Gflops real world meaning

rik81h

rik81h

rik81h

Associate
Joined
4 Nov 2009
Posts
81
Location
Stoke On Trent
I've had a quick blast with ibt lately, just got v2.51 although don't think it has any impact on my processor (555BE unlocked to x4) with avx?? anyway was just wondering if the result which obviously can be compared to other processors, how much impact this result actually has in real world usage. I average just over 50gflops no matter the stress level.
Cheers
 
rik81h said:
I've had a quick blast with ibt lately, just got v2.51 although don't think it has any impact on my processor (555BE unlocked to x4) with avx?? anyway was just wondering if the result which obviously can be compared to other processors, how much impact this result actually has in real world usage. I average just over 50gflops no matter the stress level.
Cheers
Click to expand...


You have to first of all understand what does Flops actually mean?

Flops stand for Floating Point Operations Per Second.

In simplistic terms Floating point operations involve multiplication, division, subtraction, addition, trignometric functions etc of real numbers involving decimal points.

Afaik all modern CPUs are native integer processors thus can't carry out real number calculations on their own. What does carry out flops calculations is known as Floating Point Unit (FPU) or math co-processor. FPU may well be part of CPU or separate chips on mobo.

Flops can involve many different kind of instructions such as division, trignometric, square root etc.

Intel Burn Test makes use of multi-add instructions. So multiply two real numbers and then add the final result to an initial value.

http://en.wikipedia.org/wiki/Multiply-accumulate_operation

Intel Burn test is also a double precision (64-bit) stress test program. Double precision in simplistic term just means that very large numbers with more decimal points can be computed compared to single-precision (32-bit) programs.

Modern cpus excluding Sandybridge such as Core 2, nehalem, amd phenom can carry out 4 Flops or 2 multi+2add instructions in a single CPU cycle in double precision.

The speed of a cpu such as 3GHz means it can operate 3 billion cycles per second.

If one core can carry out 4 flops per cycle then 4 cores processor can carry out 16 flops per cycle.

So a 3 Ghz Quad can theoretically carry out:

4 cores x 4 flops per core per cycle x cpu speed (total no of cycles generated per second) = 4 x 4 x 3Ghz = 48 Gflops or 48 billion floating point operations per second.

AMD 555BE doesn't support AVX instructions which just double the flops from 4 to 8 in double precision and is supported in Sandy Bridge and upcoming AMD Bulldozer.

Flops is a measure of cpu processing speed and is often used as a benchmark to compare the processing speeds of different cpus.

Intel Burn Test stress level just means the amount of ram you want to use for the test. IBT ram is used to store matrix problem size which then gets solved using Gaussian Elimination method and results are outputted in Gflops.

The higher stress level or more ram you use, the bigger matrix size will be and during each run, your cpu will be under greater duration of load. Ram usage is independant of Gflops ;). Higher stress level won't get you more Gflops for the same cpu speed.

I hope it hasn't been too complicated :p although other members such as xsistor can explain you better although he might make it even more complicated to understand :p
 
Last edited:
Thanks for that WingZero30. Although i was aware of a portion of that information you're mathematical explanation definitely clarified some things. However rather than a technological definition i was looking for a more lamen term to compare system speeds i.e would a 2600k system that can output 140gflops be over twice as fast as said spec in everyday use, encoding etc given that the measurement is of system computational power not just cpu such as smaller die faster clock etc etc obviously they'd be two different systems anyway one more up to date hence would the benefits be as great as one might first think?
Thanks, Richard
 
rik81h said:
Thanks for that WingZero30. Although i was aware of a portion of that information you're mathematical explanation definitely clarified some things. However rather than a technological definition i was looking for a more lamen term to compare system speeds i.e would a 2600k system that can output 140gflops be over twice as fast as said spec in everyday use, encoding etc given that the measurement is of system computational power not just cpu such as smaller die faster clock etc etc obviously they'd be two different systems anyway one more up to date hence would the benefits be as great as one might first think?
Thanks, Richard
Click to expand...

For AMD [email protected]

4 cores x 4 flops x 3.755Ghz = 60.08GFlops (theoretical peak at 3.755Ghz)

For i7 [email protected] (AVX instructions set)

4 cores x 8 flops x 4.375Ghz = 140GFlops (theoretical peak at 4.375Ghz assuming HT is disabled)

2600K can only produce 140Gflops with higher clock speed and inclusion of AVX instructions.

No doubt 2600K is faster clock for clock. However theoretically both cpus still can only execute 4 flops per cycle per second in double precision if no avx is included. It may be the case that Sandybridge includes additional instructions that can somehow speed up the execution rate.

Perhaps someone with computer science background can explain it more concisely.

However Ram latencies and speed can also affect Flops.

As for encoding I am not sure how cpu processes that sadly.
 
[email protected] without AVX

4 cores x 4 flops x 4.375Ghz = 70GFlops


So looking at it this way, it won't be much faster than AMD 555BE in terms of Gflops output although clock for clock it will still be ahead.

2600K will dominate AMD 555BE in those applications that make use of AVX instructions set.


Edit: In this comparison you can see 2600k miles head of amd 555BE

http://www.anandtech.com/bench/Product/120?vs=287
 
Last edited:
I like that comparison on anandtech, shame it doesn't account for the unlock but i'll have a look at a similar quad variant tomorrow. I haven't swatted up on the sandybridges yet but how are the avx and ht linked? i'm sure i found a bench that had ht on on a 2600k and it still hit 140gflops? Maybe it was a typo or my mistake. I suppose the point of this is i love my b55 for its performance per pound but as all men i yearn for the shiny new tech of the 2600 and everything seems slow after a while doesn't it lol. I have noticed myself on clocking other systems that at 4ghz and beyond does have significant performace improvements its just my chip will never do that so there's an immediate benefit to the upgrade.
 
rik81h said:
I like that comparison on anandtech, shame it doesn't account for the unlock but i'll have a look at a similar quad variant tomorrow. I haven't swatted up on the sandybridges yet but how are the avx and ht linked? i'm sure i found a bench that had ht on on a 2600k and it still hit 140gflops? Maybe it was a typo or my mistake. I suppose the point of this is i love my b55 for its performance per pound but as all men i yearn for the shiny new tech of the 2600 and everything seems slow after a while doesn't it lol. I have noticed myself on clocking other systems that at 4ghz and beyond does have significant performace improvements its just my chip will never do that so there's an immediate benefit to the upgrade.
Click to expand...


With hyperthreading enabled, cpu clock speed can be lowered and yet still achieve 140Gflops. Some people have managed 150GFlops or higher on 2500K simply by overclocking further.

But GFlops is just one way of measuring your PC performance and isn't the only benchmark. As seen in anandtech comparison tets, there are other benchmarks aswell and utilising different benchies give a broader overall perspective of your system perofrrmance. Though anadtech comparison involve tests at stock speed but still it gives a rough idea.

If you want to upgrade, I would suggest going for more cores like 6 or 8 cores cpu in the future instead of just clock speed. Wait for AMD Bulldozer aswell and see how it fares against SB then you can make a better judgement.

I am on Q6600 which I purchased last year and intend to use it for 2-3 years more as I mainly game
 
Last edited:
Well thanks for your help and info WingZero30, i did briefly contemplate an amd hex core however a friends 1090t didn't inspire me much for the build cost but i shall wait to see how bulldozer fairs against SB. Intel hex cores are too much money for my liking currently. I'm certainly happy with my current rig i just want more lol.

As for biuro's post that's over my head unfortunately :)

Thanks again guys
 
You must log in or register to reply here.
Back
Top Bottom