680i owners pls join in. We need to figure out FSB "strap" change performance hit

[dinos22]

Hi Guys

There's been a little talk now that at high FSB 680i chipset takes a beating from 975 much like 965 boards do past 400MHz FSB particularly Asus P5B Deluxe unless you boot <400FSB and clockgen up

Denny confirmed yesterday that there is a performance penalty past 424MHz FSB and i'd like others to join in and do some testing and determine whether all boards are the same

when you join the thread please run memtest 1.65 and do 5MHz FSB jumps from say 380MHz FSB and look at the RAM bandwidth. Once you see the bandwidth drop after increasing FSB determine at which FSB EXACTLY that happens and post some 32M SuperPI times

****MAKE SURE YOU MANUALLY SET RAM TIMINGS

once you've done that jump to this thread
http://www.xtremesystems.org/forums/...d.php?t=125418
and download latest beta clockgen for these boards and then boot at fast FSB and clockgen up to see whether the "strap" stays the same (i'm calling it strap as i don't know what nvidia would call that heh).

Also test what the highest benchable FSB in fast mode is comapared to looser/slower mode

here is Dennys finding

here is the proof that 1699 & 1700 FSB break line

1699FSB (424.7) fast 32M






1700FSB (425) slow 32M






anything above 1700+ are kinda useless now

i'll be joining in when my 680i arrives....any day now

[skiman]

This might be helpful for what was happening on P5B

http://www.bleedinedgesupport.com/oc...ad.php?t=23803

[dinos22]

This might be helpful for what was happening on P5B

http://www.bleedinedgesupport.com/oc...ad.php?t=23803

thanks for the link

yes we have a massive thread here regarding 965 chipset 1333 strap change at 400MHz FSB. Basically nvidia 680i chipset seems to have the same performance penalty so let's get people to contribute here and find out more

[Lestat]

what about linked and unlinked comparing the exact same ram timings.

so lets say 1699 @ 840mhz linked Sync
vs
1700 @ 840 unlinked Ram at 840.
(or what ever the actual setting would be)

you see what i mean. is linked or unlinked making a difference.

is like 3:45am and i am dead as hell and dont have the patience for 32m pi runs until the sun comes up but i will be trying it.. maybe.

my RMA 680i comes today. as eVGA was very happy to rma my board when i told them i was having the Raid and SATA errors.
they know about it, but kinda suprised them RMA'd the board if there isnt a fix. along with the huge FSB holes.
basically i cant boot at over 450mhz no matter what cpu multi or ram timings or anything.. nothing works.

[Lestat]

ok well i figured linked or unlinked wouldnt make any difference and i was right.

i lost nearly 27 seconds when going from 1699 to 1701

by using Clockgen, i still lost over 19 sec. so there is still something else going on.

and since clockgen uses higher increments i was even running faster cpu. (just a few mhz but still)

ram is 1:1 so its only like 850mhz
but even so that other guy has got some insane 32m scores. i mean sheesh.. he is over 3 minutes faster than me at the same cpu clocks.

[damnnit]

I notice with my board it boots up to 461 then has a hole till 470-473 then 500-520. Maybe this will help you. 1mb superpi is long enough for me, i havent tested any 32mb yet.

[lopri]

Are you guys setting memory subtimings mannually? Those change by whatever logic NV implemented, and FSB is a main factor that affects the subtimings. If you leave them auto, they will change drastically per FSB, memory speed, etc.

Edit: Never mind. I guess OP already set timings mannually.

[dinos22]

Are you guys setting memory subtimings mannually? Those change by whatever logic NV implemented, and FSB is a main factor that affects the subtimings. If you leave them auto, they will change drastically per FSB, memory speed, etc.

Edit: Never mind. I guess OP already set timings mannually.

YES it's important that ALL memory timings are manually set for this exercise including vdimm

linked and unlinked is another lot of tests that would be interesting to see

[JohnMike]

Are you guys setting memory subtimings mannually? Those change by whatever logic NV implemented, and FSB is a main factor that affects the subtimings. If you leave them auto, they will change drastically per FSB, memory speed, etc.

Edit: Never mind. I guess OP already set timings mannually.

That is true. I got my eVGA 680i this weekend and played with it for a while before i set my 24/7 setup. At first i just play with the main timings and was hard to go higher with Ram (CellShock DDR2 PC2-8000 2GB) speeds, then i start mess with the Ram subtimings and i was able to get higher speeds and stability.

I believe that Ram subtimings are one of the main keys to get higher performance from the board, i think thats a plus from nVidia 680i chipset over Intel 975/965 chipsets.

So it mayte be usefull that we share subtimings setups. I will post my once i get home.

[Lestat]

YES it's important that ALL memory timings are manually set for this exercise including vdimm

linked and unlinked is another lot of tests that would be interesting to see

dino check my piost above for a quick linked/unlinked comparison using the same ram speed.

[kapdori]

I am definitely loosing memory bandwidth going from 1699 to 1700 fsb.
I was actually running my e6300 at 1880 fsb with my memory timing of 4-4-4-9 at 940mhz and I was very disappointed with my bandwidth. I tried many different timings for the memory at different speeds but that was best bandwidth I got but still no happy. But at my cpu speed at 1699, I get more memory bandwidth even with same timings for the memory. So definitely there is a hole or something above 1699 fsb.

[illuminati-hwt]

From a fellow danish overclocker (Hessi):

424,75 (1699) x 8, Super Pi 32M @ 15:39,281
430,50 (1722) x 8, Super Pi 32M @ 15:56,390

424,75 (1699) x 9, Super Pi 32M @ 14:29,360
430,50 (1722) x 9, Super Pi 32M @ 14:36,969

411,25 (1645) x 9 = 3701,25 Mhz. Super Pi 32M @ 14:39,563
462,50 (1850) x 8 = 3700,00 Mhz. Super Pi 32M @ 15:01,219

[X.T.R.E.M.E_ICE]

From a fellow danish overclocker (Hessi):

424,75 (1699) x 8, Super Pi 32M @ 15:39,281
430,50 (1722) x 8, Super Pi 32M @ 15:56,390

424,75 (1699) x 9, Super Pi 32M @ 14:29,360
430,50 (1722) x 9, Super Pi 32M @ 14:36,969

411,25 (1645) x 9 = 3701,25 Mhz. Super Pi 32M @ 14:39,563
462,50 (1850) x 8 = 3700,00 Mhz. Super Pi 32M @ 15:01,219

so from 425fsb upwards the timings in the chipset get very loose.

[illuminati-hwt]

Yeps, in the bios it's represented as quadpumped.. So 1699 is fine and 1700 enforces loose timings.