TbredXP 1700 & Abit NF7 rev2.0 boards = Not Very Friendly

Like the topic says, it is true.

They don't like each other. They don't want to help each other out. sad sad sad.

memory used = 2x CorsairXMS3502v1.1 (which is capable of 238mhz FSB DC 4-2-2-2.5)

tbredXP 2700 chip with L12 mod(for 133mhz FSB) does exceptionally well with Abit NF7 rev2.0 boards in achieving high FSBs 128bit DC + CPU Interface@ 230mhz and higher.

I did the mod for xp1700. the bios reports i have the 3200+ speed with 200mhz FSB.

Result: tbredXP 1700 chip does poorly in high FSBs 128bit DC + CPU Interface with the L12 wire-drop-socket mod(for 200mhz FSB). 230mhz FSB DC is not stable at all, and barely loads windows. 3dmark? forget it.

my taste and expectation is to have the system CPU Interface + 128bit DC enabled for its glorious performance. The L12 mod(for 200mhz FSB) did seem to help little bit though. Before the mod, the system would not even boot at 230 FSB. After the mod, the system "boots" and loads windows, but that's it.

Based on what i've been seeing, the normally aspirated FSB monster 200mhz rated Barton chips do better in high FSBs, with all the DC and Cpu fast decode enabled. it seems that Baron chips are more FSB overclock friendly with the nforce2 rev2 boards. The tbredXP 2700 also puts up a good fight, less than what the Bartons do, but close..


The only thing I have not done yet is the L12 wire-drop-socket mod I haven't done to this xp1700 is the mod for 166mhz FSB. So, I cannot speak for all L12 mods that would/would not work. What do you think about this my friend hitechjb1?

On the surface, the FSB_sense 166/200 mod seems to help for the board that come wth low FSB overclocking (such as my case), by about 8 MHz DC with CPU Interface enabled, 17+ MHz for SC w/ CPU Interface disable. Without mod, 215 MHz is highest stable. I tested both the FSB_sense 166 mod and the 200 mod, the difference between the two was not noticeable.

With the 166/200 mod,
- FSB went from 215 to 223 (with CPU Interface enable and DC),
- to 238 (with CPU Interface disable and SC).

CPU Interface enable is important even the max FSB at 223 is low compared to SC with Interface disable at 238. The memory bandwidth and system performance gain (95-96% efficiency) with CPU Interface enable and DC out benefits the FSB at 238 with only 85-89% memory efficiency when CPU Interface was disable and running SC.

So the actual gain from the 166/200 mod is around 3.7% (from 215 to 223 MHz with CPU Interface enable).

Details are shown in these links:

NF7-S rev 2.0 FSB_Sense 166 and 200 mod (from 133) test results (page 12)

NF7-S rev 2.0 FSB_Sense 166 and 200 mod (from 133) test results (part 2) (page 12)

It looks like the mod helps some boards that are not good FSB overclockers to begin with, but it cannot make the board to become super smooth FSB overclocking. My board still has some strange behaviors after the 1660/200 mod such as:

- stable with multiplier 9, 9.5, 10.5, 11, ... but x10 hangs
- memory bandwidth efficiency dropped significiantly (from 95% to 89% or below) when multiplier is dropped from 10.5 to below 9.5 or below (10 did not work), even CPU frequency was kept roughly constant (strange ??)
- system very sensitive to certain FSB, memory timing and bios setting, trying to make system faster or slower would introduce instability. Looks like some sort of system timing issues (both fast path and slow path).
- it looks like there is some intrinsic memory bandwidth constraint inside the memory conrtoller and/or the CPU (reason TBD), regardless of various settings and higher FSB (e.g. SC to 238 MHz), there is a ceiling of around 3300 MB/s memory bandwidth. Higher FSB at 230+ would just drop memory bandwidth efficiency for similar CPU speed (strange ??).

I have seen Tbred B 1700+ DLT3C working with a NF7-S rev 2.0 at 238 MHz level with CPU Interface enabled and DC, ... But this may be just probabilistic, like any overclocking, ...

I suspect my board has certain system timing issues or when working combined with a 1700+ DLT3C (default FSB 133).
- I think for those boards that are good (w/ certain CPU) to begin with, no 166/200 would be needed and no improvement seen.
- For boards that are not good FSB overclocking (w/ certain CPU), the mod 166/200 would help a bit, at least on the surface for the 10-20 higher FSB number (in SC), but the intrinsic memory bandwidth is still limited, and system performance is still very sensitive to various timings and bios settings.

I have begun to suspect also what you suggested about the CPU FSB capability under high memory traffic demand (DC + CPU Interface), maybe indeed the higher default FSB ones such as 2500+ Barton, Tbred B 2700+ have a higher probability to do better than the lower FSB one such as 1700+ and 2100+. But this is still just a conjecture until more data seen.

PS: The 1700+ (0310 WPMW, not the same one that does 2.6 GHz on air) I am testing is running in AMD stock HSF at 2.3 GHz at 65 C, so I don't know whether the not so good FSB performance of the system has something to do with such heat stress on the CPU.

hitechjb1 said:

On the surface, the FSB_sense 166/200 mod seems to help for the board that come wth low FSB overclocking (such as my case), by about 8 MHz DC with CPU Interface enabled, 17+ MHz for SC w/ CPU Interface disable. Without mod, 215 MHz is highest stable. I tested both the FSB_sense 166 mod and the 200 mod, the difference between the two was not noticeable.

With the 166/200 mod,
- FSB went from 215 to 223 (with CPU Interface enable and DC),
- to 238 (with CPU Interface disable and SC).

CPU Interface enable is important even the max FSB at 223 is low compared to SC with Interface disable at 238. The memory bandwidth and system performance gain (95-96% efficiency) with CPU Interface enable and DC out benefits the FSB at 238 with only 85-89% memory efficiency when CPU Interface was disable and running SC.

So the actual gain from the 166/200 mod is around 3.7% (from 215 to 223 MHz with CPU Interface enable).

Details are shown in these links:

NF7-S rev 2.0 FSB_Sense 166 and 200 mod (from 133) test results (page 12)

NF7-S rev 2.0 FSB_Sense 166 and 200 mod (from 133) test results (part 2) (page 12)

It looks like the mod helps some boards that are not good FSB overclockers to begin with, but it cannot make the board to become super smooth FSB overclocking. My board still has some strange behaviors after the 1660/200 mod such as:

- stable with multiplier 9, 9.5, 10.5, 11, ... but x10 hangs
- memory bandwidth efficiency dropped significiantly (from 95% to 89% or below) when multiplier is dropped from 10.5 to below 9.5 or below (10 did not work), even CPU frequency was kept roughly constant (strange ??)
- system very sensitive to certain FSB, memory timing and bios setting, trying to make system faster or slower would introduce instability. Looks like some sort of system timing issues (both fast path and slow path).
- it looks like there is some intrinsic memory bandwidth constraint inside the memory conrtoller and/or the CPU (reason TBD), regardless of various settings and higher FSB (e.g. SC to 238 MHz), there is a ceiling of around 3300 MB/s memory bandwidth. Higher FSB at 230+ would just drop memory bandwidth efficiency for similar CPU speed (strange ??).

I have seen Tbred B 1700+ DLT3C working with a NF7-S rev 2.0 at 238 MHz level with CPU Interface enabled and DC, ... But this may be just probabilistic, like any overclocking, ...

I suspect my board has certain system timing issues or when working combined with a 1700+ DLT3C (default FSB 133).
- I think for those boards that are good (w/ certain CPU) to begin with, no 166/200 would be needed and no improvement seen.
- For boards that are not good FSB overclocking (w/ certain CPU), the mod 166/200 would help a bit, at least on the surface for the 10-20 higher FSB number (in SC), but the intrinsic memory bandwidth is still limited, and system performance is still very sensitive to various timings and bios settings.

I have begun to suspect also what you suggested about the CPU FSB capability under high memory traffic demand (DC + CPU Interface), maybe indeed the higher default FSB ones such as 2500+ Barton, Tbred B 2700+ have a higher probability to do better than the lower FSB one such as 1700+ and 2100+. But this is still just a conjecture until more data seen.

PS: The 1700+ (0310 WPMW, not the same one that does 2.6 GHz on air) I am testing is running in AMD stock HSF at 2.3 GHz at 65 C, so I don't know whether the not so good FSB performance of the system has something to do with such heat stress on the CPU.

Click to expand...

wow, yes, thank you for your input. All this helps to clear out about the wonderings of the FSB overclockability with certain types of chips and nforce2 A1 ultra400 of the abit boards even further. I've enjoyed reading all of your post.

Very very interesting points you have brought to me here. Hmmmmm........................... interesting.. and on some of the things, i have similar thoughts as yours!

i think I might just leave my mod alone or maybe take it off later, i haven't decided. I can tell you this, it's not any worse than before so, i can't complain..

I haven't pushed the limits like you guys have. I can't get past 2200 even at 1.7v. I'm going to push the mem/fsb until memtest fails. Then I'll work the core frequency. I don't have a blessed chip but most likely just an average one. I'm still pleased with what I've gotten out of it so far.

Audioaficionado said:

I haven't pushed the limits like you guys have. I can't get past 2200 even at 1.7v. I'm going to push the mem/fsb until memtest fails. Then I'll work the core frequency. I don't have a blessed chip but most likely just an average one. I'm still pleased with what I've gotten out of it so far.

Click to expand...

With PC3200 memory, would be interesting to see how your NF7-S rev 2.0 overclocking on the FSB, regarding to:

- how high the FSB can go with a Tbred B 1700+ DLT3C
- how high the FSB can go with a Tbred B 1700+ DLT3C under 2.2 GHz
- how high the FSB can go with CPU Interface enabled or disabled
- what is the Sandra memory bandwidth efficiency with CPU Interface enabled or disabled

Take a look at my sig. 11x233 at 2573 mhz mem at 2-2-2-6 But I have to agree with you guys. It has been a VERY frustrating experience to get here. But I can link you to my 3DM01 results for proof. I run that 24/7 folding.

climbski said:

Take a look at my sig. 11x233 at 2573 mhz mem at 2-2-2-6 But I have to agree with you guys. It has been a VERY frustrating experience to get here. But I can link you to my 3DM01 results for proof. I run that 24/7 folding.

Click to expand...

refering to 3dmark is no good, because we can't tell if you have enabled 128bit DC + CPU Interface or not. Did you get to that FSB with both of these enabled all together? These make a lot of difference in FSB overclocking.

CPU interface enabled but only running single stick OCZ PC3500EL.

climbski said:

CPU interface enabled but only running single stick OCZ PC3500EL.

Click to expand...

that's pretty good... but you still gotta(not saying you have to) run in 128bit DC mode too. These two are what makes it difficult to hit higher FSBs with stability.

hitechjb1 said:

With PC3200 memory, would be interesting to see how your NF7-S rev 2.0 overclocking on the FSB, regarding to:

- how high the FSB can go with a Tbred B 1700+ DLT3C
- how high the FSB can go with a Tbred B 1700+ DLT3C under 2.2 GHz
- how high the FSB can go with CPU Interface enabled or disabled
- what is the Sandra memory bandwidth efficiency with CPU Interface enabled or disabled

Click to expand...

I can only run with CPU Interface enabled. Without it, all multipliers under 11 fail to even post. BIOS 1.4

I'll keep pushing the FSB until memtest fails or the computer won't bootstable. Next I'll try 215 and then 220.

FAH is ripping through WUs that take a stock Tbred @ 137/1500 24hrs to run, and with my 210fsb knocking them off in less than 15hrs.

Update that bios to 17 if you havn't already. I've had great results with it and been frustrated with all other versions including 18.

climbski said:

Update that bios to 17 if you havn't already. I've had great results with it and been frustrated with all other versions including 18.

Click to expand...

Nothing it fixes is a problem for my setup. I'll wait until something is fixed that is a problem for me. 1.4 is just peachy so far.

Audioaficionado said:

I haven't pushed the limits like you guys have. I can't get past 2200 even at 1.7v. I'm going to push the mem/fsb until memtest fails. Then I'll work the core frequency. I don't have a blessed chip but most likely just an average one. I'm still pleased with what I've gotten out of it so far.

Click to expand...

thats interesting. I have an 1800+ dlt3c that stops dead at 2200mhz aswell, yet my 1700+ dlt3c is doing 2.4ghz like there is no tomorrow.

very strange......also, i too can't boot unless the interface is enabled - can't say im worried though

Oh, btw, my board has been doing 227mhz dc with the interface enabled and without the L12 mod for the last 3 weeks or so - not a single crash.

james.miller said:

Oh, btw, my board has been doing 227mhz dc with the interface enabled and without the L12 mod for the last 3 weeks or so - not a single crash.

Click to expand...

Enable 128bit DC and come back here. See what FSB you'll be at and how stable it is.

I think some people are not understanding the topic discussion here. The objective here isn't about who reaches highest FSB. The objective here is about having both CPU Interface and 128bit Dual Channel enabled and to get high FSBs, that is what we were talking about. There is a reason why we are talking about these two things.

excuse me mister arrogant. read my sig:

2x256mb twinmos pc3200. DUAL CHANNEL.

damn it. if i wernt such a nice person, i'd call you an idiot you're attitude towards people really irritates me.

james.miller said:

excuse me mister arrogant. read my sig:

2x256mb twinmos pc3200. DUAL CHANNEL.

damn it. if i wernt such a nice person, i'd call you an idiot you're attitude towards people really irritates me.

Click to expand...

I don't think I've said anything bad in my previous reply. I don't understand why you take it so badly. Holding grudges and spitting out like that in my thread is not a nice thing either. Thanks a lot for ruining my thread.


back to the subject, post some of your benches. Report your clockspeed/ FSB configuration and run some tests. Run superPi, piFast for now. I'll post mine as well. I wait for your reply.

[OC]This said:

I don't think I've said anything bad in my previous reply. I don't understand why you take it so badly. Holding grudges and spitting out like that in my thread is not a nice thing either. Thanks a lot for ruining my thread

Click to expand...

then read your last post again. DO NOT call me a liar. I have nothing to prove to you, and nothing to gain from lying.

My system configuration still stands.

227mhz x 10.5=2383mhz
1.7v vdd
1.75v cpu
2.8v ram
ram timings i cant remember off the top of my head.

right now, im not doing any tests. The heat in this room is off the chart - its approaching low 30's ambient now and im idling at 43c even at 1.8ghz.

I'll do the tests later on when it cools down. for now, here are some links:
http://forum.oc-forums.com/vb/showthread.php?s=&threadid=220019
http://service.futuremark.com/compare?pcm=1437843

....Enable 128bit DC and come back here. See what FSB you'll be at and how stable it is......The objective here is about having both CPU Interface and 128bit Dual Channel enabled and to get high FSBs, that is what we were talking about. There is a reason why we are talking about these two things....

Click to expand...

So you said nothing wrong? Sorry, my mistake. what did you mean by that then? calling me a liar? telling me i am wrong?

From seeing what have been reported about many of the NF7-S rev 2.0, I begin to have this theory/conjecture, yet to be proven.

There are two camps or sets of boards (rev 2.0), separated around the time of the 3200+ debut and support needed. There seems to be some differences, such as hardware, timing tweak, BIOS ... (exact details and reasons unknown), between the older rev 2.0 board and the newer ones.

Older rev 2.0 board:
- can do 220-230 MHz FSB at stock with CPU Interface enable and DC
- with CPU Interface enable, memory bandwidth efficiency (from Sandra) is around 95%
- the FSB_Sense 166/200 mod (or L12 mod) does not help FSB going higher
- CPU Interface disable does not help FSB higher, may get things worse

New rev 2.0 board (after 3200+ debut and support):
- can do 210-218 MHz FSB at stock with CPU Interface enable and DC (norm is 215 MHz)
- with CPU Interface enable, memory bandwidth efficiency (from Sandra) is around 95% (same as older board)
- the FSB_Sense 166/200 mod (or L12 mod) helps FSB going higher by about 5-10 MHz with CPU Interface enable
- CPU Interface disable helps FSB higher, may help by as much as 10-15 MHz, SC/DC (but this won't help actual max memory bandwidth obtained, reasons TBD)

CPU Interface enable is important for both set of boards
- with CPU Interface enable and DC, memory bandwidth efficiency is around 95% (compared to 85-89% with it disable)
- this memory bandwidth efficiency gain with CPU Interface enable out benefits the 10-15 MHz FSB gain even with CPU Interface disable
- this applies to both DC and SC, with DC having 2-3% more efficiency and SC few MHz higher (with CPU Interface enable)

So pls post your results (FSB, memory bandwidth efficiency, CPU multiplier/frequency, 3DMark 01/03, ...) and when was the board bought and impression, ...

hitechjb1 said:

From seeing what has reported about many of the NF7-S rev 2.0, I begin to have this theory/conjecture, yet to be proven.

There are two camps or sets of boards (rev 2.0), separated around the time of the 3200+ debut and support needed. There seems to be some differences, such as hardware, timing tweak, ... (exact details and reasons unknown), between the older rev 2.0 board and the newer ones.

Older rev 2.0 board:
- can do 220-230 MHz FSB at stock with CPU Interface enable and DC
- with CPU Interface enable, memory bandwidth efficiency (from Sandra) is around 95%
- the FSB_Sense 166/200 mod (or L12 mod) does not help FSB going higher
- CPU Interface disable does not help FSB higher, may get things worse

New rev 2.0 board (after 3200+ debut and support):
- can do 210-218 MHz FSB at stock with CPU Interface enable and DC (norm is 215 MHz)
- with CPU Interface enable, memory bandwidth efficiency (from Sandra) is around 95% (same as older board)
- the FSB_Sense 166/200 mod (or L12 mod) helps FSB going higher by about 5-10 MHz with CPU Interface enable
- CPU Interface disable helps FSB higher, may help by as much as 10-15 MHz (but this won't help actual memory bandwidth)

CPU Interface enable is important for both set of boards
- with CPU Interface enable and DC, memory bandwidth efficiency is around 95% (compared to 85-89% with it disable)
- this memory bandwidth efficiency gain with CPU Interface enable out benefits the 10-15 MHz FSB gain even with CPU Interface disable
- this applies to both DC and SC, with DC having 2-3% more efficiency and SC few MHz higher (with CPU Interface enable)

So pls post your results (FSB, memory bandwidth efficiency, CPU multiplier/frequency, 3DMark 01/03, ...) and when was the board bought and impression, ...

Click to expand...

Interesting find you have there.

I've also forgot to mention yesterday. I found out one of the possible factors that could be limiting from high FSBs with the rev2.0 board and xp1700. the member muzz has reported the issue and how to correct his problem. The BIOS version could be the factor that is making the difference. The abit nf7 series rev2 boards with the BIOS 10 could do the trick. Muzz has gone back from his default BIOS 12 to BIOS 10. I remember one of my earlier rev2.0 boards that I was running xp2700 was the BIOS 10. That's where I got easy 230mhz FSB DC + CPU Interface enabled. There are other members over at xtreme had the similar experience by going back to BIOS 10. My current rev2.0 board came with the BIOS 13 however. BIOS version may differ depending on the date of purchase, as we know. It is possible that the old BIOS such as 10 could be a good pair with the older chips such as xp1700 or xp2700. This is just speculation as I have no proof yet, but this could be one of the factors that would make the difference in FSB overclocking to the high 220s to low 230s FSB DC fast cpu decode.

mine was shipped with 1.0, meaning it could well be the "earlier version" that does 220-230 with the interface enabled.