Memory questions

Do you have a question? Post it now! No Registration Necessary.  Now with pictures!

Threaded View

First, greetings to the group, it's my first post here :).

I own the following (outdated :)) rig:

Intel celeron 1.4 (tualatin core, 100 FSB default)
Soltek SL-65KV2-CT (VIA Apollo pro133T chipset)
3x256 PC133 SDRAM
Leadtek GeForce 6600 (running at AGP 4x)

The processor with all three sticks of RAM (1 stick with Xerox chips, 1
Hyundai, one no name) is running perfectly stable at 115 FSB. The mobo has
the option to set the memory clock at FSB+33 Mhz, and the option to change
the CAS latency. The Vcore cannot be changed (unfortunately). Declared
memory timings are 3-3-3-6 at 133 Mhz. However, all three sticks refuse to
work when FSB is set to 115 and the memory clock to 150 Mhz (I have
identified the Hyundai stick as the weakest link :)). All three work
together when running synchronously at 115 Mhz.

I'll now list the following stable setups:

- When I run at 115 FSB and 150 memory clock (with 512 megs of RAM,
3-3-3-6), average memory read speed is 855 mb/s, write speed is around 190
mb/s and latency around 120 ns.
- When I run at 115 FSB and 115 memory clock synchronously (with 768 megs of
RAM, 2-3-3-6), average memory read speed is 840 mb/s, write speed 145 mb/s,
latency 116 ns.

My question is what's better out of these two setups - faster memory (only
512 megs) or slower memory but 768 megs total? Also, are there any voltmods
for this board, or any other suggestions to try and overclock the processor

Thanks in advance,

Re: Memory questions

Quoted text here. Click to load it

So your core is running 1.4*1.15 = 1.61GHz
That is pretty good.

One of the issues with motherboards of that era, is the divider
problem. The PCI bus should be kept less than or equal to 37.5MHz,
and you can see in this sample table, that some CPU clock choices
lead to too high a PCI bus clock. Your disk drive may be corrupted,
if the PCI bus clock goes too high. When doing your overclocking
experiments, it pays to have a full backup of the boot
disk, so you can fix it if the disk is corrupted.

1   1   1   1   133 33.3   0   1   1   1  100 33.4
1   1   1   0   124 31     0   1   1   0  133 44.3
1   1   0   1   150 37.5   0   1   0   1  112 37.3
1   1   0   0   140 35     0   1   0   0  103 34.3
1   0   1   1   105 35     0   0   1   1   67 33.4
1   0   1   0   110 36.7   0   0   1   0   83 41.7
1   0   0   1   115 38.3   0   0   0   1   75 37.5
1   0   0   0   120 40     0   0   0   0  124 41.3

Your 115MHz choice, assuming a clockgen divider value of
3, gives 38.3, which is a bit higher than 37.5. The above
example clock generator, switches to the 4 divider, when
it hits 124MHz. For the above clockgen, say 112MHz and
124MHz would be safe choices, but that is a huge jump for
the processor, and the processor would be at 1.24*1.4=1.736GHz.
I don't think that is going to happen.

The AGP bus clock also has the divider issue, but older
video cards were able to accept the overclock. Later
cards, like say an ATI 9700 or 9800, would crap out at
75MHz (which, magically, is the same CPU clock choice
as that which gives you a PCI clock of 37.5). So that
can be another reason it may decide not to start, depending
on your clock choice.

To learn about your clockgen, you need to read the part
number off the top of the chip. On your board, the
clockgen looks to be near the DIMMs. A typical place
to look for a datasheet, for an ICS clockgen, would be . (There are other brands of
course.) The only pictures I can find of your board,
are too low res to make out any part numbers.

As for the memory, the question of more-slower memory or
less-faster memory, is the same today, as it was back then.
If all the work you do, fits into the 512MB space without
a problem, then the 512MB config is the way to go. If you
do stuff like Photoshop, then 768MB is the way to go, as
more memory helps Photoshop avoid swapping to disk. The
answer to your question, depends on your knowledge of your
usage pattern for the machine. If you keep a lot of hungry
applications open, then 768MB might be the better choice.
(The last running config I had on my Tualatin, was 512MB.)

Another thing you can check, is the temps of your Vcore.
There are seven capacitors in an L-shape near the CPU socket.
Next to them are a couple of MOSFETs, with three legs on
each. Feel them with a fingertip, while running Prime95
( torture test. If the MOSFETs run so hot,
as to burn you, you are pushing your board too far. I had
to turn my board down a bit, because things were getting
a bit toasty. It all depends on how long you want it to
last. If you were to volt mod, the temp situation only
gets worse (because increased volts lead to increased
amps). If all is running cool, you can do what you like.


Re: Memory questions

Quoted text here. Click to load it

That's correct.

Hm, I was aware of the PCI bus clock problem, but I thought the clock
divider changes every 33 mhz. Obviously I was mistaken. There were no
corruption issues running at 38.5 PCI bus clock. Anyway, I looked up the
datasheet of my clockgen
( ),
and appearantly it has a breakpoint at 124 Mhz as well. I'm going to try and
toy around to see if I can get the FSB up that high :).

Quoted text here. Click to load it

The temperatures aren't a problem yet. Max Vcore is 39 degrees (and it's
really hot here in Croatia at this time of year). During winter it never
passes 34 degrees celsius. The mosfets are only slightly warm to the touch,
so there is no overheating of the mobo. Do you have any suggestions as how
to increase the Vcore, since my mobo doesn't support it. I have the mobo's
manual, and the bios is capable of controlling the Vcore (it's the standard
Award's bios), however the mobo doesn't support it. I've googled for several
days and haven't found any mention of any voltmods for this board. I've also
found cases where this particular processor (I'm not sure about the revision
though, mine is B1) running at 14*124.

And thanks very much Paul, you have been very helpful and informative as it
is :).

Re: Memory questions

Quoted text here. Click to load it
I was hasty, the divider increases to 4 at 120 Mhz :)

Re: Memory questions

<< snip >>
Quoted text here. Click to load it

The bare minimum to make a working motherboard, looks like this:

   Processor             Vcore regulator
   (VID pins) ---------- VID pins

On an overclocking board, the connectivity looks like this:

   Processor             Vcore regulator
   (VID pins) -----+---- VID pins
              Overclock_controller_chip or
              GPIO pins on Super I/O, Southbridge, etc.

The GPIO pin method is able to override the signals from
the VID pins, via the clever use of series resistors. The
signals from the processor, have a series resistor, and the
GPIO can easily overrule the processor value. Thus, for the
BIOS to gain control of Vcore, means asserting the right
value on the GPIO (general purpose input output) pins.

The simplest, guaranteed to work method, is bend up the VID
pins on the Vcore regulator. Then assign logic 0 or logic 1
to each signal. The Vcore regulator datasheet will have a
table of VID value, versus voltage. Your motherboard
ends up looking like this:

   Processor              Vcore regulator
   (VID pins) ----X   X-- VID pins

A second method used, involves sourcing or sinking a DC
current, into a FB pin on the Vcore circuit. This is not
always going to guarantee a good solution, for a number of
reasons. Sometimes, the feedback pin also plays a part in
the dynamic response of the Vcore regulator. So when a
step load happens, the Vcore might overshoot more than it
used to. And sometimes, the Vcore chip has an
overvoltage / undervoltage detector internally, which cannot
be altered or modified, and too much tweaking causes the
chip to shut down (it is not damaged).

To properly engineer a feedback method, you should use a
potentiometer (adjustment knob) plus a fixed resistor. The
value of the fixed resistor is selected, so that the Vcore
will not go higher than the absolute max for the processor.

The feedback method typically is an easier mod to make, but
the design issues are very complicated. Engineers who design
such circuits, normally work with a bench prototype, of
just the Vcore circuit. That way, if they make a mistake,
the processor is not burned. They would use an oscilloscope,
check the ripple and step response of the regulator, after the
modification was made, and so on. When most hobbyists do such
a modification, they "pray to the gods of overclocking",
that nothing will be damaged.

So, lifting the VID pins at the Vcore circuit, is the simplest
method. It requires very little knowledge of electronics, but
the unsoldering can be difficult. I am experienced in soldering
such things, and I managed to damage one pad underneath my
Vcore chip. Fortunately, the damaged pad had no signal connected
to it, which is why it was damaged so easily.

I have only managed to convince one person in a USENET group,
to lift the VID pins. He did it, in order to undervolt his
processor, and he was successful. Usually, the pins are
very close together, and a magnifyihg glass is required to
check for excess solder, or shorting and the like.

I think my recommendation to you, would be to enjoy the
1.6GHz you have already got :-) Your motherboard will last
longer, if you avoid the temptation to voltmod.


Re: Memory questions


Quoted text here. Click to load it

Well, I tried raising the FSB to 121, got it to POST but not much else :). I
will leave at 115, especially since the rumour has it that raising the
voltage on tualatins doesn't help very much. Also I've decided to go with
768 megs of ram.

Do you have any idea as to why all three sticks work perfectly together at
100 and 115, but refuse to work at 133 (their default!) or more? Especially
since I tested each of them, and they can work at 150 in pairs (any chosen
pair out of the 3 sticks).

Re: Memory questions

Quoted text here. Click to load it

That is a normal thing. On some motherboards, the user manual
will say, you can use one DIMM at DDR400, two DIMMs at DDR333,
or three DIMMs at DDR266. It is a loading versus operating speed
thing, and both SDRAM and DDR share those same issues. The
effect is caused by the address bus - a DIMM with 16 chips,
places 16 electrical loads on the address bus, and the more
DIMMs you add, the more sluggish the driver gets. The clock
has to be slowed down, to give the address values time to
settle. Using DIMMs with fewer chips (like single sided with
a total of eight chips) reduces the loading, and can coax
more MHz from the memory bus - but not all chipsets will
support higher density memory chips. (My 440BX is limited
to 256MB DIMMs, with 16Mx8 chips on them. Sometimes the
VIA chipsets allow larger capacity DIMMs to be used, than
on an equivalent Intel chipset.)

Now, I would have expected the three sticks to run at PC133.
Two sticks might go to PC150. One stick to PC166. Four sticks
would be PC100 or perhaps a little bit more (mine takes four sticks).
These are rough rules of thumb, and some memories might behave
better than others. One of the aims, of standards bodies like
JEDEC, is to try to make all modules behave the same, so that
there is maximum compatibility when you mix different brands
of RAM. But there will still be small differences between the
products and how they work - you've just been a little unlucky.

Do you know if the memory voltage is good or not ? I suppose
each stick draws a bit more power from the Vdimm supply, and
maybe Vdimm is dropping a bit when the third stick is added ?


Re: Memory questions

Quoted text here. Click to load it

Well, according to Via hardware monitor and speedfan, the 3.3 V value is
good. I'd measure the DIMM voltage, but I have no voltmeter :).

Anyways, I'm going to leave it as it is. Thanks for all your help Paul, I've
learned some new things :)

Site Timeline