Tungsten T3 benchmark results baffling

[falafel]

I'm using speedy 3.0 to benchmark my tungsten T3 and it seems that the Tungsten T3 is almost 25% slower than the Tungsten C! Can anyone confirm this?

[Vidge]

That would make sense, as the T|C uses an Intel processor and the T3 uses an XScale processor. I've heard in many places that the XScale processor is not as fast as the Intel one.

And while I haven't tested it myself, I had a T|C for a short while & it ran my programs faster than my T3 does. :(

[slinger]

Originally posted by Vidge
That would make sense, as the T|C uses an Intel processor and the T3 uses an XScale processor. I've heard in many places that the XScale processor is not as fast as the Intel one.

I don't mean to be rude, but the above statement is almost completely incorrect.

First, Intel makes the Xscale processor series so there is no Xscale vs Intel difference. Both units use Intel Xscale processors - the Tungsten C uses the PXA-255 at 400 MHz, and the Tungsten T3 may use either the PXA-255 or the PXA-263 at 400 MHz.

As for the performance difference between the two models, I have definitely noticed (as well as benchmarked) the 25% difference. It seems like the Tungsten T3 is running at closer to 300 MHz *most* of the time. I have seen some instances where the T3 benchmarked at exactly the same performance as the TC, but usually the T3 is slower. I'm not sure what is going on here. It may have something to do with the Xscale's built-in power management, but I haven't had a chance to poke around in the internals and figure it out yet.

The good news, however, is the program "PXA Clocker" works with the T3, and I was able to use it to overclock the T3 well past the performance of a stock TC!

-Robert

[Univil]

Actually, XScale is an Intel processor. From the palmOne website: "Handle spreadsheets, wireless applications and multimedia tasks with blazing speed and boundless performance with the Tungsten™ T3 handheld’s ultra-fast 400MHz Intel® XScale processor". XScale isn't a brand, it's a model. I'm not sure if the T3 is slower than the T|C. If it is, it would probably be because the screen is larger on the T3, so the processor has more work to do.

-Roberto

[falafel]

Univil.. I'm not sure that screen size makes the difference, because the benchmark test I conducted was with the Tungsten T3's slider closed..

[Univil]

Originally posted by falafel
Univil.. I'm not sure that screen size makes the difference, because the benchmark test I conducted was with the Tungsten T3's slider closed..

I remember reading somewhere that the larger screen size might be the cause of some slower benchmarking. I believe that, although the bottom of the screen goes blank when you close the T3, it doesn't actually shut that part of the screen. I think it may have been on the T3 review right here at PalmOne City.
I'm no expert on the subject though. It's very possible that I'm wrong.

-Roberto

[ztupid]

That's correct, the screen goes black, but you still see a white rectangle on the edges.

[Vidge]

Originally posted by slinger
I don't mean to be rude, but the above statement is almost completely incorrect.

First, Intel makes the Xscale processor series so there is no Xscale vs Intel difference. Both units use Intel Xscale processors - the Tungsten C uses the PXA-255 at 400 MHz, and the Tungsten T3 may use either the PXA-255 or the PXA-263 at 400 MHz.

-Robert

I am no engineer so I only repeated what I have read and heard elsewhere. Regardless, the processor in the T3 is apparently not the same as the TC and runs slower. :(

I'm looking forward to overclocking my T3 in order to get it as fast as the TC is (once I get some other things sorted out :()

[foghead]

Actually, based on my tests with Speedy, the T3 gives virtually the same results as my TC.

However, the T3 appears to slow down with use - maybe memory leaks or something. When I run the tests after a reset of the T3, the results are the same as the TC. If I run the test after running other software, using it for a day or two, etc., the T3 is signifigantly slower.

On the version of Speedy that I am running (v2.6 - not the latest), the TC gives a value of 5556. The T3 after going a few days without resetting just gave 4286. After a reset, the T3 gives a 5357. This is less than 2% slower. Two percent is basically the same speed.

[Toffin]

With Speedy 3.0, just after a reset Speedy gives 375MHz, and after Off/On Speedy gives now 300MHz :(

[brolewis]

I used a T|C and switched to the T3 and do feel a differerence in overall performace. I have come to believe that this is caused by the increased screen size, and I'll take a little bit slower feel for a larger screen any day :-)

[hotpaw4]

There seems to be some sort of bug where a background task is eating up 25% of a T3's performance after a power-cycle, but not after a soft reset or use of the voice recorder to record. So please recheck all your performance measures after a soft reset, but without letting the unit shut-off, and see if that makes a difference. If your unit slows down, try voice recording for a second or two.

[brolewis]

while this may improve benchmarking, what does it mean for the end user? We're stuck with a slow-down in performance unless we always soft reset?

[hotpaw4]

Or record a 1 second voice memo.

I'm just guessing that there will be a patch for this eventually (either from PalmOne or from a 3rd party).

And just to be on the super safe side, I might also want to soft reset before running any battery life tests, or before writing data to an expensive SD card.

[brolewis]

I hope a patch comes sooner than later. That seems to be a bit annoying

[hotpaw4]

In fact here it is! (3rd party beta version it looks like):

http://www.escribe.com/computing/pcpqa/m80264.html

[Vidge]

hotpaw4 - have you tried this yet?

[hotpaw4]

Has anyone else tried the above patch for the T3 slowdown problem yet? I'm curious about whether this affects the battery life problem also.

[foghead]

I tried it. Now I am getting the same (fast) Speedy results before and after cycling the power. I also verified that the slow down was happening after cycling the power before installing the patch.

It can't hurt and might help power consumption.

[hotpaw4]

Make sure to write the author of the patch and tell him. I think he might be gathering statistics (today 10-18-2003 GMT)

[n2ifp]

It worked for me!:)

[brolewis]

Patch works great for me. I've noticed a difference :)

[tanker_bob]

Patch worked great for me. I get the same results in Speedy 3.0 after a soft reset and again after power off/on. Super job on this.