GIGABYTE Summer Spectacular 2017 Discussion

[Lucky_n00b]

@ websmile:

thanks for the info bruh, I'm already worried that I'm missing a tweak to get those epic raymarch hahah^^

[FireKillerGR]

@ websmile:

thanks for the info bruh, I'm already worried that I'm missing a tweak to get those epic raymarch hahah^^

 

nah high clocks and/or voltage (unstable card / artifacts in general) will cause this. Apparently it doesnt render properly so you get massive numbers.

[mllrkllr88]

Well, I see raymarch is higher than fur test thus my result is deemed glitched or invalid?

 

Edit: Looks like most of my backups follow this pattern...guess that pretty much takes me out.

Edited May 25, 2017 by mllrkllr88

[FireKillerGR]

Well, I see raymarch is higher than fur test thus my result is deemed glitched or invalid?

 

Edit: Looks like most of my backups follow this pattern...guess that pretty much takes me out.

 

how high was ur lowest score (vcore and freq)?

Soon I will check mine as I have to get home first.

[kimandsally]

Well not sure about the frequency or artifacts as y best score is slightly higher than the fur test but at very low MHz on the card absolutely no artifacts ran smoothly but my score is 18 pts higher raymarch than the fur so I think it can be higher without any glitch for sure.

[Lucky_n00b]

nah high clocks and/or voltage (unstable card / artifacts in general) will cause this. Apparently it doesnt render properly so you get massive numbers.

 

Thx for the explanation, I was getting instability and almost full screen artifact a lot around 1950-ish @ 1.55-1.58v, but never caused the raymarch score to 'boost' like that. If the errors was too much it will simply gave 'driver-revert' error

[FireKillerGR]

Thx for the explanation, I was getting instability and almost full screen artifact a lot around 1950-ish @ 1.55-1.58v, but never caused the raymarch score to 'boost' like that. If the errors was too much it will simply gave 'driver-revert' error

 

Might be voltage then as on my 6.8k+ run, vgpu was at 1.7~v

 

I uploaded the 2nd best I had which is 6.7k and has raymarch < fluid.

 

Guess I might need to revisit it.

[mllrkllr88]

I was running 1.70v-1.73v for my entire DICE session and it seems all my good scores show the pattern of ray being higher than fur. I will check all my cold results when I get home to be sure, but all my good ones are like that.

 

As for artifacts, yes, I was seeing some during the bench but nothing ridiculous.

[mllrkllr88]

hmm.. High Vgpu voltage would not explain why @kimandsally is also achieving a ray score higher than fur. I know he is using water and I would guess he is running somewhere between 1.3v-1.4v. (mind didn't scale past 1.33v on water)

[websmile]

This is mainly a thumb rule, if you have 300 raymarch normally there can be no wrong anyway. We have problems with catzilla Fliud and raymarch bugs for years now and this is hard to determine where bug starts after lucky run with tweaks ends. If your fur tests su... and you have 200 raymarch there will be no need to block, the problem is that this is a competition and because of the prices people always search for bugs when they cannot beat scores - but a normal raymarch is 300-350 max for this card, 450+ is normally not doable unless you have a ton of artefacts plus partially blackscreen. This happens when card is on the edge, and the driver doesn´t revert. So around 400-450 is max we can accept depending on card clocks.

lilchronic`s Catzilla - 720p score: 4469 marks with a GeForce GT 730 (GK208, GDDR5) It can also happen if your card is no good overclocker and you are on the edge at 1200^^

Edited May 25, 2017 by websmile

[mllrkllr88]

I checked out all my runs from my 1 DICE session. It looks like I have one (my worst haha) that will work since Raymarch is low 300's and over 100 points below fur test. I also have one that looks incredibly bad with a raymarch over 500, but with oddly low core clocks.

 

Anyway, its pretty obvious that some of the runs are bugged for whatever reason. I never saw any full or partial black screens, only a bit of artifacts in that bench.

[websmile]

Load up what makes sense for you, I check and remove the creepy ones^^ - I tend to handle this in dubio pro bencher, but what is obvious cannot be accepted some benchmarks are hard to moderate

[GtiJason]

Why do we have known buggy benchmarks Point/Comp enabled, and why is this "bug" that has been known about for ages not mentioned in the comp and benchmark rules for those unfamiliar with this situation ? It seems bad for All involved, moderators have more un-needed work to do, many overclockers do not know about this even when they've been primarily benching 3d for many years and the Competition organizers/sponsors look bad for a few reasons. First being subs getting pulled so late in the competition that have correct screenshots can appear like someone has an agenda about who will get the prizes, doesn't really matter if the scores were bugged as it is also the perfect excuse to pull any sub. Second is they look like they don't know what they are doing when it comes to throwing a fair competition, they have several of the greatest overclockers ever on staff that have run this benchmark. Don't they meet with HWBot staff to discuss the benchmarks to be used, you know of this bug and they should know of this so why chose a benchmark with these problems and more. Like the fact that a verification link that not normally needed be added for the comp when everyone knows this is broken and there appears to be minimal support for this.

 

I just think that while not technically shady, this just looks bad for our community and does nothing but frustrate the new guys in ambient class and may turn them away and aggrevate some of the veterans on lin to the point of walking away. We need these guys to stay, I probably would have left myself 3 years ago if it wasn't for the knowledge and helpfulness of several of these guys, one of whom recently left the scene. Who will pass the knowledge on then and be there to help when a fresh face is ready to toss his/her $3k rig out the window. Without experience like this on things like hardware, OS and running benchmarks efficiently we are all but doomed. Instead of trying to cram as many things we can into the schedule/budget every year maybe we need to step back and make sure our foundation is solid before we start trying to stack these cards as high as we can. It's simple, quality over quantity. Lets get things right and stop waiting until 5 minutes before leagues/comps start to decide what benchmarks to run and update an existing background from Rd 2 2016 to Rd 2 2017

Edited May 26, 2017 by GtiJason

[KaRtA]

Man, I spent hours last night trying to replicate the Raymarch scores of others but wasn't possible for me. Thought I was just missing something as it is a bench I have never run until now. Every run for me is super consistent and a good gap to the fur test score.

 

Raymarch of over 200 on this card on ambient looks far from achievable. Cold yes, ambient no, unless I still haven't found a way to boost.

[websmile]

Fact is that these benchmarks were selected not by hwbot, and that all benchmarks are vulnerable to bugs.

A score of raymarch at 400 even on air is simply the result of if you get the card run at the limit before the driver reverts, artefacts are part of the game at 3ds. Unless we ask for videoproof of all results we get for competitions we have to put a limit somewhere by subtest and general score for results to avoid problems after the competition ended, and some cards are more vulnerable obviously for this kind of "boost" than others.

What we face last years is another problem. If I take the catzilla and ice storm part here for example, a lot of people always cry "cheat" or "bug" when they cannot beat scores others can run. When they get the tweaks then or find the solution, suddenly all is OK. This development leads to benchmarks now that are barely tweakable like Time spy on which LOD and so on are completely forbidden, but even then people complain. I would not have bothered at all even about the 600 of OGS normally because it maybe 20-40 points in overall score and as stated above it is result of being able to run the card at absolute limit without getting an unrealistical score like posted above, which is what OC should be about, but the drama always starts then during and after the competition.

So if you want to solve this problem, the only chance would be to stop making competitions for prizes, ban all tweaking and also ban volt mods and maybe even cold which push cards above the limit. I am not sure this is what competetive overclocking should be about.

[GeorgeStorm]

I think the issue here is for those who aren't aware of things in advance, they might try their hardest, maybe use up their DI/LN2, and end up with a bugged run that they simply thought was a good run, since as you said artifacts are a thing in 3D.

 

If there are known bugs they should be disclosed in advance in my opinion.

[websmile]

We cannot take the responsibility for all that members do - there are dozens of results of catzilla in the database for 730GT, if you check these you get a landmark of what is possible and what not. With a couple of thousand of videocards of which a lot of have different characteristics and vulnerabilities people have to check this themselves, it is their responsibility and it is easy.

All benchmarks can cause bugged runs, staffs job is only to weed out the worst ones because we usually decide in favour of benchers if we have dounts at borderline results. It is futile to discuss this anyway, each time we handle this a bit more strict we get criticized as well, even if someone asks like Leeghoofd recently did with a member, we get reactions we act unfair or this is not ok.

[GtiJason]

I am not being critical of you Michael or Leeghoofd, I understand this makes more work for you guys in terms of moderating results and trying to be fair while doing so. I think it puts you in a no win situation and as an overclocker yourself knowing the effort put in to compete and also knowing that many guys can't just easily refill there dewars and give it another go. There has got to be some way to make things fair, especially when prizes of this magnitude are involved. Maybe just stick to newer or benchmarks with better support from the developer like Geekbench 3/4, GPUPi, Futuremark (Firestrike etc), Unigen Superposition. Situations like this do nothing to help our community and we need to do a better job making sure they don't keep happening. Sorry, I've only gotten involved in this discussion because I am not competing in it but understand a lot of the frustration involved

Edited May 26, 2017 by GtiJason

[Johnd0e]

Isnt it also the staffs job to disclose in advance ALL the rules of a competition before the competition begins. Like you confirmed these "bugs" were known about for a long time, why wait until last minute to change the rules thus causing some people to have wasted maybe hundreds of dollars in cooling trying to strive for what they believed was an acceptable score since rules said nothing about a points cap or whatever.

 

Thats like playing an entire card game, and being one hand away from winning the prize thanks to a bunch of wild cards you luckily got and then dealer says "oh yea we decided wild cards wont be counted at all".

 

In my opinion, stop trying to blame benchers for mistakes that the staff makes. If there are known "loopholes/bugs" make it known in the rules that things wont be accepted. If i tried blameing customers for oversights i made that caused them to have the ability to cheat the system, my boss would have a stern talking to with me.

 

.....and thats all i have to say, i think. Enjoy your day/night.

[websmile]

With all respect,this is completely wrong.

It is your responsibility to watch your results, and if you have a black screen or the picture is rendered in a way that it is not visible anymore, this is part of rules of all 3d mark and catzilla benches for example that this is an invalid result. It is also your responsibility to check the plausibility of a result, and if you do not bother to check for example catzilla 730GT scores and ask yourself why you have higher raymarch, fur or whatever on your result than people with cpu and card on LN2 and hundreds of MHz more on both it is not the staff that has to take responsibility for this, but you.

Sure it is easier to blame someone else when you simply do not do your job, sh.. on fairplay and simply upload what you have and hope that staff overlooks it. But then blaming people who adress this is nothing but a lame excuse

 

P.S. Jason, the choice of benchmarks will not change a single issue. We used time spy for Galax qualifier and had big drama because some guys showed impressive scores and others claimed they are bugged. As long as some tweak, max out or prepare OS better than others, we will always have this problem. As said, I would not have bothered with raymarch here because margin is real low, but I had first some requests and then a report of one of the benchers that he thinks his score might be bugged. The problem is simple, when does tweaking stop and bugging starts, even with videoproof this would be impossible to judge if we accept artefacts for example...

Edited May 26, 2017 by websmile

[GeorgeStorm]

With all respect,this is completely wrong.

It is your responsibility to watch your results, and if you have a black screen or the picture is rendered in a way that it is not visible anymore, this is part of rules of all 3d mark and catzilla benches for example that this is an invalid result. It is also your responsibility to check the plausibility of a result, and if you do not bother to check for example catzilla 730GT scores and ask yourself why you have higher raymarch, fur or whatever on your result than people with cpu and card on LN2 and hundreds of MHz more on both it is not the staff that has to take responsibility for this, but you.

 

The issue isn't with the extreme results though, yeah it's up to a bencher to check their scores against others, and if they see they're beating people with far higher clocks etc, then something might be up. That's not the issue though, it's the small gaps, where there might be some artifacting but nothing crazy, and so your score is slightly better than someone at the same clocks as you, who are you to know (especially as a novice in 3D) that your score is bugged rather than you happened to find a good series of settings etc that got you that extra 1%?

 

I completely agree with you on the big stuff, but it's the rest I'm concerned about.

[websmile]

This is why I gave the warning but also said 400-450 will be ok depending on clocks and cooling for me. Noone intends to kill the work of people who invest time, effort and money for the competition. Decisions will be made in favour of benchers, but 500+ is something I have confirmed information about partial black screen runs and this is not the way a 3D benchmark is meant to be run. Handling is not easy, but I will try my best

[mllrkllr88]

I think the issue here is for those who aren't aware of things in advance, they might try their hardest, maybe use up their DI/LN2, and end up with a bugged run that they simply thought was a good run, since as you said artifacts are a thing in 3D.

 

If there are known bugs they should be disclosed in advance in my opinion.

 

My thoughts exactly!!

 

If there is a bench with a common potential issue, like Raymarch or Fluid benches in Catzilla, it would have been nice to have a heads up about it before we waste time and money benching with cold.

 

No disrespect to the mods, they have a BIG job to do and they are not responsible for informing us. However, from the sounds of it this is a pretty well known issue, even back in the X99 Gigabyte competition.

 

A simple one sentence statement at the start of the competition would have been easy, effective, and ultimately resulted in a lot less work for the mods...

[Strong Island]

man this comp was such a fail for me.

 

It's like my card is gt710 performance in a gt730 body and label.

 

i thought my issue was not knowing ice storm tweaks.

 

but last night at 2k core 6.5cpu my cat score was around 2900, which actually lines up with gt 710. Not sure what is going on.

 

After ice storm fail I did a new clean os and I had all tweaks applied but that wouldnt make my score double anyway. All gpuz info lined up with gt 730. But that seems wierd 2 different benches with 2 different os's were about half. And all 2d performance is fine.

 

Really disappointed, spent so much time modding and benching, but I guess it's all part of the game.

 

so dumb I used 2 epowers, didnt even gain like 1mhz with epower on memory.

[mllrkllr88]

Bummer deal SI, sorry it didn't work out for you I had the exact same results with external VRM on memory, not even 1 MHz more. I never tested my core with a normal FB vmod but I have a feeling the gain from external VRM on the core was pretty small.

 

Come to think of it...I binned a few cards for someone and I noticed the score was WELL below what it should have been on one of them. I didn't pay close attention to the score since I was binning for clocks, but I wonder, something is going on with these cards....

 

Too bad you were not able to at least get a token sub in for the lucky draw...

Edited May 31, 2017 by mllrkllr88