I was curious about tweaking PCI Latency to speed up my Reactor
card, so I did some experiments. PCI Latency is related to the amount
of time a device can use the PCI bus -- higher values allow longer
transfers (picture one long uninterupted transfer, instead of several
smaller ones [and the overhead involved]). I've got a Supermicro
P5STE with a P133, 32 meg EDO, and Reactor. The P5STE/AMI BIOS allows
you to set PCI Latency in increments of 32 PCI clocks from 32 to 248.
The default value is 64.
I altered PCI Latency and did my high-load test in NASCAR2's
RENDDMA (full graphics, cars ahead/back, etc.) but didn't see any
change. Most likely the resulting frame rate is so low (13-14, and an
integer value), that it would be hard to see a small change (< 5-10%).
So, I decided to use ICR2-3D in Turbo mode to run at much higher frame
rates, perhaps a more sensitive indicator of performance.
For the test, I ran INDYFAST in DOS7 in Preseason Testing mode at
the old (8-bit) Michigan track. All graphics are turned on, frame
rate indictor (Alt-R) is on, turbo mode (Alt-T) on. Two points were
selected for reading frame rate: when the pit board was displayed and
on the back straight.
Here are the results...
PCI Latency Pit Board Back Straight DMADiag DMA Results
32 34-35 fps 46-47 fps 79.1 / 79.5 m/sec
64(default) 39-41 fps 51-52 fps 97.5 / 98.1 m/sec
96 39-42 fps 51-52 fps skipped :(
128 39-42 fps 51-52 fps 109.2 / 109.2 m/sec
248 39-42 fps 51-52 fps 115.4 / 117.9 m/sec
Gee, is this a new record? ;) ^^^^^^^^^^^^^^^^^^^
Very interesting. DMADiag keeps climbing, but the frame rate
levels out very quickly. I would guess that the Verite is maxxed out
around 100 m/sec, so any further increases in DMA speed do not
translate into increased frame rate. Comments from more technically-
qualified parties are welcome...
Also, I have stressed how important the usability of DMA mode is to
the Verite boards. In my case, my DMADiag FIFO is ~21 m/sec, so the
difference between the default 98 m/sec of DMA mode *is* a big deal.
For those of you with closer FIFO/DMA readings, you will probably find
a less substantial benefit with DMA (ie: good, but not crucial).
Finally, you guys should try out ICR2-3D in Turbo mode. Not very
realistic, but a tremendous sensation of speed. I was running 235 mph
laps at Michigan in 21.5 seconds (335 mph!). The timing is just like
running a mile oval, except you're going twice the speed. NASCAR2
arcade mode ain't got nothin' on us! ;)
Pick one or more: Model Rockets (competition-NERCB) / PCs (even Atari!) /
Papyrus ICR-ICR2-NCR / Who needs a life when you have multiple non-lives?
I did roughly the same tests on my machine (ASUS P55T2P4, AMD K5-133,
Intergraph Reactor) with VQuake and noticed the same results. After
my DMA went over ~90MB/s, the framerate stayed the same. This seems
to support your theory that the Verite chip is the bottleneck, not the
software since both VQuake and Indycar reflect the same results.
Yep, good DMA is a must have. I noticed an increase of ~2 FPS
(640X480, TIMEREFRESH) in VQuake going from a DMA of 70MB/s to 90MB/s,
which is quite substantial, IMO. I haven't done any tests yet, but
I'd expect the performance drop to be linear for poor DMA speeds.
Gonna have to try that one. Contrary to other opinions, I've always
thought that normal mode provided a good sensation of speed,
especially the way the road 'blurrs' at the front of the car at high
speeds. Try switching views to the front spoiler for the best effect.
>Pick one or more: Model Rockets (competition-NERCB) / PCs (even Atari!) /
>Papyrus ICR-ICR2-NCR / Who needs a life when you have multiple non-lives?
Thanks for the post,
Darin
Sorry if this is a repost, but my Newserver has been having troubles
lately...
I did roughly the same tests on my machine (ASUS P55T2P4, AMD K5-133,
Intergraph Reactor) with VQuake and noticed the same results. After
my DMA went over ~90MB/s, the framerate stayed the same. This seems
to support your theory that the Verite chip is the bottleneck, not the
software since both VQuake and Indycar reflect the same results.
Yep, good DMA is a must have. I noticed an increase of ~2 FPS
(640X480, TIMEREFRESH) in VQuake going from a DMA of 70MB/s to 90MB/s,
which is quite substantial, IMO. I haven't done any tests yet, but
I'd expect the performance drop to be linear for poor DMA speeds.
Gonna have to try that one. Contrary to other opinions, I've always
thought that normal mode provided a good sensation of speed,
especially the way the road 'blurrs' at the front of the car at high
speeds. Try switching views to the front spoiler for the best effect.
>Pick one or more: Model Rockets (competition-NERCB) / PCs (even Atari!) /
>Papyrus ICR-ICR2-NCR / Who needs a life when you have multiple non-lives?
Thanks for the post,
Darin
