Hak Foo

I had an entry flagged today. A forum post describing a HD3650 was identified as a 6600GT. In the past, I've also had various issues where it would decide a Katmai PIII was a Coppermine, an E6750 was a Q6700, and such. The annoying part isn't the mis-detection, it's the game of "Adjust the forum post, wait for the re-scan, check, repeat if necessary." It seems like sometimes, the parser is getting stuck on things that you don't realize are the key factor, and if you have to do it several times, it does get frustrating. Would it be possible to have a "Parser test" page? Where you can enter a string, as you'd post between the hwbot tags, and it would show you how HWBot would digest it, so you can know if you're submitting correct data BEFORE you submit it the first time? This would also avoid many of the "3.2MHz" and "3800GHz" entries, I'd expect.

I'm amazed nobody's made a Linux-based bootable CD which runs a suite of benchmarks on any PC you cram it into, then writes the results to a USB keychain. That would enable easy "bulk benching" too-- just set up CPU/mobo/minimal memory/CD drive/USB stick and go, no "two hours to install enough OS to bench that K6"

As I understand it, at least on the AGP side, many of the 6800 series cards were locked down versions of each other-- with soft-mods, you could unlock a 6800LE to run like a slow 6800 or even a slow 6800GT. Not sure about the PCI-E side, by the time I switched, the card I wanted was a fanless 7600GS.

http://www.hwbot.org/result.do?resultId=775847 I reported it; I changed it and it changed back; I reported it again. What do I have to do to change the post so it recognizes "This is a Katmai PIII, not the much rarer and less competitive Coppermine PIII class?" It's not that I don't like the points, but it isn't fair to people who actually have those to bench.

Don't you have to cross some bridges on an XP to convert it to a MP, and then it identifies as a MP (thus not necessarily putting you in the category you wanted to compete in)

Even if they did pass around the same card, the celing of performance may vary because of differences ranging from "ambient temperature here is higher" to "my PSU produces worse voltage quality" to "I'm too chicken to water-cool"-- so it's still a test of the skill and configuration. In fact, it might be interesting to pass the SAME part around between a handful of elite OCers and see the differences in their results.

If you're benching a processor type which was available in a MP configuration, you can score major wins easily that way by throwing extra sockets into it. Stock clocked dual > overclocked single. But I suppose that sort of stops being an option after the PIII. Don't forget to kill off any parasitic load (Explorer, for example).

Well... http://www.ocworkbench.com/2007/articles/sli-using-pcie-agp-7600gs/g1.htm Apparently it's possible to SLI a PCI-E and an AGP 7600GS with the right drivers and the right board (The ASRock 939Dual-SATA2 - famous as one of the few boards to have a REAL PCI-E and a REAL AGP slot) Can't say about two PCI ones.

Evidently you don't look at the misbalanced PCs they sell in big-box shops. EXTREME GAMER MODEL: Core 2 Quad 9xxx + upgraded graphics to a HD2600 Pro! Plus, that's so the wrong way to succeed at this. Throwing 500MHz more CPU at 3DMark doesn't do nearly as much as getting a graphics card that clocks up well. As for the "Gentleman's Agreement" - to an extent, there's a "what's stopping them from matching you" aspect. Especially in the older hardware classes, the gear is almost "disposable". I'd sort of love to get a really good Socket A board and a nice Super 7 board, and see what I can really wring out of my old K6-233, XP 2000+, and Athlon 1.2C.

Is there a way to make it stress for accurate clock readings, like it does when it tests for itself, but from when it shells CPU-Z until when CPU-Z finishes out? If you have Cool 'n Quiet or EIST turned on, you sometimes get wacky submissions-- "Athlon X2 4600+ at 1000MHz" or "Core 2 E6750 at 2000MHz"

I assume he means the dual- versus single- versus quad-core logic. It's a valid benchmark because benchmarks are meant as proxies for real application performance. A multi-threaded real application would be able to recongifure for the number of live cores. You could get the same results by rigging wPrime's advanced settings to 1 core for all CPUs. Of course, that wouldn't accurately reflect real-world multi-threaded performance.

I noticed that some of my scores have been arbitrarily mis-assigned to the wrong CPU. Looking down my profile: PiFast: C2D E6750 became C2Q E6700. SuperPi 1M: Ditto. SuperPi 1M: Duron 1600 became Athlon XP-M 1600+ (now that's a stretch...) The scan date doesn't seem to make much bearing. I know there was some E6750/E6700/Q6700 confusion in the past, but the Duron/XP-M one is new to me.

http://www.hwbot.org/result.do?resultId=663366 What do I have to write in the submission to make it identify my E6750 as an E6750 and not a Q6600?!

Man, that brings back memories. You want BIOS version 25. It solves all problems with the AV8. (Warning: This is a joke. BIOS 25 was a fiasco; it bricked the boards of the poor sods who tried it). I had an AV8 for a while, then traded it for an ASRock ULI board which overclocked less but had PCI-E. I seem to recall a lot of people whining about 3X LDT not working, but my concern was always the never-constant temperature monitor values.

Considering that many users are still daily driving such machines, it would be fairly discriminatory to penalise users on those machines. There's a lot of effort in score-gathering that way. You can also get many more points per bench if you DO have a record in a competitive sector. The boints are motivation for some participants to bench like mad and fill in a lot of useful data. For example, my main box is a 4600+ on Socket 939. It can't go much over stock on my current board, because I can't give it more voltage. Thanks to people fighting for hwboints, though, I can know exactly what to expect in terms of overclocking capacity and performance if I were to lay out the money for a board with more voltage options.