XScale-Pocket PC marriage lacks oomph

[Ed Hansberry]

<a href="http://zdnet.com.com/2100-1103-939365.html">http://zdnet.com.com/2100-1103-939365.html</a><br /><br />ZDNet has picked up on what we've known for a week or two. "A generational mismatch between Microsoft's Pocket PC 2002 operating system and Intel's XScale PXA250 processors included in new handhelds is resulting in devices that don't deliver all their potential performance, according to Microsoft."<br /><br />There is not much new here, other than discussions here and other PDA sites have garnered the attention of the mainstream computing press. <img src="http://www.pocketpcthoughts.com/forums/images/smiles/icon_smile.gif" />

[Arne Hess]

Also the German online press links to Thoughts :D (news from yesterday):

Original: http://www.golem.de/0206/20488.html
Google Translation: http://translate.google.com/translate?sourceid=navclient&hl=en&u=http%3A%2F%2Fwww%2Egolem%2Ede%2F0206%2F20488%2Ehtml

[PJE]

This issue seems to be done to death:

1. XScale has 100MHz bus and StrongARM 103MHz

This means StrongARM can access memory slightly more quickly than XScale.

2. Current designs use 16-bit memory busses.

This results in the StrongARM being able to access memory at the same rate as the XScale which can handle double the data if available... ALL XSCALE DEVICES MUST HAVE 32-BIT BUSSES!!!!!

3. Toshiba 740 has external ATI Imageon graohic chip, which is slower at some MultiMedia than embedded graphics.

This may be due to the eneficiencies in the device driver and/or increased conjestion on the already over taxed 16-bit bus.


Conclusion.

I don't blame Intel (Or Microsoft) for the problem. But anyone designing these devices should have realised that the memory bus was probably limiting the StrongARM at 206MHz, so why think it would work at 400MHz!!!!

Redesign the 740 to use the 32-bit bus and watch the thing fly....


My 2c

PJE

[jpaq]

Yes.

[Scott R]

There is not much new here, other than discussions here and other PDA sites have garnered the attention of the mainstream computing press.
That was the same thought I had when I read this yesterday. Of course, the mainstream press will rarely ever give credit where it's due. Instead, they cite their sources as being the generic "industry analysts" and other such nonsense.

Scott

[Ed Hansberry]

This issue seems to be done to death:

1. XScale has 100MHz bus and StrongARM 103MHz

This means StrongARM can access memory slightly more quickly than XScale.
True, but why does the HP 720 run as fast as an iPAQ? They both have the 206MHz StrongARM, but iPAQ's have the 103MHz bus and the 720 has a 51MHz bus. I am not so sure anymore that the bus alone is the culprit in any X-Scale device - not to the extent we aren't seeing performance.

[vetteguy]

I think the lack of XSCALE optimized code is as much to blame as the system bus. If PPC2002 was recompiled to natively support the new extensions, I think we'd see a boost right away,

[DJR]

There is not much new here, other than discussions here and other PDA sites have garnered the attention of the mainstream computing press.
That was the same thought I had when I read this yesterday. Of course, the mainstream press will rarely ever give credit where it's due. Instead, they cite their sources as being the generic "industry analysts" and other such nonsense.

Scott

Well, credit where credit is due. It was silicon.com that "broke" the story as far as I recall.

That being said, major kudos to PPCT for getting the interview with Ed from MicroSoft :)

[PJE]

True, but why does the HP 720 run as fast as an iPAQ? They both have the 206MHz StrongARM, but iPAQ's have the 103MHz bus and the 720 has a 51MHz bus. I am not so sure anymore that the bus alone is the culprit in any X-Scale device - not to the extent we aren't seeing performance.

Do HP use a 32-bit bus on the StrongARM in the 720? This would result in the same data transfer rate as the iPaq (and hence performance) even though the bus runs at half speed.

If this is the case, it shows how important the 32-bit bus is...

Regards,

PJE

[Ed Hansberry]

Do HP use a 32-bit bus on the StrongARM in the 720? This would result in the same data transfer rate as the iPaq (and hence performance) even though the bus runs at half speed.

If this is the case, it shows how important the 32-bit bus is...
Something else - onboard processor cache. Does the SA1110 have one? Is it enabled on the iPAQ? On the J720? How large? Is it on the X-Scale? same questions -enabled? size?

Remember how horrid the initial Celeron chips were sans cache? Once they added a cache to the 2nd generation, they were only slightly slower than their Pentium II/III counterparts in most things.