Microsoft Please Read! Re:XScale optimization

[imported_jeff]

a clarification for pixelator:

Jeff said:

But this has zero to do with XScale or MS optimizing for it. The people who bought devices that forced them to change storage devices made their own beds -- nobody made them purchase those machines. No sympathy for them, sorry.

To which pixelator replied:

That's nice. So I guess you think it's OK to advertise devices as new, improved and faster (yes, claiming a device runs 400MHz compared to 206MHz is doing just that), when they're actually slower? And it's the fault of the end consumer that they believed that?

I think that you missed my point, even though you even quoted the relevant part. I've italicised it here to make it easier to see.

To recap, my lack of sympathy is not for those who merely bought XScale devices and didn't find them to run as fast as they hoped, but rather those who paid high prices for XScale devices because they had to get new storage or peripherals and the like. Adding the price of those peripherals into the cost of a new XScale device is hardly fair in the argument that Wired was making. His original assertion was this:

Many multimedia people who upgraded had to upgrade their entire system including storage. That is because most longtime users started with either a MicroDrive, PCMCIA Drive or Compact Flash Memory and some XScale PPCs only allowed for Secure Digital Cards without a sleeve. If you were one of the first to go this route in June, $700 was a lowball figure for a 3950 and a 256 - 512MB SD card. Try $1,000.

Again, I say that nobody forced these people to buy high cost technology. If you is compelled to be an early adopter, then you will be paying top dollar, and that is the price of your compulsion. This point has nothing whatsoever to do with XScale performance.

[pixelator]

Originally posted by Gamma Ray

?If SONY's next PDA works as promised, reviewers even in the general press will go nuts and many purchases will swing to SONY.

I don't think Sony is claiming any big new features with the NX except the WiFi-"only"-slot. Advance word is it's not much, if any, faster overall. The reality of it is the MSRP of $600. The reality of the PPC is the new $200-$300 pricing for many new models.

This indicates some serious encroachment on the PalmOS market by PPC by this time next year... Not a big downfall.

Now, if Sony were to release the NX series at $300 or $400, I might be inclined to worry. But with a dual-slot transflective Dell just around the corner for 200 bucks? Eh, I think Palm is in trouble, not the other way around. Even with XScale being a step back (after all, Palms are using them, too).

b

[Digitaldoc]

>>Palms aren't snappier.<<

I think that Palm's are indeed faster depending on the application of course. I agree that on native applications like Contacts, Calendar, Tasks, the difference is negligible. When I run similar medical applications that access tables on both Operating Systems, the Palm is faster to respond. I have assumed, perhaps incorrectly, that this is because the OS is more non complicated and simple, at least compared to the PPC OS by MS. I have been told by programmers for both Operating Systems that that the Palm OS is more linear and simple with less software layers before reaching the hardware. These same programmers will tell you that the MS OS is more bloated with code, further removed from the hardware, and consequently slower yet more robust for certain specific applications.

Analogy: I think of the PalmOS as a sleek cool, simple and fast motorcycle.

The PocketPC is more like a car. Slower off the line, but with more options and utility.

[Bandung]

I find it hard to accept statements that "MS doesn't support XScale". Even allowing for a good deal of liberties, the facts are that Microsoft's code does run on the new chip taking advantage of the existing ARMx instruction set. Yes, the inference behind these statements is that Microsoft will not be releasing an operating system update that takes advantage of the additional instruction set any time soon. But for the number of devices that we are talking about, it is not a big market fiasco.

What would you have MS do? A compiler optimization as an example?

1. A compiler optimization does nothing for that "out of the box" faster experience. The apps would have to be a) re-written or b) recompiled in order to call the additional functions.

2. Any programs written in year 2001 before the release of the XScale would not have known about the additional instructions within the operating system or chip.

Most of the goodies and baddies within the XScale architecture are accessible and callable from the apps. The OS doesn't block them. Even if the Operating system were optimized for the new processor, the apps would have to be re-written anyways. For those instructions calls which run slower on the XScale than the StrongARM, the apps developers are the ones who need to remove them and replace the offending code with a better work around. Not the operating system.

Blame Intel, not Microsoft for the performance issues. And take the extra $200 -$300 (plus longer battery life) and run.

[pixelator]

Originally posted by jeff

Again, I say that nobody forced these people to buy high cost technology. If you is compelled to be an early adopter, then you will be paying top dollar, and that is the price of your compulsion. This point has nothing whatsoever to do with XScale performance.

The point YOU are missing is that these devices are tagged as having faster processors when they in fact are not faster. That includes the OEM advertising AND Microsoft. It's misleading, even if you disregard the XScale support issue.

b

[pixelator]

Originally posted by Digital-Doc
>>Palms aren't snappier.<<
I think that Palm's are indeed faster depending on the application of course.

Which should illustrate to you that it's the app, not the Palm, that's faster or slower.

Anyway, this is REALLY OT.

b

[toshtosh]

I choose to identify it as the 250 Mhz processor, others simply say, its not behaving like a 400 Mhz processor. To each, his own.

You have a strange way of looking at things .

Do you think of the Pentium4 1800MHz as an Athlon running at 1232.5MHz? Or you could say that the PXA250 is a Pentium3 running at 122MHz!

As a engineer, I feel that it's not correct to identify it at as a 250MHz processor since that's simply not true. Why not just say that "a 400MHz Xscale processor runs Arm V4 as fast as a 250MHz StrongARM"?.

PS: I wish that were true .

[Bandung]

To say that the XScale is slower than the StrongARM is just not true. Most apps do run faster. Not 2 x as fast but faster.

There are some instructions and certain routines which execute slower but Intel has provided good work arounds. When implemented at the application level, the few apps which saw a performance decrease actually run significantly faster. Witness Pocket TV. If one downloads the latest code, this "issue" about the XScale being slower, doesn't exist.

Who broke "performance compatibility" with a given instruction carried over from the StrongARM than now runs slower on the XScale? Intel or Microsoft?

The other problems were created by both the hardware developers as well as Intel. Take the point about the difference in core clock speeds vs external memory clock speeds.

Quote
--------------------------------------------------
2. CACHE MISS COST
The Intel XScale core performance is highly dependent on reducing the cache miss rate. Note that this cache misss penalty becomes significant when the core is running much faster than external memory. Executing non cached instructions severely curtails the processor's performance in this case and it is very important to do everything possible to minimize cache misses.
----------------------------------------------------

Want the XScale to perform faster? Clock external memory faster. Not at 51 Mhz.


The core problem with performance rests with Intel's tradeoffs. But the tradeoffs were in fact good ones. Better battery life and significantly lower costs in return for modest 30% performance improvements.

[toshtosh]

I'm not sure if that was directed towards me, since it doesn't seem relevant.

But, I'll comment that I don't think memory is clocked at 51MHz, I think it's running at 100MHz on our units.

To say that the XScale is slower than the StrongARM is just not true. Most apps do run faster. Not 2 x as fast but faster.

Not true, same or slower on average.

[Bandung]

The fault is not with the OS. Its with the processor. And the processor can be made to perform "if the apps are re-written"

In every case where the code was re-issued, the apps ran faster. Code is code. Wether you want Microsoft to re-write their entire operating system code base and developers tools or you want the Apps Developer to do his bit. That fact remains.

Better performance = software rewrites.

The XScale is capable of delivering some performance improvements over the StrongARM. So start typing, somebody.

toshtosh, can you find a specific reference for your device to support the fact that the data into and out of external memory is clocked at 100 Mhz? (Even Chris Herrera was wrong on memory clock speeds for specific StrongARM devices. He was assuming and guessing and not checking spec sheets - which is a little more work.)