yosarianilives

I think leeg is right on this one, 2d probably still has more categories when you look at all the popular multithreaded categories worth big pts whereas gpu maxes at 4 categories except in gpupi. Also can say that while legacy 3d doesn't need more gpu performance past single card there definitely is a lot of tweaking you can do to get maximum score and lower cpu overhead from xfire/sli since even single card is a cpu bench. Not sure it's a good thing but definitely more than just set cpu clocks and use pcie switches on board for ez 1-4 way pts within an hour.

Wanted to start a discussion, and hopefully there's no drama around this, more of a what should we do with ES models of cpu (not qs). A good start of the discussion is distinguishing ES vs QS. A QS would be like a slightly pre-production 10900k or highly binned chip from a vendor. I think QS is straightforward as they basically always line up to an existing retail SKU of cpu so I'd think you'd just sub as that like people currently do. ES is a bit different as they are much earlier in the development cycle and often don't line up nicely with a retail cpu. So for those should we try making categories specifically for them as often they do have their own sku, either an amd PN or an intel code but are their own model. Sometimes they're even their own stepping. For example I have some phenom II and opteron ES chips that are C1 stepping and the first retail stepping is C2 so I'd venture to say it's basically a completely different cpu than any retail chips. But it's also really tough to prove what cpu is what based on screenshot as usually they just say engineering sample in cpuz. So maybe institute an obscure and confusing rule like you have to include a pic of the IHS? Obviously not a great solution but idk what else to do. A third part of the discussion, what about existing ES categories for chips that might not have even had a proper ES pn, basically chips that might have only had 1x ever made? There seem to be a few that knof subbed with in AM3 era and I've never seen any es part number that lines up. Not saying we should or shouldn't remove those categories but I think maybe there should be a discussion. Example: https://hwbot.org/hardware/processor/phenom_ii_x4_615_be_es/

It says generic vga card, if the hw is "unrecognized" then score will always be hidden to prevent leaks of es hw

Did you reach out to fm?

Need to test it again, but in middle of tc it stopped unlocking

Personally I think the biggest thing that determines cheat vs tweak is the workload of the bench changed. This dll tweak does not do that, it makes the cpu run the code better. Essentially uses more of its ability to do the math. No different from Intel opencl allowing avx to be used. Lod alters the workload but I see it as more grandfathered in as at one point it was not detectable so not fair that modern scores compete with old scores with potential lod unless modern can lod too. Anyways if you're really worried about this tweak being legit then drop ALL legacy 3d, ALL wprime, ALL superpi and pifast. Or at least all scores done without BM. All of those benches get tweaked to the same level. The only difference here is a few long term benchers didn't know about this tweak until now. There will always be something new in effi, and I think people should be willing to rebench when it happens or accept that they can't hold a record forever. Without allowing new major tweaks to come out for benches it becomes gate keeping to make sure the old breed can hold golds forever and it's impossible for new members to break in.

New idea, let's ban new Intel opencl for gpupi, Avx2 is too strong, remove all score. Let's remove ALL 3dmark Score with lod enabled, it's changing workload and clearly a cheat. Tess disable as well. Also let's remove waza superpi, any ycruncher that has large page files enabled, all catzilla results especially those with audio enabled or 3dmark11 in background. Let's make hwbot 2021 about setting clocks and hitting run. Tweaking will now be illegal. And tbh this would benefit me plenty, my clocks are usually not too bad it's just my os effi that's weak. So I say, no more tweaks, let's go to just hitting run on bench.

Maybe I misunderstood the tweak, but you take the dll from a different version of windows and put it on an older version. It's not like you're changing the work that's being done, youre still doing the same work just more efficiently with lower os load. I don't really see how it's different than finding the best service pack of the best release of windows with the right updates. Nobody says running a special service pack of xp that was only released in China is cheating when running superpi so not sure how this is particularly different. It's not dll injection.

in US can't really get any from other places, and tbh I suspect that most of the oem only skus came from china originally. What I do find interesting is that 565, 570, 970, 975, and 980 on ebay almost exclusively come from eastern europe.

Yeah, I usually start with imc at 800 and work up for that reason. Mem and especially cpunb help but core is king. Dunno if there's any useful information here and it's a bit messy but here's basically all the data I've recorded for binning on am3 https://docs.google.com/spreadsheets/d/16J4XStjGzraN0VhfVXPXNSZswKIRNg_2s4LIrfKxadI/edit?usp=sharing Also found that x265 is very hard on some chips, my 555 clocks amazing for wprime and r15, but no matter the voltages set it can't break 5.8 x265 even on trex pot. Whereas that 960t does probably only 100-200mhz more r15 than x265.

Helped him bench and bin, and can say was top 2 out of 10 from china. Dunno if the other chip is better for x265 as we binned on r15. This should be about the max clockwise you can get. Mem and nb pushed about as far as can go, maybe subtimings could go a little tighter but don't know if it's enough to change score and didn't really rerun much as it's 15m bench.

2.5b is supported by ycruncher and even from the pre BM Java launcher, just would have to add hwbot support and support from BM side

Agreed, I say pull whole ranking as the global ranking makes zero sense and I suspect it's 100%about unintentionally bugging

7zip has always been from inside Benchmate so could have mat do version control Suppose it'd be up to mat to add support but can we add 2.5b category for ycruncher?

The one who doesn't die of covid wins

If there is a tie then the two teams select a champion to battle it out in ritual combat

The hyperpi score that the bench spits out is an average of the times. Bench launches an instance of superpi on each thread. Skydiver should be graphics score

At least it seems to have been fixed for newer comps, as I did get pts for gskill. So perhaps in 2021 it won't be an issue at all

I notice that still don't have pts from comps earlier in year. At some point something was fixed because I have a few pts from gskill tweaker sub, but no comp points from team cup etc

You just have to support benchmate on patron to get it. There are no tiers so any amount

Reach out to fm, they probably haven't added processor to their dB yet so it shows as prerelease

Nooo double mem

Didn't even know that there was an option to rank by "thread count", in old ranking 9700k would be op I think

Just means he has to make a better score

Its screwing everyone equally so not so bad a thing to want fixed lol, that is kinda an odd average though. Perhaps it's doing per core score but that doesn't seem to math out right either.