Objective

The purpose of this experiment is to answer the following questions:

What are the hottest parts of a GeForce board? Which parts need the most cooling?

Do some memory chips run hotter than others? Is this why some GF 2 cards have heatsinks on the front memory chips but not the back?

How do the demands of the software (or game) effect the temperature of the GeForce? Does the GeForce increase in temperature the harder it works like a CPU does?

The Experiment

I will be testing a Creative Labs GeForce DDR in these four conditions.

• In the BIOS setup screen.
• Running a DVD (Robocop) in Windows.
• Running Quake 3 with S3 Texture Compession Enabled.
• Running Quake 3 with S3 Texture Compession Disabling.

For each test I will measure the temperature of each of the eight memory chips and the GPU.

The BIOS test will give temperature readings of the card while it is doing the least amount of work. At this point the card will be doing nothing more than displaying some text, no textures of 3D involved here, although a 3D texture mapped BIOS set up would be pretty cool.

The DVD test will show the card doing some fairly intense 2D work, as well as motion compensation and color space transforms. However, the load on the memory chips should remain minor and most of the geometry components of the GPU will be unused.

Quake 3 will be performed in 800x600x32 resolution with all eye candy features (brass, sky, etc.) on. This should stress the GPU quite a bit. By enabling S3 texture compression we can stress the card but keep the memory requirements low. By Disabling S3 texture compression and using 32 bit textures we can really stress the cards memory. For these tests I used the demo that came with the Area 15 map.

Equipment

The tests were performed on a Creative Labs GeForce DDR, P2-450, and ASUS P3V4X motherboard in an Inwin Q500 case. The Asus board supports an optional 3rd temperature sensor via a thermal sensor cable. This is what I used to conduct the tests. Temperatures can be read in the BIOS setup screen, which is how I conducted that test. For Windows tests I used a tool called Motherboard Monitor, which can be downloaded from Tweakfiles.com. The DVD software was Power DVD 2.55 which supports motion compensation.

Limitations And Other Factors

There are some limitations to this testing method that can affect the results. First, I did not remove the heatsink off my GPU, so what I actually measured was the temperature or the right hand edge of the chip. This is likely to be a few degrees lower than the true internal temperature of the GPU.

Second, since I measure each component of the board sequentially there is a chance that the readings could chance in the time it took to finish one complete board reading. I compensated for this by reading the board twice (yeah, this got very tedious) and repeating the experiment if the readings were off by more than 1 degree. This didn't happen though. If the results were off by 1 degree I took the highest value.

Thirdly, in order to see the temperature reading software, I had to run both the DVD and Quake 3 in a window, so I won't be testing the card in full screen mode.

Last, there are environmental conditions around the card. The ambient temperature of the room has the potential to vary some, and also airflow within my case may keep some parts of the card cooler than others. The back of the card in particular gets a fair amount of airflow from the 80mm fan that I have attached to my CPU. This could have been fixed by removing any fans in my PC, I didn't think that would be a good idea though. So there is a margin for error in the results I will gather, probably around +/- 2 degrees C. At the start of the tests the ambient case temperature was 24/75 (C/F) and the motherboard temperature was 25/77.

The Tests

For each test I will label the memory chips 1 - 8 as shown below, this probably does not reflect the real order in which these chips are addressed.

Test 1: BIOS

For this test, all the card was doing was displaying the BIOS setup screen.

As we probably expected, the hottest part is the GPU. Memory temperatures seem fairly even across the chips.

Test 2: DVD

For this test I played the Robocop DVD, this should stress some 2D functionality of the card but memory and geometry load on the GPU will be light.

GPU temperature has risen slightly, although not by much. Memory chips 6 and 7 seem to be running hotter than the rest by a couple of degrees C.

Test 3: Quake 3 with S3TC

For this test I played the Area 15 Quake 3 demo. The settings where 800x600x32, high geometry, high texture res, tri-linear filtering, all eye candy extras on. During this test I enabled S3 texture compression to reduce the memory load.

GPU temperature is actually less than with the DVD. Memory chips 3 and 4 seem to be running the hottest.

Test 3: Quake 3 without S3TC

For this test I played the Area 15 Quake 3 demo. The settings where 800x600x32, high geometry, high texture res, tri-linear filtering, all eye candy extras on. This time S3TC was switched off and 32 bit textures enabled. This should fill up the card's 32 meg of ram easily.

During this text my harddrive thrashed like crazy, indicating that a lot of texture swapping was taking place. Readings seem more consistent, perhaps because all the memory was being used, each chip had the same load on it. GPU temperature was the highest so far, yet still only 6 degrees C above what it was in the BIOS.

Conclusion

The results were a bit of a suprise. Considering how my CPU can almost double its temperature when its really busy I expected to see my GPU vary more in temperature than it did. It seems to be that GPUs do not heat up to the same extent that a CPU does. The memory temperatures showed a couple of spots, chips 3, 4, 6 and 7 that heat up a little more than others but only by a few degrees. This makes you wonder how much difference putting heatsinks on your memory chips really makes.