by FastSite
05/26/2002 | 12:00 AM
Time passes, new graphics chip generations come to replace the older ones, however, the human desire to get more for the same money is as strong as always. When the first graphics cards on NVIDIA GeForce4 Titanium chips appeared and the first questions about their performance, price and quality were answered, a vital interest towards their overclocking potential and resulting performance gain started growing rapidly.<%BANNER[article]%>
So, the question was set, everybody waited for an answer. When I first saw NVIDIA GeForce4 Ti4400 based graphics card from VisionTek, my hesitations about the best choice for extreme overclocking experiments vanished in a wink: VisionTek was the best thing I could find :)
In this article I will share with you my experience in overclocking VisionTek Xtasy GeForce4 Ti4400 graphics card by modifying the chip and memory voltages. We carried out a number of tests modeling more or less successfully the upcoming games and showing how the graphics card performance depends on the working frequency of the chip and memory under heavy workloads.
Xtasy GeForce4 Ti4400 graphics cards are shipped in very stylish boxes with the design passing very well to the NVIDIA graphics chips logos:
The package includes the graphics card, of course, the user's manual, two CD-disks with the drivers, utilities and a software DVD player from Cyberlink:
The reverse side of the box shows the description of the graphics card features:
Frankly speaking, the features of this graphics card are pretty standard, however, they are described in such an unusual way that I simply couldn't help quoting this text here for you. Just have a look:
Wow, this is something really cool for real gamers! :) I have never come across such a juicy-fruity description of the graphics card features flavoured with strong slang. Really, after reading it I felt like putting on the helmet, fastening the seatbelt and starting my favourite simulator. Congratulations, VisionTek!
To a certain disappointment, the graphics card itself didn't look that impressive. VisionTek Xtasy GeForce4 Ti4400 is based on the NVIDIA reference design and can hardly be distinguished from NVIDIA reference card samples. The only thing that indicates the difference is a small sticker with VisionTek logo on the cooler:
The graphics card is equipped with NVIDIA GeForce4 Ti4400 graphics chip and 128MB DDR SDRAM from Samsung with 3.6ns access time:
The nominal working frequencies of the card comply with NVIDIA's recommendations making 275MHz for the chip and 550MHz (275MHz DDR) for the memory.
The TV-signal is formed by a special Conexant CX25871-14 chip:
Well, the appearance of the graphics card turned out a little bit boring, although we expected something outstanding after reading the box. However, it is all within our power to improve the situation and to eliminate this discrepancy. Let's pass over to extreme overclocking.
Before we start, I would like to mention how greatly we managed to overclock the card without any tortures. During regular overclocking we managed to get VisionTek Xtasy GeForce4 Ti4400 to work at 300MHz core and 660MHz memory frequencies, which is quite an average result, to tell the truth.
However, this graphics card will not remain just an average solution: we are going to use our soldering iron to make it fly :)
There is a special SC1175CSW chip on the card responsible for the graphics chip voltage. If you are looking for more detailed instructions for voltage modifications please see our article called "NVIDIA GeForce3 Voltage Tweaking and Extreme Overclocking". Here I would like to offer you only a typical application circuit scheme for this chip:
To increase the graphics chip voltage, I shunted R1 resistor (see the scheme) with an additional one soldered to pins 18 and 20:
The R1 and 12 resistors on the typical application circuit scheme above correspond to 130Ohm and 390Ohm resistors on the graphics card, so that the default chip voltage equals to 1.66V. As soon as we shunted R1 with an additional 620Ohm resistor, the graphics chip voltage grew up to 1.92V.
The graphics card is equipped with Samsung memory chips with 2.8V internal circuits and input/output buffers voltage (VDD/VDDQ). This voltage is formed by SC1102CS chip from Semtech. This chip works absolutely the same way as SC1175CSW chip mentioned above, with that only difference that it has one channel instead of two independent channels of SC1175CSW. And one more important difference: the reference voltage of SC1102CS equals to 1.265V instead of 1.25V as by SC1175CSW. So, the Vout of the voltage regulator can be calculated with the following formula: Vout=1.265x(1+R8/R7), where the resistors R7 and R8 are taken from the following typical application circuit scheme:
As we have already mentioned in our article on "NVIDIA GeForce3 Ti500 and Ti200 Based Graphics Cards Extreme Overclocking", NVIDIA GeForce3 Ti200/Ti500 based graphics cards use chips with two independent channels for internal circuits and input/output buffers (3.3V/2.5V - VDD/VDDQ). By NVIDIA GeForce4 based solutions, unlike their predecessors, the VDD and VDDQ electric circuits are connected with one another. Moreover, the input/output buffers of the GeForce4 chip work with the same voltage as input/output buffers of the graphics memory chips. This is a serious obstacle on the way towards voltage increase. For example, when the VDD/VDDQ voltage grows up to 3V, you can notice some ripples in text modes, which is out of the question, of course. So, we could have failed the extreme overclocking of the graphics memory, if…
If we hadn't been so lucky to discover that the graphics card reference design allows installing an additional voltage regulator for the input/output buffers of the graphics memory chips if necessary, that is when the internal circuits and input/output buffers have different voltages. In order to install the additional voltage regulator, you need to remove the VDD and VDDQ circuits first by unsoldering the bridges on the front side of the PCB (see the picture below):
And on the reverse side of the PCB:
Then you should solder the voltage regulator for the graphics memory chips input/output buffers (the spots for components are perfectly shown on the photos). Of course, it is not difficult at all for an experienced person, however, it would save you even more effort if you use a simple diode. Let me explain. As is known, in case of direct connection the voltage of silicon diodes usually drops down by 0.7V-0.8V. Therefore, you can painlessly increase the voltage of the graphics memory chips internal circuits (VDD) up to 3.5V, for instance and transfer the power to input/output buffers from VDD via the diode.
This is what I have done. First of all, I increased the Vout on the voltage regulator up to 3.51V by shunting the SC1102CS chip with a 270Ohm resistor (it was soldered to pins 11 and 14):
This way, the voltage of the internal circuits of the graphics memory chips was increased from 2.82V to 3.51V.
Secondly, I searched through my collection of PSUs and found a powerful silicon diode. I immediately soldered this lucky find instead of the bridges:
Now the voltage of the input/output buffers of the graphics memory chips equals 2.77V, as the voltage on the diode dropped by 0.74V.
Now that we have increased the voltages, we have to make sure that our graphics card is supplied with appropriate cooling. As you know, overclocking makes the lifetime of our hardware shorter, and voltage modifications nearly exhaust our poor hardware. Certainly, efficient cooling can not only allow getting the card to work at higher frequencies but also ensure longer reliable functioning.
In the first place, I worked on the graphics chip, remembering pretty well that the surface of the metallic chip lid is not that flat and needs good rubbing with polishing paper in order to provide appropriate contact between the chip surface and the heatsink. Here is the result of my hard work:
After that I borrowed a cooler from ATI RADEON 8500 graphics card, which has already undergone extreme experiments of ours (see our article on ATI RADEON 8500 Extreme Overclocking Experience):
The cooler was fastened to the heatsink ribs with a couple of nylon buckles having stretched them through the holes in the PCB.
The last thing was to cover all memory chips with heatsinks. I used heatsinks from Thermaltake Memory Cooling Kit having fastened them to sticky thermal pads also taken from the kit.
Now that the cooling system has been enhanced, VisionTek Xtasy GeForce4 Ti4400 looks as follows:
We managed to overclock the card and make it work stably at the maximum of 340MHz chip and 710MHz (355MHz DDR) memory frequencies. So far, it is the highest frequencies ever achieved on any NVIDIA GeForce4 Ti4400 based graphics card. The frequencies have even exceeded those of GeForce4 Ti4600.
For our tests we used the following testbed:
Software:
We ran these applications in the following modes:
We tested the card in Extreme Quality mode: "Quality" graphics quality settings, 32bit color modes, we ran "GFX: Extreme Quality" add-on before each test.
For this test we launched a standard "The Grand Cathedral" demo.
32bit frame buffer, 32bit textures and 32 (24) bit Z-buffer, D3D Pure Hardware T&L mode.
We ran the tests with default settings.
I believe that it is worth telling you a bit about Codecult Codecreatures test based on an engine developed by Codecult team. The guys are doing their best to promote the engine and offer it for licensing. The benchmark imitates nature: meadows, mountains, a lake, grass, it all very popular now :)
Of course, the first thing that comes to my mind is the Nature benchmark from 3DMark2001 testing set. IMHO, Codecreatures test doesn't yield to MadOnion's offspring from the technical point of view, however, the visual impression is much weaker: no sun in the sky, dim colours, grass bending unnaturally in the wind and too calm water surface… But each of you has a unique perception of the natural beauty, so I will not impose my personal opinion here. :)
If you haven't yet had the chance to see this benchmark, here are some screenshots:
Well, I've got the feeling that overclocking makes some sense here. Codecreatures loads the graphics card really heavily, which you can notice even without the exact fps rates. However, I don't think you mind getting some numbers to back up this statement:
The scene consists of a great number of polygons (over 500,000) and sets very high fillrate requirements. Here graphics chip overclocking appeared much more efficient than graphics memory overclocking.
The remarkable thing about another benchmark imitating nature is that it appeared more sensitive to graphics memory overclocking rather than to core overclocking. The explanation of this phenomenon is evident: the Nature scenes have fewer polygons, and the polygons with transparent textures imitating grass "are planted" not so close to one another.
In Serious Sam: The Second Encounter graphics memory overclocking doesn't have any visible effect, which is actually not at all surprising, as the performance is limited solely by the graphics core. It happens mostly because of the Extreme Quality mode enabling high-level anisotropic filtering (by 32 samples). This filtering technology requires additional clocks from the graphics chip: 8 clocks instead of 1 (for more details see our Anisotropic Filtering Investigation for more details).
To give a more exact evaluation of the graphics chip overclocking efficiency and its influence on the performance, I combined the performance gains obtained in all the three benchmarks on a single diagram (the increase of the graphics chip frequency in percents is also shown on the graph):
Codecreatures and Serious Sam: The Second Encounter prove really "hungry" for graphics chip resources, while the Nature benchmark hardly depends on the graphics chip frequency increase.
Now let's have a look at the same graph built for the overclocked memory:
Here we see the opposite situation: the results in Nature and Codecreatures have become considerably higher as the memory frequency grew up. However, Serious Sam: The Second Encounter seems to remain indifferent to the changes in the graphics memory working frequency.
The next graph shows the performance gain during extreme overclocking of the graphics chip and memory:
We see a really impressive performance increase, and now we have every right to state that Codecult Codecreatures benchmark loads VisionTek Xtasy GeForce4 Ti4400 graphics card perfectly well, using the whole of its potential even in overclocked mode. The performance gain even in 1600x1200 made 27% with the graphics chip working at 23.6% higher frequency and graphics memory - at 29.1% higher frequency.
As we saw the performance gain in Nature and Serious Sam: The Second Encounter benchmarks appeared a little bit lower, which implies that it is no longer the graphics card itself that limits the performance, but also the CPU and even the entire system.
Well, the experiment was a success. VisionTek Xtasy GeForce4 Ti4400 graphics card stood the extreme overclocking test and showed pleasingly high performance improvement. We have to admit that the percentage of the frequencies increase is not so impressively high as it used to be in case of NVIDIA GeForce3. However, you shouldn't forget that both: NVIDIA GeForce3 and GeForce4 are manufactured with the same 0.15micron technology. Therefore, no wonder that GeForce4 based graphics cards working at higher frequencies from the very beginning feature lower overclocking potential. It looks as if we were about to see new "well-overclockable" chips only when NVIDIA starts using finer manufacturing technology for their production. And in the meanwhile we have to be happy with a 20%-25% frequencies growth.
Although extreme overclocking of GeForce4 Ti4400 is a very troublesome task, even a 20%-25% performance increase appears very pleasing, especially taking into account that the performance in Codecreatures and Nature tests at nominal frequencies is very low. Overclocking allows improving the situation. On the other hand, you can always use the extra performance even in weak systems or in "light" games. You can enable FSAA, anisotropic filtering, set higher resolution, etc.
Returning to the tested VisionTek Xtasy GeForce4 Ti4400 graphics card itself, I can conclude the following:
Highs:
Lows:
Reminder: