All
modifications described henceforth will void yourwarranty
and if not done with proper care can
result in rendering the product non functional.
Neither
the author of this article or
TechSpot shall be held responsible for any injury incurred
or hardware damage when performing
these
modifications. So proceed at your own risk...
Some Background
information
Although
with significant differences, one could say that GeForce4
Ti4200 cards are based on
very similar hardware than most Ti4600 cards (our
overclocking target). The first difference you will notice
when looking at both cards is that Ti 4200’s are much
smaller; being a slower product, most manufacturers save on
production costs by reducing the power circuit
significantly. This card actually has a
power circuit more comparable to
that of the GeForce3 than the GeForce4 Ti4600. The memory
used is also slower to reduce costs; this is actually what
makes the “high-end” cards so expensive since memory
accounts for much of their total cost.
Exact
specs among Ti4200’s can vary, you will probably find
memory in the 3.5-4ns range from most of the Taiwanese
manufacturers, and if you paid a little extra for your card
perhaps 3,3ns memory. You divide 1000 by the nanosecond
rating of your memory, in the case of 4ns memory this is the
same as 250 MHz, or, if you prefer, 500 MHz DDR which
happens to be the clock speed NVIDIA has officially set for
the Ti4200.
Just for your reference the Ti4400 usually uses 3,6ns memory
and the Ti4600 2,8ns memory.
If
we are to compare the similarities between the cards we can
take a look at the GPU, these are just the same on all Ti
boards, the only thing that will differ is in the text
printed on top of them, but it’s really the same chip. As
you might be aware the Ti4200 is clocked at 250 MHz for the
core while the Ti4600 is clocked at 300 MHz, as you might
guess this leaves quite a lot
of
headroom for overclocking.
The
card I will be using for this modification is Gainward’s
GeForce4 Ti 4200 64MB variant (not their “gold model”.)
I paid $150 for this card here in Sweden.
Overclocking
before Volt-modding
Using
standard cooling I was able to get this card right out of
the box up to a stunning 320mhz for the core and 580mhz for
the memory that’s a 28% increase in GPU speed and 16% in
memory speed, quite impressive.
In
an attempt for achieving a higher speed I swapped the
standard cooler for a watercooler generously provided by CPUFX
INC. Though, to my disappointment, I was not able
to overclock a single MHz more, the reason for that can be
seen below.
As
you can see the GPU is not flat and it did not make contact
with the cooler, a quick measurement above the GPU showed an
egg cooking 80°C temperature! To remedy this I took some
600-grit sandpaper and sanded the core down until I got it
flat, the result shown below
After
this the core came down to a more humane temperature of 65°C
using standard air-cooling when tested at 320mhz just like
before I lapped the core. This gave me an 8,5mhz increase in
overclocking resulting in 328,5mhz.
Now
that the GPU had been pushed to the limit without
volt-modding I turned to the memory. My card came with 64mb
of 3,6ns EtronTech memory without heatsinks, the maximum
speed I managed to get out of them artefact-free was 580mhz,
the “Point Sprites” test in 3D Mark 2001 was used to
reveal when overclocking
was causing visual glitches
(see pic below).
