(In reply to Kevin Brace from comment #0)
> Of course, Pentium III is pretty slow by today's standards (even too slow
> for Lubuntu), but that being said, in terms of 2D rendering latency, I felt
> like the XAA-based r128 x.org DDX device driver had less latency than the
> rewritten EXA-based r128 x.org DDX device driver.

That essentially means pure-software rendering had less latency than the EXA-based code. With the XAA code, I saw how many operations were actually performed and it was something tiny like a few per minute.

> That being said, I felt like the rewritten EXA-based r128 x.org DDX device
> driver had better block drawing performance when a window is being closed,
> so in some ways, something got better but other things got worse.

It would seem desirable to disable some aspects of EXA but not all of them. Some options for this are in "man exa" but I think only Option "EXANoComposite" "true" will have an affect.

>     I was curious as to whether or not the rendering performance / latency
> is really CPU dependent, so I decided to test the EXA-based r128 x.org DDX
> device driver with Intel Pentium 4 2.53 GHz, ECS ECS L4S5A/DX+ mainboard
> (SiS 645DX north bridge and SiS 962 south bridge), ATI Technologies Rage 128
> 16 MB AGP.
> This is the same computer used for Bug #91346.
> Note that Pentium 4 2.53 GHz is probably the fastest processor I can mate
> with an AGP 3.3V supporting graphics card (Rage 128 does not support AGP
> 1.5V signaling, hence, no AGP 4X mode support).

That's good to know... thanks!

> According to dmesg log located under /var/log/dmesg (cat /var/log/dmesg), my
> Rage 128 AGP is running at AGP 1X mode, not 2X mode it supposedly supports
> (I did not use any manual setting to force it into 1X or 2X mode.).
> AGP 1X mode effectively means that AGP can transfer at 2X the speed of 33
> MHz 32-bit PCI.
> Even with such a fast system (compared to Pentium III 450 MHz), I feel like
> the rendering latency is noticable.
>     When I say rendering latency, this is particularly noticable when I move
> a window with a mouse.
> When a window is moved across another window, the rendering to cover up the
> reminants is definitely noticable, even with a Pentium 4 2.53 GHz computer.
> I am wondering if this is occurring due to EXA or the lack of 2D accelerator
> hardware use.

Comparing the speed between EXA and software (or XAA) should be done with benchmarks. Look into "x11perf". The test where EXA was the clear winner was "aa10text" but maybe there are some where it loses.

>     Regarding the Linux distribution I used, for Intel Pentium III 450 MHz
> and Transcend TS-ABX mainboard computer, it was Lubuntu 10.04 and Lubuntu
> 12.04 with 384 MB of RAM (PC100 SDRAM).
> For Intel Pentium 4 2.53 GHz, ECS L4S5A/DX+ mainboard computer, it is Ubuntu
> 10.04 with 768 MB of RAM (DDR266 DDR SDRAM).
> I will install Lubuntu 10.04 and Lubuntu 12.04 to this computer shortly for
> testing purposes.
> I do not have access to the specific Pentium III-based computer used for
> this bug report for many months, but I do own several comparable Pentium III
> computers, so I can easily replicate the environment.
> 
> Regards,
> 
> Kevin Brace

I seem to be the only one CC'd here anyway.

*** This bug has been marked as a duplicate of bug 88767 ***