Discussion PPL - Performance Per Liter - Round 3 - Compute Edition...?!?

[DrHudacris]

I think you should read more carefully. I didn't say the formulas are equal. I said, and I quote "rankings are equivalent". That means both formulas would produce the same "ranking" of systems.



I'm not sure what you think is incorrect in what I wrote... I said to use directly Volume (measured in liters, as you know I hope?) and not any power of it. [Edit: My second edit in my previous post was not in for some reason. I fixed that now. There I was offering to use Volume directly]

And I hope you understand what a liter is?
1 Liter = 1000000 mm^3
So, yeah, liter is not a distance measure, it is a measure of volume.

Such anger! I would suggest you crunch the numbers yourself, as your conception of how powers can be moved across a quotient (and how that affects rank order) and the concept of a dimension such as liters (or cubic mm or any unit of volume) being weighed out of proportion "because its cubic" are shaky. In your second edit using volume directly is fine, but should not be miscontrued as "already having a cubic power". The individual dimension (L, W, H) are cubed, but the resulting volume is used. Ahh, I see someone has crunched the numbers for you to show that rank order is easily affected by different powers being used! So not equivalent either

Yes. I was wrong by ignoring the power on volume thinking it didn't matter but it does have a very major impact.

Geometric mean for 2 terms = (a * b)^0.5

Current: CPU*GPU / Volume ^0.5
Correct: Geometric Mean / Volume
Incorrect: Geometric Mean / Volume ^ 1.5
Simple: CPU * GPU / Volume ^ 2


Ranking wise they are nowhere equivalent. I only looked at the top 5 but there's a very large difference in how systems are ranked based on the equation used. I was wrong to ignore the power on volume. Like you said, we're measuring performance per liter and putting a power other than 1 on volume makes a drastic difference. My initial suggestion was wrong by ignoring the power on volume. The correct method is to use the geometric over volume with a power of one. Or as you said, the simpler and equivalent equation would be cpu*gpu over volume with a power of 2. The equation in this format may seem like we're giving volume more weight but mathematically speaking.. what's actually happening is we're getting the geometric mean of the cpu and gpu benchmarks.

In conclusion, what we want to measure is performance per liter. If we want to respect that, we have to use either the equations (Geometric Mean/Volume) or (CPU*GPU/Volume^2)

Thank you for illustrating your formulas! The OP of the thread may consider the new formulas in a new round or something, as it might be too complicated to change everything now.

Edit: I think it best to avoid derailing the topic further; this is kind of off topic.

 

[theGryphon]

Such anger! I would suggest you crunch the numbers yourself, as your conception of how powers can be moved across a quotient (and how that affects rank order) and the concept of a dimension such as liters (or cubic mm or any unit of volume) being weighed out of proportion "because its cubic" are shaky. In your second edit using volume directly is fine, but should not be miscontrued as "already having a cubic power". The individual dimension (L, W, H) are cubed, but the resulting volume is used. Ahh, I see someone has crunched the numbers for you to show that rank order is easily affected by different powers being used! So not equivalent either

To begin with, I'm not angry, just trying to tell you that you came off too quick and in fact wrong by telling me that what I said is incorrect. It is clear that you do not know what you're talking about, or insisting on misunderstanding what I said.

I'll walk you through it.

First off, @PLLovervoltage suggested the following:

Geometric mean: ((CPU*GPU)^0.5/(Volume^1.5))

In return, I said, this formula would yield the same ranking as the following:

((CPU*GPU)/(Volume^3))/100

This is accurate, and here is why:

Let's call the outcome of what @PLLovervoltage said "A". That is, A = ((CPU*GPU)^0.5/(Volume^1.5)).
Let's call the "numerator" of what I said "B". That is, B = ((CPU*GPU)/(Volume^3))

Therefore, A = ((CPU*GPU)/(Volume^3))^(0.5) = B^(0.5)

I sincerely hope you can see this, because otherwise we're not going anywhere.

Now, by saying that both formulas yield the same ranking, I'm merely stating that B^(0.5) and B/100 give you the same relative placement of systems with respect to each other. This is a mathematical fact as long as B > 0, which is true.

I won't say anything else, but leave you to your means.

Just don't come and claim that I said anything other than what I actually said. Of course, if you change the power of Volume (without changing anything else) you would change the ranking. I guarantee you that I know my math... just be sure to check yours and/or your reading comprehension

 

[theGryphon]

Yes. I was wrong by ignoring the power on volume thinking it didn't matter but it does have a very major impact.

Geometric mean for 2 terms = (a * b)^0.5

Current: CPU*GPU / Volume ^0.5
Correct: Geometric Mean / Volume
Incorrect: Geometric Mean / Volume ^ 1.5
Simple: CPU * GPU / Volume ^ 2


Ranking wise they are nowhere equivalent. I only looked at the top 5 but there's a very large difference in how systems are ranked based on the equation used. I was wrong to ignore the power on volume. Like you said, we're measuring performance per liter and putting a power other than 1 on volume makes a drastic difference. My initial suggestion was wrong by ignoring the power on volume. The correct method is to use the geometric over volume with a power of one. Or as you said, the simpler and equivalent equation would be cpu*gpu over volume with a power of 2. The equation in this format may seem like we're giving volume more weight but mathematically speaking.. what's actually happening is we're getting the geometric mean of the cpu and gpu benchmarks.

In conclusion, what we want to measure is performance per liter. If we want to respect that, we have to use either the equations (Geometric Mean/Volume) or (CPU*GPU/Volume^2)

Thank you for actually getting it!
When you keep the numerator constant and keep changing the exponent of the Volume in the denominator, you will certainly impact the rankings.

I truly believe PPL measure should be Performance / Volume, as the name implies. Then, one just needs to define what "Performance" is.

I totally see your point in using the geometric mean now. By simply multiplying the CPU and GPU performance metrics, we are in a way using "Performance^2" in the numerator. If "Performance" is supposed to be a singular "summary" measure, using the geometric mean actually makes great sense.

So, yeah, you convinced me, (Geometric Mean/Volume) or (CPU*GPU/Volume^2), which are equivalent (right, @DrHudacris ?) should be used.

Digging deeper, "CPU" itself can be a geometric mean of two different CPU benchmarks, as well as "GPU" being a geometric mean of two different GPU benchmarks. It would be more work to submit results, so maybe this will not be preferred...

 

[DrHudacris]

Thank you for actually getting it!
When you keep the numerator constant and keep changing the exponent of the Volume in the denominator, you will certainly impact the rankings.

I truly believe PPL measure should be Performance / Volume, as the name implies. Then, one just needs to define what "Performance" is.

I totally see your point in using the geometric mean now. By simply multiplying the CPU and GPU performance metrics, we are in a way using "Performance^2" in the numerator. If "Performance" is supposed to be a singular "summary" measure, using the geometric mean actually makes great sense.

So, yeah, you convinced me, (Geometric Mean/Volume) or (CPU*GPU/Volume^2), which are equivalent (right, @DrHudacris ?) should be used.

Digging deeper, "CPU" itself can be a geometric mean of two different CPU benchmarks, as well as "GPU" being a geometric mean of two different GPU benchmarks. It would be more work to submit results, so maybe this will not be preferred...

Ahhh it's my reading comprehension! Thank you for the explanation, I see where I made the error (equivalent rank order vs mathematically equal). And I misunderstood the comparators (the suggested formula vs your simplification of it not current formula vs suggested formula).

 

[AURMEND]

You guys and your maths. I'm just here waiting for Petricor to beat my score.

 

[CubanLegend]

i mean isn't this a measurement of 3 things, cpu GPU and liters? so shouldn't all 3 be weighted equally? thus allowing someone to have fewer liters to be rewarded for packing it all into a smaller case? we're rewarded for CPU and GPU but are we rewarded to the same degree for volume?

And another thought.. isn't this performance PER LITER, then liters of volume should certainly be considered to be weighed just as equally as CPU/GPU power, otherwise only those with more powerful setups will always win? idk just thinking out loud

 

[theGryphon]

i mean isn't this a measurement of 3 things, cpu GPU and liters? so shouldn't all 3 be weighted equally? thus allowing someone to have fewer liters to be rewarded for packing it all into a smaller case? we're rewarded for CPU and GPU but are we rewarded to the same degree for volume?

And another thought.. isn't this performance PER LITER, then liters of volume should certainly be considered to be weighed just as equally as CPU/GPU power, otherwise only those with more powerful setups will always win? idk just thinking out loud

CPU, GPU and Volume? Well, not exactly...
It's about "Performance" per "Liter". Liter is easy. So, we need a good definition of what "Performance" is.

Brainstorming: If we start counting the performance of components in a computer system, we should then start talking about Memory and Storage as well? Is multiplication of individual component benchmarks a measure of "performance"? Each being "components" of how a system "performs", shouldn't we bring them together in a smarter way?

Thinking along these lines, taking the geometric mean does appear to be a very sound method. The beauty of it is, one can think of more dimensions/components of what system "performance" is and still put them all together in a geometric mean: (multiply N measures)^(1/N)

 

[PLLovervoltage]

i mean isn't this a measurement of 3 things, cpu GPU and liters? so shouldn't all 3 be weighted equally? thus allowing someone to have fewer liters to be rewarded for packing it all into a smaller case? we're rewarded for CPU and GPU but are we rewarded to the same degree for volume?

And another thought.. isn't this performance PER LITER, then liters of volume should certainly be considered to be weighed just as equally as CPU/GPU power, otherwise only those with more powerful setups will always win? idk just thinking out loud

The power of 2 on volume makes it seem like we're weighing volume more than cpu and gpu. But it's not. This is basically like how 26% of 50 is the same as 50% of 26. Mathematically it's the same but on face value to us it seems like completely different values. Using the geometric mean IS weighting the three equally. How it is currently is not. CPU*GPU/Volume^2 is equivalent to the geometric mean.

 

[CubanLegend]

The power of 2 on volume makes it seem like we're weighing volume more than cpu and gpu. But it's not. This is basically like how 26% of 50 is the same as 50% of 26. Mathematically it's the same but on face value to us it seems like completely different values. Using the geometric mean IS weighting the three equally. How it is currently is not. CPU*GPU/Volume^2 is equivalent to the geometric mean.

wait so right now all 3 arent being measured equally? hmm.. yeah i don't like the sound of that lol.

and @theGryphon i see your point, but when it comes to performance, enthusasts usually use 3DMark and a good measure outside of game benchmarks... and those synthetics only measure CPU and GPU power.. hence why we use those 2 metrics, they have the most impact on performance, well above memory and hard drives.

 

[theGryphon]

@CubanLegend No, I'm not saying we have to include memory or storage benchmarks. The point is, taking the geometric mean of whatever number of benchmarks is the way to go. At this point, CPU and GPU measures are sufficient, I agree.

TimeSpy is totally relevant and should be used IMO.

But CB r20 is not doing a cover-all evaluation of CPU performance. Plus, it's too short IMO. A longer, more intense test would be ideal. Such a test would also make sure the system is "stable" at that performance.

Linpack Extreme (current 1.1.3) benchmark is something that comes to mind. Take the average of 5-run benchmark there, for example. Score is given as GFlops...

 

[Aux]

PPL June Update

Regarding the math discussion . . . it would be a simple matter to change the power if we all agree . ..

 

[deprecatedcoder]

Regarding the math discussion . . . it would be a simple matter to change the power if we all agree . ..

Given I go from 4 to 3, I am down ?

 

[petricor]

You guys and your maths. I'm just here waiting for Petricor to beat my score.

Your wish is my command
First attempt with vanilla settings for everything apart from the asrock bios in "performance" mode:

((9454*13832)/(5.0^1.5))/100=116,962

CPU: R9 3950x
GPU: RTX 2080ti
Case: NFC Skyreach 4 mini

 

[Aux]

Your wish is my command
First attempt with vanilla settings for everything apart from the asrock bios in "performance" mode:

((9454*13832)/(5.0^1.5))/100=116,962

CPU: R9 3950x
GPU: RTX 2080ti
Case: NFC Skyreach 4 mini

Running open air as pictured?

 

[Boil]

Your wish is my command
First attempt with vanilla settings for everything apart from the asrock bios in "performance" mode:

((9454*13832)/(5.0^1.5))/100=116,962

You need to include CPU-Z with the Cinebench screenshot & GPU-Z with the Timespy screenshot...

 

[CubanLegend]

So I was recently able to concoct something delightfully devilish



Yup, your eyes do not deceive you. I've managed to fit the new HDPLEX 400W AC-DC v2.0 in my S4M-C, so it's now a BRICKLESS, S4M-C! So, I'm down from 4.9L to possibly 4.3/4.2L.. I will remeasure and double check the case measurements once im done doing some wire cutting, heat shrinking, and wire cramming/tucking, and after i remove the "horizontal" feet i never used on the case.

Do yall need to to resubmit scores or just update my volume once i'm done re-measuring?

 

[Aux]

So I was recently able to concoct something delightfully devilish


Yup, your eyes do not deceive you. I've managed to fit the new HDPLEX 400W AC-DC v2.0 in my S4M-C, so it's now a BRICKLESS, S4M-C! So, I'm down from 4.9L to possibly 4.3/4.2L.. I will remeasure and double check the case measurements once im done doing some wire cutting, heat shrinking, and wire cramming/tucking, and after i remove the "horizontal" feet i never used on the case.

Do yall need to to resubmit scores or just update my volume once i'm done re-measuring?

Hi

very nice ..

yes please do.

thanks

 

[petricor]

Running open air as pictured?

Yes - all as pictured... will post another one with case closed and cpu/gpu-z!

 

[petricor]

You need to include CPU-Z with the Cinebench screenshot & GPU-Z with the Timespy screenshot...

Here we go - this time with the case closed, slightly different results:
Today it's a bit cooler than yesterday so things run faster - ambient temps on a hot day (it's brutal at the moment...) play a role as the build is surfing on the edge of thermal capacity.
Will tinker a bit more to see where I can take this - I feel that the GPU has more potential looking at some of the other scores here!

((9529*13833))/(5.0^1.5)/100=117,899

CPU: R9 3950x
GPU: RTX 2080ti
Case: NFC Skyreach 4 mini

 

[Aux]

Yes - all as pictured... will post another one with case closed and cpu/gpu-z!

thanks