I'm afraid you're mixing a couple of things there...

AGP 8x initially failed due to the limited RAM bandwidth of common Memory, which it additionally has to share with CPU and the PCI Bus.

Also, even working at a 100% of its theoretical performance, it would still reduce the effective Memory bandwith available to the GPU to a mere 2.1GB/sec , which today is far below what the onboard VRAM can offer. Add heavy usage (amount of Memory supplied via AGP Texturing) and complex Multitexture manipulation, and the framerate drops down to a crawl.

CPU-wise (64bit), the problem is not the CPU itself, but the long process of fully optimizing the Software for it.
Quickest progress is commonly seen with isolated, hand-optimized Assembler code, but widespread (useful) usage is only achieved when the Compilers have reached a mature state of Optimization (which takes years)...

And having OS-specific commands on a CPU would mean a waste of tons of valuable Transistors (e.g. integrated FlashROM), which are entirely proprietary.

Proprietary is something noone could possibly want nowadays, the flexibility (and compatibility) to run Linux (or any other) Software just as fine, was the key to the Success of today's PC's.

And finally, attempting to speed up something on CPU level, that is known to be among the most inefficient (memory-footprint) Code ever written for the x86 Architecture, would be almost useless.
Windows hardly uses a few % of today's high performance CPU's 'as is' already.

Helping Microsoft manifest their monopoly even further is probably the worst thing anyone could possibly want. (unless you seek unlimited, corrupted power and money)  

...there is so much more out there than Windows, far more efficient and powerful. Over the last years, increasing numbers of Users have realized that (including myself, phasing out Windows on almost all Systems over time)

Quote:


maybe that's one of the causes win longhorn takes so long (expected to come 2005)
imagine the ai and the physics system made possible by 64bit cpus.  

beside: i heard Microsoft wants to implement a rom on mobos w/o you can't run win longhorn...i bet that'll mean microsoft won't rule the market the next years, 'cause nobody's interested in a fully microsoft depending pc system (actually that's not what i expect from a personal computer).

to andrei & agp8x: keep in mind you have the possibility to upgrade your mobo's work-around area some months later when other hardware components like 3d accellerators will make use of such developments. then you don't have to buy a new mobo, so, not a bad idea yet.

--adding--

when thinking about pu design this one came to my mind: is it possible to use..let's say a graphics chip (and its board) for sound output? keep in mind the board's environment (not the output links, but the chip's interface etc.)...i don't think so.. 

First, I don't see an economical viable project to do 256 Mb FlashRoms. Next, I would add one extremely important thing: when you read from a ROM, you read slowly, that is why you need shadowing to speed things up, and make the slowly ROM to be readed only once.

A conclusion: rather use a fast RAM, which should keep all the needed OS instructions inside it, as long as the Computer is running, and don't let anything else enter the "cache". THIS way we would hit the amazing efficiency of true Programable Processor, the most efficient design there could be, to take fully advantage of the calculation power of the CPU, without passing some useless cycles through the OS--> it would directly execute an instruction, not decode it, and translated over and over again, each time it goes from the USER to GDI to KERNEL to Actual CPU instruction set...

At the somehow equivalent comparison between 16 bit apps running in a 32 bit environment, I agree until one point: what is the cost of running a 16bit app on a 32bit system? Around 50-75% of the resources needed to run a true 32 bit app. In DOS apps, yes, you can find out a performance increase, but in WIN16 apps, I never found a program to run faster on WIN32 that it did on WIN16. Again, I wouldn't say that it is a true performance gain, when on Windows 3.x you use the MS-DOS SMARTDRIVE cache, and on Win98 you use a faster, more performant disk access method (32 bit). Still, reading/writing performance (max MB sended) is not doubled up when using 32 bit Vcache on Win 98.

Finally, to be totally cruel, a 16 bit app should never run faster on 32 bit environment, than on 16 bit, since, until now, THERE WERE NO APPS to be forced by the system to split in two ther code, and run 2 paralel  processes. A 16 bit app runs as a 16 bit app, it only uses the full 32 bit bus, and that's all. Wasted space. It's rather the enhancements, the device drivers that make the difference.

Also, when a typical 32 bit app is using 4Mb, and a more common one is using 12Mb, then multiply this with 2, and what's the catch to run them on a 64bit environment, as long as they would take a full 64bit allocation unit, and nothing else? The answer: to buy more memory, faster CPU's, you know the rest. 64 bit is a must to justify the new 1Gb RAM modules, and 3Ghz processors..

I'm sorry, but I think I'm wasting my time again...

*omg*, You have my blessing to try find someone that builds you a proprietary, Windows-only, 5 Billion Transistor heavy CPU then

(I imagine having a ~500W power-hungry CPU doesn't fit everybody's ideas about efficiency though )

PS.
Reading everything from such a "Boot FlashROM" into Memory first (for speed), is effectively exactly what any OS is already doing today, just from HD.
You can install your Windows onto a Flash Drive, if you wish to see the effect of your Ideas for yourself...

Building a truly, programmable CPU has also already been done. Buy a Transmeta based Board, and see how it performs (power-efficient, but bad performance).
---------------
I, for my part, will in the meantime continue to take advantage of real Ideas built into CPU's and OS'es

And apparently, you have never observed that there are computing Tasks out there, that take even a 3GHz CPU several weeks to complete.
Saying that nobody needs higher performance (or 64bit, for that matter), is about as wise as the Industry once stating "noone will ever need more than 640kB of Memory!".
(we all know how realistic that prognosis ended up, don't we?)

Quote:


I never said that there are computing tasks that don't take months on a 1Ghz processor, I haven't questioned the speed, but the lack of connection between "theoretic speed" and "actual speed".

More, I was never against 16 bit or 32 bit or 64 bit, when everything around was 8 bit. But the key is to use them wisely! Frankly, there were no doubles-up in speed when 32 bit appeared (and there was 16 bit at that time),  the gain in speed was 5%.

Technology obviously evolves, but there's a catch: time and money. 32 bit, as it evolved until now, proved that it can handle a greater deal of data, and an improved performance. But most of the time, when out of the drawing board, with tons of papers with calculations on how fast will perform, 32 bit is trully performing 75% faster than  16 bit. It never gets to twice the speed, due to various "out-of sync" processes.

64 bit will be a must in the future, but I was referring to the "frenzy" that was in 1995 to make everything "32 bit". The results: every program was compiled in 32bit code, but the results were not so amazing: 5% increase, or even no at all. This "frenzy" will appear again: and then I wonder... What's the use of having a program like Winamp written in 64 bit code, just to take up more processing power and bandwidth when there is a 3D rendering on the run? A 32 bit Winamp wuldn't be more than enough?

There are apps that qualify for 64 bit, and there are apps that frankly, are wasting space and time even in 32 bit... So, the key is being wise, and use 64 bit where is needed at most, not just to make everything 64 bit, to "suposedly" make everything faster. Rather make 2 really concurent 32 bit (processes) to run on a 64 bit bus, and then, having 2+2 processes (with the HT addition), we will have a computing arhitecture that is definitevely faster and a big step forward.