> Also note that the specific video driver you use
> can drastically affect video bus performance (by a factor of 5x in some
> cases!).
Some more info on this. I recently installed one of the new-fangled
latest-crop high-end 3D lots of VRAM graphics cards to do some testing
on it. I downloaded the latest drivers from the manufacturer's web site.
At some point we wrote a micro-benchmark which did 8-bit, 32-bit, and
memcpy copies of system memory to the display and the card gave us a
measly 44MB/sec transfer rate. This seemed to concur with results we
were getting from GDI and 2D. I chalked this up to a hardware design
that catered to 3D game performance over direct pixel access performance.
But, then I ran the tests under Windows 98 where I simply installed the
drivers from the CD that came with the card and didn't want to bother
with downloading the latest from the web site. Suddenly the card was
running a 220MB/sec transfer rate and the 2D applications were running
noticeably faster. I went back to NT and reinstalled the drivers from
the CD and the times on NT also went back to about 220MB/sec.
JDK 1.1.8 performance measurements were about the same with either
driver.
So, a 5x performance difference in blit speed due to drivers? It was
hard to believe, but that is what I saw. Swing performance was improved
as well, but not by the same 5x factor since it is doing a lot more than
simply blitting the offscreen buffer to the screen.
With the faster drivers, blit performance of system-to-VRAM copies was
only about 4x slower than VRAM-to-VRAM performance (220MB vs. 850+MB).
With the slower drivers, there was a 20x difference (44MB vs. 850+MB).
Was it due to a change in the tuning philosophy between the 2 drivers?
Was it due to a botched installation or download of the slower set of
drivers? I would need to do more testing to be sure, but the slower
numbers were consistently reproducible before I reinstalled the drivers
and the faster numbers were consistently reproducible afterwards. I
don't want to mention the particular card until I can be more sure of
what was going on...
...jim
===========================================================================
To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
of the message "signoff JAVA2D-INTEREST". For general help, send email to
[EMAIL PROTECTED] and include in the body of the message "help".
