than seek hardware combination too. Contrast that to the PS3 and Xbox 360 where Sony and Microsoft are likely to agressively seek scaling and silicon combination to achieve their ongoing cost reduction goals.

Extrapolating out to the mid-point and conclusion of this new console life-cycle, we can forsee points where Microsoft and then Sony achieve non-loss price points on their respective consoles, where they're selling for more than it cost to make (which clearly isn't the case right now). However, the fact that the console vendors are seeking to enhance their offerings in ways that didn't effect the last generation, be that better optical technology, larger hard disks, better display output options and more (all while keeping the base spec constant), means that certain SKUs might never cross over that break-even point.

The vendors therefore still hang a good chunk of their money-making endeavours on the games and value-add online services. We watch any new console variants with great interest nonetheless, since they'll have influence no matter how small on how the vendors attack a brand new generation come 2010-11.