arstechnica.com — NVIDIA's Fermi reveal shows clearly that the company is pursuing a two-pronged approach to staying alive: at the high-end, target supercomputing, and at the low-end, target mobile and embedded. This might eventually leave the PC gaming in the hands of AMD and Intel, but with the state that market's in, they can have it.
Oct 1, 2009 View in Crawl 4
kanojo1969Oct 2, 2009
The idea that modern OS's can be run an ancient processors, and that this indicates a lack of evolution in CPU design, is interesting and leads me to wonder why the GPU market hasn't fallen into the same trap.After all, we've been through quite a few complete redesigns of the GPU architecture and at the client application level older applications can't tell the difference.So, it seems to me that the difference is in DirectX. Is a similar concept for CPU's realistic? The OS developer creates an API that applications use to get access to the CPU hardware, and the CPU manufacturers produce CPU drivers that map this onto their particular implementation. If their architecture has weaknesses in certain areas, then those aspects of the API will be slower than other architectures might.This would allow complete changes in CPU architecture without any real problem of backward compatibility. A new super-parallel CPU could come along, and people using it would know that in regular desktop apps, it would be around the same or 20% slower than the existing X86 CPU. But for certain tasks it would be hundreds of times faster.Sounds like a no-brainer to me, and you should then be able to run the same binary on any CPU with the appropriate API Driver. Due to CPU differences in memory etc you would probably have to have managed runtime stuff like .NET as the most raw environment for applications, C++ would be out. But these days there's very little performance difference and for target tasks the CPU is hundreds of times faster anyway, using C wouldn't give the benefits it currently does.Why hasn't this already been done?
trellsaracenOct 2, 2009
Yeah, just like game developers avoided DirectX and went with OpenGL instead. Oh, wait.Sadly, the better standard rarely wins out.
curiousnightowlOct 2, 2009
Great Idea
ghostwoOct 2, 2009
@kanojo1969This has been done. A major improvement in the last few years is multi-core processors. Newer apps can utilize the resources of multiple cores for significant speed increases, while older apps get relegated to single core. Not to mention 32/64 bit computing. These apps are fundamentally different at low levels, yet they all run just fine. As far as Differences in CPU's, it mainly comes down to architecture. Should memory transfers be handled by the processor itself, or should this task be off loaded to a separate chip? A FSB is easy to implement due to it's simplicity, and allows the processor to do other stuff, as well as less heat, fewer defects, and lower costs. Of course the lower cost comes at the expense of a small amount of speed, so there is a trade-off; However, as tech improves, such as less heat, fewer defects, more efficient algorithms, the trade off is no longer necessary, so you end up with an i7.With all of the billions invested, if it can be done, it likely has been studied. What we get as users is the best option at the time.