havli

And what about HW that doesn't work with latest version of CPU-Z, GPU-Z, etc? Also what is wrong with posting older scores?

Yes, but most likely different clocks than you actualy have when running 3DMark. Also the rules say just "A verification screenshot is required, including motherboard and GPU sensor tabs ". I take it as - sensors must be present on the screenshot but doesn't matter when you run it and what it shows.

Hmm, why is GPU-Z sensors tab required for CC? Cooling methods are unlimited as far as I know, so not because of temps... this is not LCC, so no clock limit either. Most people (me included) open the GPU-Z after the run is finished... so there is nothing to see besides idle status of the GPU. I don't mind putting extra window on the screenshot, just don't see the point as it shows no relevant info.

I was developing the benchmark on SB-E platform and at the time I performed a lot of testing. AVX had zero inpact on performance when running on SB (Core and Xeon series - those do support AVX). You can try it yourself - either by using Windows 7 non-SP and SP1... or the built-in function of x265, the CPU features override (which is much more convenient). AVX2 on Haswell and later improves the performance a lot. AVX1 perhaps helps too on some architectures, but certainly not on SB/IB.

AVX makes no difference on Sandy Bridge (in x265)... SSE4 is the key.

Here is cpu-z validation link. http://valid.x86.fr/i37akj Thanks

Actualy I didn't try versions other than 3.2 and 3.3.2. Those behaved the same - appcrash. Thanks fro the tip, I'll try the 3.1 next time.

I think the singlecore stage will be very exciting... especially when you get to the point of having a great score and then you flip the coin whether you get the data file saved, or appcrash. I have a feeling there will be a lot of smashed keyboards.

GPUPI 3.x is a hit or miss it seems. Today I was running it on C2D and always appcrash when saving data file. On the other hand, A64 AM2 was running just fine http://hwbot.org/submission/3927537_havli_gpupi_for_cpu___100m_athlon_64_le_1640_(orleans)_8min_27sec_791ms

I'm not really sure... but isn't s939 too old to support HPET? AM2+ boards definitely support it, never tried it on older platforms.

This is weird. Do you have the same Windows install for both systems? Java version shouldn't matter - x265 is using built-in portable Java... in order to run as a portable benchmark without any special requirements on SW installed.

Buggy bios maybe? At least my UD3P is really pain to use, especially with K8 CPUs.

Thanks for the link, if simple standalone patch is sufficient, then it is worth trying.

Ok, I take it as no other benchmarks. You realy don't need to teach me what overclocking is about I'm here long enough to know a few tricks here or there myself. The thing here is to evaluate whether it is worth the extra time spent or not. In other words - what is better - quantity or quality? For example SuperPI 32M - we all know it is best to run in on tweaked XP (for most platforms)... but is it worth the effort to install XP just for one benchmark, when I'm running the other 20 on W7 64? For most cases it isn't. And the same applies here, on top of that the uncertainty of GPUPI future (what version will have points... 3.2, 3.3, 4.0... or which one will ger merged / deleted) might be discouraging people from putting too much work in it.

Yes, sure you are. Killing HW is not exactly proof of skill, is it? But i guess it is understandable considering you started overclocking just a year ago

@Mr.Scott That is very nice offer and I really appreciate it. For now I'll stay in my team. But if they won't change their attitude to team competitions for the next season, then the chance to join W9 is very tempting (if the offer still stands by then). @_mat_ Well, then show me another benchmarks that gain performance from this... besides GPUPI.

Cheapest? No, not for me anyway... that would be s478 or 754. Maybe some people don't have problem doing hardmods on $600 GPU, but I do. Simple look on my paycheck is discouraging enough. Also I don't like killing HW in the process of overclocking - I have thousands of scores here on Bot... and yet I can easily count components that died on my fingers. $800 NUC or dual Titan is no concern to me, because I don't have that kind of HW and never will. Also I don't really care about TC standings at all - one-man-team can never reach decent position in competition like this. Sure and then desolder the old chip and put new one in... doable of course. But not bricking the original one is far more time/money effective.

Well, it is simple evaluation of risks and benefits. Bios flashing is far more risky than OC. And since money doesn't exactly grow on trees (at least not here ) then it is better to get the most of it by focusing on the 20 other benchmarks that doesn't require such modifications.

No, but it is not needed for anything but GPUPI... so waste of time in my book. Also the risk of bricking the MB is always there if something goes wrong.

Or the other possibility - skip the FM1 stage completely... because it is pointless to run it at normal conditions and many people won't bother flashing bios to get this tweak working.

I also saw this error from time to time on older platforms, but it seemed to be random and eventually x265 started properly. When the error message shows up, there should be file named hardware.txt in the x265 working folder if you don't dismiss the error (the file is deleted automatically after succesful detection or benchmark exit). Please upload it here and I'll take a look. Btw - 64-bit windows should be faster for x265. XP is good for 32-bit only CPUs... as far as x265 is concerned.

I agree, Strunkenbold is doing a great job to keep the bot running. Sometimes when problem like this CPU-Z situation shows up... action simply must be taken. And when we don't know if the scores are false or real with those certain CPU-Z versions, then removing all scores possibly affected is the only way to keep the DB in good state. Of course it is some amount of time and effort wasted for those people affected by this. But this isn't really anything new - several benchmarks were removed completely in the past (for example UCBench and others which I don't even remember anymore). Other had points removed (PCM05, Hwbot Prime). I alone lost hundreds points by this... and I'm still here. Cleaning the bugged results from the DB is a good thing and everyone shoud understand it.

If the scores are bugged, I don't see another way.

Did someone try to measure power consumption during these slow / fast runs? It could give some hint - if the extra work is being done, then the CPU also should draw more power.

This is why I picked 45 vs 65nm Core2, these differ only in SSSE3 / SSE 4.1 (we can disregard extra cache here). There must be some use of SSE 4.1, otherwise 4.6 GHz Conroe would always beat 4.1 GHz Wolfdale.