The performance per clock formula for SPi?

[KTE]

KTE: The formula you state is right, but used completely out of context here.

Did you read what GHZ and Gautam mentioned? That's the formula they use. It's not out of context at all and I know this from Penrose who was one of my professors for 2 years. They used it to analyze CPU efficiency when comparing one CPU to another at the Uni. If you understand what I'm saying, then it'll be quite clear: works out processor cycles taken to complete a fixed benchmark.

That will always stay the same for a given processor as long as you keep every single other thing constant. It can be used in reverse to find out what clock speed someone ran the benchmark at as long as you know the time taken.

That's what these guys are working out but giving it a different name (PP) and meaning to (etc).

But it has flaws like you said: the processor is not the only thing in the equation here. You have dozens of variables which can effect SPi time.

SPi does not rely solely on CPU MHz. If it was, there would be 1 single time for a given clockspeed.

That's exactly what I've been saying on why the formula is not accurate to find a cheater from non-cheater.

[massman]

But it has flaws like you said: the processor is not the only thing in the equation here. You have dozens of variables which can effect SPi time.
That's exactly what I've been saying on why the formula is not accurate to find a cheater from non-cheater.

It gives an overall image, you can easily spot cheaters if the PP is too low.

[massman]

how about 45K at 5GHz

With the new yorkies, close to 45k is quite good

[KTE]

Yep, it's an estimate. Can only work for scores very high up but not always.

Meaning: any person can use it to work out a combination to get a specific PP and then run higher clocks, show that time with clockgen/setfsb down but actually have run it at faster clocks, so their own PB might have been much higher (slower time). Badly loopholed. If I switch on my Intel system soon I'll show you what I mean.

[Gautam]

Whether someone beats a time and he's called a cheat and so on, is usually up to the masses to lobby against. If they like and know the person well, it'll be accepted and praised and if not, it won't be and the person slandered instead.

Indeed. No arbitrary number is needed to see the issue at hand...one look at the graph T_M posted is enough to see where the issue is. But I'd rather we steer away from that.

Btw, for OPB, while the Gregory-Leibniz series does converge to Pi/4, SuperPi does not use it to calculate Pi. It uses the Gauss-Legendre algorithm, which is much more efficient in practice.

[tiborrr]

With the new yorkies, close to 45k is quite good

It's hard to compare well tweaked run done at performance level (e.g. FSBRAM strap) 5 and one done at PL 7... Some memory dividers are just bugged and cannot work as low as others.

Imho the performance ratio is the only certain indicator of an 'healthy' run when knowing the performance level ran at.

[dduckquack]

predictive at most, nevertheless, helpful. thanks for the explaination gautam

[massman]

It's hard to compare well tweaked run done at performance level (e.g. FSBRAM strap) 5 and one done at PL 7... Some memory dividers are just bugged and cannot work as low as others.

Imho the performance ratio is the only certain indicator of an 'healthy' run when knowing the performance level ran at.

That doesn't change the fact that 45K is very good. Try it, you will have trouble to beat it

[CapFTP]

tuned....i started being interested in this topic after I ran this pi and it seems it's quite a good efficiency....

what do you think about it ?



this means 8.656x5220=45184 ?

i ask because if you take a look in HWbot seems quite good related to neighbourhood

http://www.hwbot.org/hallOfFame.do?t...pplicationId=3

[Rol-Co]

but can be better....



Efficiency Asus Maximus Formula.

[CapFTP]

yes , fine....

could it be a trend for highest clocks ?

[Rol-Co]

just not to compare to 5ghz+ , higher clocks you loose some efficiency, it always was...

[dinos22]

just not to compare to 5ghz+ , higher clocks you loose some efficiency, it always was...

r'ly?

not tweaked at all

ram nowhere near maxed out



this would easily be below 45K with maxmem and faster RAM or even 6-5-5-x timings

[massman]

Performance Product of the Yorksfield cpu's

Dino, would you please provide me a run with a PP < 45k ?

[Rol-Co]

Dinos22
i acualy ment when you maxed out on cpu ....on 1 core , look at top score's hwbot,nobody came close to 45k round.

nice eff for ddr3 btw...do you think x38 is faster with ddr3 instead of p35?

[dinos22]

Performance Product of the Yorksfield cpu's

Dino, would you please provide me a run with a PP < 45k ?

i haven't got a yorkfield any more

just waiting for E8500 chips to show up so we can start getting into dual core superpi action

but as i've shown going below 45K will not be a problem even at 5GHz+

44K is not possible for me but that doesn't mean it is not possible for others (or i should say not possible with me with the current tweaks or lack of them)

with the CDT tweak working the way OPB demonstrated in 1M 44K flat looks possible. The only problem is OPB isn't explaining it properly or not showing all the info even to closed forum sections as i have not seen a single person repeat 1M performance at his levels

i haven't made any major ground on normal coppy wazza compared to CDT tweak instructions either

[dinos22]

Dinos22
i acualy ment when you maxed out on cpu ....on 1 core , look at top score's hwbot,nobody came close to 45k round.

nice eff for ddr3 btw...do you think x38 is faster with ddr3 instead of p35?

you are right about efficiency dropping off and DDR3 helps in any case

however i feel with the RAM timings hipro had on his run he should be running more efficiently (maybe it's the mobo )

for those that don't know he had DDR3 running at 1800MHz 6-6-5-x and super tight subtimings with 58xxMHz

[content removed due to thread cleaning] - STEvil

[massman]

but as i've shown going below 45K will not be a problem even at 5GHz+

I'm not convinced , sub 45k is a problem

[Gautam]

Nah, I think its best we not discuss it all anymore, period. I guess I was more exasperated than anything. This thread should deal simply with performance product calculations. I would prefer seeing it cleaned myself, but that's not up to me.

[STEvil]

I agree, cleaning would be nice.. so i'm going to lock it for a few minutes and clean it up some.

edit

Thread cleaned, where you guys go from here is your decision.

[Rol-Co]

I'm not convinced , sub 45k is a problem

Closest i could get ....Asus Maximus+ddr2

45016

used eram/realtime/memset only
win xp sp2
With a bit luck though i must say, most of the runs are 11.265..

[T_M]

Im still floating around in the mid 46k's, but that was rushed testing on unoptimsed system

[dinos22]

Closest i could get ....Asus Maximus+ddr2
[IMG]http://82.173.172.10/Rol-Co/4ghz-pi-air-eff.JPG
45016

used eram/realtime/memset only
win xp sp2
With a bit luck though i must say, most of the runs are 11.265..

eram doesn't work that well for me for 1M

why not try maxmem instead

[KTE]

Thanks for cleaning the thread. Makes very good sense.

Let's wait for some Wofldale testing.

So what's the "maximum PP" or "efficiency" as some of you call it for a 450x8 1:1 4-4-4-4 2T on C2D?

[dinos22]

it cannot be predicted man
fastest times around 3.6Ghz are around 13.8xx which is below 50K