Are WR points really necessary for Legacy 3D benchmarks (01, 05, 06 and AM3)?

[Splave]

[websmile]

Sure, didn´t you tell algorithm is ok? Then you might take in that there are nearly 10 times as much subs with 4770k than 5960x and legacy benches are old with tons of results^^- your example is far from legit or you simply change your algorithm^^ and btw, where do you get 2d wr points with 4 core cpu? at cb, wprime or hwbotprime against 8, 16 or 24 core cpu- the 2d wrs are very much concentrated on server systems? lmao - fail, Roman, and this is why I said that removing legacy 3d wrs is needing small compensation at 2d - other option to avoid the bad influence of cheap hardware to hwbot might be lower cap on global points, and I also see no reason to give globals to multi-cpu benches that are lower than single scores like on old 3d mark 01 for example

Edited February 3, 2015 by websmile

[Gunslinger]

 

is classic HWBot

[der8auer]

Sure, didn´t you tell algorithm is ok? Then you might take in that there are nearly 10 times as much subs with 4770k than 5960x and legacy benches are old with tons of results^^- your example is far from legit or you simply change your algorithm^^ and btw, where do you get 2d wr points with 4 core cpu? at cb, wprime or hwbotprime against 8, 16 or 24 core cpu- the 2d wrs are very much concentrated on server systems? lmao - fail, Roman, and this is why I said that removing legacy 3d wrs is needing small compensation at 2d - other option to avoid the bad influence of cheap hardware to hwbot might be lower cap on global points, and I also see no reason to give globals to multi-cpu benches that are lower than single scores like on old 3d mark 01 for example

 

No clue where you got the 4770K 2D WRs from.

 

To quote myself:

 

 

4770K:

 

wPrime32m: 119.4 p

wPrime1024m: 111.9 p

SuperPi32m: 139.2 p

PiFast: 126.9 p

Cinebench11.5: 124.7 p

CinebenchR15: 128.5 p

XTU: 166.4 p

HWBotPrime: 159.4 p

 

That's a total of 1076.4 Global Points + 0 WR Points you can achieve with a golden 4770K!

 

 

...there are nearly 10 times as much subs with 4770k than 5960x and legacy benches are old with tons of results...

 

That's exactly the point of the calculation. 4770K is capable of much more points than 5960X. Even if we take away the WRs of the legacy benches.

 

And algorith ≠ WR points. The algorithm only calculates the amount of globals and hardware points - not the WR-Points. So like I said before - the algorithm is fine.

Edited February 3, 2015 by der8auer

[websmile]

What´s the problem with benches awarded more points that are benched more often and hardware awarded more points that is used more often? The hardware, the benchmarks or your algorithm? What will happen in five years when other 3ds are benched to the max like vantage or 11, and you see the same problem with other hardware? Will you remove simply wr points or will you start the solve the real problem? Taking away the only wr points possible for most used hardware at community is not solving the reasons of the problem but only making cheap work at the symptoms

 

P.S. the wr points were my bad, I read lower facit were there was no zero inserted - and to your edit about algorithm, it is not the wr points why you get more points for the 4770 and for legacy 3ds, but it is your algorithm - so why remove wr points when they are not the problem and your algorithm is fine? Then justice wants 4770k to get more points, it is the system aka algorithm and not the wr points that cause this

Edited February 3, 2015 by websmile

[der8auer]

There is no problem with benches receiving more points Just pointing out that the WR-Points were meant to compensate the lower competition in multi GPU-Benchmarks like FireStrikeExtreme so the actual "world record" with 4 cards receives more points than single card record which is awarded with the highest amount of globals.

[websmile]

How about stripping legacy 3ds except 03 from wr points an making small compensation to give to an old 2d benchmark like Spi or Pifast? I know these are already awarded great, but I see this as a symbol for community that you can get honour also without using the real big gun, it would be a sign of appreciation also for the non pros and a small compensation and sign, we can´t ignore that majority of the community uses these and it should be appreciated in a more direct way than now via the old 3ds which are out of vga limit now for some time. What is a bigger problem for me, even though I like benching old cards in cf or sli at legacy benches and know others like this as well, is the fact that you get globals for score with 4 cards that are nearly half of best single card scores (3d01 for example) - on this I see absolutely no sense, and if we discuss the future of these benchmarks, this is a bigger injustice for me than the wr points because logic fails on this. I am for sure not against changes that make sense, and I know that removing wr points from am3 for example would be OK for other reasons than stopping haswell dominate the ranking

[Massman]

We can try out removing the WR points for those benchmarks that no longer scale with multi-GPU on our UAT environment.

[Massman]

Okay, long overdue but I finally managed to run a simulation on the UAT test server. Note: this is a simulation which includes the competition points. For this simulation I disabled WR points for the 3D benchmarks where the world record is achieved with Single GPU: Aquamark, 3DMark01, 3DMark05, and 3DMark06.

 

ELITE (1-20) (left with WR, right without WR)[/b]

 

 

ELITE (21-40) (left with WR, right without WR)[/b]

 

 

EXTREME (1-20) (left with WR, right without WR)[/b]

 

 

EXTREME (21-40) (left with WR, right without WR)[/b]

 

Edited April 28, 2015 by Massman

[websmile]

Good example - I guess it might indeed be time to let it go when ranking isn´t changed much - some of the ranking examples with and without are nearly identical, so I personally think this convinced me that it might be the right time for the change at legacy 3ds

[Mr.Scott]

Good example - I guess it might indeed be time to let it go when ranking isn´t changed much - some of the ranking examples with and without are nearly identical, so I personally think this convinced me that it might be the right time for the change at legacy 3ds

Or time to leave them alone. As you said, not much changed.

[websmile]

Yes, nontheless the weird situation remains that you have 3d benches that have wrs with one card because they are so cpu limited that more cards don´t give you more points - I guess anyway it is handled, it is OK for me

[ObscureParadox]

I would also argue that the older benchmarks shouldn't lose their world record points until some replacements can be found.

[Massman]

It's not so much a matter of finding replacements as much as it's a matter of re-evaluating which global categories need correction. The reason why we have WR points is because some 4xGPU 3D records receive a low amount of global points (compared to the effort it takes). This is due to the low participation in the 4xGPU rankings. For example, 8 Pack's Fire Strike WR only has 48.8 global points.

 

For benchmarks like Aquamark, the WR already has the correct amount of points. For example TeamAU's Aquamark WR has 125 global points from the 1xGPU category. There is simply no need apply a correction for this benchmark.

[der8auer]

To keep the logic we should actually remove them. We saw that there is only a very small impact on the ranking anyway.

[K404]

It's cool that this is being discussed semi-seriously, but i'm in two minds about the approach.

 

Making changes when they have little effect makes the transition easier to "cope" with and get used to... no-one feels s***-on

 

but only making changes when nothing changes is...fearful of the backlash? Things evolve, some people take a short-term setback in the name of progress- man up and stop being selfish

 

 

This post has no point

[Mr.Scott]

 

This post has no point

 

Oh yes it does.

[Wizerty]

Damn, I'm late....

 

My 2cent : Remove them for 01-05-06-AQ3, too much point for 1 submition, and 3d benchmark only cpu dependent... You don't even have to use LN2 on VGA... global point are enought, don't need to give bonus anymore for these benchmark

For 03 maybe you can keep it, or removed it. 03 is more VGA depend so...don't know for this one.

 

Roman show us the perfect example, with the 4770k vs 5960x.

And no one can't agreed that you need MORE SKILL, TIME, LN2 to achieve WR one 4way ring than on 1way with card on air... so giving 2time more point for 10time less effort...

 

---

 

the good thing is that you (hwbot crew) try to remove excess in point attribution, and I think it's the way to go, no excess = more fair for everyone.

[tiborrr]

I agree with Wizerty about 01, 05, 06 and AQ3 but not so sure about '03 at the moment. Would keep it for another year.

[Massman]

03 is fine for the moment since it's still scaling (a little bit) with multi-gpu.

 

As a general rule of thumb we can disable the WR points as a fix for a benchmark once 1xGPU is the optimal choice for WR.