Some of the performance tests performed during the evaluation of bug 4394782
suggest that video cards with little or no vram performe much worse in jdk1.4
than in jdk1.3. Following are some of the results from that bug:


Pentium 166MHz, S3Trio 2MB
                     8bpp 16bpp 24bpp
-----------------------------------------------------------
1.3.0 ~19fps ~40fps ~22fps
Merlin b42 ~80fps ~6fps ~4fps

Pentium 750MHz, ATI 8MB
                     8bpp 16bpp 24bpp 32bpp
----------------------------------------------------------------------
1.3.0 ~89fps ~100fps ~90fps ~90fps
Merlin b42 ~250fps ~240fps ~27fps ~150fps

Tested on PIII-800 with Intel i810 graphics Windows NT4.0 machine
Results:
   jdk 8bpp 16bpp 24bpp
------------------------------------------------
jdk1.3.0fcs 170fps 103fps 69fps
jdk1.3.1-b15 135fps 83fps 60fps
jdk1.4.0-b50 44fps 38fps 28fps

From these results, it looks like:
- The S3 card was able to perform hardware acceleration at 8 bpp. But
at higher bit-depths, it probably did not have enough vram to support
the back buffer in hardware and thus punted to software. These software
operations performed significantly worse in jdk1.4 than in jdk1.3.
- The ATI card had sufficient VRAM to support hardware buffers and thus
got hardware acceleration at all bit depths (except 24 bpp, which was the
cause of bug 4394782).
- The Intel i810 has, I believe, no VRAM (it stores all surfaces in system
memory and accesses them via AGP). So it is not able to accelerate
surfaces at any bit depth and thus goes through software for all operations.
It's performance relative to jdk1.3 is much worse in jdk1.4.

----------------------------------
This is a cross-platform issue: you can get the same dismal results
by running the test app on Solaris with the -Dsun.java2d.pmoffscreen option.

chet.haase@Eng 2001-03-01