One of the most interesting aspects to new console design is only indirectly related to the device’s game performance. Internal design — how components are laid out, integrated, and cooled — is vital to building a reliable machine. Until quite recently, this aspect of design only attracted consumer interest if something went wrong. Each generation of consoles has had issues, but the Xbox 360′s famous Red Ring of Death catapulted design problems into mainstream discussion. The surging popularity of websites like iFixit, meanwhile, has shone a spotlight on design decisions that were previously hidden under plastic shells.
Nintendo has chosen to one-up such popular sites by releasing its own data on how the Wii U’s components were designed. The new information, released through an “Iwata Asks” explainer, sheds light on the difficulties of designing a system as compact as the Wii U, which isn’t much larger than the Wii, while substantially increasing its overall performance. Nintendo remains tight-lipped on many details, but we know that the CPU inside the system is a multi-core, POWER-based IBM design. Nintendo went with IBM for the combined module, as shown below:
Without any sort of scale, the only thing we can conclude from the image is that the second on-die package is much smaller than the first. Nintendo offers a few additional hints with the following:
This is rather odd. Rumors around the Wii U’s GPU initially pointed to an RV770-derived (HD 4000-series) chip with DX10.1 support, while later statements claimed the AMD component as actually a Radeon 6770. The Radeon 6770 was itself a re-badged Radeon 5770, and at 170mm2 on TSMC’s 40nm process, it wasn’t all that small. Toss in the 32MB of eDRAM we know the Wii U has on-package, and that’d be an awfully large piece of silicon. If the smaller die is actually the CPU, it means that Nintendo, for all its praise of AMD and IBM’s knowledge of CPU+GPU integration, chose to keep the two packages separate. There’s noAPU to be seen here, alas.
It also implies that the Wii U is a fundamentally a GPU with a CPU hanging off the side.
Here’s the Wii’s heatsink in comparison to the Wii U’s. Iwata says the new console generates 3x more heat than the old one did, which necessitated a larger fan and more fins. The heatsink design is still technically passive; the system uses a single fan to intake and exhaust air over the heatsink. Again, while we don’t have absolute scales, Iwata’s demonstrations put some hard thermal limits on what sort of GPU could be inside the system. Under load, the Wii drew ~16W as measured at the wall; Iwata’s comments imply that the Wii U will draw between 40-60W.
A 5770 dissipated 108W and measured 63.5mm tall, 165mm wide, and 267mm long. It’s true that this referred to total board power and that AMD doubtlessly refined its base GPU to Nintendo’s specifications, but a 50% reduction in active power seems unlikely.
Nintendo offers a few additional hints when it writes that “An MCM [multi-chip module] is where the aforementioned Multi-core CPU chip and the GPU chip are built into a single component. The GPU itself also contains quite a large on-chip memory.” There are, in short, plenty of reasons to think that an integrated CPU+GPU with on-package cache would serve Nintendo better than a GPU with on-die eDRAM + on-package CPU. The company’s phrasing points towards the former, but image labeling states the latter.
What can we tell from this new data? The “Juniper” GPU at the heart of the 5770 was almost certainly too large and too hot for Nintendo’s targets. Redwood, the GPU inside the Radeon 5670, is a much better bet. Not only is it smaller, at 104mm2 on 40nm, it also shipped in a 39W flavor. Additional optimi ations and tweaks could easily squee e a GPU of that caliber into the 40W-60W dissipation window we previously mentioned.
Nintendo’s reveal also supports developer statements that while the Wii U’s GPU is significantly more advanced than anything the Xbox 360 or PS3 can offer, the CPU is on-par or slightly less powerful than the Xbox 360 or PS3′s. Nintendo may have gone for a POWER-derived architecture this time around, but the smaller package in the image above implies three lean cores without much muscle.
The Wii U’s unusual controller and secondary screen means that a great deal may be riding on how many CPU cycles are dedicated to controller synchroni ation and processing. In theory, it might be possible to offload certain tasks to the system’s GPU, but that depends on which core Nintendo used. Both the RV770 and the Evergreen (HD 5000) GPUs supported OpenCL, but RV770 was limited to OpenCL 1.0, whereas the HD 5000 family supported OpenCL 1.1.
It goes without saying that Nintendo has dedicated a great deal of effort to building a small system that would retain the quiet operation and comparatively low power draw that made the Wii stand out in the 2005-2006 product cycle. Will lightning strike twice? That’s a good deal murkier. Nintendo is moving first this time around, but Sony and Microsoft are both planning their own next-gen consoles at much lower price points than the Xbox 360 and PS3 debuted at.
Nintendo tears down Wii U to show off single-chip IBM/AMD CPU + GPU
0 comments:
Post a Comment