How about 3D

[steponz]

2d is taken over by mediocre benches.... ill put 20 benchmarks tonight running default... lets give it more points.... its what hwbot is coming too...

 

bring cpu cores to 3d... let more people bench and it will come..... but right now as I see it...... 2d is what hwbot wants to concentrate on...

[Massman]

2d is what hwbot community wants to concentrate on...

 

Let me specify this for you

[newlife]

If you want to help engage the community to bench 3d a good place to start would stuff like low end card comps and the likes which I'm perfectly happy to help with. Also if you manage to get people interested in having fun and enjoying what they are doing rather than focusing on point you might find people are a lot more willing to put the time and effort into 3d

 

Personally I don't think separating GPU's based on high end and low end would encourage many people to bench because I reckon it goes much deeper than that

[buildzoid]

Do you have any idea what your talking about.. 1 core? Get a clue man... Submit some stuff first then.... talk about what ya know.

 

I may have exaggerated with it only needing 1 core however it certainly doesn't benefit enough from core count for a 5960X to beat a 6700K in it. Clocking a 6700K 11% higher than a 5960X and getting 9.9% more points even with lower clocked GPUs makes Unigine a low core count CPU benchmark.

[Bruno]

Eh, does that not break the hardware sharing rule?

 

Anyone with golden gpu that wants to be my friend please pm me.

 

 

I was not talking about a team mate. I personally have friends that don't have anything to do with HWBot, but they were willing to borrow me their cards to bench. So no hardware sharing involved and that's exactly what i meant.

 

Seriously though, more points to 4 way than 1 way etc, that's just making a corner for money bags to compete in with little competition.

 

Time to tweak the platform, drivers, settings etc applies whether or not using one gpu or 10. In fact you probably need to spend more time with single gpu as theres actually some competition there, not just raw horsepower. And how many cards do you think the top person had to bin to get that top spot? Probably more than 4.

 

It may drive to a corner, but also it may drive others to start getting more cards. And not necessarily newest cards. They can get 4x 4870 for example, and if the HW points for them would be higher than Globals for 4x 980 TI, the competition may be better and the money will not count that much. And modding will be revived. Remember that HW points come as top 20 while Global only 15, so giving more points for older (and cheaper) hardware may balance the points against the money.

 

However, let's be fair about the money. Overclocking is an expensive hobby. And no one is forced to get in. But if you do, just stand for it We have to buy pots, TIM, insulation, etc. There are still a lot out there that can't afford to buy LN2, while others can spend hundreds of dollars per month to bench whatever/whenever. Is it fair for all? Will we ever see an enthusiast in world wide top 10? I don't think so. It's just impossible to beat LN2 with water cooling and the points are received according to the higher score. It's the money advantage. How can that be balanced? I'd say split HWbot in 2: 1 for subzero and 1 for air/water cooling. Two separate sites. It would be nice to see WR points on air/water.

 

 

Then comparing time and skill 2d v 3d isn't valid. How many hours do guys spend benching ram to within an inch of its life along with system tweaks (cue 32m) compared to some of the 3d? Pot kettle /apples oranges.

 

I think it's just a mute point tbh and points should be allocated based on level of competition. Everyone has the same opportunity to bench whatever they want, it's up to the individual. Pretty sure the majority of the 'top 10' are balanced across 2d/3d which is what put them in that position.

 

I agree with that. 2D and 3D should be separate in points. So get some memory benchmarks with a lot of points so memory geeks can get what they've worked for.

 

I said from the beginning that the amount of work should be important, not the money spent, because there always be some richer than others.

 

To get a good score in 3D you need a good CPU and memory, so 2D it's solved anyway. That's why 2D should get lower points.

[Bruno]

I may have exaggerated with it only needing 1 core however it certainly doesn't benefit enough from core count for a 5960X to beat a 6700K in it. Clocking a 6700K 11% higher than a 5960X and getting 9.9% more points even with lower clocked GPUs makes Unigine a low core count CPU benchmark.

 

Some of the 3D benchmarks are more CPU dependent (in number of cores) than others. I'd say it's a good thing because that diversity balances the importance of CPU/GPU over the rankings and it gives us another learning curve in overclocking.

[buildzoid]

Some of the 3D benchmarks are more CPU dependent (in number of cores) than others. I'd say it's a good thing because that diversity balances the importance of CPU/GPU over the rankings and it gives us another learning curve in overclocking.

 

Yes but Unigine is at the point where running LN2 on the CPU and Air on the GPUs is enough to get second place in the world. Unigine IMO needs a new preset which is so Heavy that a 980Ti running 1500mhz scores about 2000 points in it. That should be pretty simple to do because the current preset is only 1680x1050. A 4K preset would absolute make Unigine much more GPU focused again.

[Bruno]

Yes but Unigine is at the point where running LN2 on the CPU and Air on the GPUs is enough to get second place in the world. Unigine IMO needs a new preset which is so Heavy that a 980Ti running 1500mhz scores about 2000 points in it. That should be pretty simple to do because the current preset is only 1680x1050. A 4K preset would absolute make Unigine much more GPU focused again.

 

3DMark 2K1, 03, 05 are also very CPU dependent. Yet it's not a bad idea to have a new preset for Unigine. It may be interesting.

[speed.fastest]

I think the problem hardware point is too low because hardware is not popular, even if your score is very good for the hardware. I think if we got more hardware point based on score, not based on popularity we can grow 3D bencher. Sorry if you dont understand my English...

[buildzoid]

Ok crazy idea here but what if the amount of points for a GPU submission was some how affected by the % overclock. Maybe even in relation to the average OC. This could allow you to get say 10 extra HWpoints which would solve the current 2 HW point top score that most GPUs suffer.

You get 1 bonus point for every 10% you go over stock clock or something. If you're second you do not get as many high clock points as the 1st person. The down scaling system for this needs some more thought but could spice up low end GPU 3D. Sure it wouldn't get people on the front page but I bet there would be more 3D subs overall which would pump up the global 3D points because there would be more subs overall.

[Massman]

One of suggestions that keeps coming back (saw it recently on Facebook too) is to give points based on the amount of pots necessary for a certain benchmark. That idea was explored and quantified in Post #11.

 

So I spent a little bit more time exploring this appoach.

 

The problem with 3D is two-fold. 1) Its decreasing popularity and 2) no "difficulty" parameter in the Points equation. Let's focus on the second problem.

 

It's not easy to define the difficulty of a certain benchmark. There are many aspects that we have to take into consideration to conclude benchmark A is more difficult than benchmark B. Let alone compare the efforts of Person X in Benchmark A to the efforts of Person Y in Benchmark B. For an algorithm to work on mass-scale (ie. a database of 1.5 million submissions) we need objective parameters. Adding more or less points based on a vote is highly subjective and seeds for even more discussion in the future.

 

So, here's another try defining the difficulty of a benchmark as: amount of LN2 pots used in the top-20 of each benchmark category. It's a slightly different approach than what was tried earlier. The main differences are:

• Top-20 instead of top-50
  
• No normalization of difficulty by not counting GPU LN2 pots for CPU benchmarks
  

Choosing top-X is arbitrary. The only reason why I would choose 20 is because we include 20 results in the 'first page' of every benchmark ranking. Not normalizing has a bigger effect: 3D benchmarks will typically be assigned a higher difficulty level because you can have CPU+GPU+MEM on LN2 whereas 2D benchmarks have only CPU+MEM.

 

The table below is what the new point spread would look like. The algorithm is: maxPTS = MAX(P+(250-P)*D;10), with

• P = current maxPTS ~ popularity of benchmark category
  
• D = [sUM(IF(CPU=LN2;#CPUs) + IF(GPU=LN2;#GPUs) + IF(MEM=LN2;+1) / SUM(MAX(#CPUs,20) + MAX(#GPUs,20) + 20)] ~ difficulty of benchmark category
  

D will be 100% if all CPUs, GPUs amd memory of in the top-20 of a benchmark ranking is cooled with liquid nitrogen. Caveat: 2D benchmarks can never have 100% difficulty since they don't have LN2 on GPU.

 

(Note; might want include all extreme cooling and not just LN2)

 

[table=head]Benchmark | Category | P index | D index | MaxPTS

3DMark - Fire Strike|1|138.6|0.633|209.2

XTU|4|166.4|0.5|208.2

3DMark11 - Performance|1|124.7|0.633|204.1

3DMark03|1|116.7|0.633|201.1

3DMark - Fire Strike Extreme|1|119.2|0.617|199.9

Aquamark|1|120.9|0.6|198.4

3DMark2001 SE|1|107.9|0.635|198.1

3DMark05|1|113.2|0.617|197.6

3DMark Vantage - Performance|1|118.3|0.583|195.1

XTU|6|165.7|0.333|193.8

HWBOT Prime|4|162.9|0.333|191.9

SuperPi - 32M|4|132.6|0.483|189.3

Catzilla - 720p|1|125.6|0.483|185.7

SuperPi - 1M|4|142.1|0.4|185.3

Unigine Heaven - Xtreme Preset|1|119.3|0.483|182.5

3DMark06|1|118.1|0.467|179.7

CPU Frequency|8|148.8|0.267|175.8

PiFast|4|125.1|0.383|173

3DMark - Fire Strike|2|102.7|0.475|172.7

Cinebench - R15|4|127.6|0.367|172.5

3DMark03|2|77.4|0.55|172.3

Cinebench - R11.5|4|120.3|0.367|167.9

XTU|2|166.4|0.017|167.8

Geekbench3 - Multi Core|4|114.4|0.393|167.8

HWBOT Prime|2|125.8|0.333|167.2

XTU|8|115.4|0.35|162.5

GPUPI for CPU - 1B|4|106.8|0.383|161.7

3DMark Vantage - Performance|2|87.7|0.438|158.7

wPrime - 32m|4|118.1|0.3|157.7

3DMark11 - Performance|2|88.3|0.425|157

wPrime - 1024m|4|109.7|0.333|156.5

Cinebench - R15|2|105.6|0.35|156.1

Catzilla - 1440p|1|83.3|0.433|155.5

Cinebench - R11.5|2|107.6|0.333|155.1

wPrime - 32m|2|106.9|0.333|154.6

Unigine Heaven - Xtreme Preset|2|82.5|0.41|151.2

3DMark - Fire Strike Extreme|2|90.3|0.375|150.2

HWBOT Prime|6|100.6|0.317|147.9

Cinebench - R15|6|99.3|0.317|147

GPUPI - 1B|1|123.9|0.183|147

HWBOT Prime|8|101|0.3|145.7

wPrime - 1024m|2|97.8|0.3|143.5

wPrime - 32m|6|81.3|0.367|143.2

Cinebench - R11.5|6|92.3|0.317|142.2

wPrime - 1024m|6|75.2|0.377|141.1

Aquamark|2|79.5|0.361|141.1

Cinebench - R15|8|93|0.3|140.1

3DMark05|2|77.6|0.346|137.3

wPrime - 32m|8|80.4|0.333|136.9

Cinebench - R11.5|8|87.3|0.3|136.1

Geekbench3 - Multi Core|6|78.5|0.333|135.7

Catzilla - 720p|2|88.7|0.288|135.1

wPrime - 32m|1|77|0.333|134.7

GPUPI for CPU - 1B|8|75|0.333|133.3

Geekbench3 - Multi Core|2|97.8|0.233|133.3

Geekbench3 - Multi Core|8|74.5|0.3|127.2

wPrime - 1024m|1|64.2|0.333|126.1

wPrime - 1024m|8|76.8|0.283|125.9

Cinebench - R11.5|1|71.3|0.3|124.9

3DMark06|2|78.9|0.25|121.7

3DMark Vantage - Performance|4|48|0.353|119.4

3DMark - Fire Strike|4|51.2|0.342|119.1

GPUPI for CPU - 1B|2|82.5|0.217|118.8

Unigine Heaven - Xtreme Preset|3|44.9|0.36|118.7

3DMark - Fire Strike Extreme|4|48.8|0.342|117.5

GPUPI for CPU - 1B|6|74.3|0.233|115.3

3DMark2001 SE|2|64.2|0.266|113.6

Unigine Heaven - Xtreme Preset|4|41.1|0.339|112

3DMark11 - Performance|4|46.3|0.302|107.8

HWBOT Prime|1|85.6|0.133|107.5

wPrime - 32m|3|35|0.333|106.7

3DMark Vantage - Performance|3|52|0.258|103

wPrime - 1024m|3|28.6|0.333|102.4

Catzilla - 1440p|2|63.1|0.188|98.1

Cinebench - R15|1|58.3|0.2|96.6

3DMark - Fire Strike|3|57|0.19|93.7

3DMark03|3|40|0.25|92.5

wPrime - 32m|12|28.6|0.288|92.3

HWBOT Prime|3|38.4|0.233|87.8

Catzilla - 720p|3|53.1|0.175|87.6

GPUPI - 1B|2|82.8|0.026|87.2

3DMark11 - Performance|3|50.4|0.181|86.5

Catzilla - 1440p|4|40.6|0.216|85.7

3DMark - Fire Strike Extreme|3|53.5|0.16|84.9

Cinebench - R11.5|3|36.2|0.2|79

3DMark06|3|39.5|0.186|78.6

3DMark06|4|39|0.186|78.3

Cinebench - R15|3|38.4|0.183|77.2

wPrime - 32m|5|14.2|0.267|77.1

Aquamark|4|37.3|0.185|76.6

Geekbench3 - Multi Core|1|60.5|0.083|76.3

3DMark05|4|35.6|0.181|74.4

Aquamark|3|36.2|0.175|73.7

3DMark05|3|38.4|0.165|73.3

3DMark03|4|33.8|0.164|69.2

wPrime - 1024m|5|13.2|0.233|68.5

CPU Frequency|1|55.3|0.033|61.8

3DMark2001 SE|3|32.6|0.13|60.9

Catzilla - 720p|4|42.1|0.086|60

GPUPI for CPU - 1B|1|56.6|0.017|59.8

Catzilla - 1440p|3|39.5|0.093|59

Geekbench3 - Multi Core|3|34.4|0.1|56

wPrime - 1024m|12|19.8|0.154|55.2

CPU Frequency|4|51.3|0.017|54.6

GPUPI - 1B|3|48.4|0.03|54.4

HWBOT Prime|5|31.9|0.1|53.7

Cinebench - R11.5|5|13.2|0.15|48.7

3DMark2001 SE|4|27.1|0.092|47.6

GPUPI - 1B|4|37.9|0.042|46.9

Cinebench - R15|5|12.1|0.117|39.9

XTU|16|37.3|0|37.3

Cinebench - R15|12|30|0.027|35.9

Cinebench - R11.5|12|29.3|0.025|34.8

GPUPI for CPU - 1B|3|29.3|0.017|33

HWBOT Prime|12|26.4|0.025|32

HWBOT Prime|16|24.1|0|24.1

Geekbench3 - Multi Core|12|23.3|0|23.3

XTU|1|22.5|0|22.5

XTU|12|21.7|0|21.7

Cinebench - R11.5|16|20.8|0|20.8

Cinebench - R15|16|20.8|0|20.8

GPUPI for CPU - 1B|12|20.8|0|20.8

wPrime - 1024m|16|20.8|0|20.8

Geekbench3 - Multi Core|32|12.1|0.033|20

HWBOT Prime|32|12.1|0.031|19.5

wPrime - 32m|16|19|0|19

Cinebench - R11.5|28|11|0.033|19

Geekbench3 - Multi Core|16|18.1|0|18.1

XTU|18|18.1|0|18.1

wPrime - 32m|24|12.1|0.022|17.2

wPrime - 1024m|24|12.1|0.021|17

Geekbench3 - Multi Core|5|16.2|0|16.2

Geekbench3 - Multi Core|18|15.2|0|15.2

GPUPI for CPU - 1B|18|15.2|0|15.2

Cinebench - R15|48|14.2|0|14.2

GPUPI for CPU - 1B|5|14.2|0|14.2

GPUPI for CPU - 1B|16|14.2|0|14.2

HWBOT Prime|10|14.2|0|14.2

HWBOT Prime|48|14.2|0|14.2

wPrime - 1024m|48|14.2|0|14.2

XTU|32|14.2|0|14.2

Cinebench - R11.5|48|13.2|0|13.2

Cinebench - R15|18|13.2|0|13.2

Cinebench - R15|20|13.2|0|13.2

Cinebench - R15|36|13.2|0|13.2

Geekbench3 - Multi Core|48|13.2|0|13.2

GPUPI for CPU - 1B|24|13.2|0|13.2

HWBOT Prime|20|13.2|0|13.2

HWBOT Prime|64|13.2|0|13.2

wPrime - 1024m|36|13.2|0|13.2

wPrime - 32m|36|13.2|0|13.2

wPrime - 32m|48|13.2|0|13.2

Cinebench - R11.5|10|12.1|0|12.1

Cinebench - R11.5|18|12.1|0|12.1

Cinebench - R11.5|20|12.1|0|12.1

Cinebench - R11.5|24|12.1|0|12.1

Cinebench - R11.5|36|12.1|0|12.1

Cinebench - R15|10|12.1|0|12.1

Cinebench - R15|24|12.1|0|12.1

Cinebench - R15|28|12.1|0|12.1

Cinebench - R15|32|12.1|0|12.1

Geekbench3 - Multi Core|24|12.1|0|12.1

Geekbench3 - Multi Core|36|12.1|0|12.1

GPUPI for CPU - 1B|48|12.1|0|12.1

HWBOT Prime|18|12.1|0|12.1

HWBOT Prime|28|12.1|0|12.1

wPrime - 1024m|18|12.1|0|12.1

wPrime - 32m|18|12.1|0|12.1

wPrime - 32m|32|12.1|0|12.1

XTU|10|12.1|0|12.1

Cinebench - R11.5|32|11|0|11

Cinebench - R11.5|64|11|0|11

GPUPI for CPU - 1B|32|11|0|11

HWBOT Prime|24|11|0|11

HWBOT Prime|36|11|0|11

wPrime - 1024m|14|11|0|11

wPrime - 1024m|28|11|0|11

wPrime - 1024m|32|11|0|11

wPrime - 32m|14|11|0|11

wPrime - 32m|28|11|0|11

wPrime - 32m|64|11|0|11

Cinebench - R11.5|14|9.9|0|10

Cinebench - R11.5|15|9.9|0|10

Cinebench - R11.5|40|9.9|0|10

Cinebench - R11.5|45|9.9|0|10

Cinebench - R11.5|60|9.9|0|10

Cinebench - R15|14|9.9|0|10

Cinebench - R15|15|9.9|0|10

Cinebench - R15|30|9.9|0|10

Cinebench - R15|40|9.9|0|10

Cinebench - R15|45|9.9|0|10

Cinebench - R15|60|8.7|0|10

Cinebench - R15|64|8.7|0|10

Cinebench - R15|80|8.7|0|10

Geekbench3 - Multi Core|10|9.9|0|10

Geekbench3 - Multi Core|14|9.9|0|10

Geekbench3 - Multi Core|15|9.9|0|10

Geekbench3 - Multi Core|20|9.9|0|10

Geekbench3 - Multi Core|28|9.9|0|10

Geekbench3 - Multi Core|30|9.9|0|10

Geekbench3 - Multi Core|40|9.9|0|10

Geekbench3 - Multi Core|45|9.9|0|10

Geekbench3 - Multi Core|60|9.9|0|10

GPUPI for CPU - 1B|14|9.9|0|10

GPUPI for CPU - 1B|15|9.9|0|10

GPUPI for CPU - 1B|20|9.9|0|10

GPUPI for CPU - 1B|28|9.9|0|10

GPUPI for CPU - 1B|30|9.9|0|10

GPUPI for CPU - 1B|36|8.3|0|10

GPUPI for CPU - 1B|40|9.9|0|10

GPUPI for CPU - 1B|45|9.9|0|10

GPUPI for CPU - 1B|60|9.9|0|10

HWBOT Prime|14|9.9|0|10

HWBOT Prime|15|9.9|0|10

HWBOT Prime|30|9.9|0|10

HWBOT Prime|40|9.9|0|10

HWBOT Prime|45|9.9|0|10

HWBOT Prime|60|9.9|0|10

wPrime - 1024m|10|9.9|0|10

wPrime - 1024m|15|9.9|0|10

wPrime - 1024m|20|9.9|0|10

wPrime - 1024m|30|9.9|0|10

wPrime - 1024m|40|9.9|0|10

wPrime - 1024m|45|9.9|0|10

wPrime - 1024m|60|9.9|0|10

wPrime - 1024m|64|9.9|0|10

wPrime - 1024m|80|8.7|0|10

wPrime - 32m|10|9.9|0|10

wPrime - 32m|15|9.9|0|10

wPrime - 32m|20|9.9|0|10

wPrime - 32m|30|9.9|0|10

wPrime - 32m|40|9.9|0|10

wPrime - 32m|45|9.9|0|10

wPrime - 32m|60|9.9|0|10

wPrime - 32m|80|8.7|0|10

XTU|28|9.9|0|10

XTU|30|9.9|0|10

[/table]

Edited December 3, 2015 by Massman

[speed.fastest]

Or maybe need baseline score for every hardware and benchmark. Then score percentage from baseline score is what hardware point come from, more improvement from baseline score more hardware point. maybe possible.

 

If score based on difficulty, maybe all benching with Extreme Cooling even the score is not good for Extreme Cooling.

[buildzoid]

I think benchmark length is also a big part of the difficulty of a benchmark. Also I do not believe in LN2 being consistently difficult. AMD 32nm with LN2 is incredibly easy to do. Doing AMD 32nm with LN2 properly is hard but the use of LN2 on it's own is not a good indicator of difficulty. So any kind of "difficulty" system will IMO not work. I think it should be purely based on %OC over stock or maybe in relation to the cooling type average.

 

So say that 100 people submit 1 CPU benchmark leading to an average OC of 4.5Ghz on air cooling. Then you come along and manage to do 5Ghz on air cooling. Assuming the 100 people before you weren't really really bad at overclocking, getting 20% more than average is really really hard. The same would work with LN2 except maybe the LN2 would award more points because most of the time it is harder than air/water.

 

So the end result of this would be that some who is at the top of the table for their cooling category will get more points leading to something like this:

 

CPU XYZ

LN2 average: 6Ghz

Water average: 5Ghz

Air average: 4.5Ghz

 

The world record holder who got 7.2Ghz on LN2 gets 20 points for being the best in his cooling class

Rank number 20 who ran 6.3Ghz LN2 gets X bonus points along with normal points //I do not want to write essays about the effects of how X scales

Rank number 30 who ran 6Ghz on LN2 gets 1 extra point because he's average for his cooling class

Everyone running bellow average clocks on LN2 gets 0 extra points

 

The nut job who threw 1.8V at his WCed chip and hit 5.5Ghz gets 10 extra points.

The guy who managed 5Ghz on water gets 0 extra points because being average on water is pretty easy

 

Air cooling would follow the water cooling extra point system

 

The only thing I don't like about this system that I'm proposing is that it doesn't account for efficiency. Beating someone who is clocked XX/XXXmhz higher than you should be worth something.

 

For efficiency there could be extra points based on benchmark score divided by clock so that you get score per mhz which is basically the best efficiency measurement I can think of.

 

These changes would drop a massive amount extra points into the 3D benchmarks with low submission quantities and make them much more competitive because instead of getting 2 points with your volt and cooling modded R7 370 you get 10-20 points for winning the cooling category and 10points more from efficiency. So an air cooler could get up to 22 points with that card. Suddenly it's worth benching it and then the HWpoints will naturally go up as more people try to get first with that GPU even though it's not competitive in the globals.

Edited December 4, 2015 by buildzoid

[Taloken]

It would require to always add a rig photo, as many people leave cooling at "Standard stock cooling" whatever they are on air/water/ln2.

 

And the scaling of these "cooling points" have to be studied so ln2 keep the lead in global ranking. Not interesting to see a watercooling guy be #1.

 

Or these points can be "league points".

[Alex@ro]

Who came with the idea that who has 4-was setup e.g has more money should be rewarded more? I can guarantee that doing that will cause a lot of people frustration and leaving hwbot.

 

The main idea is to find a way to reward skill and knowledge and effort and focus less on money/sponsorship/binning. How that should be done?Looong discussion...

[buildzoid]

It would require to always add a rig photo, as many people leave cooling at "Standard stock cooling" whatever they are on air/water/ln2.

 

And the scaling of these "cooling points" have to be studied so ln2 keep the lead in global ranking. Not interesting to see a watercooling guy be #1.

 

Or these points can be "league points".

I thought system photos are always required for submissions? I always add mine. Maybe make it so that if there is no system picture the score isn't eligible for extra cooling points.

 

In my system LN2 is worth 2x the points of water/air and it just adds onto the current system. It would have little to no impact on global WRs because those are 100+ points for first place. You would need to get a top 5 score using water to even come close to getting more points than the best running LN2. So that's not really a risk. Sure there are times where 4x air/water 290X was best in Unigine however that was only for about a month after launch and that happened with the current system.

 

Due to my suggestion being a fixed amount of bonus points it will massively increase the value of being the very best on something not benched all that much but would also have little impact on the top 5 scores in any becnchamrk.

[Massman]

Or maybe need baseline score for every hardware and benchmark. Then score percentage from baseline score is what hardware point come from, more improvement from baseline score more hardware point. maybe possible.

 

Using a baseline score is very difficult to manage on a large scale. Keep in mind that there are 539 global rankings and 58246 hardware rankings at HWBOT. Maybe it's possible to find an appropriate baseline in the active rankings, but how do you define a baseline in rankings with 1, 2, or 3 scores?

[Massman]

I think benchmark length is also a big part of the difficulty of a benchmark. Also I do not believe in LN2 being consistently difficult. AMD 32nm with LN2 is incredibly easy to do. Doing AMD 32nm with LN2 properly is hard but the use of LN2 on it's own is not a good indicator of difficulty. So any kind of "difficulty" system will IMO not work. I think it should be purely based on %OC over stock or maybe in relation to the cooling type average.

 

So say that 100 people submit 1 CPU benchmark leading to an average OC of 4.5Ghz on air cooling. Then you come along and manage to do 5Ghz on air cooling. Assuming the 100 people before you weren't really really bad at overclocking, getting 20% more than average is really really hard. The same would work with LN2 except maybe the LN2 would award more points because most of the time it is harder than air/water.

 

So the end result of this would be that some who is at the top of the table for their cooling category will get more points leading to something like this:

 

CPU XYZ

LN2 average: 6Ghz

Water average: 5Ghz

Air average: 4.5Ghz

 

The world record holder who got 7.2Ghz on LN2 gets 20 points for being the best in his cooling class

Rank number 20 who ran 6.3Ghz LN2 gets X bonus points along with normal points //I do not want to write essays about the effects of how X scales

Rank number 30 who ran 6Ghz on LN2 gets 1 extra point because he's average for his cooling class

Everyone running bellow average clocks on LN2 gets 0 extra points

 

The nut job who threw 1.8V at his WCed chip and hit 5.5Ghz gets 10 extra points.

The guy who managed 5Ghz on water gets 0 extra points because being average on water is pretty easy

 

Air cooling would follow the water cooling extra point system

 

The only thing I don't like about this system that I'm proposing is that it doesn't account for efficiency. Beating someone who is clocked XX/XXXmhz higher than you should be worth something.

 

For efficiency there could be extra points based on benchmark score divided by clock so that you get score per mhz which is basically the best efficiency measurement I can think of.

 

These changes would drop a massive amount extra points into the 3D benchmarks with low submission quantities and make them much more competitive because instead of getting 2 points with your volt and cooling modded R7 370 you get 10-20 points for winning the cooling category and 10points more from efficiency. So an air cooler could get up to 22 points with that card. Suddenly it's worth benching it and then the HWpoints will naturally go up as more people try to get first with that GPU even though it's not competitive in the globals.

 

I like the way you're thinking and a lot of the points you bring up are valid arguments in the discussion. The problem with many of the suggestions is that they are hard to quantify in an objective manner on a large database with a wide variety of hardware and activity. In addition, there are problems with verification.

• % OC: the highest percentage overclock is +340% on an obscure Geode NX-1250+. Percentage overclocking is difficult to compare across architectures
  
• Cooling average: this will work for very popular hardware only as an average is only meaningful with a large enough dataset. In addition, it also largely depends on the cooling context. Hitting 4.5GHz with 15C ambient is easier than hitting 4.5GHz with 35C ambient. Also, doing better or worse than average could say more about your luck in the silicon lottery than your overclocking skill.
  
• Per cooling avg: there are 539 global rankings and 58246 hardware rankings at HWBOT. In theory we would have to calculate 1617 and 174738 averages.
  
• Efficiency: downclock for screenshot or during benchmark run = higher points
  

In this discussion, a lot boils down to the ability to put the ideas into an equation which is rigid enough to deal with outliers and simple enough to scale across a wide variety of benchmarks, hardware and popularity.

[buildzoid]

I like the way you're thinking and a lot of the points you bring up are valid arguments in the discussion. The problem with many of the suggestions is that they are hard to quantify in an objective manner on a large database with a wide variety of hardware and activity. In addition, there are problems with verification.

• % OC: the highest percentage overclock is +340% on an obscure Geode NX-1250+. Percentage overclocking is difficult to compare across architectures
  
• Cooling average: this will work for very popular hardware only as an average is only meaningful with a large enough dataset. In addition, it also largely depends on the cooling context. Hitting 4.5GHz with 15C ambient is easier than hitting 4.5GHz with 35C ambient. Also, doing better or worse than average could say more about your luck in the silicon lottery than your overclocking skill.
  
• Per cooling avg: there are 539 global rankings and 58246 hardware rankings at HWBOT. In theory we would have to calculate 1617 and 174738 averages.
  
• Efficiency: downclock for screenshot or during benchmark run = higher points
  

In this discussion, a lot boils down to the ability to put the ideas into an equation which is rigid enough to deal with outliers and simple enough to scale across a wide variety of benchmarks, hardware and popularity.

 

I assumed that the bonus points for % OC would always be hardware specific. So things like that 340% OC on the Geode NX-1250+ wouldn't impact the points awarded say a 1600mhz GTX 950.

 

The silicon lottery already dominates most of the HWbot rankings. 100mhz more core clock is pretty much an auto win setting for everything except SuperPi. Giving it points probably won't do much harm but would hugely raise the popularity of the obscure hardware categories. Also why couldn't you just use the current average clock stats. AFAIK those are derived from the average across all the submission with that cooling type so they get slightly inflated by CPU-Z/SuperPi submissions but are otherwise sound.

 

If you added a CPU/GPU clock log as part of required verification for efficiency points eligibility it might work. However the logger running would probably wreck efficiency. This isn't really a problem if everyone has to run it to get the extra points however it would make the old score look better than the new ones. So of all my suggestions this is the one I think is completely impossible.

 

The problem with dropping the efficiency points is that people will artificially try to inflate the points they get from % OC. So IMO the happy medium would to find a balance where cheating on efficiency to get extra points will lead to an equal loss of points in the %OC department. However that's probably going to be a nightmare to balance. People with bellow average OC will on purpose focus on trying to cheat on the efficiency score to get extra efficiency points and vice versa. This might turn out to be really hard to catch when it occurs and I can't think of any convenient ways to solve it. Maybe allow negative points for exceptionally bad efficiency/%OC that only eat into the extra points. So say that some hack job says he can beat a 4.5Ghz score(average OC and better than AVG efficiecny) with a 4Ghz score. However he gets -10 points for his exceptionally bad clock so the fact that he was IRL running 4.6Ghz is lost on him getting an over -X extra points. If you get bellow 0 extra points they do not affect you over all score(we don't want people with negative points that's just silly). This will force silicon lottery winners to learn to tweak if they want all the extra points they can get. So mister I binned 200 chips won't have the most points in the i5 6600K air cooled rankings with his 5.3Ghz i5 unless he also knows how to atain above average efficiency.

 

If you keep throwing more stick in my master plan for points calculation HWbot will need to be hosted on a super computer to accommodate the extra math needed to calculate points.

 

BTW I mostly focus on water/air so if anyone has any input on LN2 I would love to hear it.

 

BTW2: Sorry for kinda derailing the thread

[speed.fastest]

Using a baseline score is very difficult to manage on a large scale. Keep in mind that there are 539 global rankings and 58246 hardware rankings at HWBOT. Maybe it's possible to find an appropriate baseline in the active rankings, but how do you define a baseline in rankings with 1, 2, or 3 scores?

 

Ok thats understandable. Here my little thought about HWBot Scoring system :

*Global Points = Its good system nothing wrong.

*Hardware Points = Its hard to make good score with many submission in same hardware category.

So whats make some submission is not valuable with this scoring system? Performance Points

Let's add Performance Point to scoring system

Performance Points is based on % score improvement from baseline score. Baseline score is lowest score in the hardware category. HWBot Staff can choose baseline score if the lowest score is too low (lets say lowest clock in cpu clock category). Hope this can help.

[rbuass]

Nothing changed...

10 x more scores 2D than 3D...

One more first page with 100% 2D scores... similar to yesterday, ... before yesterday... etc etc etc...

I am not able to change nothing, but only gave my opinion that`s something is not right for 3D benchers...

 

If I am part of development..

 

1- not allowed ES CPU... and even not allowed non public BIOS, tools or weapons (if I can not buy... I cannot worth points/ if BIOS not public... I cannot worth points ... as few examples ... in order to be fair with ALL)

2- create separate league for 2D and 3D, and even create something similar to AMD and Intel (so we can see more people with AMD, not only looking for CPU frequency)

3- create some kind of achievements for the overclockers (like before)... Overclocking Legend... Overclocking Star... Revelation of the year and so on

4- give more power to lower hardware (find a way to consider and worth more to lower cards and CPUs), and even drop relevance of the MONSTER number of points in some benchmarks people competing alone (256 cores Superduper Xeons in Wprimes... as example)

5- create some kind of "old school or legacy" for new discontinued stuff (Aquamark, 01, 03, 05 and 06 even discontinued by FM ... Wprime, Spi, Pifast and old XP benches).... even some people not consider good that idea, in order to keep relevance to very old benchmarks, even Microsoft no more provide any kind of support for XP... no more drivers, etc etc... keep the relevance, but in different category

6- create some "light area" for newcomers... that can play and show gaming benchmarks, like GTA V, Metro Last Light, Tomb Raider, etc etc... in order to bring gaming comunities and grow OC community (can be followed by videos, for example, in order to keep reliability... there are very complex and detailed benchmarks, for example, Metro Last Light that read very detailed information)

Please don`t tell we will have Photoshop... because nothing hardest to Photoshop 2D scores where we even don`t have validation files

7- Include some good benchmarks to plus (Fire Strike Ultra, 3Dmark 11 Extreme, etc)... and 2D (Winrar, Passmark or any benchmark you decide)

8- Do not give highest relevance to 4 way (xfirex), because there are VERY FEW competitors, and is very easy to get a lot of points just running (same than superduper xeons 256 cores)... even more in 3 cards... 1 card is most competitive... 2 cards is ok, because we even have dual GPU videocards... but to worth A LOT of points to 4 way, in my way of view, is much more about $$$ than about the effort (PLEASE, I DO NOT CONSIDER "not amazing" top dogs overclockers running 4 cards at 2000 Mhz, ... but just the SAME GUYS can do similar running 2 cards 2100 Mhz or 1 card 2200 Mhz (one option is to worth ONLY 1 score per benchmark... or max 2 scores per benchmark... for sure... the 2 best scores).

 

I think that if I go deep inside, I can find more "FROM MY WAY OF VIEW", with respect for all guys that think different than me.

The reason I have type this post is not to conflict with people that don`t have same opinion than me, but just to share a point of view that maybe still not deep analyzed... and sure, to try to help with my cent.

 

I create this topic, and I think I gave all opinion I can here.

For sure, for me... the things is not good for 3D benchers... and maybe will never be... since looks for today 90% of the overclockers give more importance top benchmarks like XTU than Fire Strike.

 

Regardless all... with my opinion, or no... best wishes for all and sorry my awful english.