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

[mmills]

I'm working on my first box. My NF7-S rev 2.0 was purchased 8/7/03. I'm using a single stick of TwinMOS PC3200.

I've got the external clock (FSB/2 ?) up to 211. At 215, it reboots as Windows 2k starts up. At 220, it required a full power-off before I could get a reboot.

Using the 211 external clockA multiplier of 10.5 works, 11 fails at Win2k startup. According MBM, that's a 2220 mhz CPU speed.

My CPU voltages are stock. I haven't changed them at all.

My BIOS is 18.

mmills

 

[Audioaficionado]

These are pretty much the same limits I'm hitting with my NF7-S rev2. I can boot and run 215 for a day or two but eventually it freezes up and I have to reset it. I've maxed the RAM , AGP & chipset to the max BIOS voltage settings. vcore was 1.75v. 1.8 vcore or above won't post. Makes no differance either what anything else is set to.

Now I'm stable @ 210fsb/2200/1.675vcore. Memtest86 is good to 220 at 2.6v 2.5-3-3-7 Samsung PC3200.

BIOS v 1.4

 

[mmills]

Audioaficionado,

Have you tried BIOS 10 and the L12 mod?

(BTW, I"m assuming the terms 'L12 mod' and 'sense 166 mod' are interchangeable)

mmills

 

[hitechjb1]

Summary on overclocking the NF7-S rev 2.0 (with Tbred B 1700+ DLT3C and ...)

From my experience on two NF7-S rev 2.0 boards and reading results of many posts, it seems to me that there is no definite combination in terms of the following

- bios version: 10, 14, 16, 18
- L12 mod or FSB_Sense 166/200 mod
- bios setting CPU Interface enable/disable

to achieve highest FSB and memory bandwidth. It has to be tried on a board by board basis (w/ memory modules used) to achieve its max FSB potential, which averaged around 220 MHz, some lower at 210 MHz and some higher to 230 MHz, with stock chipset cooling and without chipset Vdd mod, using good 3200/3500 CAS2 memory with Vdimm at 2.8, 2.9 V if necessary.

I am going to list 6 different situations, no clear trends:

1. Some boards can go to its highest FSB with bios 14, 16, 18, without the L12 and FSB_Sense 166/200 mod (in this situation the mod may help very very little, few MHz (1-3)).

2. Some boards may need the L12 or FSB_Sense 166/200 mod to get stable higher FSB, by as much as 10-15+ MHz.

3. Some boards may need to go back to bios 10 to achieve highest stable FSB. But bios 10 is without the bios fail-safe saving feature of bios 14 or higher. One could get bios corruption w/ bios 10 and need a bios recovery kit to recover.

4. The boards may also require faster CPU above 2.2 GHz to achieve high memory efficiency around 95%.

5. Further, some boards may have stability problem and cannot run at high FSB at certain CPU multipliers (above or below are fine, strange, seeems to be some timing issues ??).

6. Some reported CPU Interface disable made a noticeable higher FSB, as much as 10-15+ MHz. But some can go to highest FSB with or without CPU Interface enabled. But CPU Interface disable has significant negative impact on memory bandwidth performance (6-8%, see below).

It seems to me that the timing of the chipset memory controller, CPU FSB interface, memory modules as a whole making the FSB overclocking very tricky and needed to be done on a case by case basis, without a definite clear recipe.

I think there is an intrinsic limit on FSB based on the combined system timing of the above mentioned components. The board's FSB overclocking would max out at certain level of memory bandwidth (MB/s), regardless of the bios versions, L12 or FSB_Sense mod, CPU Interface, dual channel or single channel, and how well the settings are tweaked and means to achieve the max FSB.

Regardless of the FSB MHz, the number to watch is the effective measured memory bandwidth and memory efficiency, which determine the overall system performance (instead of just FSB MHz). I found that

CPU interface enable (fast decode) and dual channel, integer memory efficiency ~ 95 - 96%

CPU interface disable and dual channel, integer memory efficiency ~ 88-89%

For single channel, memory efficiency would drop by 2-3% but FSB may go few MHz higher, hence single channel and dual channel is roughly a wash on effective memory bandwidth.

Main emphasis for performance should be getting CPU interface enable to work. Since the 6-8% gain of memory efficiency (as described earlier) for memory bandwidth w/ CPU interface enable would be equivalent to 12-18 MHz at the level of 200 MHz FSB. A lower FSB with higher memory efficiency may be better than a higher FSB with lower memory efficiency for both memory bandwidth performance and stability (chipset, memory controller, memory, CPU FSB interface).

Memory bandwidth (MB/s) and memory efficiency can be obtained from the latest Sandra (MAX3) memory bandwidth test, or calculated by

memory_efficiency = measured_memory_bandwidth / (16 x FSB)

 

[Audioaficionado]

Audioaficionado,

Have you tried BIOS 10 and the L12 mod?

(BTW, I"m assuming the terms 'L12 mod' and 'sense 166 mod' are interchangeable)

mmills

I don't want the 10 BIOS as the extreme OC isn't as importaint to me as ultimate system stability and integrety.

Can you do the L12 with a socket wire trick or do you have to connect/blow a bridge?

Overall I'm happy with my system. My original goal was to get 200/2000 using a sane vcore when I speced the parts prior to ordering everything.

It would be like winning the loto if I could have hit 2400 or higher but I'm just at the knee of the probability curve and it seems to be more like a sheer cliff instead of just a steep hill.

 

[Bengan]

metal to the ppl ^^

 

[mmills]

From my experience on two NF7-S rev 2.0 boards and reading results of many posts, it seems to me that there is no definite combination in terms of the following

- bios version: 10, 14, 16, 18
- L12 mod or FSB_Sense 166/200 mod
- bios setting CPU Interface enable/disable

...

OK. I can deal with it.

Regardless of the FSB MHz, the number to watch is the effective measured memory bandwidth and memory efficiency, which determine the overall system performance...

OK. Here's my memory bandwidth and efficiency.

Which numbers do you recommend watching? Can this tell me why Prime95 has a 'rounding error' almost as soon as it starts?

Main emphasis for performance should be getting CPU interface enable to work.

How does one know if it is working? For example, it is 'enabled'. Does that mean it is working?

BTW, why check efficiency as a metric? Isn't all that matters the 'effective bandwidth'?

mmills

 

[hitechjb1]

The actual metric for FSB should be the memory bandwdith as measured by a benchmark program such as Sandra.

memory_bandwidth = FSB x 16 x memory_efficiency

I suggested to use memory efficiency is that it is normalized to FSB (MHz), so one can evaluate the various setup and settings such as
- single channel vs dual channel
- bios setting such as ras/cas timing, CPU interface, ...
instead quoting the absolute bandwidth number in MB/s.

Sometimes, higher FSB but with a much lower memory efficiency may not be as good as a lower FSB but with a higher efficiency. But for a given setup, you may just use bandwidth (MB/s) to evaulate the FSB settings. I find that efficiency is easier to do the distinction instead of the bandwidth in MB/s. But either one is fine to use, but NOT simpy using the FSB MHz.

Your prime95 crashed as soon as it started, seems to indicate either
- the FSB or the CPU clock frequency is too high, or
- Vcore is not enough, or
- die temperature is too high (50-60 C) leading to instability, or
...
to keep the system stable. You have to raise Vcore, or if Vcore is high enough (1.8-1.9V), back down the FSB or the CPU clock frequency or both to find out the limit of the FSB (mainly related to motherboard and memory, and some tricky setup for the NF7-S rev 2.0 as described in previous posts), and the limit of the CPU (mainly due to Vcore, cooling) first. Then increase them (w/ sufficient Vcore, chipset Vdd, memory Vdimm) to find out the combined stable operating point for FSB and CPU clock.

Your previous post said you have been using stock Vcore, definitely raising Vcore to 1.8 V should be safe and should help overall system and prime95 stability.

This link may help:
Summary for overclocking CPU and FSB

 

[mmills]

hitechjp1,

Wonderful post. Thanks.

Have you thought about these issues in general system terms, not just motherboards, CPUs and FSBs?

mmills

 

[hitechjb1]

hitechjp1,

Wonderful post. Thanks.

Have you thought about these issues in general system terms, not just motherboards, CPUs and FSBs?

mmills

Can you be more specific about what do you mean by general system?

MP? 64-bit system? ... Probably they are beyond the scope of the hardwares that I can have my hands on, ...

A NF7-S motherboard, FSB, CPU, chipset, memory, hard drives, ... forms a pretty complex single processor system, lots of things to experiment, learn and analyse in details.

 

[mmills]

hitechjb1,

> Can you be more specific about what you mean by general system?

I was thinking in terms of operations reseach and/or Turing Machines.

I've got my machine running stable under Prime95 load. 2.4 gig, 218 FSB, 93% bandwidth efficiency, 3269 mem bandwidth. Using 7-3-3-2.5, vcore 1.85, vdimm 2.9, CPU temp 48 C, case 33 C.

I can't get 220 FSB to work for me.

mmills

 

[hitechjb1]

hitechjb1,

> Can you be more specific about what you mean by general system?

I was thinking in terms of operations reseach and/or Turing Machines.

I've got my machine running stable under Prime95 load. 2.4 gig, 218 FSB, 93% bandwidth efficiency, 3269 mem bandwidth. Using 7-3-3-2.5, vcore 1.85, vdimm 2.9, CPU temp 48 C, case 33 C.

I can't get 220 FSB to work for me.

mmills

I can see lot of interesting relationship between operations research and overclocking of CPU, FSB, memory, ... Operations research deals with mathematical programming and combinatorial optimization problems, such as linear, non-linear and integer programming to solve very large, complex real-life problems such as air traffic scheduling, communication network scheduling, oil exploration, ...

To solve large problems, the computers and programs have to handle very large, special data structures of matrices and graphs stored in hard disk, memory and cache. So faster disk, larger total memory size (GB), higher FSB and memory bandwidth (MB/s) and larger CPU cache would help these applications to move data through these different levels of memory hierarchy: cache - FSB - memory - hard disk (paging). Further, a higher raw CPU power would be desirable to number crunch the inner loop of those problems, formulated as linear programming, combinatorial search, in cache.

Combinatorial search deals with discrete optimization problem such as the travelling salesperson problem which is to find a travelling path with the shortest distance to visit all of a given number of cities (once). It is mathematically difficult to find the optimal solution, if the number of cities is very large; but it can be approximated and number crunched by using mathematical programming technique.

As for Turing machine which is a simple computation model with a very simple processor and infinite memory to study computational complexity in time and space (memory), I can hardly see any relevance with overclocking. If you have some ideas, pls post.


Back to your overclocking, running at 218-220 MHz FSB is about the norm for NF7-S rev 2.0. If you are using 3700 memory (from your sig 256 MB, too little?), have you tried setting memory to CAS2, it would help to bring the memory efficiency to 95-96% which is the top end number for the memory bandwidth for this board. What is your chipset Vdd at? A Vdd mod may help, if you want to try higher than 218 MHz. If you use 3700 memory, it should not be limiting yet.

If you are in the northern hemisphere, as days go on, we are getting better and better seasonal help for overclocking, you may be able to get above 2.4 GHz towards 2.5 GHz soon as a drop of 5 C in ambient temperature would help the CPU overclocking frequency by about 2% (~ 50 MHz), assuming HSF being kept the same.

 

[Audioaficionado]

I just put a Y-S Tech 10-50CFM varible speed fan on my Vantec Areovent HS and dropped temps 8c. Now I can run stable at 1.75vcore/215/2260/52c/vdimm 2.9v/chipset 1.7v/AGP 1.7v. It's relatively quiet even at full speed.

I'll try for more as the weather cools off. Temps were inspite of a 30c ambient enviornment.

 

[mmills]

hitechjb1,

As to Turing Machines, most think Turing machines very slow affairs. They are actually infinitely fast at 'seeking' memory values on its infinitely long memory 'tape'. In the context of this conversation, a Turing Machine has infinite bandwidth and zero latency (100% efficiency). Even if the 'symbol' sought is infinitely far off on the 'tape', the tape is rewound instantly and the symbol 'read' without pause between 'computation' and 'read' steps.

(I just wanted a chance to talk about Turing machines with these new vocabulary words. Please excuse the digression.)

Getting back to my OC efforts, I have not tried CAS 2 ( 7-2-2-2, I guess). Based on the results posted at http://www.vr-zone.com/reviews/TwinMOS/PC3700-3200/page5.htm , the more relaxed settings off better FSB. I compared my effective bandwidth at 7-3-3-2 against 7-3-3-2.5 and got a slightly higher number with 7-3-3-2.5. That's as far as I've taken it (logically or in testing).

mmills

 

[RedDawg41]

This is with CPU Disabled option, ram timings [email protected]
vcore v1.95

SiSoftware Sandra

Int Buff aEMMX/aSSE (Integer STREAM) Results Breakdown
Assignment : 3523MB/s
Scaling : 3513MB/s
Addition : 3507MB/s
Triad : 3506MB/s
Data Item Size : 8-bytes
Buffering Used : Yes
Offset Displacement Used : Yes
Bandwidth Efficiency : 95% (estimated)

Float Buff aEMMX/aSSE (Float STREAM) Results Breakdown
Assignment : 3524MB/s
Scaling : 3343MB/s
Addition : 3180MB/s
Triad : 3158MB/s
Data Item Size : 8-bytes
Buffering Used : Yes
Offset Displacement Used : Yes
Bandwidth Efficiency : 90% (estimated)

Test Status
Memory Used by Test : 322MB
NUMA Support : No
SMP Test : No
Total Test Threads : 1
SMT Test : No
Dynamic MP/MT Load Balance : No
Processor Affinity : No

Chipset 1
Model : Abit Computer Corp nForce2 AGP Controller
Front Side Bus Speed : 2x 230MHz (460MHz data rate)
Width : 64-bit
Maximum Bus Bandwidth : 3680MB/s (estimated)

Logical/Chipset 1 Memory Banks

Features
MMX Technology : Yes
SSE Technology : Yes
SSE2 Technology : No
SSE3 Technology : No
EMMX - Extended MMX Technology : Yes
HTT - Hyper-Threading Technology : No

 

[Audioaficionado]

Well I tried again after it cooled off here and my CPU temps were 45c under 100% F@H load running under forced SSE boost. This application heats things up to the max.

Anyway I can't budge past my Abit stats on my sig. The Samsung DDR400 is SPD rated 3-5-5-9 ?T but I've set it to 2.5-3-3-7. I don't know what T it is set for but I'm assuming 2T. Nothing in the BIOS lists it as far as I can see and CPU-Z v1.19b doesn't know either.

BIOS v1.4.

I threw 1.8v at the CPU and maxed the volts on the other components where I could. It booted into XP at 2300 but froze soon after and the CPU was still under 50c. It won't even post if I try to put anymore on the CPU. Vcore above 1.825 just causes the mother board to do the siren alarm. I tried to relax the RAM timings and still no go. Memtest86 shows no RAM errors @ 220fsb and the tighter timings.

I'm rock solid stable at my current settings but just a smidgen more and it hard locks after a few minutes.

CPU-Z 1.19b

WCPUID v3.1a

 

[Audioaficionado]

Installing XPproSP1 and S2k3 with APIC disabled in the BIOS has no effect on overclocking abillity. One good point is that they can run both APIC enabled or Disabled after the install.

However S2k3 has a weird querk about it not recognizing any partitions not formated during its installation. I installed it on partition 3 of a HD and it could only see the first FAT32 partition and the next two NTFS partitions that 98, xp and s2k3 created durring the initail install. XP could see all the formated partitions on both HDs but s2k3 couldn't. It can see them in the computer management -> diskmanagement utility but won't recognize them as formatted file systems. It couldn't even see the storage partions on the same HD it was on that were formated by XP on the other HD that was APIC enabled. So installing without APIC seems to be the problem for s2k3.