by FastSite
11/20/1999 | 12:00 AM
We continue discussing one of the most acute topics, which is devoted to the peculiarities of graphics cards basedon today's most powerful 3D-chipset NVIDIA GeForce 256. In the current article we won't discuss the nominal characteristicsof this chipset and the testing results, so, if you want you may take a look at our Creative 3D Blaster Annihilator Review,where we particularly dwelled on these things. Besides, soon we are going to offer you more reviews on the graphics cards ofthis family that's why the information on the matter will hardly be insufficient. This time we will focus on the performanceof NVIDIA GeForce 256 based graphics cards when working at higher frequencies, i.e. when overclocked.<%BANNER[article]%>
Most owners of the latest generation graphics cards as well as those who have the last but one generation graphics cardsare very well aware that if the card works at higher chipset and memory frequencies the performance in 3D-graphics maysignificantly increase. Especially, since the chipsets from the "spring draft" have a pretty high safety margin, so thateven some graphics cards manufacturers felt like taking advantage of this attractive marketing move: to rise the chipsetand memory working frequency and to announce their products as the fastest of all their competitors. For example, thenotorious Hercules Company famous for its bankruptcy used to manufacture NVIDIA Riva TNT2 Ultra based graphics cards withthe chipset and memory default frequencies 18% higher than those by their analogues. Moreover, Gigabyte produces GA-660 andGA-660+ based on NVIDIA Riva TNT2 and NVIDIA Riva TNT2-A chipsets also working at the frequencies exceeding the nominal ones.Of course, this helps to present the graphics card from a more attractive point of view.
However, speaking about companies we could count such situations on the fingers of one hand, while as it comes to individualusers, numerous attempts of 3dfx Voodoo3 or NVIDIA Riva TNT2/TNT2 Ultra owners to overclock their chipsets and memory became aneveryday routine. Only a very lazy person won't try to squeeze the maximum out of his graphics card.
Thank goodness that we are living in the epoch of progressive fans used for the chipset cooling that is why the experimentswith higher frequencies are no longer as dangerous as they used to be some time ago (although a certain risk of burning thecard still remains). Of course, we do understand that if we want to succeed in overclocking we should also take into accountthe environment temperature. Therefore if you overclock your graphics card in summer, it appears more complicated to achievestability at higher working frequencies, than in winter.
However, let's return to our today's hero - Creative 3D Blaster GeForce graphics card based on NVIDIA GeForce 256 (just incase you didn't know that ;-)).

It has an AGP 2x/4x constructive, 32MB SDRAM 5ns memory:

The chipset is equipped with an active cooler. The graphics card we had didn't have a TV-out although there were allthe necessary chips.
And now a few words about the testing system configuration:
We would like to particularly mention that we used Inwin Solist Q500 PC case equipped with a demountable section for themainboard with all the expansion cards.
The majority of our readers probably know that the graphics cards based on NVIDIA GeForce 256 work at 120/166MHz asdefault frequencies (the first one denotes the chipset frequency and the second one - the memory frequency). Comparethese values to the parameters of NVIDIA Riva TNT2 Ultra: 150/183MHz. Surprised? However, the restrictions like thatare far not occasional: GeForce 256 chipset heats too much that is why if there is no additional cooling (except thealready provided chipset cooler), the microchips and even the whole card get so hot that the temperature almost reachesthe human threshold of pain. So, it is more than evident: this graphics card will be a hard nut to overclock.
But is the game worth the candle? Well, we think it is. You shouldn't forget that NVIDIA GeForce 256 can boast not 2(as we saw by NVIDIA Riva TNT2) but 4 pipelines, which means that if the frequency gets only 1MHz higher the effect willbe twice as big as that by the previous generation graphics cards. And bearing in mind GeForce's bottleneck - narrow memorybandwidth, which very negatively tells on the performance in 32-bit color regime, memory overclocking may have a reallyimpressive effect on the performance in True Color.
But let's take a look at real practice. We carried out two experiments in different conditions. They can be calledas follows:
Now let us specify our idea. We resort to normal overclocking every time we test a graphics card at higher frequencies.The system block is situated in a room (+23oC) and two rotating fans cool the graphics card.

Having tried to overclock GeForce 256 in these conditions we managed to achieve stable work at 140/200MHz. The mainboardgot +28oC warm, and the graphics card chipset heatsink heated up to +42oC (we took this temperature with the help of anexternal thermal sensor).
However, this time we decided to continue our investigation and to go a bit further. So, we took the system block withall the cards to ... the freezer of our fridge:

In these conditions we managed to see our graphics card working stably at 150/210MHz. The mainboard temperature was -1oC,and the graphics chipset heatsink heated up to +12oC. Of course, it was just our pure curiosity that made us arrange thistesting experiment. Besides, we had to be pretty quick and didn't have much time at our disposal because we were afraid thatthe appearing condensate could damage some devices.
And now we suggest that you should take a closer look at the results of our extraordinary testing. The maintools we utilized were:
First we will take a look at the results obtained in Quake3:



You can see that in almost all modes the performance gain percent is the same (with the exception of Fastest modebecause of the restrictions imposed by the processor frequency). Let's try to find out if this overclocking was justifiedwith the help of High Quality mode (32-bit color).

When overclocking in normal conditions at the resolution set to 1024x768 (chipset frequency increase made 16%,memory - 20%) we got a 17% performance gain, which completely made up for our "violent" handling the graphics card.
As for extreme overclocking (chipset frequency increase made 25%, memory - 27%), the performance gain equaled 28%.Looks quite impressive, what do you think? It is a perfect example showing the great effect achieved due to the graphicscard working frequencies increase, which so positively told on the performance. We would also like to mention that at150MHz the peak fillrate reached 600Mpixels per second (compare to 480Mpixels per second we have at nominal frequencies).And it is none other than a 25% performance gain!
And now Expendable:


This game shows that at lower resolutions the performance gain is limited by the CPU power, however, as it comes tohigher resolutions the situation gets more optimistic.
Let's sum up the obtained results and draw certain conclusions. First we will take 16-bit color. At 1024x768 in normalconditions (chipset frequency increase made 16%, memory - 20%) we managed to improve the performance by 5%, but at 1280x1024this value reached 10%. In fact, the performance gain was absolutely inadequate to the frequencies increase.
In extreme conditions (chipset frequency increase made 25%, memory - 27%) the card performed 18% faster at 1280x1024,which was also less than we had expected. Why did it happen like that? We think that it is again the system CPU that slowsdown the whole thing (note: there were no games supporting T&L among those considered).
And what about 32-bit color? Theoretically, the CPU should be powerful enough here and the only thing that may setcertain limits is the graphics card and its potential. In normal conditions (chipset frequency increase made 16%,memory - 20%) at 1280x1024 we got a 24% performance gain (at 1024x768 - 12%). Well, now we start to like the situation:it really looks as if such overclocking was worth it.
In extreme conditions (chipset frequency increase made 25%, memory - 27%) at 1280x1024 the performance grew by 38%(at 1024x768 - 18%). To tell the truth, the result was beyond our expectations, however, the gaming performance of thecard at 1280x1024 still left much to be desired.
What conclusions can we make here? First of all, we should say that better cooling (especially if it is connectedwith some additional expenses) isn't sometimes that efficient. Why so? The main reason is the processor frequency,which very often restricts the gaming performance and brings to naught all the overclockers' efforts. Besides, thereare no games now supporting T&L and reducing the CPU utilization.
But practice shows that overclocking methods and means available to general public (namely overclocking with thehelp of additional coolers) are very often quite fruitful.
In conclusion we would like to stress that the ability of the graphics card to work stably at higher frequenciesgreatly depends on the individual peculiarities of this or that card. That's why we would like you not to regard thismaterial as a proof that all Creative 3D Blaster GeForce cards will be so perfectly overclockable. Besides, you shouldn'tmiss out the fact that graphics card overclocking is also connected with the power consumption increase, which is also alimiting factor, because not all mainboards can provide the needed power. And the last remark: we are not responsible forthe damage caused to the cards during overclocking. Moreover, the manufacturer is not responsible for the card's performanceat higher frequencies.