Step6 faster then Step5

[kiwi]

Same OS, clean install, same drivers (Intel released new inf file a few days ago), same memory subtimings, same tweaks, largesystemcache? And voltages, cooling too pretty much similar

Looking at icons one can conclude it is not fresh install

[dinos22]

deathmetal(great username) heh is making it too difficult

it's freaking simple

one system
two CPUs (1xB2 + 1xB1)

tests at any setting, record results

post findings

SIMPLE

[ArcTan]

deathmetal(great username) heh is making it too difficult

it's freaking simple

one system
two CPUs (1xB2 + 1xB1)

tests at any setting, record results

post findings

SIMPLE

you can add do a couple of runs of each and take an average

[dinos22]

you can add do a couple of runs of each and take an average

with a stable system there should not be much variation at all

if there were and you needed to take an average i would find a more stable testing setup/settings

[mike]

In the past there have been speed variations clock for clock between various steppings. It has been true for Preslers, Yonahs, Merom and Conroes.....

Some steppings ( and obviously there are many steps from A0s to B/Cx ) have no appearant speed variations others show significant improvements.

I have ran test between most of the Conroe steppings and seen great differences - and am currently looking into step 5/6 myself

For example here is a posted comparo from Sierra_Bound/Fredyama which has a variation that on first glance appears to be rather CPU specific and not stepping specific - yet too early to tell what the real cause is :

http://www.xtremesystems.org/forums/...4&postcount=59

Vcore etc. is one of many influences that can cause variations - not to mention temps etc. - now if the differences between steppings are tiny - it's very hard to tell what you're looking at. 1M downclocked is a good starting point imho to see if something is worth investigating in...

[dinos22]

true about vcore it impacted runs on AMD for me

[death metal]

deathmetal(great username) heh is making it too difficult

it's freaking simple

one system
two CPUs (1xB2 + 1xB1)

tests at any setting, record results

post findings

SIMPLE

Yep, I never said different systems right? I just made it too wordy though =)...

[mascarilha]

with E6400 its same thing >> step6 faster than step5 ???


btw: great JOb


regards

[elmor]

Take a look here: http://www.xtremesystems.org/forums/...4&postcount=59

[eva2000]

LOL B1 are NOT slower than B2.. that B1 time is slow as snail though and unsual as 9x400mhz 4-4-4-8/12 on 975x platform unsually completes 32M in around 14min to 14min 30ish seconds

[Gautam]

Look again at this chart by fredyama (yes I know its been linked to twice already)



If he doesn't know how to optimize Pi, I don't know who does. Like Mike said, tiny difference, but its there.

Also Mike's word is trustworthy-I don't think there's a single Conroe stepping that hasn't passed through his hands. He's mentioned that this same phenoma occured with Yonahs. No surprise its happening again.

[eva2000]

using 1M to test isn't accurate enough, the swing in 1M completion times are very minor but they exist even on subsequent runs of 1M with the same cpu.

[ChkDsk]

Aren't newer steppings supposed to be "better" ?

[Gautam]

using 1M to test isn't accurate enough, the swing in 1M completion times are very minor but they exist even on subsequent runs of 1M with the same cpu.

I find it very hard to believe that fredyama wouldn't be aware of this. I'm sure he must have run multiple times. That .047 occuring nearly every time doesn't look coincidental. If it were just random variation, they Step 5 would pull ahead at least once or twice.

[Pandamonia]

did u say one was 965? other 975?

the latency of 965 is slower in super pi.

for this test to be accurate u wud need 1 system 2 CPU's one step 6 and other 5

[M.Beier]

using 1M to test isn't accurate enough, the swing in 1M completion times are very minor but they exist even on subsequent runs of 1M with the same cpu.

Seriously Eva2000...
Fredyama is one of the very best pi-calc tweakers, do you really think he havnt considered that?
Show a bit respect man.

death_metal, no, but, if you want multiple systems too compare, well, write a letter and send it to Tomshardware.com or Anandtech.com..

But I belive the difference is much clearer in the 32M run, but seems like the 2*512MB vs. 2*1GB had a large impact I didnt consider... Thats why I got another result from GOESA.

[freecableguy]

$20 says one has LargeSystemCache = 1 and the other doesn't. B1 = B2 (except for a small twist that can't be discussed yet).

[Gautam]

While I never tested whether there is such a performance difference between step 5 and step 6, it certainly appears to me that there are differences in their thermal characteristics. 6 seems to run cooler at the same voltages and speeds. Furthermore, it seems quite a bit more responsive to voltage. IMHO this is why many of the retail OC's don't look so impressive. The retails tend to need more voltage do well, but once its ramped up they can more than compete. B1's being somewhat the opposite.

[syne_24]

To have a good comparison, you have to test 32M on the same system and just switch the cpu.

exactly..I think the ES will shine when you have a more control setup.

[eva2000]

M.Beier, can you do a rerun of 32M at those speeds but this time screen capture the full super pi window rather than cover the top portion of the the window with cpuz screenies...

Also do a tweaked pi run to see how much faster than 14min 06.672s you can complete 32M in.

[M.Beier]

Sure tomorrow, hmm, screencapture, like some software or just takin' some pictures during the run?

Its 1am, so Im on my way too bed, didnt get any sleep last night.

EDIT:
Im not gonna change OS
But sure I'll give it a shot, MAXMEM 3-2-2-4 and some winmemtweaks (they worked on A64, dont know if they work for C2D, I'll figure out) and ofcause, memset

[dinos22]

there's nothing wrong what eva2000 is saying................i say 32M times as well......

[eva2000]

Sure tomorrow, hmm, screencapture, like some software or just takin' some pictures during the run?

Its 1am, so Im on my way too bed, didnt get any sleep last night.

just screen captures using print screen at end of 32m time NOT during the run though.. just would be nice to see the times are each of the 24 iterations..

so do rerun of

untweaked 32M

then do a run of 32M tweaked

[Absolute_0]

Well who's got both B1 and B2 here? This should be settled on 1 system.

If there is a difference, it's going to be very small, like Fredyama's results. 2 minutes difference is due to the system.

[G H Z]

G H Z: It isnt

The title of your thread is step 6 is faster than step 5, yet later you fall back on Fredeyama's numbers which suggest the exact opposite. Which is it?