Another X61T/s overclocking thread: BSEL + Full VRM

[xiphmont]

No, this was with 2133 MHz, so no dual IDA.

Its not a big deal, I did it mainly to try it out and to lower the temperature.

Oh, hm. I saw reports in the German forum at one point that the L7500 is closer to the voltage limit at the 0x8 multiplier than anywhere else. What's your VID at 8? I've seen the processors report anywhere from 0x18 to 0x1c though usually 0x1c, and I've been speculating that Intel only reduced the requested VID when the processor didn't hit the rated TDP at the usual voltage. If it was stable at reduced VID and hit TDP, it still went into the L7500 bin...

[el-sahef]

What's your VID at 8?

CPU-Z shows 1.063 V at full load with multiplier 8. I have another X61t board here at the moment where I tried the BSEL mod. It also has the L7500, but the voltage is only 1.025 V at full load. I suspect that this is the reason why it crashes almost immediately with BSOD after starting prime95 (tested with 533-flashed RAM known to work with the BSEL mod).

[wileE]

CPU-Z shows 1.063 V at full load with multiplier 8

Same here.
On dual IDA it can jump to 1.175 V for a second ot two. Only with a light load.
My L7500 never throttled, but I cannot undervolt either. Machine freezes when I try that.

[xiphmont]

What's your VID at 8?

CPU-Z shows 1.063 V at full load with multiplier 8.

so 0x1c, the full normal voltage. If you drop to just one active core, will it undervolt? If not, unlucky processor

I have another X61t board here at the moment where I tried the BSEL mod. It also has the L7500, but the voltage is only 1.025 V at full load.

So a FID/VID of 0x819. That's quite low! The lowest I've seen L7500s report is 0x818; The two mobos with those were high-middle part of the pack in terms of power draw too. Those were not part of the BSEL experiment, I should try them to see if they instacrash under load too.

BTW, I tested the actual supplied core voltages on about half my mobos at idle and found the regulators to be really remarkably precise. None was off by more than 5/1000ths of a volt from the selected VID! I have not tried to measure it during load, partly because the ripple figure is probably more relevant than average/sag.

I suspect that this is the reason why it crashes almost immediately with BSOD after starting prime95 (tested with 533-flashed RAM known to work with the BSEL mod).

It lends support to my theory that the processors reporting low VIDs are closer to being overbinned.

[xiphmont]

CPU-Z shows 1.063 V at full load with multiplier 8

Same here.
On dual IDA it can jump to 1.175 V for a second ot two. Only with a light load.
My L7500 never throttled, but I cannot undervolt either. Machine freezes when I try that.

WileE: Is that with the full VRM mod or stock?

[wileE]

Is that with the full VRM mod or stock?

Stock pinmod. No way I can do the VRM mod with my present skills.

Did you look at the VID of your boards before the pinmod? Maybe noticed a pattern of differences that could say something about the probability of success before doing the pinmod.
Most people (me included) have done it by cutting the trace on the board. Knowing which ones should definitely not be modded might save a few boards.
My untouched L7500 boards seem to have some differences in the voltage.

[xiphmont]

Did you look at the VID of your boards before the pinmod? Maybe noticed a pattern of differences that could say something about the probability of success before doing the pinmod.
Most people (me included) have done it by cutting the trace on the board. Knowing which ones should definitely not be modded might save a few boards.
My untouched L7500 boards seem to have some differences in the voltage.

I didn't! I still have to repeat some of the overclock testing because of my memory screwup... it was time consuming ( a few days) first time around.

I did record voltages on all the mobos, though only at idle. No pattern jumped out at me first time around.

[xiphmont]

I've finished re-collecting more detailed BSEL overclock data across the 15 test motherboards. ...is there a way to do tables in the board markup? (I wish I didn't have to make everything a link to an outside site-- it seems too often that I Google something to see if some one has done something cool before, and they have! But all the results and images are broken links and have been for years, and there's no longer any record of how it was done).

BTW, el-sahef: Now that I looked at all the multipliers before and after, I see that on the L7500 the 8x multiplier is in fact the usual problem point. The 9x multiplier (IDA) usually has much more voltage headroom, and although the VRM mod is necessary to get the necessary current for dual IDA, voltage is seldom a problem. The VRM mod only helps very slightly with the 8x multiplier (only one or two additional VID steps if any). On my full-VRM modified board, I hadn't checked what my margin on the 8x multiplier was, just that it appeared stable at 8x as it had been pre-mod. Even though I can undervolt the 9x multiplier on that board, the 8x is actually pretty close to the edge.

Anyway, all the raw data coming as soon as I format it!

[xiphmont]

...and here it is! I recollected the minimum 'stable' VID settings for more conditions this time around to get a clearer picture of what's going on. The recollection agreed closely with my first run, it's just more detailed. All the data I've collected is there including some likely irrelevant stuff. Perhaps it can help sort out which motherboards are likely to take the mod well ahead of any permanent changes. I don't see any hard-and-fast rules in the data though

<-- HTML table here

I'd misremembered badly how many successful overclocks there were; of the 12 L7500s on the list, half successfully completed the quick load stability test I ran after modding (there was one more success not on the list that got further modded before my detailed recollection). But even the successful conversions don't have a ton of voltage margin.

None of the L7700s tested had a prayer of staying up for long if they booted at all. I did test two with a +4%, +8% and +12% hardware voltage boost; it didn't help much, but that was before thinking of trying the VRM mod. Comparing the one-core vs. two-core numbers, the load regulation seems way beyond rated wattage. On mobo 10 (my VRM mod test), the one- and two-core numbers were ~ identical at all FIDs before and after overclock with the VRM mod. That alone still probably isn't enough to make the L7700 stable, but VRM + voltage boost together will likely work. I'll try that eventually.

[RMSMajestic]

Maybe a bit off-topic, but feel free to check the X61 northbridge overvolt method here http://forum.thinkpads.com/viewtopic.ph ... &start=120

While I'm still waiting for any1 to modify the X61t BIOS for P8800 and better CPUs

[xiphmont]

Maybe a bit off-topic, but feel free to check the X61 northbridge overvolt method here http://forum.thinkpads.com/viewtopic.ph ... &start=120

Thanks! I had not seen that hack before. It might come in handy later. Does it increase the voltage used by the NB, or the DRAM bus drive, or...?
I'm not currently limited by memory speed, at least, it seems when a stick throws errors it's the sticks themselves that are the limiting factor. At least that's been my assumption. I have 2x2GB sticks that are running happily at DDR888 w/ 1066 static timings on all the boards without trouble.

While I'm still waiting for any1 to modify the X61t BIOS for P8800 and better CPUs

....are you capable of reballing!?

[RMSMajestic]

Thanks! I had not seen that hack before. It might come in handy later. Does it increase the voltage used by the NB, or the DRAM bus drive, or...?
I'm not currently limited by memory speed, at least, it seems when a stick throws errors it's the sticks themselves that are the limiting factor. At least that's been my assumption. I have 2x2GB sticks that are running happily at DDR888 w/ 1066 static timings on all the boards without trouble.

Well, I don't think that X61 has separate voltage control for NB, DRAM controller and Integrated graphics......

....are you capable of reballing!?

Not me, but BGA reballing is quite cheap in China, about 20-25GBP each time

[wileE]

The testing for that table was probably a lot of work, but you do know that it is utterly incomprehensible for all those of us who do not write code every day?
At least I understood that you did not find any patterns for predicting a successful pinmod.

And could you please tell how you got your OS to recognize and use the faster FSB?
There is not a single piece of solid information on overclocking in linux to be found. Maybe it is as simple as writing new FID:VID for the phc-intel driver to use?

[xiphmont]

The testing for that table was probably a lot of work, but you do know that it is utterly incomprehensible for all those of us who do not write code every day?

Well, complicated things are complicated. I suppose I really do need to write a little guide to reading the chart and what's useful about it. I was a bit worried about explaining stuff everyone already knew, especially when I'm just learning it myself and am in no way any kind of authority. The guys on nb51 and the German forum are the real pros, they'd probably roll their eyes at my rambling.
OTOH, I was also considring making some nice bar charts

At least I understood that you did not find any patterns for predicting a successful pinmod.

Correct, although you can try undervolting the processor before modding the motherboard, figure out the amount of voltage margin you're starting with, and then make a pretty good guess if overclocking via FSB will work without fuss. If you've got less than 13 VID steps spare on multiplier x8, it probably won't go without a voltage boost. Actually, every machine I tested has only a miniscule number of VID steps to spare, makes me leery of runing any of them without a mild boost for x8.

And could you please tell how you got your OS to recognize and use the faster FSB?
There is not a single piece of solid information on overclocking in linux to be found. Maybe it is as simple as writing new FID:VID for the phc-intel driver to use?

I suppose that's a solid request for 'write a guide'
Writing new FID/VID combinations to phc-intel is for having the in-kernel speedstep management set custom voltages automatically. It's roughly equivalent, once set, to the processor having declared those settings itself, and the kernel treats the new values you give it as authoritative when changing speed. It has nothing to do with setting frequencies, just what voltages match what multipliers.

There's no need to inform the OS about the faster FSB; it already knows. The machine is perfectly aware it's being overclocked, and the OS probes FSB speed from the hardware. The only reason it's necessary to mess with the DDR2 timings is because the BIOS apparently never checks. It appears to simply assume 'I'm an X61T, my FSB is 200, off we go!' It's a mixed blessing-- apparently it also means the BIOS doesn't try to lock you to low multipliers, which I've read elsewhere is a problem on the T61 (don't know for sure, never had a T61).

[xiphmont]

Well, I don't think that X61 has separate voltage control for NB, DRAM controller and Integrated graphics......

I meant, which of the voltage rails does the hack affect? I suppose I should go consult the schematics.

....are you capable of reballing!?

Not me, but BGA reballing is quite cheap in China, about 20-25GBP each time

Argh, Chinese hackers have access to all the cool stuff that _used_ to be here in the States... but isn't anymore...

OTOH, at that point, it's almost getting to 'let's just design/fab a new planar'... All of this effort will feel silly if the X62 ever drops, and if it doesn't someone else will just have to make it.

[el-sahef]

Thak you for your testing efforts! It think it would have been better to write the actual voltage value instead of the VID, but this is high-level moaning.

The only reason it's necessary to mess with the DDR2 timings is because the BIOS apparently never checks. It appears to simply assume 'I'm an X61T, my FSB is 200, off we go!' It's a mixed blessing-- apparently it also means the BIOS doesn't try to lock you to low multipliers, which I've read elsewhere is a problem on the T61 (don't know for sure, never had a T61).

The three BSEL[0:2] pins on the CPU determine the FSB. They are connected to the clock generator and the chipset on X61 and T61. So the chipset reads the FSB according to the BSEL value, it does not assume anything. Because intel 965 chipset does not support 266 MHz FSB, you have to make it think it is running at 200 MHz FSB. Some notebooks will still start with the BSEL pins at the chipset @ 266 MHz setting but lock the multiplier down to 6x. Others (including T61 and X61) do not start at all. I do not know why there is this difference. Maybe it depends on the BIOS implementation.

With a CPU made for 200 MHz FSB that you want to overlock, you cut the connection CPU BSEL --> clock generator BSEL and set the according pins at the clock generator manually by a pullup or pulldown resistor (for 200 --> 266 it is just one pin). If you want to use a CPU made for 266 MHz FSB, then you cut the connection CPU BSEL --> chipset BSEL and set the BSEL pins at the chipset manually to 200 MHz FSB so that the chipset thinks it runs at 200 MHz FSB although it is in fact running at 266 MHz FSB (for 200 --> 266 it is just one pin). This is also the reason for the somewhat odd memory speeds when overlocking.

The images linked in this post show what I mean.
Please ignore the ground connection for BSEL2 on the image for 333 MHz FSB. It is very likely that it is not necessary.

[xiphmont]

Thak you for your testing efforts! It think it would have been better to write the actual voltage value instead of the VID, but this is high-level moaning.

I did measure voltages, but they're simultaneously uninteresting and difficult to interpret-- At idle (near-zero current draw), voltage was within 5/1000th volts of the level selected by the VID on all of the motherboards.

As the CPU started drawing current things got more complicated. The Intel Core2Duo spec sheets indicate regulation should be designed for a loadline voltage sag of 2.1 mV per amp drawn, so the 'correct' (or at least expected) voltage at the processor depends on what the processor is doing at that moment. I had no way of measuring the exact instantaneous current (and every processor had a measurably different power draw!), but given a nominal average draw of 20-30A at full load peaking at 50A, everything appeared approximately nominal on all motherboards running the stock config.

The three BSEL[0:2] pins on the CPU determine the FSB. They are connected to the clock generator and the chipset on X61 and T61. So the chipset reads the FSB according to the BSEL value, it does not assume anything. Because intel 965 chipset does not support 266 MHz FSB, you have to make it think it is running at 200 MHz FSB.

Right. After cutting the BSEL1 trace, the chipset and CPU are still connected on all three BSEL lines, only the clockgen is lifted. I did try pulling the uncut line hard low (like a real CPU pinmod would do) and the machine would not even boot to the splash screen.

What I meant was that if you ask software like memtest86 or the linux kernel what the FSB is after cutting the trace, it will happily tell you '266' despite what the CPU or chipset thinks. I had assumed, and I could be very wrong, that it was checking the clock generator configuration directly as opposed to simply going by what the CPU asked for.

With a CPU made for 200 MHz FSB that you want to overlock, you cut the connection CPU BSEL --> clock generator BSEL and set the according pins at the clock generator manually by a pullup or pulldown resistor (for 200 --> 266 it is just one pin). If you want to use a CPU made for 266 MHz FSB, then you cut the connection CPU BSEL --> chipset BSEL and set the BSEL pins at the chipset manually to 200 MHz FSB so that the chipset thinks it runs at 200 MHz FSB although it is in fact running at 266 MHz FSB (for 200 --> 266 it is just one pin). This is also the reason for the somewhat odd memory speeds when overlocking.

Yup.

[el-sahef]

I did not mean that you should have measured the actual voltage on the mainboard but that you should have written the voltage value that corresponds to the VID instead of the VID itself for better readability. I do not know them inside out .

And I have problems understanding them. For example, you said that 1.063 V (closest step is 1,0625 V) corresponds to VID 0x1c. But according to the VID table in the PDF for the L7500 linked here (page 22ff) 1,0625 V corresponds to 010 0011 which I would write as 0x23.

What am I doing wrong here?

[Qing Dao]

....are you capable of reballing!?

Not me, but BGA reballing is quite cheap in China, about 20-25GBP each time

Quick question on reballing in China. Where? I see infrared reballing machines and hot air rework stations in many stalls at "technology markets." Is that where, or are there better alternatives? Something like 200rmb doesn't sound bad for the job, but how high is the risk? While I don't deny their technical expertise to do a reball, in general they do seem to handle a lot of sensitive hardware very wantonly. After watching some youtube videos on it, I've thought about getting a high end hot air rework station from Taobao, but I feel that it might be biting off more than I can chew.

Argh, Chinese hackers have access to all the cool stuff that _used_ to be here in the States... but isn't anymore...

You would be surprised. Just because you don't see signs on the street offering these services doesn't mean they don't exist. Put up an ad on craigslist or something like that saying what you are looking for. At my father's work they have a team of guys whose job is to fab and repair delicate electronics. It is their full-time job to do that kind of stuff and they have all the latest and greatest tools. They can take anything they want with them there to work on in their spare time and they do tons of stuff like this for people. I'm fairly certain that they aren't the only such people in the entire United States! Just put yourself out there and ask around.

[RMSMajestic]

Quick question on reballing in China. Where? I see infrared reballing machines and hot air rework stations in many stalls at "technology markets." Is that where, or are there better alternatives? Something like 200rmb doesn't sound bad for the job, but how high is the risk? While I don't deny their technical expertise to do a reball, in general they do seem to handle a lot of sensitive hardware very wantonly. After watching some youtube videos on it, I've thought about getting a high end hot air rework station from Taobao, but I feel that it might be biting off more than I can chew.

risk isn't that high, If you have a BGA-to-PGA CPU that has the PINs soldered directly onto the pads, according to plclife, 99%.
The problem is, once you have a fail, it adds about 100-200RMB extra cost to every other motherboard in that batch of motherboards.
And, another big problem is the number of times of BGA reballing that a cpu/motherboard can handle. plclife has access to the factory full-time repair guys and machines, he gives the number: 3 on average. Having more than 3 BGA reballing, the CPU/motherboard will have fallen pads and bumps

If you wanna contact plclife, I'm happy to be the liason person

Alternative place in Shanghai is Qiujiang Road (虬江路) , right next to Baoshan Road Station (宝山路站) of Line 3 and 4. Price there is about 200 RMB, but success rate is about 70-80%. In Huaqiangbei(华强北) in Shen Zhen (深圳), success rates is above 90%.

I've spent about 500 RMB on my X61s P8800, but I happened to be one of the lucky guys. Normally X61 T9900 is around 800-1000 RMB in cost (including the cost for BGA)

[RMSMajestic]

Quick question on reballing in China. Where? I see infrared reballing machines and hot air rework stations in many stalls at "technology markets." Is that where, or are there better alternatives? Something like 200rmb doesn't sound bad for the job, but how high is the risk? While I don't deny their technical expertise to do a reball, in general they do seem to handle a lot of sensitive hardware very wantonly. After watching some youtube videos on it, I've thought about getting a high end hot air rework station from Taobao, but I feel that it might be biting off more than I can chew.

There is a X61 T9900 on taobao here:
http://2.taobao.com/item.htm?id=4268702 ... bbucket=14

TBH with you, 850 RMB isn't expensive

[xiphmont]

I did not mean that you should have measured the actual voltage on the mainboard but that you should have written the voltage value that corresponds to the VID instead of the VID itself for better readability. I do not know them inside out .

OH! Ha ha, of course. That's a good idea. I'll go add that.

And I have problems understanding them. For example, you said that 1.063 V (closest step is 1,0625 V) corresponds to VID 0x1c. But according to the VID table in the PDF for the L7500 linked here (page 22ff) 1,0625 V corresponds to 010 0011 which I would write as 0x23.

What am I doing wrong here?

The physical lines are active-low, and the bits get 'flipped' when reading the values from the registers. VID6 isn't represented in the register at all... perhaps because it's not under software control? in any case, 010 0011 on the physical lines (where 0 is low and 1 is high) -> 101 1100 (where 0 is 'inactive' and 1 is 'active') &= 011 1111 (mask) -> 001 1100 or 0x1c, the value that appears in the hardware register.

[xiphmont]

You would be surprised. Just because you don't see signs on the street offering these services doesn't mean they don't exist. Put up an ad on craigslist or something like that saying what you are looking for. At my father's work they have a team of guys whose job is to fab and repair delicate electronics. It is their full-time job to do that kind of stuff and they have all the latest and greatest tools. They can take anything they want with them there to work on in their spare time and they do tons of stuff like this for people. I'm fairly certain that they aren't the only such people in the entire United States! Just put yourself out there and ask around.

There was a time about 20-25 years ago when those kinds of things did exist here (signs on the street corner, you could walk right in), but the electronics, semiconductor, and computer industries substantially pulled up their roots and left New England after DEC collapsed, and the surrounding hobbyist scene fueled by local engineers similarly evaporated when the engineers left. Not BGA specifically (BGA is a more recent development , but similar kinds of stuff. There was a brief resurgence in the dot-com bubble, but it wasn't deep and only lasted a few years. The same decline has happened on the west coast, though it's been slower. The hardware industry in the US has been substantially replaced by software.

You still find plenty of hardware hackers (and being plugged into the MIT maker scene, I can find most expertise I'd want), but it's a collection of individuals working on individual an usually crazy projects. Everyone is doing something different and custom and one-off, and even these guys outsource everything they can to overseas manufacturers. Eg, we all know _how_ to reball, but no one is set up to _do_ it because it's cheaper now to go to china on the rare occasions you need it. Positive feedback loop.

The maker community is still vibrant, but at least here it's not self-sufficient anymore. OTOH, production quality standards have also gone way up since the 80s! So there have been some benefits too.

[khseal]

Hello)
Very Nice post. I'm used ThrottleStop for overlclock.
My config http://kodabra.ru/pictures/x61t_1.jpg
I made a hole that had been closed.
http://kodabra.ru/pictures/x61t.jpg
I want to make the second phase of the VRM.
Thanks for the info.

[khseal]

Where order non-standard SMD resistors?
I need to 143K, 165K, 240K, 182K.
I have purchased a broken board from X61s. That's my fault. I should have buy a motherboard from X61. =(

[khseal]

Trying to do today Full VRM. I made a mistake. I forgot to remove 26 R634. Laptop works, but the second phase does not work.
Perhaps I damaged ADP3419.
In FETs U2 and U6 no control signal.

[xiphmont]

Trying to do today Full VRM. I made a mistake. I forgot to remove 26 R634. Laptop works, but the second phase does not work.
Perhaps I damaged ADP3419.
In FETs U2 and U6 no control signal.

I'm guessing somewhat, but looking at the spec sheet and schematics, I think you likely did not damage anything by forgetting R634 for a while. If holding the IN pin of the ADP3419 high dumped too much current into VCore, the voltage sense circuitry should have noticed that and shut down the second channel. Also, there's a bootstrap sequence and not feeding at least a cycle or probably does nothing. Also a guess, but the ADP3207 should have noticed PWM2 was still pulled high and asserted shutdown on the second channel anyway. Power circuits like this tend to be multiple-fault-tolerant, so perhaps there is some other conversion mistake preventing it from running.

BTW, the second channel is only active under load-- the master controller will shut it down to save power when there's insufficient current demand. Make sure the machine is crunching hard before looking for activity on the second channel.

[khseal]

False alarm and facepalm =)

BTW, the second channel is only active under load-- the master controller will shut it down to save power when there's insufficient current demand. Make sure the machine is crunching hard before looking for activity on the second channel.

Yes, under the high load everything works fine.
Just been triggered protection with a resistor and a laptop does not turn on. It scared me =)

[khseal]

Everything works fine, but there is overheating. Overheating is not the processor. I installed radiator 42X3805 by x61.The power supply system has a low CPU temperature.
I think it is overheating Northbridge or something near it.
I installed the drivers for the video card. http://brenton.nom.za/forum/showthread.php?tid=1
They are better decoded video.
The problem is that the North Bridge in the BIOS is the low threshold temperature(some 65-70 degrees) shutdown.
Even without a load on the processor shuts down notebook sometimes occurs. Saves a laptop cooling pad =(

Overheats BUS or PCI.

[wileE]

I made a picture of five of the fans xiphmont tested in this post: http://forum.thinkpads.com/viewtopic.ph ... 85#p744267

Text is in german, but the pics might be usefull to some.
Very large picture warning: http://thinkwiki.de/X60_%26_X61_K%C3%BC ... %C3%BCfter

Still missing decent pics of the X60 full voltage 42W2522.