l12 mod and cpu interface

Allright, so a while back i did the vcc mod and got my nb voltage up to 1.9. that took me from a stable 215 to a stable 220 (with cpu interface disabled/enabled, no difference). not that impressive. Finally got tired of 220 so trying something else i yanked my chip out this morning and connected the 3rd L12 bridge. When i got this t-bred i was using a kt333 board, and stuck @ 166mhz, so i modded the L3's to have it default to 13X multi. So now my chip defaults to 13X166 or xp2700 . Sweet. Ok so i found that i still couldn't get above 220. I disabled cpu interface and was able to boot at 233 . i havn't tried for higher yet, but i'm wondering what is anyones experience with having cpu interface disabled? is it a big performance hit? and if so where.. cpu instructions or mem bandwidth? i wonder if the 13 extra fsb would make up for it? bios 19 btw.


thanks,

J.

um, WOW



this is stable, with ram @ 6, 3, 3, 2.5 and 2.9vdimm
i can post at 245 but can't quite get into windows. I can load windows seemingly stable @ 240 but i get some mem errors in memtest so it's not truely stable. I need to do the vdimm mod and get my vdimm up to 3.2ish and see if i can get it stable @ 245 or maybe even higher . Since i got no replies to my question if i have time tonight i will bench it @ the highest fsb i can do with cpu interface enabled (220) and again @ 233. I was hoping someone could maybe save me the time .



J.

well, i have heard that there is a performance loss w/ CPU Interface disabled...i am just not sure how much...ill be interested in your results...im current stuck @ 208 due to my RAM...would be interesting to see how high i could w/ cpu interface disabled....im always short of time but maybe Ill have a chance to tinker w/ it later...ill post results if i can.

-TriX

L12

Im trying to do the same thing as you however I wont be doing the Vmods or L12 AGAIN until this weekend.
Im stuck at 225 right now, funny thing is if I disable CPU interface I cant even get into windows at this speed.
It seems that most people report the opposite.
In fact its not stable over 200fsb with it disabled.

There is a stickied thread over at xtremesystems.org that has alot of info for L12 mods.
They went over your question but it escapes me now what the outcome was but Ill look again.

Later

Wardog

theres a huge loss witht the interface disabled. i can do 227 enabled, 235 disabled. 227 enabled is definatly faster.

james.miller said:

theres a huge loss witht the interface disabled. i can do 227 enabled, 235 disabled. 227 enabled is definatly faster.

Click to expand...

i knew there was quite an effect. but i'm talking from 220 to 240. One thing i'm really curious bout, where do you notice the performance hit? memory benchies? or cpu instructions? I also followed a thread at abit forums, someone claimed that with bios 19 the difference was lessened. I will check it out a little right now i think, i'll post back with some results.


J.

*edit*




Ok, i just had myself a little "reboot, sandra, reboot, sandra,reboot.. " u get the picture.

here are my results regarding cpu interface. I didn't take screenies because i got other things to do tonight. I wrote everything down and will jus copy it here. .


all three tests were done with memory timings at 6,3,3 cas 2.


TEST 1:
FSB = 220
cpu clock = 2420 (220X11)
cpu interface = Enabled
memory bandwidth = 3123/2873
CPU ALU = 8835
CPU FPU = 3594

TEST 2:
FSB = 220
cpu clock = 2420 (220X11)
cpu interface = Disabled
memory bandwidth = 3035/2784 (hmm.. not such a big diff there)
CPU ALU = 8925
CPU FPU = 3595 (no decrease in performance there at all)

Test 3:
FSB = 233
cpu clock = 2446 (233X10.5)
cpu interface = Disabled
memory bandwidth = 3225/2953
CPU ALU = 9046
CPU FPU = 3624

so, there's what i found. a bunch of hype about nuffin. . looks like fsb is still king. I would be interested to see if other people get the same sort of results.

J.

ba bump. thought u guys might be interested in that ^

thats just my point though. at 225 enabled i still get better memory benchies than you do at 233 disabled.


Test
FSB = 225
cpu clock = 2362 (225X10.5)
cpu interface = ENABLED
memory bandwidth = 3346/3107
CPU ALU = 8814
CPU FPU = 3538


at the end of the day, you should use whatever gives you the highest overall performance. 99% of people will find that enabled @ a lower fsb is still better, but that depends on how far you can push the board with the interface disabled.

james.miller said:

thats just my point though. at 225 enabled i still get better memory benchies than you do at 233 disabled.


Test
FSB = 225
cpu clock = 2362 (225X10.5)
cpu interface = ENABLED
memory bandwidth = 3346/3107
CPU ALU = 8814
CPU FPU = 3538


at the end of the day, you should use whatever gives you the highest overall performance. 99% of people will find that enabled @ a lower fsb is still better, but that depends on how far you can push the board with the interface disabled.

Click to expand...

what timings are u using for your ram? I relaxed timings a bit for this test just to be sure my mem would stay stable.

7,3,3,2.0

juliendogg, try using another benchmark to get an idea of the total system performance. Use 3dMark01 for instance to compare diff settings. Sandra tests specific elements of a system, whereas with 3dMark you will notice overall results. Overall, its been proven that on AMD/nForce2 platform, you want CPU interface inabled, and tightest memory timings.

Depending on the kind of RAM you are using, find the tightest timings it will run, (try 2-2-2-11) and find the highest stable FSB with CPU interface inabled.

um. Mostly i was interested in the memory bandwidth effects of cpu interface. I used looser timings for the test intentionally. . Thanx though.

Overall, its been proven that on AMD/nForce2 platform, you want CPU interface inabled, and tightest memory timings.

Click to expand...

You think that you could produce this "proof"? Actually, it has been shown that memory timings do not make as much difference with modern DDR systems as they did with SDR platforms. This is of course within reason. IE: 5-2-2-2 is considerably faster than 12-4-4-3, but the difference between 5-2-2-2 and 5-3-3-2(or even 2.5) is very small. Easily made up for by a few mghz fsb. As far as CPU interface, i've been all over the net, and i see no proof. So i'm curious how you came to believe that this has been "proven" .


J.

G|-|oST said:

juliendogg, try using another benchmark to get an idea of the total system performance. Use 3dMark01 for instance to compare diff settings. Sandra tests specific elements of a system, whereas with 3dMark you will notice overall results. Overall, its been proven that on AMD/nForce2 platform, you want CPU interface inabled, and tightest memory timings.

Depending on the kind of RAM you are using, find the tightest timings it will run, (try 2-2-2-11) and find the highest stable FSB with CPU interface inabled.

Click to expand...

aye, i believe also that with it enabled, even though there is a fsb hit, its drasticly increases stability.

pwnt by pat said:

aye, i believe also that with it enabled, even though there is a fsb hit, its drasticly increases stability.

Click to expand...

what?

Ok so anyway. with vcc @ 3.2 i can load windows @ 250fsb! too bad it's not very stable. Not memtest stable with timings at 7,3,3,2.5 and my usb lan and onboard sound dissapear above 245. I'm still playing with it, right now @ 240 6,3,3,2.5. Memtest stable.


J.

That's pretty incredible...my max FSB with 2 sticks of Corsiar XMS PC3200 CL2 is 198 in DC. Timings at 2.5-3-3-11, 2.8v, 1.7v on the chipset. CPU Interface enabled/disabled makes no diff. No VDD or VDimm mods...I've tried them before and it fried my board. I'm going to try the L12 probably next weekend to see if it takes me anywhere. *Crosses fingers*

Memtest stable, with cpu interface ENABLED. Woot! Seems i could get those high fsb's with cpu interface enabled and lower multi's, just not with 10 or 10.5. So i tried the modded bios 10. Worked like a dream. I tested for corruption by compressing a large folder of many small files with winrar, seems all good. No corruption yet.

again.. Woot!

any higher and i get errors in memtest. so for now i think i will have to be happy with 240. Even if i lax up the timings i still get errors at 245, and anythin above 245 kills my onboard peripherals anyway so..


J.