Adjustment Global Points - General Discussion

[Rasparthe]

Not EVERY HW-points bencher wants more HW points

 

 

Aye, probably vice versa too.

 

There will never be a good answer that will please everybody. Just leave it alone.

 

I still agree with this wisdom when it was posted when the original thread was started.

[Strong Island]

man people love subtly taking shots at people around here.

[Rauf]

Just some quick numbers for your suggestion @Rauf

 

//edit: what do we do with benchmarks like GPUPI where CPU doesn't matter and participation is also not like the 2D benchmarks?

 

3DMark Fire Strike Extreme Categories by GPU count (4 total)

 

GPUs	Max	Users

1	14432	1040
2	22824	542
3	27910	157
4	34252	108

 

3DMark Fire Strike Extreme Categories by GPU and CPU count (30 total)

 

GPUs	CPUs	Max	Users

1	1	236	1
1	2	8280	113
1	3	2224	1
1	4	12614	759
1	6	10440	375
1	8	14432	231
1	10	3383	1
1	12	3813	4
1	16	6597	3
1	20	4711	1
1	64	9896	4
2	2	2452	3
2	4	19048	302
2	6	17528	247
2	8	22827	134
2	12	10015	3
2	16	5304	1
2	20	8667	1
2	24	10262	1
2	64	5027	1
3	4	12887	33
3	6	20244	88
3	8	27910	64
3	12	4302	2
3	18	14626	1
3	20	11857	1
4	4	15600	23
4	6	25332	60
4	8	34252	55
4	64	25020	2

Well, statistics shows that the collective wants to bench 1 GPU with 4 core CPU Done deal!

Problem with this is that it will result in A LOT of benchmarks that gives globals if we have unrestricted CPU category as now with 1,2,3,4 GPUs. And then add 4 core CPU with 1,2,3,4 GPUs...

 

GPUPI is 2D so it's not affected by CPU cores

[Guest cowgut]

man people love subtly taking shots at people around here.

 

I hope you do not mean me I was just saying.

I ,unlike some others have respect for other benchers if you took offence sorry.

cus I know you worked very very hard for the score that was used a an example.

I did edit my last post

Edited January 11, 2016 by cowgut

[Rauf]

Unfortunately this is telling as to the real agenda. Increasing HW points in relation to Globals would do nothing but help out 3D, since all those low level cards would suddenly become much more attractive. It will make Global 3D and 2D less attractive since they won't be as powerful in terms of moving yourself up the rankings. So why wouldn't someone that wants to see more 3D interest not agree that HW points should be worth more in relation to Globals? Or is it that only the 3D that give globals should be worth more? In other words, make sure those that spend the most get the most points?

 

Don't know why you talk about "the real agenda". If you have followed this discussion you would know it has always been about the global points for 3D vs primarily XTU. Your agenda is quite clear by the way as you only bench for HW-points...Try reading the topic of this discussion...

 

HW points (top spots) is mostly about running average GPU clocks and pair it with a golden last gen CPU. It is not so much about the HW the points indicate (the GPU).

 

This may be true in 3D, since every generation brings new efficiency (when your Intel), but not so in 2D. Trying to take down gold cups in Socket 462 is as hard today as it was in 2004. Even in 3D, after a certain point of competition it becomes much, much, much more about the GPU.

 

Example: This recent sub in 8800 GTS 512:

http://hwbot.org/submission/3048348_strong_island_3dmark03_geforce_8800_gts_512_mb_74160_marks

 

Its gotten to the point that cards need to be zombie modded and extensively soldered to make it into the top spot. Certainly using the newest generation of CPU helps but that is the nature of 3D. A ton of work still went into that sub. Worth only 49.7pts. Some have even argued that the level of 'hard' should count for giving out points. Shouldn't this perhaps be worth more than this?

 

Yeah, he is nr1 because of newest Skylake CPU, vs nr2 who runs higher GPU clocks but on haswell... Still a great achievement, but it actually proves my point.

Don't know why you bring up 2D HW-points. Of course a new generation of cpus is not gonna help an old E8500 clock higher...2D HW-points is all about binning or being very lucky. Don't see the "hard" part but more on that below.

 

http://hwbot.org/submission/3060316_rauf_unigine_heaven___xtreme_preset_2x_geforce_gtx_980_ti_8917.17_dx11_marks

 

Unless the argument is that running a couple of 980Ti in SLI on stock coolers is harder than the first example? Still this thread is littered with calls from overclockers that want to see even more points added to globals for this sort of submission.

 

In my sometimes solitary opinion, I would love to see more 8800 GTS subs with guys have to add caps, replace components, using voltage regulations cards running under LN2 than stocker coolers taking down more total points. But only my opinion, everyone decides in their own head what makes a great overclocking submission, mine happens to be the first example.

 

I don't understand why you have to belittle my result to make cheap points in your argumentation. Or maybe you're just mad about Country Cup

I don't want points to be given out based on how "hard" a benchmark is. That is very artificial and can never be made fair. I want 2D and 3D to get similar global (and HW) points. Or maybe you don't think there is anything wrong with XTU points and that a GFP XTU is much "harder" than a GFP in Fire strike and therefore, by your own argumentation, deserves more points?

[Strong Island]

Don't know why you talk about "the real agenda". If you have followed this discussion you would know it has always been about the global points for 3D vs primarily XTU. Your agenda is quite clear by the way as you only bench for HW-points...Try reading the topic of this discussion...

 

 

 

 

 

Yeah, he is nr1 because of newest Skylake CPU, vs nr2 who runs higher GPU clocks but on haswell... Still a great achievement, but it actually proves my point.

Don't know why you bring up 2D HW-points. Of course a new generation of cpus is not gonna help an old E8500 clock higher...2D HW-points is all about binning or being very lucky. Don't see the "hard" part but more on that below.

 

 

 

I don't understand why you have to belittle my result to make cheap points in your argumentation. Or maybe you're just mad about Country Cup

I don't want points to be given out based on how "hard" a benchmark is. That is very artificial and can never be made fair. I want 2D and 3D to get similar global (and HW) points. Or maybe you don't think there is anything wrong with XTU points and that a GFP XTU is much "harder" than a GFP in Fire strike and therefore, by your own argumentation, deserves more points?

 

but meanwhile you have no problem belittling my result for your argument. And it's so funny that you said what you said because you are ahead of people in top 10 with lower gpu clocks, so I guess your score is because of skylake also.

Edited January 11, 2016 by Strong Island

[Rauf]

but meanwhile you have no problem belittling my result for your argument. And it's so funny that you said what you said because you are ahead of people in top 10 with lower gpu clocks, so I guess your score is because of skylake also.

 

You did read the part about a great achievement?

Of course my score is because of skylake, never said otherwise...

Edited January 11, 2016 by Rauf

[Strong Island]

You did read the part about a great achievement?

 

I dont want to make this about me, I should never have opened my mouth, I just get a little frustrated sometimes.

 

I worked on that sub for months and binned like 8 cards, killed probably about 4 cards/1 epower learning how to solder, binned cpu's, tweaked LOD forever and I actually had a higher shader clocks and the 3024mhz shader clock I had in 06 is the highest I can find anywhere. So it's a little annoying when someone just says "Great Achievement but it's because of skylake" My cpu was also 1.5ghz less.

 

Anyway lets get back on track, I try never to post negatively but sometimes it's hard to hold back.

Edited January 11, 2016 by Strong Island

[Massman]

Well, statistics shows that the collective wants to bench 1 GPU with 4 core CPU Done deal!

Problem with this is that it will result in A LOT of benchmarks that gives globals if we have unrestricted CPU category as now with 1,2,3,4 GPUs. And then add 4 core CPU with 1,2,3,4 GPUs...

 

GPUPI is 2D so it's not affected by CPU cores

 

GPUPI is an extreme example, but there are plenty of benchmarks that fall in between. For example 3DMark06 scales with multi-threaded CPU but only up to a certain point. It doesn't make sense to have a distinction to 8C and higher if the benchmark only scales to 4 CPU cores. (this is also the case for multi-gpu in benchmarks like 3DMark01)

 

Another consequence of this system is that the popularity index for each of the rankings will hurt, meaning a lower amount of points for each category compared to what 1xGPU is now receiving. But that can be addressed with a threshold adjustment as discussed in the technical thread.

 

Another consequence, but more related to perception, is that the "top single gpu" score will most likely not have the highest points anymore since it would fall out of the category with the highest participation. Technically it can be solved by creating an intermediate ranking of "single gpu", but this adds an additional layer of complexity. Not exactly a prime example of the KISS principle.

 

Also, this will not address any of the concerns that the most expensive graphics card is required to be competitive in 3D benchmarks. If anything, it opens up more global rankings for the GTX 980 Ti to dominate. For the GTX 970 owners, this has no effect whatsoever. Certainly the cost has been reduced from $1000 to $400 for the CPU, but the $1000 graphics card is still a factor. $1400 for CPU+GPU in 3D is still triple the $400 for CPU in 2D. In that regard, the Class idea was addressing this problem much better (data output here).

[Rasparthe]

Don't know why you talk about "the real agenda". If you have followed this discussion you would know it has always been about the global points for 3D vs primarily XTU. Your agenda is quite clear by the way as you only bench for HW-points...Try reading the topic of this discussion...

 

 

I actually don't bench for HW points, I sometimes take them when convenient, but I'm not a HW point bencher, nor a global point bencher. No offense taken, I don't follow you either.

 

 

Yeah, he is nr1 because of newest Skylake CPU, vs nr2 who runs higher GPU clocks but on haswell... Still a great achievement, but it actually proves my point.

Don't know why you bring up 2D HW-points. Of course a new generation of cpus is not gonna help an old E8500 clock higher...2D HW-points is all about binning or being very lucky. Don't see the "hard" part but more on that below.

 

I don't understand why you have to belittle my result to make cheap points in your argumentation. Or maybe you're just mad about Country Cup

I don't want points to be given out based on how "hard" a benchmark is. That is very artificial and can never be made fair. I want 2D and 3D to get similar global (and HW) points.

 

 

I wasn't trying to belittle your result, in fact, as was pointed out, I thought it was smart benching. I just know how much went into that submission that strong did, how much binning and testing and mistakes were made. To me it was the epitome of mastering an category. It was one hell of an journey. I'm sorry if you felt your submission was just as tough but it shows what I'm talking about.

 

Everyone has different degrees of what is 'hard' or even what they believe is 'hard'.

 

The original discussion as you point out, was the fact that XTU was 'easy', 3D is 'hard'. More points must go to 3D because it is 'hard'. I guess my mistake was believing that what was wanted was a way to increase popularity of 3D through naturally increasing interest in lesser cards which would in turn increase interest in high end cards. XTU is worth so much because its popular, I thought what was wanted the same popularity for 3D cards. This was the reason I was talking about increasing HW points in relation to globals, not because I'm a hardware bencher.

 

The 'real agenda' I was talking about was the idea that HWBOT should just tack on globals to 3D because its 'hard'. As you said its artificial and can never be fair. So I will wait and see how increasing 3D globals in relation to 2D ones is done without being artificial and without being unfair.

 

 

Or maybe you don't think there is anything wrong with XTU points and that a GFP XTU is much "harder" than a GFP in Fire strike and therefore, by your own argumentation, deserves more points?

 

 

No, I think it deserves whatever points the benchmarks popularity decides. Each benchmark is treated the exact same by the algorithm, that is fair. That is the definition of balanced. If Futuremark ever integrates with HWBOT and suddenly Firestrike is so popular that it gives out massive points, will you be arguing to change the 2D points to match?

 

I certainly won't, if 3D is more popular I'm fine with that. If XTU or 2D is more popular, I'm fine with that. Right now, the submissions do the talking, more popular = more points, in essence the points go where the interest lays. I just have an issue when a very, very small percentage of total members that have a great deal of interest in 3D start moving this balance in their favour.

 

I hope that explains my position better, perhaps things got muddled in there somewhere.

 

At any rate, I've given my opinion, I hope I've cleared up my position, and I don't have any skin in the game, so I will let the powers that be sort it out.

[Rauf]

GPUPI is an extreme example, but there are plenty of benchmarks that fall in between. For example 3DMark06 scales with multi-threaded CPU but only up to a certain point. It doesn't make sense to have a distinction to 8C and higher if the benchmark only scales to 4 CPU cores. (this is also the case for multi-gpu in benchmarks like 3DMark01)

 

Another consequence of this system is that the popularity index for each of the rankings will hurt, meaning a lower amount of points for each category compared to what 1xGPU is now receiving. But that can be addressed with a threshold adjustment as discussed in the technical thread.

 

Another consequence, but more related to perception, is that the "top single gpu" score will most likely not have the highest points anymore since it would fall out of the category with the highest participation. Technically it can be solved by creating an intermediate ranking of "single gpu", but this adds an additional layer of complexity. Not exactly a prime example of the KISS principle.

 

Also, this will not address any of the concerns that the most expensive graphics card is required to be competitive in 3D benchmarks. If anything, it opens up more global rankings for the GTX 980 Ti to dominate. For the GTX 970 owners, this has no effect whatsoever. Certainly the cost has been reduced from $1000 to $400 for the CPU, but the $1000 graphics card is still a factor. $1400 for CPU+GPU in 3D is still triple the $400 for CPU in 2D. In that regard, the Class idea was addressing this problem much better (data output here).

I don't see any real problems, I see opportunities. The 4 core cpu category should be for the true 3D benchmarks which steponz listed earlier.

 

As for the popularity the "unrestricted CPU" category should not be affected as all GPU submissions (incl. 4 core cpu) should count towards the points in that category. Will be the same as now. As for 4 core CPU it is the most popular CPU so it will probably get enough points. I don't think it should get as much as the unrestricted category anyway, doesn't seem fair.

 

Yeah, you still need the highest end GPU, and from an economical point of view both 4 core CPU + low and mid end classes of GPUs would be better. But that would mean an impossible amount of rankings. If I have to choose one I would go with 4 core CPU over GPU classes. 4 core CPU systems must be the most popular systems on hwbot and that means a 700€ GPU to compete for top rankings. Otherwise it would mean X99 MB, 5960X, extra mem + GPU = >> 700€.

[nickolp1974]

Hasn't this thread gone on long enough now?? Whenever people start having diggs its time to end it! @Massman do you have enough info/ideas to run a poll yet? Lets get voting and lay this to bed and get benching

 

Maybe not a constructive post but a Poll would give an indicator of where the majority want this to go instead of it going round in circles.

 

My vote, all 3D subs when benched on LN2 should get more points for the difficulty and skill level, should not matter if its a £20 or a £1000 vga, the skill needed is still pretty much the same with more points awarded for using multiple cards(LN2)

[Massman]

Yes, plenty of ideas. Now I only need to find the time to make the formulas and get the devs to try something on UAT ...