nf7-s rev2 fsb sense L12 mod (24 july 2003)
To see whether L12 mod affects FSB overclocking on the nf7-s rev. 2, I've ran some test with modified 1st L12 (ag31) and/or 3rd L12 (ah30). The mod can be used to force 166 and 200fsb on a 133fsb CPU, and 200fsb on a 166fsb CPU. Note that these mods can also be used on other boards that auto-set FSB using the L12 settings.
The ag31 and ah30 pins can be connected directly to ground, as shown below, with U-shaped wires made from a telephone cable. The wire is 6mm long, 1.5mm wide for the ag31 (1st L12) to ground (ah32) connection, and 6mm long, 3mm wide for the ah30 (3rd L12) to ground (ah28) connection.
For example, to force 200fsb on a 166fsb CPU, connect ag31 to ah32, or to force166fsb on a 133fsb CPU, connect ah30 to ah28, as shown in the table below. Ah32 and ah28 (blue) are ground pins; ah30 can be connected to either of these pins, e.g., to force166fsb on a 133fsb CPU, ah30 can be connected to ah32 instead of ah28. Ak30 is a vcc/vcore pin (red), and is connected to ah30 to force 133fsb from 166fsb.
| CPU type | L12 mod | FSB_Sense[0]
ag31 (1st L12) |
FSB_Sense[1]
ah30 (3rd L12) |
| 166fsb
CPU
1st L12 opened + 3rd L12 closed |
force 200fsb | connect ag31 to ah32 | no connection |
| force 133fsb | no connection | connect ah30 to ak30 | |
| 133fsb
CPU
1st + 3rd L12s opened |
force 200fsb | connect ag31 to ah32 | connect ah30 to ah28 |
| force 166fsb | no connection | connect ah30 to ah28 | |
| Mobile1
1st L12 closed + 3rd L12 opened = reserved |
force 200fsb | no connection | connect ah30 to ah28 |
1 The 1st and 3rd L12 bridges on the mobile CPU are closed and opened to give low (grounded, logic level 0) and high (logic level 1) signals by default. Based on the front-side bus truth table (p.75, AMD 26237.pdf data sheet, rev. C, May 2003), the bus frequency for the XP-M is technically reserved. The combination of a closed (grounded) 1st L12 plus the ah30-ah28 (ground) mod would produce low signals on the XP-M L12s, similar to a 200fsb CPU where the 1st and 3rd L12s are closed by default. Unlike a 133fsb CPU which has opened 1st and 3rd L12s, the same ah30-ah28 connection would force 166fsb L12 from 133fsb.
To measure the voltage levels on pins ag31 and ah30, I've modified the multimeter probe by connecting one end of the probe to a 21g needle.
The test conditions were:
Test 5 stability results
| CPU | 3ag31
(mV) |
3ah30
(mV) |
4Bus frequency | 5Highest error-free FSB in test 5 |
| 12500+ | Grounded
(9.3) |
Grounded
(8.8) |
200 | 228 |
|
6Grounded
(11.9) |
Default/low
(179.3) |
200 | 228 | |
|
Default/high
(2609) |
Grounded
(11.5) |
166 | 228 | |
|
Default/high
(2609) |
Default/low
(178.4) |
166 | 228 | |
| 22100+ |
7Grounded
(13.2) |
7Grounded
(13.3) |
200 | 223 |
|
Default/high
(2609) |
Grounded
(13.3) |
166 | 223 | |
|
Grounded
(12.6) |
Default/high
(1828) |
reserved | 219 | |
|
Default/high
(2609) |
Default/high
(1832) |
133 | 220 |
1 With the default setting for the 2500+, and the CPU seated in the socket, the resistance of ag31 (opened 1st L12) and ah30 (closed 3rd L12) to ground were 1.362k ohm and 0.897k ohm, respectively. On the CPU itself, the resistance of ag31 to ground was unmeasurable, .i.e., ag31 is not connected on the CPU to either Vcore or ground when the L12 is opened. Pin ah30 (closed 3rd L12) gave 0.986k ohms resistance to ground. Several tests were ran for 166fsb and 200fsb L12 settings to see whether the L12 voltage levels (in parentheses) affect of FSB overclocking.
2 With the default setting for the 2100+, and the CPU seated in the socket, the resistance of ag31 (opened 1st L12) and ah30 (opened 3rd L12) to ground were 1.398k ohm and 10.01k ohm, respectively. Both ag31-ground and ah30-ground resistances were unmeasurable on the CPU itself. Thus, the motherboard provides the voltage required to set these L12s to logic level 1, i.e., high signals.
3 Default means no connection.
4 Front-side bus truth table, p.75, AMD 26237.pdf data sheet, rev. C, May 2003.
5 The tests were ran at least twice.
6 With a 166fsb CPU, a single ag31 to ah32/ground wire connection was used to force a 200fsb L12 setting.
7 With a 133fsb CPU, ag31 to ah32/ground, and ah30 to ah28/ground wire connections were used to force 200fsb L12. A single ah30-ah28 wire connection will force a 166fsb L12 setting.
update, 2 aug 03
I've ran more tests comparing the 200fsb ag31-ground L12 mod vs. the 166fsb default L12 on the 2500+. Here're the results with different bios versions, tested using prime95 ver. 23.4.1 (ran for at least one hour) and 3dmark2001 (ran for at least four loops). Conditions were identical to the above except all peripherals were connected, 1.75vcore, and ram timing was 13-3-3-2 to maintain stability at 230fsb. The numbers below represent the highest stable FSB with the 2500+, where both prime95 and 3dmark ran problem-free for the time indicated above.
| Bios version | 166fsb default L12 | 200fsb ag31-ah32 L12 mod |
| 1.0 | 216 | 218 |
| 1.1 | 215 | 217 |
| 1.2 | 215 | 218 |
| 1.4 | 228 | 228 |
| 1.6 | 228 | 228 |
| 1.7 | 228 | 228 |
| 1.8 | 230 | 230 |
update, 24 aug '03
Using bios 1.8, I've ran some tests with forced 133fsb L12 vs. the default 166fsb L12 on the 2500+. Conditions were identical to the above.
| Bios 1.8, fsb x8 | 166fsb default L12 | * forced 133fsb L12 |
| 230 | prime95 ran ok for at least 30 min. | no post |
| 228 | not tested | no post |
| 227 | not tested | winxp pro freeze during startup |
| 225 | not tested | prime95 freeze within 2 min. |
| 223 | not tested | prime95 ran ok for at least 30 min. |
* Using flat-head clips, pin ah30 was connected to the filter capacitor vcore pin on the back of the board, giving the same electrical connection as an ah30-ak30 wire trick. The integrity of ah30-vcore connection was verified by resistance and voltage measurements.
With the 2100+, the highest FSB I've obtained was with 166fsb and 200fsb L12 settings on my board. With the 2500+, the 200fsb L12 mod has no effect on the stability of FSB overclocking with the later bios versions in my system. However, the mod gave FSB increases using earlier bios versions. Forcing 133fsb on the 2500+ reduced the fsb oc by at least 7mhz (from 230 to 223fsb).
For anyone would like to experiment with the L12 settings, I'd suggest trying out the 166 and/or 200fsb L12 settings, depending on your CPU. Since the L12 mod may or may not yield any increase in FSB overclocking, it might also be useful to experiment with different bios versions to determine which one(s) work the best in your system.