OK your point is sound. But how about ARM chips getting used on their own merit?With the lack of consistency of SIMD support, cache architectures etc. etc. across ARM platforms I don't think low level familiarity matters one iota.
x86 maybe, ARM no.
Now this is awesome. SSD Controler + 64GB of memory on a single chip made by Sandisk. Perfect for nextgen consoles.
http://semiaccurate.com/2011/01/05/sandisk-shows-single-chip-sata-ssd
Anybody think Nvidia's project Denver ARM chip might have any console implications? If so for who?
http://www.anandtech.com/show/4099/nvidias-project-denver-nv-designed-high-performance-arm-core
I guess the obvious guess would be Sony, given they use Nvidia in PS3.
Doubt it will be a factor in a home console. Maybe a portable.
The portable end already has plenty of ARM implementations for higher end handset graphics (including Tegra). I think a console would make more sense to be honest, but I don't think ARM would be the best solution for the next round, but I can see an ARM core or two in the GPU used like Starlett in the Wii, especially if it's an established GPU already available on the PC market.
I don't expect to see ARM being competitive at the level of performance required for a next gen console, which are probably at least a couple of years out.
Why not? Microsoft went from an OOO CPU last generation to a very simple IO CPU this generation, and made up for it by packing lots of them on die and clocking them highly. Where's the evidence that they'll go for a much more complex CPU design next generation, especially now that GPUs are becoming more general purpose?
This gen everybody went from AOS to SOA programming.
I think they should equip the consoles with an embeded SD chip of like 1gb that is used to store game saves and other user data. Then provide 2 SD slots that the user can install thier own cards into with the ability to run those chips in a raid 0 configuration. The small embeded chip insures that the game saves are protected if a user decides that they need to upgrade the storage capabilities of their console. While the SD slots (protected via a door much like the PS3's HDD drive) allow for greater speed to help offset the increased amount of Ram these new consoles will have to combat the slow media transfer rates afforded by BD and DVD.
No they didn't...
What evidence do you have to suggest such a thing..?
Naïve question (Arcangelmorph your insights are welcome too ) if you have a single chip as in Intel SnB or Nvidia Tegra 2, with pretty low communication latency between the CPU and the GPU, could it makes sense to pass on SIMD on the CPU and to rely more on the GPU for those calculations?Alright, that was too strongly worded. Mea culpa.
Let me restate it as: There's a trend towards converting compute intensive algorithms to SOA, - like GPGPU physics engines.
4D vectors is an excellent fit for vertices (XYZW), textures (RGBA) and framebuffers (RGBA), however today we have many more channels for our texture/material system and multiple render-targets, where 4D vectors make less sense.
I still think a wide vector extension in next gen console CPUs is a viable way to increase compute density of your main CPU. You can still stack 4 4D vectors into a 16D register if you want to.
Cheers
Naïve question (Arcangelmorph your insights are welcome too ) if you have a single chip as in Intel SnB or Nvidia Tegra 2, with pretty low communication latency between the CPU and the GPU, could it makes sense to pass on SIMD on the CPU and to rely more on the GPU for those calculations?
For the tegra2 Nv states that implementing simd unit would have inflate CPU die size by ~30% hence my question.