Back in the day I used PowerStrip to tune my FPS in games. I remember RivaTuner and discarding it, but I could not remember why.

Today, in front of me is a Radeon 9200 with 4ns VRAM rated to 250MHz, which the VBIOS clocks at 200MHz.

So, naively, I used RivaTuner to set 250MHz. Nothing much happened. Obviously, what I observed is that 3DMark performance generally scales with MHz set using RivaTuner but its not linear. Furthermore, what surprised me is that every frequency set by RivaTuner produced lower FPS than the stock 200MHz.

Anyway, my nagging old memories made me dig out an old copy of PowerStrip. The results are very different. I stupidly went straight to 250MHz because 250MHz had already been exercised in RivaTuner with no artefacts. And, what surprised me is that with PowerStrip artefacts immediately appeared (quickly ESC that).

Now for the control test: What happens if RivaTuner is used to manually set the same default 200MHz frequency? Why would anyone do that? Anyway, my test today shows the frame rate performance drops by almost 15%!

As the RivaTuner set 200MHz performs significantly worse than the VBIOS set 200MHz, my hypothesis is that although RivaTuner can set higher frequencies, it can do so only because it is triggering looser CL timings that worsen performance and lower FPS. This feels entirely consistent with my memories of tuning game performance decades ago. What I could not see then, and that I do see now, is that degradation is not linear.

Radeon 9200 128-bit, Catalyst 4.1, 3DMark2001SE

VBIOS
250/200 -> 6861

RivaTuner
250/250 -> 6843
275/250 -> 6767
230/250 -> 6747
275/200 -> 6722
275/220 -> 6716
250/200 -> 5917
(other frequencies tested and they did not seem interesting enough to write down)

ATi Tools
262/210 -> 5921 (5929 with clock lock disabled, but not)
(consistent with RivaTuner results and alert prompt suggests Catalyst applies a safety mode, see next post)

PowerStrip
275/250 -> cancelled/artefacts

More testing needed, but can someone please help bring me up to speed on where this has been explored before?