Replacing onboard crystal
 

Hi,
I've been modding a cheap nVidia 6600 (non GT) some time ago. Picture of the card (warning high res): http://home.scarlet.be/wittekakker/6600/IMG_0937.JPG

Since pc parts always work with a clock, there must be a oscillator somewhere. Cystals have been used for many years now when it comes down to high precision oscillators. But just because it is a crystal, it is not possible to switch it's frequency just by soldering an extra resistor on your hardware or anything like that. You will have to replace the crystal. And so I did ;-D

My nVidia 6600 videocard has one crystal with a frequency of 27MHz, but yesterday I came across an old ATI Rage wich had a 29MHz crystal. I took both crystals off and soldered the 29MHz crystal back but on the nVidia 6600 this time. And... the card booted :woot:

First thing I noticed was that in fact some changes were made. My TFT is set to it's maximum resolution: 1600*1200 60Hz. After the mod I got a message: "frequency out of range". With Windows safe mode I could set a lower resolution and get into Windows XP again. So changing the crystal atleast did something.
After that I opened a bunch of software tools to overclock my videocard. The programs all red default card clocks. Before I did this mod I got a 3D Mark 06 score of 981 3D marks. After the mod I gained about 100 points with the same clock applied in software. I'm still further investigating what is happening here, but it's remarkable that I got a very decent performance boost. I will come back with a lot of info, but until then, does anyone know what this mod exactly does? Will it only change some basic clocks like the core clock without the software being able to read it, or would the whole card get a little boost (even the core's internal clocks wich I can not read)?

Here is what I got:

at overclocked speeds it doesn't seem to offer a noticeable performance boost, but at stock speeds... wow.. where did you get the idea & info to do this? :)

By changing the crystal the 6600 got about 5~10% faster in synthetic benchmarks. But wih overclocking we see something differant. The 6600 with 27MHz crystal seem to overclock much further then with the 29MHz crystal. In our 3D Mark 2001 Nature test this is enough to almost equilize the 29MHz crystal equiped 6600. The small margin is neglictable.

I have a 32MHz crystal ready to pick up next week friday, stay tuned.

I saw Hipro5 @ XS talking about this. I've never heard of anything like this, but my thought: why not just try it out :)

Crystal is before PLL days; cheaper to produce too. I remember a similar talk with my youngest brother when he was in electrical engineering school back in 1983. Harmonics in frequency response -
http://www.glenbrook.k12.il.us/GBSSC...nd/u11l4d.html

Good to experience with things around :)

Crap, with a 32 MHz crystal my screen stays blank even after boot. Nothing wrong: after some seconds I here my speaker playing that nifty Windows logon sound, but to bad I just don't get anything on my screen. Well... back to the 29MHz it is :)

not sure it's a good thing to OC that part of the vga;)

9% performance boost in every synthetic benchmark. Using Rivatuner might be a better smarter thing to do :p Heck, I know what it does now :)

9% at stock speeds, once you OC that difference at same OC speeds disappears :)

Can you indicate the crystal on your picture? i might be interested to trie this on my 6200 64bit.
It might make it a bit less crappy :)

They often look like this: http://www.sure-electronics.net/engl...ors/full31.jpg

When you look at the picture from my first post, it's right in the middle on the left side if the heatsink.

I thought it would be that one.
I'll check my collection of dead cards next weekend and report back if i was able to do the mod.
Tnx for the info geoffrey!

I asked myself a guestion yesterday on XS, and I'm going to ask that same question here too...

As you can see in my above posts, changing the crystal affects the whole VGA card without being able to read it. Now... what if I did the same with my mainbord. Assuming it would work, my system would be faster, though Windows would indicate me running @ stock speed like I showed above. Now the thing is: the whole system is based on time/oscillators, and yes, even Windows depends on that, it doesn't communicate with the atomclock in Germany for example. If I changed a specific crystal, would that make 'Windows Time' run assynchronic with the real time? Would it be possible to travel back in time |D

if I buy a faulty watch, does that make me a time traveler ?

I was joking about that last part, but the fact is, if I can change the speed of Windows time, I can fake benchmarks also. I'm curious if it's possible, though, if it is possible, who will believe me when I ever set a high score. Kinda like a double edged sword eh.

I doubt it will impact benchmarks that much:)

superpi is about ms though

Need some help for my next little project.

Someone can find 'ICS 9250CF-08' datasheet?

Closest I could get till now was ICS 9250-10 datasheet.
E-mailed them, I hope I get some respons.

EDIT: ICS = IDT nowadays, www.idt.com

http://www.idt.com/products/files/24...ics9250-08.pdf

this is without CF , but it looks like the one for early pentium boards...

Thx buddy. It's indead used on the older Pentium boards, to be exact, on my ASUS P3B-F. It is the only hardware I can mess around with when visiting my father. I'm planning on taking it the extreme tour.


Voltage setting thrue BIOS doesn't work for example. I did a simple V-mod on the power regulator IC but it didn't really like it. Further research learned me that the CPU sends a 5 bit TTL signal to the power regulator. This digital code is internal converted to an analog signal, for each code the analog signal reads a differant voltage. This analog signal is then compared against the output voltage of the power regulator, 3 times.
1 -> to check over voltage (110%)
2 -> to check under voltage (90%)
3 -> over voltage protection (115%)
If the output voltage stays withing the marges, the power regulator will create a POWERGOOD signal, and if all goes well your system will boot.

The trick we used to do so much when voltmodding is fooling the feedback part of the power regulator. The output voltage is being send back to the PR through a series of resistors. If we change the resistance of that circuit, the voltage at the feedback pin will change also, causing the power regulator to regulate it's output to a higher/lower voltage. But the thing is, there is no resistor circuit used with this power regulator, so I don't really know how to fool the feedback circuit.

But I do know what digital code has to be set to the power regulator to set a certain output voltage. I have to losen those pins so that the CPU can no longer do what's suppose to do and then set the code myself with some dip switches or stuff like that.

I needed this datasheet to find a way to get above the 166MHz limit the BIOS allows me, and to add some hardlocks on the PCI bus/AGP bus.

Recently I scored almost 31.000 3D Marks with the 6600 I modified in the first post. This was enough to take the single card world record for 6600's with DDR1. Have a view: http://www.hwbot.org/result.do?resultId=565214

Though, I proved that the crystal has about zero impact when it comes down to highest overclock, people still are questioning my top spot score. They say that the framerate might be higher due to the timescale that changes when replacing the crystal. Wouldn't I have show that on my graph?

Feel free to share any knowledge about crystals/PLL, it's usefull and could maybe be used for further investigation.

yes that would have shown in your chart, + build in 3dmark anti cheat, think these people who accuse you of cheating a just jealous

I got the mobo voltmodded this morning. 5 jumpers allow me to set CPU voltage over 3V |D

2,6V volt got the P3 500 @ 667MHz at 45°C in bios, and still rising. Its big heatsink is aircooled though, what a power eating beast. (ambient is below 14°C)

To bad I didn't take my camera with me, but I promise they will be online anytime soon ;-)

Cool :) This reminds me of the "old" days. Overclocking wasn't that easy as it is now and overclocking was very limited with only jumpers. To squeeze more performance out of their PC's, some replaced the crystal of their motherboard.
Never done it though so I'm not sure if it also influenced "time". Maybe in some cases it did but I think in most cases it didn't because "time" was calculated by a second crystal (32768 Hz) and a real time clock.

Today I came across an article about BIOS chips, and with some google searches I came across the following things about the Real Time Clock (RTC)...

In the older days the RTC was build with an IC, nowadays it should be added in the mainboards' southbridge. As Laagvliekerke mentiond above, the Real Time Clock is based on a 32.768kHz oscillating source (crystals for example). Reading the Intel ICH8 datasheet I came across this:

5.11 Real Time Clock (D31:F0)
The Real Time Clock (RTC) module provides a battery backed-up date and time keeping
device with two banks of static RAM with 128 bytes each, although the first bank has
114 bytes for general purpose usage. Three interrupt features are available: time of
day alarm with once a second to once a month range, periodic rates of 122 μs to 500
ms, and end of update cycle notification. Seconds, minutes, hours, days, day of week,
month, and year are counted. Daylight savings compensation is available. The hour is
represented in twelve or twenty-four hour format, and data can be represented in BCD
or binary format. The design is functionally compatible with the Motorola MS146818B.
The time keeping comes from a 32.768 kHz oscillating source, which is divided to
achieve an update every second. The lower 14 bytes on the lower RAM block has very
specific functions. The first ten are for time and date information. The next four (0Ah to
0Dh) are registers, which configure and report RTC functions.
The time and calendar data should match the data mode (BCD or binary) and hour
mode (12 or 24 hour) as selected in register B. It is up to the programmer to make
sure that data stored in these locations is within the reasonable values ranges and
represents a possible date and time. The exception to these ranges is to store a value
of C0–FFh in the Alarm bytes to indicate a don’t care situation. All Alarm conditions
must match to trigger an Alarm Flag, which could trigger an Alarm Interrupt if enabled.
The SET bit must be 1 while programming these locations to avoid clashes with an
update cycle. Access to time and date information is done through the RAM locations. If
a RAM read from the ten time and date bytes is attempted during an update cycle, the
value read do not necessarily represent the true contents of those locations. Any RAM
writes under the same conditions are ignored.
Note: The leap year determination for adding a 29th day to February does not take into
account the
end-of-the-century exceptions. The logic simply assumes that all years divisible by 4
are leap years. According to the Royal Observatory Greenwich, years that are divisible
by 100 are typically not leap years. In every fourth century (years divisible by 400, like
2000), the 100-year-exception is over-ridden and a leap-year occurs. Note that the
year 2100 will be the first time in which the current RTC implementation would
incorrectly calculate the leap-year.
The ICH8 does not implement month/year alarms.


This basicly means that my actions have no influince to the real time clock, and so I didn't do anything wrong with replacing the VGA's crystal, it's just another way to overclock your system. Thanks Laagvliegerke for your input here ;-)