Alright Jow Forums, I finished my multicore CPU performance formula (GMP). It represents the maximum performance of a CPU when using all cores, and it turns out to be pretty precise, here it is: > MF(2C + T) = GMP where: > MF: Maximum Frequency > C: Number of Cores > T: Number of Threads > GMP: Generic Multicore Performance
As discussed last time this thread happened, you have to account for architectural differences.
Nolan Gonzalez
What even if this? This makes absolutely no sense, you just assign a value to a given CPU, there is zero worth in your formula, it means absolutely nothing. Any other equation is just as meaningless.
>and it turns out to be pretty precise, here it is: It measures NOTHING, at what could it be precise??
I don't even want to think about why you thought adding cores times 2 and threads together would make any sense at all.
Jason Richardson
>check my finely crafted formula to shill ryzen again with self-made proofs the absolute city of AMDrones
First learn how CPUs work and how they're used in programs. Don't forget to learn how OS distributes CPU's power between programs.
Jacob Collins
instructions per clock(clock=hertz)
Benjamin Stewart
I like your idea, but that is not the solution.
Evan Fisher
You're literally just adding the threads and cores with separate weights to them.
No IPC is taken into account.
Charles Edwards
To be clear, something like this will only work within identical architectures.
Henry Lee
>what is CPU architecture and optimizations and cache and fucking everything
Adrian Gray
this.
also, since Ryzen and current Intel have similar IPC, it should work for simple comparisons
Owen Garcia
> The number of instructions executed per clock is not a constant for a given processor; it depends on how the particular software being run interacts with the processor
Austin King
> FX > 8c/8t technically, more like 4c/4t
Oliver Hill
Terrible. Doesn't even account for IPC, not to mention architectural differences. According to your metric, AMD's FX processors are better than CPUs that absolutely rape them.
Christopher Thompson
how do one account for IPC? It's subjective > FX cores not real cores
Blake Jackson
They're about as real as your rating system. It doesn't account for a multitude of factors that affect performance. You just fell for the "muh gigahertz" and "muh cores" memes. What's the point of this when you can effortlessly find actual performance ratings?
maximum frequencies are very important, although I would agree that the maximum all core clock-speed would be much more precise.
Jayden Parker
Real benchmarks would be much more precise.
Carter Clark
> 37 different benchmarks > 37 different results > much more precise I'd rather have one formula that averages every single benchmark difference
Adrian Rivera
go post it on Jow Forumsamd
John Myers
Damn, you mean I should spend more than 5 minutes researching an important purchase? It's not like the OP formula accomplishes what you're asking anyway.
Kevin Martin
OP thinks he's "got it." He doesn't realize the only reason his equation "seems" to work is due to Ryze/Coffeelake has just recently reached some parity, and thus his "neat" equation "seems" to work right now. It breaks down when you take into account of any future CPU or past CPU.
Its a non-relevant formula at this point.
Ryder Powell
Didn't you start this last year?
Julian Barnes
He started it last year
David Johnson
Yeah I remember a retarded thread like this from quite a while ago. I don't think he's really changed his "formula" since then either.
Mason Clark
I remember someone told him to do this and that and he'd consider hit but your right. It doesn't look like anything has changed
lol this is useless wtf an ancient xeon from 2005 with high frequency and many cores would look amazing by this equation, yet it is absolute shit in reality and outdated. also why the 2, just for larger looking numbers?
Joshua Moore
Poopoo >Literal braindead retard. I think you guys are being too hard on the guy. Relax I remember he was planing to have like a co efficient or something for each architecture or series/gen I mean he's thinking at least
Adrian Walker
that coefficient would need to be calculated or else it can be said to have bias since it otherwise would be pulled from nowhere, thata flawed too. It takes no effort to factor the things that matter in like l2 cache, pcie lanes, hardware ram support, etc.
Asher Nguyen
>It takes no effort to factor the things that matter in like l2 cache, pcie lanes, hardware ram support, etc. No I'd think that precisely would make it more difficult
William Cruz
add fucking variebles in for each holy shit!
Julian Walker
that would make the formula huuge and basically have little impact on the results. Frequency, cores and threads are the most advertised data for a reason.
Chase Scott
>let's just sum and multiply all the numbers in the spec sheet to see which cpu is better!
> tfw 2014 Macbook with 54.4 Niggers that's enough niggers for me.
Alexander Cooper
>thats exactly what the fucking original equation is doing, its randomly put together clearly, and its not even accouting for anything besides the now meaningless clockspeed. An ancient cpu from decades ago that has shit peformance can get a higher score then modern top tier cpus. wEW, autism lad.
Ryan Young
>why the 2 Because a core is 2 times more important than a thread for multicore performance.
Austin Phillips
this is wrong, threads scale differently depending on the workload
Sebastian Parker
its an abitrary addition, no point. you can literally make a better cpu perofrmance equation by multipling what mentions together. There is no requirement for how to make the equation up. It can be fucking negative numbers or in the googleplex's. and the fucking units maken no sense.
David Kelly
how much is that in Hitlers though?
Grayson Collins
>It can be fucking negative numbers or in the googleplex's. and the fucking units maken no sense. Good luck, in the end it could end up being 1% more precise than the simple MC(2C + T) formula.
Andrew Nguyen
im pretty sure the threads op is trying to get us to make them a better equation and is using that whole make an erroneous statement on the internet and you get more responses then asking for help. we are not doing the work for you op, figure it out yourself.
Jason Reed
1 NIGGERS = 0.000001 Hittlers
Hunter Jenkins
what in the fuck does that mean??? read you are putting cpus that are far different and older on the exact same level with your equation. A fucking arm arch cpu would do really well since they can be 2ghz 8 core, but note, ARM architecture, vastly different from x86. list of issues goes on.
Connor Howard
>its an abitrary addition not for multicore performance. Each core has 2 threads, that's why it's weight is 2.
William Martinez
I dont think you understand how models work. Sorry.
now throw in some older hexa/quad cores and let's see
Julian White
gmp is GCC dependency
Mason Mitchell
it works with precision if you compare within the same brand/generation. I will compile enough data with those 6 benchmarks and make a new IPC constant for each brand so it will be comparable between different generations and brands.
