Socket P CPU in T60 (or in any other socket M board)

[pc2005]

Hi

Once I've got a little "bored", so I compared merom and penryr datasheets. It seemed most of the pins were compatible and there was only like 70 misplacings, so I made a pin remapping figure (for socket P CPU).



Do you think it would be possible to do? (socket M with unlocked multiplier is much more expensive than socket P with unlocked multiplier)

[dr_st]

Do you think it would be possible to do? (socket M with unlocked multiplier is much more expensive than socket P with unlocked multiplier)

Let's suppose that it's technically possible. When you factor the extra time, effort and probably cost to do this mod - do you still find that Socket M is more expensive?

Just buy a T61 board, sheesh.

Somehow an unlocked multiplier for a 10-year old, practically useless in today's world CPU does not make me excited.

[axur-delmeria]

You're free to try. The pin spacing is too small TBH.

[TPFanatic]

Easier to just swap in a T61 board.

4:3 T61 board to 4:3 T60
15.4" T61 board to 15.4" T60

and boom, Socket P capabilities.


May be easier and cheaper to just get a T61.

[Screamer]

Once I've got a little "bored", so I compared merom and penryn datasheets. It seemed most of the pins were compatible and there was only like 70 misplacings, so I made a pin remapping figure (for socket P CPU).

Do you think it would be possible to do? (socket M with unlocked multiplier is much more expensive than socket P with unlocked multiplier)

No, I highly doubt that is possible. You are looking at physically re-routing Socket P's pins to match Socket M's pin layout, which would require a vertically short interposer or a lot of wires to match Socket M's pin layout. If you are going to re-route the pins manually via wires, please take note that Socket P and Socket M has 2 pins that are not used (and are also positioned differently). Socket P has pin A1 and B1 blocked, but Socket M has pin A1 and A2 blocked. This would mean that a Socket P processor will get rejected by Socket M, due to pin A2 being blocked by Socket M. Because of this, you would have to drill or hammer a hole in the socket to 'create' a space for pin A2. This is a little dangerous though, considering that the contact pads are directly underneath the socket. You also risk damaging the motherboard, if you are not careful.

Then, you would have to manually edit the T60's BIOS to add the microcodes that suits the Socket P processor that you are trying to use on Socket M. This in turn, would also need the embedded controller to support the Socket P processor that you are trying to use.

While I can say that the first 2 modifications are somewhat possible, I don't think I can say the same for the modification of the embedded controller. The embedded controller has only been modified in the past by people who wanted to change the RPM values in the fan speed table. Say, for example a fan speed level of 7 would mean 4000 RPM, but the noise is 'annoying, loud, disturbing'. This is then changed to 3600 RPM (for example) for the said fan speed level, by editing the embedded controller's fan speed table.

Honestly, all of these steps would only make the T60 boot up with a black screen, at the very most. The reason being that the motherboard in mind was designed to power a Socket M processor, would mean that the voltage regulator module's design and revision would be very likely incompatible and so would supply insufficient power to drive a Socket P processor. Take the Core 2 Duo T7600 for example, it has an IccMax value of 41A. Then compare the Core 2 Extreme X9000's IccMax value of 57A, which would likely overload the voltage regulator module.

Other than that, let's have a look at the chipset. The T60 uses the 945GM or the 945PM chipset, though it depends on the model. (models with the ATI graphics chip are paired with the 945PM, models without the ATI graphics chip are paired with the 945GM chipset) These 2 chipsets share one important common trait in mind, which is a maximum front-side bus speed of 667MT/s. By taking this into account, any Socket P processor that utilizes a front-side bus speed of 800MT/s wouldn't run with the 945GM or the 945PM chipset. To prevent this, you would have to set the BSEL[2], BSEL[1], and BSEL[0] pins to voltage Low-High-High respectively. This would force the inserted Socket P processor to use a front-side bus speed of 667MT/s, which would reduce its memory bandwidth and clock frequency. This shouldn't be much of an issue for the Socket P processors with an unlocked multiplier though, but the ones with a locked multiplier will face a performance drop.

Going further in electrical mumbo-jumbo, the AGTL+, CMOS, and open drain signal group DC specifications has a slight change in them. While most of the specifications appears to be left unchanged, the pad capacitance values had changed for all three signal group DC specifications. For Socket M's AGTL+ and CMOS specifications, the pad capacitance is 1.6 picofarads at the minimum, 2.1 picofarads typically, and 2.55 picofarads at the maximum. In Socket P's AGTL+ and CMOS specifications, the pad capacitance is 1.8 picofarads at the minimum, 2.3 picofarads typically, and 2.75 picofarads at the maximum. Furthermore, Socket M's open drain specifications state a pad capacitance of 1.9 picofarads at the minimum, 2.2 picofarads typically, and 2.45 picofarads at the maximum. This in comparison to Socket P's open drain specifications, is even worse than the previous comparison. I'd say that this modification is not doable, and probably serves no purpose at this point.

Just buy a T61 board, sheesh.

Easier to just swap in a T61 board.

and boom, Socket P capabilities.

I suggest that you follow their advice, because a motherboard replacement is far easier than the rest of the modifications I have suggested above. Please, or I will punish myself by wearing a T60 as a hat with the word 'DUNCE' written on it.

[pc2005]

First, thank you everybody for the answers, I appreciate your opinions and ideas. I can assure you I wont try the pin remap in near future as I don't have (yet) the equipment and I'm still using my T60p for useful work, but I have another MB with i945 chipset, where the mod could be efficient even today (really good mITX board).

The microcode: It isn't a problem at all, I will soon move to the coreboot on both devices anyway. A modification of vendor's BIOS should be possible too. I've played with both anyway.

The cost: That's how I got the idea originally. T7600G (unlocked socket M) was like $200 on ebay and socket P was much lower in price. The prices went down during last two years, but unlocked socket M is still 3x more expensive than unlocked socket P (and it is comparable with unlocked quadcore extreme). I don't count my time as cost it is only for fun (and learn) and as a proof (for me) there was no reason to change socket pinout by Intel, so there is a space for material cost.

The power: yup this means I probably cannot overclock X9100 as much as T7600G. With the exception of 50% larger L2 cache and SSE4.1 instructions, the X9100 (45nm, newer generation) should be same or more efficient than T7600G (65nm). If I set X9100's multiplier to 14x, both FSB (667 MT/s) and core (2.33 GHz) will be lower than original speeds (1066 MT/s, 3.07 GHz), so the power consumption will not hit its maximal values. BTW if you look at quadcore Q9100's Iccmax at 2.26 GHz is only 64A (two penryr chips on a single socket), so only one penryr chip at the same speed will be probably a little more than half -> 30-40 A.

The pin reroute: I've got the idea from an actual LGA771->LGA775 mod, but with pins instead of pads. There are already dummy power load sockets anyway. The socket P CPU could be made into a BGA chip, but the PCB could be probably made from thin/flexible material (so it could fit into T60p) and some pins could be left soldered in the original socket. From my analysis most of the pins would be straight-through, only ~70 needs to be rerouted. The pinout isn't too much jumbled so I can route like up to 1 (if any) wire between pins and up to 1 (if any) via in the middle of 2 by 2 pin square. The pins around the key are easy. Surplus pin can be just removed (it's a ground anyway, if it is a fake ground/enable pin, it could be routed from neighbor ground). The missing RESET pin would be soldered on the new PCB.

While I can say that the first 2 modifications are somewhat possible, I don't think I can say the same for the modification of the embedded controller.

Are you talking about H8S2116? What functions does it run which are incompatible with penryr? It seems it only controls fan, charging and maybe power monitoring. It should be compatible with penryr(/socket P merom).

To prevent this, you would have to set the BSEL[2], BSEL[1], and BSEL[0] pins to voltage Low-High-High respectively.

Yes of course I took the BSEL mod as a granted thing. These signals are an almost constant value anyway, no problem with rerouting anywhere on the board.

For Socket M's AGTL+ and CMOS specifications, the pad capacitance is 1.6 picofarads at the minimum, 2.1 picofarads typically, and 2.55 picofarads at the maximum. In Socket P's AGTL+ and CMOS specifications, the pad capacitance is 1.8 picofarads at the minimum, 2.3 picofarads typically, and 2.75 picofarads at the maximum. Furthermore, Socket M's open drain specifications state a pad capacitance of 1.9 picofarads at the minimum, 2.2 picofarads typically, and 2.45 picofarads at the maximum.

I don't think these will be relevant when BCLK will be only 166 MHz (signals 667 MHz), yes on 266/1066 it could be a problem. I'm more concerned about signal length changes caused by pin rerouting or some CPU→NB protocol changes (but still, there were meroms for soc M and soc P, so it's improbable).

Please, or I will punish myself by wearing a T60 as a hat with the word 'DUNCE' written on it.

[Thinkpad4by3]

They have made in China i7-4820HQ mobile CPUs that are BGA into desktop CPUs in the exact method you mention but it looks very difficult. And don't even mention the BGA soldering of the CPU itself onto a main board.

[Screamer]

Seems like you know what to do, that is good to know.

Yes, the H8S/2116B (according to the T60 and the T61's schematics). I am not quite sure of what makes it unable to run with a Penryn installed, since the only lead I had was a different firmware that could've been used for it. I guess that this shouldn't impact the ability to run a Penryn on a T60, so you should be right.

My main worry lies in the voltage regulator module's revision, because that has been bothering me ever since the PGA478 days. History: during Prescott's era, some PGA478 motherboards that did not comply with the VRD 10.0/FMB 1.5 specification could not run with a Prescott installed (some Prescotts did comply to the earlier VRD 10.0/FMB 1.0 specification). This was made even worse when the LGA775 motherboards came out (most of them didn't say anything about complying with the VRD 10.1 specification), because every Conroe and Kentsfield derivatives required the VRD 11.0 specification to operate.

That old VRD fiasco had mainly affected desktop motherboards, but may also affect laptop motherboards. Although, I am not sure whether if it does, because Intel didn't specify or even release a VRD guideline for Socket M or Socket P. This may affect the ability to run a Socket P processor on a Socket M platform, but don't take my word for this.

[Thinkpad4by3]

While knowing what to do is great...it might not help so much here.

Unless you have one of these, I'm not sure where the project is going to go. A PCB manufacturer might solder pins on for you...but the CPU? Doubt it. Most places like that are really only equipped for the odd QFN package or maybe a small count <48 BGA package. Soldering a C2D with its pins removed onto a PCB would be a nightmare.



Also consider this: How are you going to deal with a taller CPU? Now the cooler has to be changed and it might interfere with the keyboard and.......