by FastSite
01/12/2002 | 12:00 AM
Things in the graphics world never remain unchanged. So, NVIDIA launched two new chips to follow GeForce3. These are GeForce3 Ti200 and GeForce3 Ti500. In fact, both, GeForce3 Ti500 and GeForce3 Ti200, are the same GeForce3 but working at different frequencies, while all the other "enhanced functions" of these chips is merely a marketing trick of NVIDIA, intended to create a novelty image.<%BANNER[article]%>
The arrival of GeForce3 Ti200 is a result of purely pragmatic approach implying that the standard GeForce3 could be made cheaper and consequently more popular. At the same time, it would be more successful in its competition with the new mainstream chip from ATi, RADEON 7500.
Taking advantage of the great potential hidden in GeForce3 architecture, NVIDIA launched a faster version of this chip, GeForce3 Ti500, which has every right to be called " GeForce3 Ultra". The mission of GeForce3 Ti500 chip is to take over the leadership of GeForce3 before the mystical NV25 comes out and to oppose RADEON 8500 from ATi.
NV25 is to come into being pretty soon, and the new, even more overclocked versions of GeForce3 are highly unlikely to appear. But is the potential of GeForce3 architecture totally exhausted with the launch of GeForce3 Ti500? Or is there anything else to be squeezed out of GeForce3?
We modified GeForce3 Ti500/Ti200 based graphics cards designed on the most popular PCBs and carried out some extreme overclocking experiments.
Now we will check what an overclocker can do with these cards after GeForce3 was "officially" overclocked to GeForce3 Ti500 by NVIDIA and if there are any new horizons for the GeForce3 Ti200 "outcast".
On announcing the arrival of the new chips, NVIDIA introduced a new reference design of GeForce3 Ti500/Ti200 based graphics cards. However, since the only difference between GeForce3 and these chips is the working frequencies, most graphics card makers decided not to migrate to the new reference design in order to save the costs. They simply installed the new chips on standard GeForce3 cards, which they produced before. Surely, they have also changed the labels and inscriptions on all the boxes…
In the meanwhile, there appeared a number of cards sticking honestly to the new reference design from NVIDIA. As a result, today most GeForce3 Ti500/Ti200 based graphics cards are designed in one of the three basic variations: GeForce3, GeForce3 Ti500 or GeForce3 Ti200 reference design.
In spite of the fact that all three variants feature differing layouts, core and graphics memory power regulation circuits, which are of the greatest interest to us, are nearly the same everywhere. All the three design variants involve the same microchips enabled similarly, so the entire difference restricts to the location of electronic components on the PCB.
We have already discussed in great detail the typical application circuit schemes for pulse-duration SC1175CSW controllers used on these cards in our article called "NVIDIA GeForce3 Voltage Tweaking and Extreme Overclocking", so further on we'll just remind you of what was written there.
For Vcore regulation the chip is used in a "current sharing" mode with two of its independent channels working for the same load (that of the graphics core). According to the official specs from SEMTECH web-site, a typical application circuit scheme for this mode looks as follows:
The Vout in this case is calculated with the following formula: Vout = 1.25x(1 + R12/R1) on the typical circuit scheme. There you can also see how the Vout can be increased by shunting R1 resistor.
There is one more SC1175CSW chip used to implement the memory power supply. In this case both of its channels work independently providing the memory chips with two voltages: 3.3V for all the internal circuits and 2.5V for the input-output buffers:
The Vouts for both controllers are determined by the proportion between the resistor ohmages: R15, R14 (on the typical scheme) for the controller providing 3.3V and R13, R11 for the one providing 2.5V. The Vout values for both channels are calculated with the following formulas: Vout1 = 1.25x(1 + R15/R14) and Vout2 = 1.25x(1 + R13/R11) respectively. Shunting R11 and R14 may increase Vmem.
The power supply regulators on GeForce3 Ti500/Ti200 reference cards are located on the rear side of the PCB:
The Vmem regulator is marked with "1", the Vcore regulator - with "2".
Now let us take a look at all the three layout variants separately:
As the regulator circuit schemes stay the same in all three cases, it is only the location of electronic components that allows us to distinguish between the three layouts. In all three cases the modification will be the same:
On GeForce3 / Ti500 / Ti200 reference cards the resistance ratio of the Vmem resistors makes normally 110Ohm/100Ohm for one channel, giving 2.6V in the end, and 170Ohm/110Ohm for the other channel, resulting in 3.4V. When each of the divisors is shunted with accessory 1kOhm resistors, the resistance ratio becomes 110Ohm/91Ohm and 170Ohm/90Om. As a result, the memory Vout values increase to 2.8V and 3.6V correspondingly.
For the sake of convenience, we soldered the shunting resistors directly to the pins 3, 10, 18 and 20 as it is shown on the photo:
The Vcore values of GeForce3 Ti500 and Ti200 are different.
The resistance ratio of resistors determining the GeForce3 Ti200 Vcore is 130Ohm/770Ohm with the Vcore equal to 1.46V.
On GeForce3 Ti500 based cards the resistance ratio makes 130Ohm/580Ohm with the Vout equal to 1.53V.
Regardless of the different resistances in the divisors, the Vcore can be increased in the same way for all the cases described, namely, by shunting an extra resistor, which we soldered to the pins 18 and 20 of the controller, as it is shown on the photo.
With an extra 1kOhm resistor the new divisor resistances on GeForce3 Ti500/Ti200 based cards turn 130Ohm/435Ohm and 130Ohm/367Ohm correspondingly, whereas the Vcore increases up to 1.62V on GeForce3 Ti200 and up to 1.69V on GeForce3 Ti500 based cards.
A little bit later you'll see the cards overclockability gets influenced by the modification like that, and now we'll say a few words about products from ASUS, which stand out, as usual.
Like V8200 cards built on GeForce3, ASUS V8200 T2 / T5 cards based on NVIDIA GeForce3 Ti200 / Ti500 graphics chips respectively boast a layout of their own, which is different from what NVIDIA reference design recommends.
Let's have a look at both solutions:
ASUS V8200 T5 is equipped with 2 regulators. One of them, SC1175CSW, is just the same as the one used on the reference cards (marked as "A1" on the photo). The other one is a regular AMS1505 single-channel regulator from Advanced Monolitic Systems (marked as "A2" on the photo).
One of the SC1175CSW channels is responsible for the Vcore, the other ensures 3.3V for the internal circuits of the graphics memory. AMS1505 chip provides 2.5V for the graphics memory input-output buffers. The typical application circuit scheme for this voltage regulator taken from the manufacturer's specifications is provided below:
The Vout of this regulator is determined by this formula: Vout = 1.25x(1 + R2/R1) + Iadj x R2. If we shunt R1 with an additional resistor, the Vout will grow even higher.
ASUS V8200 T2 (GeForce3 Ti200) based graphics card has no 2.5V Vmem regulator, so all the memory circuits are powered at 3.3V, in spite of the manufacturer's specs reading clearly that the memory chips should be power at two voltages: 3.3V +/-0.3V and 2.5V +/- 0.2V.
We could expect to come across a total neglect of the specs like that for the sake of one-pence cost saving from some not very well known manufacturer, but not from ASUS, which positions its products as the most reliable, stable, safe, etc. Moreover, if we take into account the traditionally high prices on ASUS cards, this economizing trick will look just disreputable.
The divisor resistances determining the Vcore of ASUS V8200 T2 and ASUS V8200 T5 are 140Ohm/780Ohm and 140Ohm/590Ohm correspondingly. The Vcore values make 1.47V for ASUS V8200 T2 and 1.55V for ASUS V8200 T5.
The divisor resistances determining the voltage of the memory chips internal circuits are equal for ASUS V8200 T2 and ASUS V8200 T5 (220Ohm/120Ohm) and result into 3.54V Vout.
When we applied 1kOhm shunting resistors, the Vcore of ASUS V8200 T2 and ASUS V8200 T5 cards turned 1.67V and 1.72V respectively, and the Vmem of "3.3V memory" got as high as 3.81V.
As we have mentioned above, there is no 2.5V voltage regulator for memory input-output buffers on ASUS V8200 T2 cards, that is why these memory chips are powered at the same voltage as the inner circuits, i.e., at 3.81V. Indeed, this looks far too much for circuits designed for 2.5V. Though the modified ASUS V8200 T2 card successfully passed all the tests, this doesn't mean that no problems will arise if the card keeps on working in such an overclocked mode for some longer period of time.
ASUS V8200 T5 doesn't face this problem, so the use of a 1kOhm shunting resistor led to 2.76V Vout of AMS1505 regulator.
Now it is high time we assessed what these modification tricks can give an overclocker in real applications.
For this test session we took Leadtek WinFast cards based on NVIDIA GeForce3 Ti500 / Ti200 graphics chips. We selected these cards for their 3.8ns video memory chips and the smart cooling system:
In our Leadtek Titanium Graphics Cards Review we have already checked the overclockability of these graphics cards. The top frequencies we manage to achieve last time were 260MHz/590MHz for Leadtek WinFast Titanium 500 TD and 230MHz/580MHz for Leadtek WinFast Titanium 200 TDH.
This time the modified cards reached the heights of 270/610MHz (Leadtek WinFast Titanium 500 TD) and 240/590MHz (Leadtek WinFast Titanium 200 TDH).
As you can see, the overclockability grew just a little bit after the modifications had been completed. There could be the whole bunch of reasons for that. First of all, we didn't change the cards cooling as we decided to do with the standard cooling solution provided by Leadtek. Secondly, every chip has technological frequency limits. When the clock frequency is getting close to a limit like that, overclocking becomes a really hard task to fulfill. Subsequently, graphics cards lose their stability at high frequencies no matter how perfect the cooling is and how high the voltage is. On the other hand, the increase in working frequencies may have a negative effect on the graphics card stability, because they're designed to work at nominal frequencies only.
The final result was no surprise for us: the maximal frequencies at which the modified NVIDIA GeForce3 Ti500 based cards worked after extreme overclocking turned out not decisively higher than the top result of a standard NVIDIA GeForce3 based card in extremely over-nominal conditions.
The overclocked core of Leadtek WinFast Titanium 200 TDH worked at 240MHz, which is the nominal frequency of GeForce3 Ti500 chip. This proves that GeForce3 Ti500 and Ti200 chips differ a lot in overclockability. Perhaps, Ti200 is just a cull from Ti500 chips.
In spite of the same 3.8ns access time, the video memory of Leadtek WinFast Titanium 200 TDH worked slower than that of Leadtek WinFast Titanium 500 TD. There could be two major reasons for that: firstly, the graphics memory chips used on these two cards are made by different manufacturers, and secondly, the cards do actually differ in design.
To test these graphics cards, we used the following hardware:
Software:
We tested the performance of the overclocked cards in 3DMark 2001 alone, as we already carried out a detailed performance analysis of NVIDIA GeForce3 based cards in Quake3 Arena and 3DMark 2001 in our review devoted to NVIDIA GeForce3 Voltage Tweaking and Extreme Overclocking.
Bearing in mind what class these graphics cards belonged to, we tested them only in "High Detail" mode of the gaming tests and only in 32bit color mode:
As you can deduce from the diagrams, extreme overclocking gives no appreciable boost to GeForce3 Ti500, which is not so surprising, actually, as the nominal core and memory frequencies of GeForce3 Ti500 based graphics solutions appear quite close to their top limit, so even extreme overclocking cannot bring about any notable frequency increase and subsequently performance gain. Of course, we don't mean low resolutions here, because in low resolutions the result is determined by the CPU and the overall system performance, but not by the graphics card.
GeForce3 Ti200 card overclocked to the maximum breaks a trifle ahead of GeForce3 Ti500 working in nominal mode, but it's only thanks to the relatively high graphics memory frequency. However, if we compare the obtained results with the nominal mode, the performance boost will be really impressive, sometimes reaching up to 50% at higher resolutions.
We can now say for sure that it hardly makes any sense to modify GeForce3 Ti500 based cards for extreme overclocking purposes: the performance growth doesn't make up for all the trouble you will have to take. Of course, you could increase the Vcore and Vmem even higher, upgrade the cooling solution with something as powerful as Dragon Orb 3 or Volcano 6 for example, or even resort to Peltier elements, but it would be really extreme overclocking then. Moreover, the pains needed to change a graphics card into such a monster are unlikely to make you completely satisfied with its performance.
Thing stand a bit differently with NVIDIA GeForce3 Ti200 based cards. On the one hand, the top working frequencies of NVIDIA GeForce3 Ti200 chips make 220MHz-230MHz during traditional overclocking and 240MHz-250MHz in case of extreme overclocking. On the other hand, extreme overclocking of GeForce3 Ti200 provides a greater increase in the core frequency compared with the nominal level (175MHz), this way providing a greater performance gain in comparison with GeForce3 Ti500.
As for overclocking the graphics memory of GeForce3 Ti200 based cards, it depends on the quality of the chips used. Accordingly, the lower the access time is (compared with the standard 5ns), the greater will the frequency overclock. So, if you buy a graphics card, we advise you to pay attention to the type of graphics memory used, even if you plan no overclocking experiments.
Reminder: