TerraRaptor

Pretest or cpu limited with 2x4GB ram?

Benchmarks kill themselves. No legacy 3d is VGA benchmark for a long time. Some 2d is also subject to inconsistencies etc. So some of these "golden era" benchmarks need globals/boints removed. All this drama is not about benchmarks;) You are talking about boints:) If you like benchmark you will bench it no matter how points are assigned - database is up and running so no recent changes will affect your ability to submit. Take boints easy. It is not what hwbot is about.

PM @ground, he should be able to give you good input

Is it measured 2Ohm with cpu in socket? Normally, the value without cpu should be higher.

There was a tool named GMAbooster that was able to control/monitor "multi" of iGPU (not sure if it applies to g31). Actually, the clock should be locked with desktop chipsets and is independent of FSB. The boost we get with cpu clock is probably because some of the 3D functions are done through software path (like software vertext shaders in earlier GMA generations). I was benching 945g+fatbodies recently, chipset is extremely sensitive to voltage and doesn't allow tight timings though mem itself is capable of tight timings if I use external VGA. Sadly, 2.5v vNB was too hot to get golds so I gave up.

Great clocks

Haven't seen new scores from you so long, glad you are still in overclocking:)

Pifast is always such a drama with 775)

Score needs editing

Serial initialisation packet contained in bios for every cpu multi - a set of bclk timings that affect performance and stability.

strong!

GFP, yet you shouldn't be the happiest person - just 0.003fps away from breaking 1fps. Great score:)

Not a surprise, well done.

Sick fsb bro:)

Good job!

Is it 4-dimm action?

I guess that will stop someone with another good tweak from sharing it (and tweaks are available for every benchmark). All that is needed then is to have just 0.1-0.2% advantage in efficiency to avoid questions - use another uber-tweak, use some sub-optimal settings and there you go. Why my score is that efficient? Oh, just some mchbar tweaking. Oh, just some subtimings tighter. I bet we won't see another waza tweak in public then (and wazza was introduced in 939 era if i'm not mistaken, so all results before that became zero value efficiency-wise). As some 3dmark01 (?) tweaks are not available etc. PS. And we also have MSI ABX that is NDA software giving advantage to a very limited number of ocers. And LN2 bios versions that are not publicly available. It relates to hardware tweaking but the idea is the same - tweaks not exposed to public. Sadly, it becomes another story about boints and not about true spirit of overclocking (that is getting it faster, higher, stronger).

Is tinyXP still xp if it has compatibility problems with games etc? I mean, not a rare case when msdn based XP works fine and those nlite mods don't. Yet, tinyxp is allowed here (to the best of my knowledge). Also, that is very difficult to replace any essential system dlls - ole32, kernel32 etc won't work even if you inject these from 2003 to xp. Another thing to consider then - official beta releases of OS. Hundreds of them - win2k sp0-sp4, winxp sp0-sp4 (yep, sp4 not official but still being used by many), Longhorn betas (those that are more on xp side and those that are more on vista side) etc etc. Should we disallow windows for Embedded if we bench desktops? Just imagine newcomer that uses daily OS with third party software to update system components from 3rd party repositories? Overall, my motivation was to match efficiency of some guys in socket 478. Believe me or not, even with that tweak and after 500GB of downloaded windows ISOs I'm still behind. That is the main reason to share good tweak - everyone should have a chance to compete on a level ground. Probably if everyone would share his knowledge, we can beat those guys or expose a real cheat.

Installing 3d drivers is also enabling compatibility with 3d:) we can nlite xp to exlude VB engine - wprime won't be able to run, incompatible with given OS setup as vb runtime is missing. Then install vb runtime to enable compatibility - voilà, same tweak done different way. Take GPUPi for CPU. It requires OpenCL driver. OpenCL driver is not a part of OS, not a part of benchmark. It won't run without it. Or you may update it to a different version. That opencl driver is just the same shit for gpupi as is vb runtime for wprime.

I remember am3 problem was different, not swapping anything? Result was shown as part of 3d, no desktop etc - copy/paste it into paint and add cpuz/gpuz. Same about 3dmarks - altering dlls in benchmark directory was always forbidden as they are integral part of benchmark. But nobody says altering 3d driver is illigal - not a part of benchmark.

Then ban all superpi runs in xp on modern hardware?

Tweaking is adjustment of benchmark environment. OS, drivers, runtimes, APIs, ACPI tables - everything is more or less different for everyone who is more or less involved into tweaking. This tweak is obviously less questionable than probably using Geek 3.4.2 vs Geek 3.4.4 for example (ehr, not sure what Geek3 gives better score?). As soon as we have a single allowed driver version for given 3D card, as soon as we can validate OS installation hash is the same for every bencher - I would then proceed with banning this tweak. It is in the wild, shared with the community and available for validation by everyone. Not one of those "secret" tweaks that are still unknown to comminity yet being used by their inventors.

Cool score!

Красава!

I'm sure Leeghoofd sent him a warning that inefficient results like previous one will lead to lifetime ban.