Alright Jow Forums, yesterday I made a thread, and you still didn't believe my formula works. Today I will be proving it actually does work, with a 2% margin of error.
TLDR: I made a simple formula that calculates the multicore performance of any CPU by only getting it's maximum clock speed, number of cores and number of threads (pic related) Right now it only works when comparing between the same brand and generation, however, I'm working on a IPC constant that's based on benchmark data, when that's over, you will be able to compare between different generations, Ryzen vs Intel etc. > why? Many CPUs, specially newer and lower end ones don't have much, if any benchmark data online. With this formula, you can preview and compare it's performance in a few seconds. It's useful to preview overclocking gains as well.
>Bump up memory clock/tighten timings >Suddenly numbers are completely different >Bump up core clock on different architectures >Behaves completely differently How much do you even know about modern CPU architectures?
>62.4 - 44.4 = 40.5 what? 62.4 - 44.4 is 18.0 the math in is accurate, but your math is still off in GMP by 0.3 what the fuck kind of math are you doing user
more like >Bump up memory clock/tighten timings on both chips >Suddenly numbers are completely different >... >however, still proportional to each other
o ok didn't even think % difference, my bad user, it's been a long day
anyways, how did you derive this formula?
Brayden Bell
>a formula that doesn't consider IPC is accurate when you compare CPUs with similar IPC >this is somehow a proof Are you fucking retarded or just pretending?
Hunter Davis
>anyways, how did you derive this formula? last year I was taking a bath, and it just popped in my mind.
IPC doesn't matter when comparing the same brand/generation since it's pretty much the same chip. As I said, I'm working on a IPC constant based on real world data so you will be able to compare it between different generations and brands, it will look something like this: > (MF * IPC)(2C + T)
Jayden Allen
Oh God user do you know something about performance evaluation?
>IPC doesn't matter when comparing the same brand/generation Then what's the point of the formula? You might aswell compare model numbers.
>adding a magic parameter to fix wrong results Do you realize you're going to need a constant for every chip? What's the point of your formula if it's derived from real benchmarks anyways? Why would anyone use your retarded formula instead? What if a new series of CPU is released? You're going to have to calculate a magic constant for that too.
Seriously, this is so fucking retarded, it's like claiming x is always equal to x^2 if you multiply every point by x^2/x
Lucas Rivera
>then what's the point of the formula? look at the proofs, it's useful to know how fast a CPU is compared to another one on multicore workloads. Read OP. > Do you realize you're going to need a constant for every chip? wrong. I already proved you can compare different chips with precision. Only CPU generations will need a constant (Like Ryzen 1 constant, Coffee-Lake constant, etc)
> What's the point of your formula if it's derived from real benchmarks anyways? It's not. Only the other formula will be.
> Why would anyone use your retarded formula instead? because it's easier than looking at several different benchmarks, some of them biased.
> What if a new series of CPU is released? Well, that's another reason for using the formula. If intel releases the specs of the new i7 9700K and i5 9600K, you will be able to know how much faster the i7 is even before the benchmarks come out.
> t's like claiming x is always equal to x^2 if you multiply every point by x^2/x wut?
Dude you don't even know how stupid this is. Just leave and we'll ignore this ever happened. If you keep going down this road you're going to become a meme.
Jayden Sanchez
>look at the proofs where are the proofs? you mean your formula fits some point? how many?6? Do you know how to validate a model? if not at least try to justify your formula. Come on user I know you are thrilled because you think you found something cool but what you are presenting here is not enough to take you seriously, it only shows, I'm sorry to tell you, lack of education in the subject
Brayden Cox
>you're going to become a meme But that's by dream user!
Blake Phillips
dude, I'm claiming to have made a scientific discovery or anything, it's just a neat trick and apparently works with some precision every single time.
Daniel Scott
I'm not claiming*
Logan Wright
Can you explain why this works and/or give a bunch more varied examples?
Xavier Moore
>look at the proofs, it's useful to know how fast a CPU is compared to another one on multicore workloads If I need to lookup the constant for the CPUs I want to compare, I might aswell search for benchmarks.
>only CPU generations will need a constant (Like Ryzen 1 constant, Coffee-Lake constant, etc) That's want I meant with "every chip"
>It's not. Only the other formula will be. So the first one is totally useless and the second one is the same as looking at benchmarks, great.
>several different benchmarks In that case, why not making a site that collects data from different sources instead?
>some of them biased But you're pulling data from there anyways.
>If intel releases the specs of the new i7 9700K and i5 9600K, you will be able to know how much faster the i7 is even before the benchmarks come out. No, you won't. Maybe you could get away with it, considering that it's most likely going to be another Skylake re-release, but for Zen2 or anything else, you're fucked.
Charles Wilson
Okay, now go back 5 generations and try it again, you fucking retard. I remember you posting this retarded shit like a year ago or something and you were mocked just as much then, because it doesn't make any fucking sense. You think legions of computer engineers for the last 50 years are so braindead they didn't see something so blatantly obvious? It HAPPENS to work in this instance. Try it with literally any other generation, go try it out with some P4s vs a c2d fairs.
Kevin Robinson
Because that isn't how clocks work you fucking retard Making it go more fasterer is only one tiny fragment of how processors have worked for more than the last decade, we've known this for a long fucking time and is why architectural changes have been more important than GO FASTER, it's only because of thermal allowances afforded by sub 40nm manufacturing that we get to conveniently also enjoy 4ghz clockspeeds as a standard now
Jonathan Sanchez
>and apparently works with some precision every single time. You're the reason /sci/ mocks us
Ryan Scott
>why this works because clock speeds, core count and threads are linked. In theory, a 4GHz 4c/8t CPU should be 2x faster (multicore) than a 4GHz 2c/4t one of the same generation.
That's exactly what the formula does. > 4GHz 4c/8t: 64 GMP > 4GHz 2c/4t: 32 GMP
Obviously, it doesn't work when IPC is different.
Benjamin Collins
>In theory, a 4GHz 4c/8t CPU should be 2x faster (multicore) than a 4GHz 2c/4t one of the same generation. ...no no it should not that is not how those things work you are an idiot go back multiple generations and keep repeating this test and you'll see exactly how wrong you are
Landon Green
I believe in you, op! don't listen to what people are saying.
Luke Taylor
>no it should not >that is not how those things work that's not what multiple multicore benchmarks say.
Jacob Nguyen
>t. OP
Levi Wright
Only problem I see here is application, I don't really see this to be any easier to just look at some real world benchmarks as not every software is made equal
Logan Johnson
6 data points is not sufficient evidence to prove any kind of mathematical model, you ingrate You also directly stated that if IPC is different it doesn't work Well /guess fuckin what/, IPC is constantly different Unless your formula exists to tell us "The new i3 will be slower than the i7" in which case, no shit?
Luis Powell
Idiot
Carter Gomez
>IPC is constantly different IPC is identical on chips of the same brand/generation, since it's the same fucking CPU. > Unless your formula exists to tell us "The new i3 will be slower than the i7" It actually tells how much faster it will be, with a 96% precision.
Carter Perez
>Gee bill, in an ideal world how much stronger will this 4 cylinder engine be than this 2 cylinder? >Well bob, I'd reckon about twice as fast, got twice as many cylinders! And again A) You have not shown any level of precision beyond a happenstance, 6 comparisons is literally fuckall data B) 96% precision is pretty fucking shitty unless you're entering the realm of "Literally just guessing at human interaction now" C) If all this """formula""" is meant to show is differences between same-generation, same-architecture CPUs but with different core/thread counts, then it's literally worthless. Will this 8 core CPU of the same architecture do better than this 4 core CPU in a well optimized multithreaded task? GEEEEE I DUNNO
Lincoln Thomas
>96% precision is pretty fucking shitty sorry, meant 98% >Will this 8 core CPU of the same architecture do better than this 4 core CPU in a well optimized multithreaded task? everyone knows it will, but by what margin? that what the formula does.
Joseph Rodriguez
More data that really makes you think > 2700X vs 8700K
Margin of error is not a measure of how off your number is compared to the "real" one.
Aaron Barnes
damn, you're right. My mistake. I'm not a scientist as you may have noticed.
Aaron Perry
it will not be arbitrary at all. It's based on real-world results.
Easton Reed
>it's based on the results I want to get
John Nelson
It's based on an average of several multi-core benchmarks.
Since the formula is meant to calculate multicore performance, I don't see anything wrong with it.
Chase Rivera
Now compare ARM and x86 CPUs with your scoring system, or even shitty old Xeons with a modern CPU. Your formula is worthless. And what's the point of calculating some unique constant per CPU based on real world results? Your just skewing actual benchmark data with your awful formula. Why would someone ever use this instead of spending 2 minutes looking up some relevant benchmarks?
Josiah Carter
Your formula is retarded and calculates nothing.
For fucks sake, at least show the correlation between your formula and actual scores, doing this is just RETARDED. This is LITERALLY 0/10 for your methodology.
Also, you have zero clue about mathematics, your formula is ABSOLUTELY worthless, since any retard can come up with a better one BY JUST DOING A BEST FIT ON THE ACTUAL DATA. There is no need to guess your Formular if you have a mathematical method of communication up with the BEST POSSIBLE FORMULAR.
2% margin of error is A DESASTER, just DO TGE FUCKING BEST FIT POLYNOMIAL.
Lincoln Reed
>I don't see anything wrong with it. You want a best fit, so do a FUCKING BEST FIT. No need to guess.
No, that's a 5.22% margin of error. (0.403 - 0.383) / 0.383 = 0.0522
Asher Perry
So you're just comparing cpus from the same brand and generation. That's fucking useless. You think you've come up with a neat formula, but some basic statistics will give you a best fit that's even more accurate. Even that would be useless though, since it tells you nothing about the differences between architectures and generations.
James Gray
calm down Steve, my formula won't take your business down.
Adrian Fisher
>2% margin of error is A DESASTER, just DO TGE FUCKING BEST FIT POLYNOMIAL. the butthurt is real