GFlops comparison

[5T34LTH]

i wanna discus something im look in right now, i wanna know if Gflops are a good way of comparing cpu, we all know that GHz is a terrible way of comparing 2 different CPU's right, now for Gflops, would that be accurate to calculate the power of CPU's, i dont want to compare different skills, the simple power, but while using the same program, what i want to know is, if its has much accurate has, MB/s for SDD or HDD

lets say i test 2 systems, one has 2 Gflops average, one has 5 Gflops average(with the same tool) does that mean that the first one is crappier, or is it nothing to look at?

 

[nna2]

a processor cant really be compared on 1 thing, although 1 scores higher in FLOPS, what about a program like geekbench or cinebench

 

[Tamz_msc]

Using just one variable to decide the performance is not correct. There are so many factors that determine CPU performance - architecture, core count, clock speed, fabrication process etc. You should decide what CPU is best for you by reading reviews from different websites, consider your requirements in terms of performance and price and only then you can come to a conclusion.

 

[esrever]

since most people don't use heavy float point calculations, having high flops is pointless in everyday computing.

you can have a 20 billion terra flop system but it wouldn't do anything cause nothing would run.

 

[5T34LTH]

to esrever, sorry, but a 20 billions terra flop? im talking about 1 gigaflop vs 5 gigaflop, from the tests i took it looks like 10-20 gigaflop is a good number for an average user, to get back to your subject, if i would have 20 billion terra flop......

to the other 2 ones up there, from what i saw, this single calculating form is more accurate then any of these forms alone, what i want to calculate is how much information can a cpu use
of course theres also other things to keep in mind, like power consumption, price range, and before all, readability

 

[5T34LTH]

but from what i saw, it says without complex things to look at how much power a computer has to give in CPU, what i would like to do is be able to compare 2 cpus and show it to some one stupid and make him understand how much more power there is to it then... GHz which is what pple stick to

 

[nna2]

^if there really that stupid let them stick to GHz

you clearly didnt understand what eserver was getting at....

but core arcitecture is a big one (pardon my spelling )
a first gen i7 @2.8 GHz can blow my phenom II propus @3.8GHz in cinebench by a whole point

 

[FinneousPJ]

I believe everyday computing is mostly integer operations, not floating point ones.

 

[5T34LTH]

he meant that if you get a computer able to power 10 times the game most of it is going to be useless, i know that, he dint get what i was getting to nether(thats why i responded this) my point was to be able to compare something to get the right processor more easily

listen, im trying to see if Flops if a good way of calculating for my clients, form my tests it looks like its more accurate then you think it is

when im buying a CPU im looking for the power/value, like in fallout or skyrim, you want the best thing at the best weight, here what im doing is trying to create my self a short cut, if you could have a number that would eliminate all the things you got to look into and that tells you directly what kinda power you have, it would be great

ghz only works for 2 simylar cpu, its like saying hourse power(does not say allot)

btw, your spelling is great, i dont get what your talking about :L

 

[ElMoIsEviL]

If you want to compare the compute power (the processors ability to hash through heavy mathematical loads) then yes... GFlops.

If you want to compare execution speed though... that's different. Then you'd want to be looking at ALU performance in 3 key areas:

1. Calculation in a single clock

2. Calculation pipeline performance

3. Iterative calculation performance

 

[esrever]

flops only work for the same cpus and will just scale linearly with clock rate.

cpus don't even do mass float point operations so giving up lots of die area to make it do more flops makes no sense.

the part of the cpu that handles floatpoint is completely different than the integer calculations parts which is what most everyday computing will be. there is no point in looking at float point performance for consumer cpus.

since theres almost no data on cpu float point calculations where it is remotely useful, you can look at gpu float point performance. the 6970 does 2703 gflop the gtx 580 does 1580 gflops. does that make the 6970 x2 as powerful as the gtx 580? no.

 

[5T34LTH]

If you want to compare the compute power (the processors ability to hash through heavy mathematical loads) then yes... GFlops.

If you want to compare execution speed though... that's different. Then you'd want to be looking at ALU performance in 3 key areas:

1. Calculation in a single clock

2. Calculation pipeline performance

3. Iterative calculation performance

what exactly would you mean by heavy mathematical loads, would you mind showing software examples or operation examples, do you mean that in no ways(almost non i guess) would flops help to compare how fast a computer would pass a scan(virus scan) and that i would be more helpful to have better ALU performance in the 3 specific points for speed?

the 2 points you discussed there, complex vs simple data? or something that in dint get

and how and with what would i calculate these 3 points you got there from the ALU? you got some special software i dont know about?(im probably gonna start a search right after this anyway....)

thx for the great answer.... hopefully i understood well.....

 

[esrever]

you can go look at a synthetic benchmark and get a general picture of how powerful a cpus.