Can you give ATI Tool a try as well? In the past I'd clock those cards with neither of those tools, but ATITool

tehsiggi wrote on 2026-08-20, 11:05:

Can you give ATI Tool a try as well? In the past I'd clock those cards with neither of those tools, but ATITool

ATi Tools gives an alert prompt, "These cards are locked against overclocking when using original Catalyst drivers"

That is misleading because we see change in MHz resulting in change in FPS but clearly ATi are trying to dissuade overclocking.

Without mitigation: I set 262/210 using ATi Tools and GPU-z read back 262/210 numbers. Whether the core and memory executed at the ATi Tools-set 262/210 or Catalyst-set 250/200 is besides the point. The point is, the actual output (5921 3D Marks) is lower than when VBIOS set 250/200. With mitigation: Disabled the clock-lock in ATi Tools and it read back 262/210 (5929 3D Marks).

Does anyone have a copy of the standalone ATi Radeon Softmod please? I have the Omega drivers but they are unsafe in a system that frequently swaps between different Radeons.

Other oldschool options (shields up on this site and only click the DL link with the military theme) https://www.majorgeeks.com/files/details/rage3d_tweak.html

Aside from memory timings, I also recall there were sweet spots between relative core and memory speeds, where you could get things happen like the benches jumping 10% by dropping RAM speed a couple of clicks, or finding the next "magic number" upwards.

ATi Tray Tools is also an option.

BitWrangler wrote on 2026-08-20, 13:50:

Other oldschool options (shields up on this site and only click the DL link with the military theme) https://www.majorgeeks.com/files/details/rage3d_tweak.html

Aside from memory timings, I also recall there were sweet spots between relative core and memory speeds, where you could get things happen like the benches jumping 10% by dropping RAM speed a couple of clicks, or finding the next "magic number" upwards.

When I investigated GeForce256 VBIOS I found in there multiple tables for memory configurations, including SDR and DDR tables in the same VBIOS. What pointed the firmware to read a particular table was an electrical signal from the PCB.

I have not looked into the Radeon firmware but I anticipate something similar. There might be one optimised 128 bit retail table, one optimised 64 bit table, a developer debug table, etc. There could be tables for other cards in there. When we change the frequencies, we might also be changing the electrical signal that is polled by the VBIOS when it reinitialises.

The Serpent Rider wrote on 2026-08-20, 15:05:

ATi Tray Tools is also an option.

I also found Raidon and Radtweak. There are others too, and they might not all use the same approach to set clocks.

Hmm.. a lot more tests to do down this rabbit hole.

Random development. I found a different Gigabyte Radeon RV280 board in the bits box; same silicon but assembly is visibly minimal (fanless, less fibreglass, fewer outputs).

Head-to-head between Blue Gigabyte 9264DH 128bit 4ns (I think 64 just means 64MB) and Red Gigabyte 9200NF 128bit 4ns (basic build):

VBIOS
Blue 250/200 -> 6861
Red 250/200 -> 5921

RivaTuner
Blue 250/200 -> 5917
Red 250/200 -> 5936 (slight anomaly, might be noise)
Red 262/210 -> 5909 (consistent with the pattern seen before)

Gigabyte clearly set more aggressive default memory latency timings on the blue DH board compared to the red NF board, and it looks like Gigabyte and/or ATi used a similar technique to throttle overclocking more generally. I should be able to dump both VBIOS and diff them; I did wonder if it's safe to simply flash the NF with DH VBIOS but I assume not because there is an extra choke on the DH - clearer signals at lower voltages?

Also, with NR or RivaTuner timings it looks like matching DH's 6861 3D Marks would need a hypothetical ~290MHz core frequency; that would be risky.

I've made small progress.

Gigabyte's own overclocking utility, V-Tuner, has a kernel driver that interfaces directly with the GPU independently of ATi video drivers.

Method: Other utilities (ATi Tray Tools, PowerStrip, ATi Tool, RivaTuner, etc.) attempt to set Core/Memory clocks and V-Tuner reports back the actual current state directly from hardware (GPUz reports the last request, not the actual current state).

Good news: The actual MHz reported by V-Tuner align very well with the actual 3D Marks observed. I like it when things make sense.

Bad news: It's clear the other utilities do not understand RV280. Attached screenshot shows ATi Tray Tools attempting to increment the core frequency (279MHz at PrtSrc) while V-Tuner shows the actual core frequency dropping (232MHz at PrtScr); memory frequency dropped off the bottom the scale while FPS in the test window dropped to S3 ViRGE levels!

Moving forward: It appears to me that there is a frequency divider table on the Radeon board that is mapping the requested frequency onto another frequency. Decoding that is more work that I anticipated.

Why not simply use V-Tuner sliders? They also don't work, but page 20 of the Gigabyte R9264DH manual says clearly V-Tuner overclocks the Radeon 9200 series. Maybe I've installed a wrong version of V-Tuner.. a period thread suggests yet another utility called Radclock may work.. more stones to turn over..

MattRocks wrote on Today, 18:45:

I've made small progress. […]
Show full quote

I've made small progress.

Gigabyte's own overclocking utility, V-Tuner, has a kernel driver that interfaces directly with the GPU independently of ATi video drivers.

Method: Other utilities (ATi Tray Tools, PowerStrip, ATi Tool, RivaTuner, etc.) attempt to set Core/Memory clocks and V-Tuner reports back the actual current state directly from hardware (GPUz reports the last request, not the actual current state).

Good news: The actual MHz reported by V-Tuner align very well with the actual 3D Marks observed. I like it when things make sense.

Bad news: It's clear the other utilities do not understand RV280. Attached screenshot shows ATi Tray Tools attempting to increment the core frequency (279MHz at PrtSrc) while V-Tuner shows the actual core frequency dropping (232MHz at PrtScr); memory frequency dropped off the bottom the scale while FPS in the test window dropped to S3 ViRGE levels!

Moving forward: It appears to me that there is a frequency divider table on the Radeon board that is mapping the requested frequency onto another frequency. Decoding that is more work that I anticipated.

Why not simply use V-Tuner sliders? They also don't work, but page 20 of the Gigabyte R9264DH manual says clearly V-Tuner overclocks the Radeon 9200 series. Maybe I've installed a wrong version of V-Tuner.. a period thread suggests yet another utility called Radclock may work.. more stones to turn over..

I'm a bit surprised, as I have never experienced the issues you described here. I've overlcocked quite a bunch of 9200s (even back when they were new) with ATITool, and did not see what you noticed (default value OK, OC will reduce performance unexpectedly).

I have a bunch of 9200s laying around, incl. the mentioned gigabyte card as well. Perhaps I'll give it a try next week.

I am btw. not sure how much they really rely on the kernel driver for overclocking. I think they might just need it to directly access the I2C bus on he Gigabyte card, as it's using a Winbond monitoring chip that's attached to it.

tehsiggi wrote on Today, 19:37:

.. not sure how much they really rely on the kernel driver for overclocking. I think they might just need it to directly access the I2C bus on he Gigabyte card, as it's using a Winbond monitoring chip that's attached to it.

Many users in the past started threads suggesting V-Tuner is only good for monitoring and does not overclock. The installer I used has a bug and I needed to manually create the service that loads the *.sys driver the installer dropped into Windows/System32/drivers/ and after that the monitoring features worked consistently (with or without video drivers) and the sliders for voltage rails work, but so far my Core/Mem sliders have not been user adjustable.

Gigabyte FAQ suggests the overclocking is enabled by using the drivers on the CD. All my tests were done with bare bones generic INF from an extracted ATi Catalyst 4.1 installer that probably originated from a magazine CD. I did not run the installer so I have no Catalyst control panels and related registry settings.

I did try a Gigabyte-distributed INF from a CD via archive.org but that caused my screen to blank and I then reverted to Catalyst 4.1 drivers. I now know the blank was caused by the display drivers defaulting to the max resolution reported over EDID - not sure why the max EDID resolution failed but it was humongous. Now that I know why the driver failed, I can retry that.