Pocket PC on XScale: a bike with little motor on the locomotive

The problem is as follows. The Pocket PC based on today's fastest 400-megahertz XScale processor produced a while ago, but the real speed we are of them never came. Moreover, and 400-megegertsovaya, and a simplified version of the 300-MHz XScale work exactly the same speed as the 206-MHz processor StrongARM ... speak out carefully - in most cases, though in fact it is an overwhelming majority. Journalists who "run" the new computers, usually written - say, until the advantage in speed is not so much because the programs are optimized for the XScale in the market. And then, with undisguised regret - but soon these programs become available, and only then ... However, some testers just swear. And they can understand. It's coming about doubled the processor speed. Clearly, this is a different architecture - not familiar to us by personal computers, CISC, but a real RISC. It is clear that too explicit dependence of handheld performance of the CPU clock should not expect ... a question - and, actually, why? Remember the appearance of the first processor Pentium II, and then their variations with a clock frequency of 400 megahertz. Jerk was very tangible. The computer could not find - a mountain deer, Hound dog, frightened rabbit hunter, not a computer ... Immediately have to talk about productivity growth in a few percent - if you have to do. Something is wrong here. The first "wrong" - where these "optimized" program? At least two? Or even one? The second "wrong" - appeared on the market first graphics accelerators for Pocket PC. These chips should speed up drawing two-dimensional images (of three-dimensional graphics language seemingly does not go - and why is it necessary for such a screen size?) And to undertake the work on the output video when playing video. Clearly, the computer graphics chip will be more expensive than without it ... but clocked XScale - 400 MHz! With such a performance take streaming video with a resolution of 240x320 pixels a snap (even StrongARM cope without strain). Suspiciously so. Very suspicious ... But not this the case when the howitzer fired on the wheel? But birds have, and fled. And the issue XScale was a pacifier. And along with it - and our optimistic expectations.

According to Professor

More recently, the internet site Pocket PC Thoughts, an article appeared Dzhostroma Andy (Andy Sjostrom) "Christmas expectations: a new tire for my XScale". This article is curious in itself, because Andy (whose name in Russian transcription I hesitate to say out loud), a good journalist. But still curious interview, included in the article. On the problem of CPU XScale spoke Sven Myhre (his name I'm not trying to translate and - let it remain in the original, not a distorted form), the head of Amazing Games and renowned expert in the field of software and hardware for computers Pocket PC. Sven Myhre said that theoretically, a 400-MHz XScale processor is definitely more productive than the 206-MHz StrongARM. But in reality, your computer's performance depends not only on the clock and "pure megahertz, but also on how the processor is loaded (that is, how to organize data exchange between the processor and other devices, PDAs). In computers, Pocket PC XScale and StrongARM processors work with 16-bit data bus. Although both processors - a 32-ryazryadnye RISC processor, designed to work with 32-bit instructions. In order to fully load the processor and to compensate for the restriction of 16-bit bus, it is necessary to double the operating frequency of the data bus. In this case, the exchange of 16-bit data will occur at the same rate as the 32-bit. In other words, increasing the operating frequency of buses will allow the processor to work with 16-bit bus as well as with 32-bit, for which both the processor and designed. But doubling the operating frequency of the data bus - it's not all.Necessary to provide the performance headroom of at least 25% to ensure the correct exchange rate with other devices to your computer when you run applications. As a result, the bus must operate at a frequency of 2.5 times greater than the CPU clock speed. This means that the operating frequency data bus for a processor with such a frequency of 400 MHz must be at least 1000 megahertz to 300 megahertz processor - 750 megahertz and 206 megahertz StrongARM - 515 MHz ... Now imagine the chasm that separates the real performance Pocket PC processors from what we have today. XScale processor with that at 400, 300 and 200 MHz-based system bus, the operating frequency is 100 megahertz. And familiar to us StrongARM processor clocked at 206 megahertz - with the bus, the operating frequency is 103 megahertz. In this case, according to Andy Dzhostroma also the reason for the similarity of the results of tests carried out on computers with 400, 300 and 200-MHz processor. Bus width and its low operating frequency are so narrow neck, which prevents the exchange of data between the processor and cache and between cache memory and a memory device. Moreover, it was noticed the superiority in performance StrongARM - in some applications - is just 3 percent of those for which the operating frequency of the data bus computers with the StrongARM processor, more bus frequency processor XScale - 103 and 100 MHz respectively. Provisions are not correct and the increase in cache memory, a built-in processors. Cache memory processor StrongARM sostavlyut 8 kilobytes instruction cache and 8 KB data cache. In the XScale cache size was increased to 32 and 32 kilobytes. But the exchange between the processor and cache memory are portions of the minimum size is 64 bytes. Therefore, when handling the processor to a single byte of information he has to process all 64 bytes. As a result, when the computer with an address database or program planning time, where user data are quite small in size, performance XScale impossible to estimate. Most of the work computer processor Pocket PC operates in vain.

And in reply - silence ...

It seems there was a smell sensation. But what - made public by the biggest blunder in the architecture of Pocket PC. Unprocessed seriously circuitry Pocket PC developers "stuck" in a computer processor, which obviously was not able to provide any significant increase in performance ... and get a bicycle with a little motor on the locomotive. Interview Sven Myhre did not cause as long as "thunder" - nevertheless it is the opinion of one person, even if very authoritative. But a host of issues, if not experts, then the interested public. Many seem obvious shortcomings Pocket PC now look different. For example, the slowness of the operating system Windows CE. In the failures and the slow formation of a family of Pocket PC is more than just blame Microsoft. Mol, the interface is overloaded with unnecessary luxury, the majority of routine operations unnecessarily complicated for the sake of similarity "pocket" of the operating system with a system for desktop computers and so on, until the excessive appetite Windows CE for handheld computing resources. But now it turns out that Microsoft is on a piece of software just does not podkachala because it works XScale in full force, the operating system, literally, "summer" would be like a rocket. However, not all rush to agree with Andy Dzhostromom and his companion. Some blame ... lazy programmers. Like, what just no idea - just would not work. But this is only partial statements which I have found on the Internet. More detailed and much, much less, official explanation yet. The authors, translation:

Nikolai Yakovlev