You weren't kidding, nice one!

is this not bugged GT2? FS can cause textures to not render properly when pushing mem, as I'm sure you know. You have more scores consistent with this? I saw it on RX 460 as well...

Ok, I checked my scores. This is obviously bugged GT2, if I were you I would upload backup score. Maybe mods will allow it, but I wouldn't bet on it to have just one score...

 

I have runs at 2835 core and mem at 2600, 2629 and 2650. There is no huge gain in GT2 from pushing just mem...

FireKillerGR`s 3DMark - Fire Strike Extreme score: 8846 marks with a GeForce GTX 1060

 

I Think a lot depends on core and mem balance - raising core brought lower gt2 for example for firekiller at same mem freq. We have atm some similar cases we investigate with FM

Edited November 10, 20169 yr by websmile

FireKillerGR`s 3DMark - Fire Strike Extreme score: 8846 marks with a GeForce GTX 1060

 

I Think a lot depends on core and mem balance - raising core brought lower gt2 for example for firekiller at same mem freq. We have atm some similar cases we investigate with FM

Not sure what you mean, Stavros score is as expected from those clocks

What I mean is that he upgraded score as you know, he did raise core - result was that he lost gt2 FPS at this despite memory being exactly the same. Seems balance of core and memory on cards makes a difference

FireKillerGR`s 3DMark - Fire Strike Extreme score: 8885 marks with a GeForce GTX 1060

core and mem are exactly same freq in those runs. What differs is mostly cpu score. And diff in both GT1 and GT2 is within margin of error when benching on LN2.

 

At 2835 core and 2650 mem i got 9493 GPU score

At 2835 core and 2629 mem i got 9466 GPU score

At 2835 core and 2600 mem i got 9430 GPU score

 

This score is 2695/2727 and 9728 GPU score (with GT1 being much slower). It just doesn't make sense. You telling me it is a valid strategy to aim for bugged scores? I think I still have a bugged score with RX460 I can show...

Bad explanation by me - the raise of memory(Dram) boosted physics for stavros, but he lost on GPU score - when the tests are in line, both GT1 and GT2, on scaling with core and memory, we cannot say a run is bugged and delete it. This is what we currently invetigate on some AMD cards for firestrike where we see similar scaling. We also have runs at the competition which have lower core and sub 2600 mem and still beat your GT2 - are all of these runs bugged or are they a result of the balance that you have at system?

I see only one other person with high GT2... but I agree it is very hard to tell for sure. Would require access to FM tools to look at frame rates through out the test. Last year FM removed a couple scores because of "bugged runs". But then again, that might not work either cause you can have consistent bugged frames through one of the subtests.

 

Only possible explanation I see would be a specific LOD which is exceptionally good for GT2, but bad for GT1 and combined...

I'll leave this be now!

 

steponz, hope you understand this is nothing personal, I know you are a great 3D-bencher and seem like a really nice guy. Maybe there is some other tweak I just don't know... I'm just sick of bugged benchmarks, cheatable benchmarks (32M) etc... Is there any benchmark we can trust?

As said, we have similar observations at 290 and 390 amd cards lately and asked FM to check this out, like you said observation of framerates with tools and plausibility check is needed for this. We have a range of ~33 - ~37 at gt2 at the competition, and around 2500-2730 mems, this is simply not enough out of line for me to throw out results without check by FM, we have this option. In spite of 32m having no monitoring, it is somehow easier to see for me because variance by driverr version, settings like LOD and overall balance of cpu, dram and card frequencies is harder to dtermine in the end

I didnt read half of the comments yet but on the latest score I got 34.xx fps due to temp. If the temp is higher than -165 (ex. -162) mems will artifact and I will get less fps.

 

Thats all. 35.2 is what I should get in normal cases; it is just that my latest gt2 was bad due to benching gt2 @-163

There is many factors that come into play with this score.

Just because you cant acheive it or dont understand it... maybe you should come to me and not trash my result.

 

FM has checked the benchmark and it is ok. But obviously you cant figure it out, you call out my sub. Pretty disappointed in ya right now.

Did ya mod your card? Have you ever ran fse before? Its all about gt2. If mem bugs, it scores lower... not too mention your on a completely different os. Many factors come into play and its consistent. I asked fm originally to look at because i knew one of you guys eould do this

Its quite a shame that i have to defend a result after clearly people know i bench the firestrikes all the time.

Just because you cant do it.. doesnt mean its impossible.

Strong score, realy matter of skills this game!

Great score bro! congrats

There is many factors that come into play with this score.

Just because you cant acheive it or dont understand it... maybe you should come to me and not trash my result.

 

FM has checked the benchmark and it is ok. But obviously you cant figure it out, you call out my sub. Pretty disappointed in ya right now.

Did ya mod your card? Have you ever ran fse before? Its all about gt2. If mem bugs, it scores lower... not too mention your on a completely different os. Many factors come into play and its consistent. I asked fm originally to look at because i knew one of you guys eould do this

Its quite a shame that i have to defend a result after clearly people know i bench the firestrikes all the time.

Just because you cant do it.. doesnt mean its impossible.

No need to get upset, it's just a discussion. I'm not saying you're cheating or something. Maybe saying it's "obviously bugged" was a too strong statement, sorry for that. But still, there is a reason you checked the score with FM right?

 

As far as I know the Fire Strikes were designed to be "non tweakable" and measure "pure performance". Of course we bypass this by LOD/tesselation etc. But FS it's not like legacy benching.

The thing that makes me wonder is that your score is not consistent across the tests, and not consistent with your Time Spy result. If it truly had been "all about mem", or to find some combination of temps/core/mem/volts to make the card perform better you would do better in GT1 and combined also, and in Time Spy. Of course the different subtests can scale differently, but your scaling in GT2 is a bit extreme in my opinion. With that being said, of course there could be other explanations than bugged score...

 

When I ran for wild card qualifier I got a few bugged scores when pushing mem really high on RX460. Once I got almost 50% higher in combined for example. You would not call this bugged? Where do you draw the line, 5%, 10%, 50%??? When I have just seen these types of bugged scores, of course it makes me wonder...

 

Also, we have this from last year's qualifier:

https://community.futuremark.com/forum/showthread.php?183665-GALAX-GOC-2015-WORLDWIDE-QUALIFIER&p=1881070&viewfull=1#post1881070

Aranha got some scores removed, due to too high GT2 results. Those scores would have taken him to the finals...

 

Maybe you can understand my point of view a little better? I'm not trashing the score, I'm discussing it. If it actually were possible to get bugged scores, would it be good to find that out?

Again you continue to say that my results are suspect on my submission. You cant compare fse to timespy at all. They do completely different things. Its very well known by majore gpu fse benchers that gt2 on fse has alot to do with memory. I pushed memory really high.. there were no bugs.. flat out raw performance becauee pascal scales very high with mem. And when your dialed in with good memory and proper voltages it will work as intended. You keep comparing your results with mine. Ehich you cant. You obviousky havent done any proper efficency testing and driver settings. Different drivers in different os's react completely different. Stop trying to say they're suspect. I went to futuremark for this simple reason. When benchers dont have a clue how to do things the proper way they scream suspect. I knew the result was valid and i only did this so i can hammer down people with their bs accusations.

 

Another thing. Stop comparing a Amd 460 with nvidia.. vompletely different thing. When mem bugs in gt2 with nvidia cards it drops or hangs.

 

Im not going to write a how to guide on how i do my results because you dont know what your doing. Next time if you questioning something go to the person privatrly. Not trash their hard work.

 

At this point im done with this conversation and i suggest you start doing single gpu and work out your bad results.

There is nothing left to say here - Mods checked the score, and more important, FM checked it and rated it a valid score, and I see no need to discuss a general issue here. If some one searches for bugs, here is the wrong place