X-bit labs

NVIDIA Riva TNT2 Overclocking Guide

Category: Video

by FastSite

[ 08/26/1999 | 12:00 AM ]

Have you ever met a computer-lover (or even a gamer), who doesn't like high speeds? We think you haven't. Everybody isreally happy to realize that his PC uses all 150% of its power. So, how can we squeeze the maximum out of our electronic fellow?

Table of contents:

• Q3Test
  
• Quake2 timedemo1
  
• Unreal 2.25 timedemo
  
• How can I choose a well overclockable graphics card?
  
• The classification of local graphics memory chips on TNT2 based cards
  
• Memory on TNT2 Ultra cards
  
• Memory on TNT2 cards
  
• Conclusions
  
• P.S.
  

Have you ever met a computer-lover (or even a gamer), who doesn't like high speeds? We think youhaven't. Everybody is really happy to realize that his PC uses all 150% of its power. So, how can wesqueeze the maximum out of our electronic fellow? The answer is more than evident: overclocking.There are several different types of overclocking. But as a rule, users usually overclock thesystem bus (FSB), which helps increase in the first place the performance of the CPU, system memoryand sometimes even peripheral devices. However, there is always another side to the picture (read: tooverclocking). On the one hand, you increase the overall performance of the system or of its separatecomponents, while on the other hand, you face some problems with the system stability and cooling,which need immediate solving. If you only dream about calm and quiet life and consider the day withoutfighting the trouble to be simply wasted for nothing, then you are a born overclocker. Usuallyoverclocking attracts those computer guys, who are quite willing to have a bit of fun after work,for instance to defeat their close friend in Quake3 Arena about 20 times, shall we say. Due to theoverclocked graphics accelerator they manage to increase the amount of the sacred fps (frames per second).What for? Well, first of all, the more fps you get, the higher is the gaming performance. And in practiceyou will see all the characters in your game moving very easily and naturally, and it means that you willfeel all the realism of the scenes. Of course, you may object saying that in fact you won't even noticeif there are a few additional fps or not. Then, what's the purpose of overclocking? Here we have todisagree with you. The more complicated is the scene you see on the monitor screen, the harder worksyour poor graphics accelerator. That's why the power reserve will be right at hand when you enter thelevel with a dozen of people fighting severely. Then you won't deny that the additional 10fps will bereally helpful. In fact, these particular fps mean the overall performance of your graphics card. Themore detailed is the scene shown, the slower it is rendered and the lower is the amount of fps. In otherwords, if the chipset utilization rises, the performance drops and hence drops the fps amount. The morefps are reserved, i.e. the higher is the graphics card performance, the higher is the guarantee that thescene rendering rate as well as the fps will remain satisfactory. In other words, you won't need to watcha slide show going on your monitor while you are firing a rocket launcher into a crowd of your friends.Well, today we are going to speak about graphics accelerators overclocking. Before we start we wouldlike to note that it is somewhat easier to overclock graphics cards than CPUs. It can be explained bythe fact that it is easier to choose a graphics accelerator with a certain power reserve (and safety factor),because local graphics memory chips are situated within sight and are clearly marked. And the graphicschipsets by one and the same manufacturer usually overclock almost the same. That is why you don't haveto lay yourself out trying all the cards to choose the one that will suit you most of all - just consultthe general overclocking stats we are offering you here.

The graphics cards based on nVidia Riva TNT2 chipsets can be very well overclocked for the followingreasons:

1. The chip and memory frequencies of the TNT2 based cards are not fixed in relation to eachother (as we see by 3dfx, for instance), which allows achieving maximum possible performance ofthe chip and graphics memory for each particular piece. It is only due to this peculiarity thatTNT2 based cards perform just unbelievably when overclocked, because due to its architecture thefastness of TNT2 based card depends mostly on the memory frequency. And so, we don't have to checkthe maximum possible processor working frequency while overclocking the memory.
2. A lot of cards are equipped with the memory chips, which maximum possible working frequenciessignificantly exceed the nominal ones.
3. The cards manufacturers usually provide their products with the memory chips by different suppliers.If they run out of the "slow" chips for the cheapest models of the family, then they simply use fastermemory, which usually goes to more expensive ones. The opposite situation is also possible though a bitmore rare. As a result, you may purchase a very coolly overclockable graphics card for less money.

To begin with we suggest taking a look at the results achieved in the following three benchmarks,which show the expected performance gain after overclocking.

• Quake3 1.07 demo1
• Quake2 3.20 demo1
• Unreal 2.25f timedemo

The testing system was configured as follows:

• Pentium II 450MHz CPU
• ASUS P3B-F mainboard
• 256MB SDRAM DIMM system memory
• 6.4GB Quantum CR HDD
• SB Live Value sound card
• Windows 98

The key to the graphs:

• X axis - the chipset working frequency
• Y axis - the memory working frequency
• Z axis - the fps

All the tests were run at the resolution set to 1024x768 with 16 or 32-bit color depth.

Q3Test

Q3Test maximally utilizes the graphics card that's why it is a perfect test to estimate the acceleratorsperformance in the future games.

Q3test 1024x768x32bit, fps		Memory frequency, MHz
		150	175	200	225	250
Chipset frequency, MHz	125	21.8	25.1	28	30.1	31.4
	150	22.2	26.1	29.7	32.5	34.6
	175	22.5	26.7	30.6	33.9	36.1

In 32-bit regime, the weak spot is the graphics memory. If the memory working frequency increases, thefps increase much more than if the graphics processor working frequency goes up. However, at 125MHzprocessor frequency, the difference between 225 and 250MHz memory frequency is quite insignificant -the processor can't catch up with the memory and hence any further increase of the memory workingfrequency won't lead to higher performance. At 150 and 175MHz processor the rising memory frequencywill make the performance increase almost linearly.

Q3test 1024x768x16bit, fps		Memory frequency, MHz
		150	175	200	225	250
Chipset frequency, MHz	125	34.7	35.9	36.6	36.9	37.1
	150	39	41.1	42.9	43.7	44.2
	175	41.7	45.3	47.8	49	49.9

If the color depth is set to 16 bit per pixel, then the main burden should be carried by the graphicsprocessor - the fps grow greatly with the growth of the processor working frequency and comparatively lesswith the growth of the memory frequency. And the difference depending on the change of the processorfrequency is quite significant even if the memory is working at its minimum, and it increases evengreater if the memory working frequency gets higher.

So, we see that in the 32 bit per pixel color depth the determinative in Quake3, which influences theperformance, is the memory frequency, and in 16-bit color it's the graphics processor.

Quake2 timedemo1

Unlike Quake3, Quake2 evenly utilizes CPU and the graphics card that is why the fps value doesn't growso fast if the graphics card components are overclocked or if the color depth is changed from 32 to 16 bitper pixel.

Quake2 1024x768x32bit, fps		Memory frequency, MHz
		150	175	200	225	250
Chipset frequency, MHz	125	51	55	57.3	58.8	59.5
	150	54.2	61.2	65.5	68	69.4
	175	56.6	64.8	71.1	74.9	77

Quake2 doesn't overload the graphics memory bandwidth therefore even if the color depth is set to32bit you will still see the difference between 125, 150 and 175MHz on the graphics processor whilethe memory is working at its minimum frequency.

Quake2 1024x768x16bit, fps		Memory frequency, MHz
		150	175	200	225	250
Chipset frequency, MHz	125	58	60.9	62.5	63.3	64.2
	150	63.4	69.2	72.5	74.3	75.6
	175	66.1	74	79.5	82.6	84.4

In 16bit color the dependence on the processor frequency is even higher. The performance gain due to theprocessor overclocking by 25MHz is much higher than that due to memory overclocking by the same 25MHz.

As we can notice, Quake2 tests showed that memory frequency exerts much less influence over the overallperformance than we saw in Quake3. It is mostly the graphics processor speed that tells on the result evenin 32bit color.

Unreal 2.25 timedemo

Unreal 1024x768x32bit, fps		Memory frequency, MHz
		150	175	200	225	250
Chipset frequency, MHz	125	30.41	34.32	36.81	37.99	38.6
	150	30.68	35.06	38.05	39.66	40.2
	175	30.83	35.22	38.5	40.52	42.1

The results of this test don't depend so greatly on the graphics card performance as in the previoustests. As for fps, this value depends on the memory and graphics processor overclocking in the same waywe saw in Quake3. In 32bit color depth the memory plays the main role for the cards fastness and in 16bit -the processor.

Unreal 1024x768x16bit, fps		Memory frequency, MHz
		150	175	200	225	250
Chipset frequency, MHz	125	39.23	39.5	40.08	40.25	40.3
	150	40.92	42.07	42.83	42.41	43.3
	175	42.47	43.52	44.6	44.79	45.2

If the color depth is set to 16bit per pixel and the memory is overclocked from 150 to 250MHz(by 100MHz!!!), then the performance gain may make less than 10%, which is quite disappointing,actually.

So, we see that it will be false to speak about some absolute performance gain of the graphics card whenoverclocked - different overclocked components in different tests show different results. For example, thedifference between the graphics card with 125/150MHz (chipset and memory correspondingly) and the one with185/250MHz in Quake3 makes 50%, in Quake2 drops down to 30% and in Unreal it hardly reaches 15-20% with thecolor depth set to 32bit.

And now, after we have discussed the general effectiveness of the overclocking as it is, let's turn to thepractical side of our investigation.

How can I choose a well overclockable graphics card?

The mere word "choose" already implies that there is a wide "choice" of possible candidates. However,we decided not to mention here all the existing graphics cards on nVidia Riva TNT2 chips. We will mainlydwell on the most popular models from ASUS, Creative and Diamond.

First of all let's find out which cards can boast highly overclockable processors.

When you buy a graphics card based on TNT2 chip you can never say for sure how well you'll manage tooverclock it. That is why all you can do is just to rely on the statistics in this situation. We took alook at about a hundred of TNT2 and TNT2 Ultra based graphics cards by the above mentioned manufacturersand summed up our impressions on the matter.

Creative has only two models:

Creative 3D Blaster TNT2 Ultra (32MB, 150/183MHz, TV-Out)

The Ultra-model is equipped with a really weak radiator, almost without ribs, and with a fan. The radiatoris stuck by means of heat conductive glue. The graphics processor, which works at 150MHz as default, as arule works at 175MHz as well. The probability that the graphics card you buy doesn't work at this frequencyis very low. We came across only one not working at the chipset frequency over 170MHz from the very firstsupplies, which arrived in our country. The chipsets of all present graphics cards work at 170-175MHz,which is significantly worse than those from the first shipments, where the chipsets supported up to190MHz.

Creative 3D Blaster TNT2 (32MB)

If the Ultra model can at least boast some active cooling system, then the younger model has just apoor 8-rib radiator. The situation is also aggravated with the fact that the graphics cards are usuallyinstalled into the PC with all the chips (and hence the radiators as well) turned downwards, which makesthe whole cooling absolutely useless. Well, if you decide to overclock the processor on one of these graphicscards you will definitely need to work pretty hard with your own hands. :-)

Here is what should be done. You take two plastic panels (here you can use the spare lids for 3.5" sections)and cut an L-shaped thing, which narrower part should be fastened between the radiator panels.

So, now you will be able to install a more powerful Pentium cooler instead of a very rare 486 one.

Besides, you may also arrange something :-) to air the upper (or reverse) surface of the card. It can bedone in the following way. You take the remains of the lid for 3.5" section we already used above and cutit into four pieces. Then you stick them to the upper side of the graphics card

and over them you stick the cooler (it would be better to take a Pentium cooler).

After the improved cooling system is activated you will be able to overclock Creative TNT2 graphicsprocessor up to about 150-160MHz. (By the way, the cooling system for "the upper surface of the graphicscard" will be also very efficient for the cards by other manufacturers :-))

Diamond offers us a choice of three models:

Viper V770 Ultra (32MB, 150/183MHz)

The card goes with the radiator as beggarly as the one by Creative TNT2 Ultra. Frankly speaking, wecan't understand the graphics cards developers: these strange guys install very expensive Ultra chipsversions and at the same time try to save a few cents on radiators. Because of this disappointing factmost chipsets of V770 Ultra cards work at 175-185MHz at the most.

Viper V770 (32MB, 125/150MHz)

This graphics card is equipped with perfectly overclockable processors and with large radiators, whichis very convenient and practical at the same time. The working surface of this radiator allows installing aPII cooler without any additional fastening mechanisms. With this cooler most chipsets of V770 cards showvery stable work at 160-170MHz.

Viper V770 (16MB, 125/143MHz)

This low-cost TNT2 based model is provided with less overclockable cores and memory, which is quitenatural. However, the situation doesn't strike as absolutely hopeless due to a good radiator, the sameas on 32MB V770. The processors of these graphics cards can be overclocked to 160MHz, not more.

ASUS can boast the largest range of models:

1. V3800 Ultra Deluxe (32MB, 150/183MHz, stereo glasses, TV in/out)
2. V3800 Ultra (32MB, 150/183MHz)
3. V3800 TVR Deluxe (32MB, 125/150MHz, stereo glasses, TV in/out)
4. V3800 TVR (32MB, 125/150MHz, TV in/out)
5. V3800 32 (32MB, 125/150MHz)
6. V3800 16 (16MB, 125/150MHz)

We aren't going to bury you under the whole bunch of photographs, since all the cards have one and thesame PCB design. Here is the picture of the most expensive model - V3800 Ultra:

And the cheapest one - V3800 16MB:

All ASUS graphics cards are designed with the same cooling system, which can be ranked as average -very short radiator cooled by a small fan. By the way, unlike Creative and Diamond graphics cards, theradiator is not stuck to the chip with some glue but pressed to it with a special fastening mechanism.Besides, the contacting surfaces of the radiator and the chip are covered with a thin layer of pastelooking very carefully done and not dried up, which was so typical of the "ordinary" TNT cards fromASUS. All ASUS graphics cards proved very stable during overclocking - 160MHz for TNT2 chipsets and175MHz for TNT2 Ultra. We can't say what made them less overclockable than their competitors, howeverwe may tell you for sure that it won't be the best choice for overclocking.

So, summing up we can state that the best card for overclocking among those based on TNT2 chip is V770,and among those based on TNT2 Ultra - Creative 3D Blaster TNT2 Ultra.

The classification of local graphics memory chips on TNT2 based cards

Since the cards based on TNT2 chips are assembled with different types of local graphics memory(SDRAM and SGRAM), let's try to find out if the memory chips type may really mean significant speeddifferences. Note that you can overclock today's graphics cards by optimizing the timings of the workinglocal graphics memory. In order to put an end to various rumors about the performance variations betweenthe cards by different manufacturers due to the timings we decided to take a closer look at a couple ofgraphics cards on TNT2 chip with SDRAM local graphics memory:

1. Creative TNT2 Ultra with 5ns memory chips from ESMT
2. Creative TNT2 with 6ns memory chips from Hyundai
3. Diamond Viper 770 Ultra with 5.5ns memory chips from Hyundai

Having applied the TNTClk utility we found out that all these cards have the same memory timings(left column). Maximum performance (or minimum timing) we managed to get is shown in the right column(for a better comparison we added the very right column with the minimal timings).

The working frequencies of TNT2 based card - 150/183MHz (chipset and memory correspondingly) -underwent optimization and the card's performance rose from 62.8 fps to 65.1 fps in Quake2 at 1024x768x32bit.

Besides, we have also tried three graphics cards with SGRAM local graphics memory:

1. ASUS V3800 Deluxe with 6ns memory chips from SEC
2. ASUS V3800 Deluxe with 7ns memory chips from SEC
3. Creative TNT2 with 7ns memory chips from SEC

Here everything is not so uniform - by ASUS cards the first column contains "1" instead of "0" as byCreative. It was probably made to provide the so highly praised reliability of ASUS cards, although wedidn't notice if this change made any influence on the overall performance. All the other parameters werethe same for all the cards. The second picture below also has two columns: the left one contains the defaultvalues and the right one - the minimum we managed to achieve.

Look: the performance difference makes 62.5 against 63.9 fps in the same conditions.

By the way, is you compare SDRAM and SGRAM memory parameters you will easily notice that SDRAM memoryvalues are more profitable, which was also proven by Quake2 test.

So, there isn't much sense in changing the memory timings because it seriously reduces the chances ofgood memory frequency overclocking.

And now let's check how far we can overclock the memory chips by different manufacturers: unlike thegraphics chipsets these ones do not hide under cooling radiators and are clearly marked. We would alsolike to point out that overclocking of the memory chips manufactured by different companies doesn't alwaysend with the same results even if the maximum achievable performance marked on them is absolutely the same.So, now all we have to do is just to find out what manufacturers make the best overclockable memory and thento pay special attention to the cards with this particular memory (keeping in mind the chipsetsoverclocking statistics, of course).

Memory on TNT2 Ultra cards

We can say for sure that only 5.5ns memory is able to work at 183MHz frequency that is why graphicscards manufacturers try to use these particular memory chips in most cases.

SGRAM:

SEC 5.5ns

On Samsung memory certified as 183MHz "nanoseconds" are marked not with the usual numbers but with the"C" letter. This memory is used in today's Creative TNT2 Ultra graphics cards and can be overclocked upto 220MHz.

ESMT 5.5ns

Strange as it might seem, but ASUS, which usually prefers using SGRAM chips from SEC, resorts to thememory from other manufacturers for its Ultra cards. This memory can be overclocked to 230MHz.

EliteMT 5.5ns

This is also the memory taken from ASUS3800 Ultra and it also works at 230MHz. Well, it looks as ifASUS decided to play safe since SEC 5.5ns memory works only at 220MHz :-)

SDRAM:

Hyundai 5.5ns

This is the best memory you can find today on the graphics cards: it works at 230-240MHz. You can seeit on V770 Ultra cards and some Creative TNT2 Ultra ones.

EliteMT 5.5ns

This memory is a bit less overclockable - it's top frequency is 225MHz. Very often it appears on thelatest (manufactured and supplied) V770 Ultra graphics cards.

ESMT 5ns

For today it is the fastest memory. The first Creative TNT2 Ultra graphics cards could boast the memory,which worked at 240-250MHz. Unfortunately, it is almost impossible to find a graphics card with thisparticular memory today, because Creative equips all its latest TNT2 Ultra cards with a slower SGRAMmemory from SEC.

Memory on TNT2 cards

SGRAM:

SEC 7ns

Can be overclocked to 200MHz, which is not that bad actually compared to SEC 7ns SDRAM. From the verybeginning this memory went only for low-cost 16MB V3800 graphics cards, however, now you can find thismemory on all "non-Ultra" ASUS cards.

SEC 6ns

The max limit for this memory - 210MHz. They are used in V3800 and modern Creative TNT2.

SDRAM:

SEC 7ns

These "slow" chips (maximum 180MHz) are installed only in low-cost V770 16MB.

Hyundai 6ns

Like in the situation with 5.5ns chips, 6ns memory from Hyundai is also ahead of all: we managed tooverclock it to 225MHz on most cards. Can be found only on V770 32MB graphics cards.

EliteMT 6ns

A bit less overclockable memory - proves reliable and stable only working at 215-220MHz. A lot of recentlymade V770 32MB graphics cards are equipped exactly with this type of memory.

EtronTech 6ns

Well, this seems to be the worst memory for overclocking: the maximum frequency we managed to get was180MHz. Although we saw only one single supply of V770 32MB cards built with this memory.

So, now we see that Hyundai appears to be the manufacturer of the most highly overclockable chips andthen this memory goes to Diamond Viper V770 and V770 Ultra cards.

Conclusions

Well, let's sum up. We have every right to state that Diamond Viper V770 graphics card with 6ns memorychips from Hyundai turns out the best choice. Most cards with the standard working frequencies set to125/150MHz (chipset and memory correspondingly) can be overclocked to 160/225MHz, which provides a 50%performance gain in applications requiring powerful graphics accelerator. V770 Ultra with 5.5ns memorychips from Hyundai can be overclocked somewhat better, but the price is unproportionally higher.

As for ASUS and Creative, we wouldn't recommend you to buy their cards from recent supplies for furtheroverclocking, because they have less overclockable graphics chipsets and slower SGRAM memory.

So, make your choice among the whole bunch of the presently available graphics cards, but don't forgetabout proper cooling. We hope our attempts to study this matter will be of some help to you. Good luck!

P.S.

In this article we posted the results for the graphics card, which memory was overclocked to 250MHz, butwe didn't mention the utility we used for this purpose. Strange as it might seem, but this super-overclockingprogram was none other than our good old body - PowerStrip. Yes, of course it allows achieving only 225MHzfor memory with the default settings, but we all know that if you really want something then there is nothingimpossible. Are we right? ;-)

So, here you are. Open pstrip.cfg file stored in Windows directory and find the line "[RivaTNT+]".Below you will see something like this:

MClk=75,142,110NVClk=70,125,90 MClk2=100,225,150NVClk2=100,200,125 

The three numbers separated by a comma are the minimum frequency, maximum frequency and default frequency.The first two keys concern TNT cards, the second two - TNT2. MClk and MClk2 keys set the memory workingfrequency, NVClk and NVClk2 - define the work of the processors.

Now you know what and where you should change that is why there shouldn't be any problems withsetting 250MHz instead of 225MHz. Unfortunately, the program doesn't respond to the values over250MHz and the regulator will still remain at 250. Actually, you don't really need to set it higher,because there are not so many cards proving stable at 250MHz anyway.

Of course, you can "adjust" PowerStrip to all graphics cards it supports and not only to TNT.