Latest topics
First article
Liquid Nitrogen Demystified
Read about the ins and outs of Liquid Nitrogen used to boost PC performance. Here are the answers to the questions you always asked yourself.
Second article
The GTX285 vgpu mod explained
This article will shed some light on the GTX285 voltage mod and a different way of using a variable resistor for voltage modifications.
Third article
Measuring voltages
Do not trust software, finding the hot read out spot, using the correct ground reference and more regarding measuring voltages.
Forth article
Voltages and Temperatures
What the actual voltages or temperatures are is a non relevant question. Get a hand on what is really happening and you are better set for success.
Fifth article
Filtering caps
Capacitors are a vital part of any voltage regulator. Here are some background information of the purpose of them and some facts to consider when beefing up the stock one.
PWM - Pulse Width Modulation
posted by Robert "crotale" Kihlberg @ 8:13am, Saturday 16 May 2009.
(The site just broke 200 unique visitors and is just about to break 2000 page views. Thanks for the interest and support. Please let me know if you have any questions or if you have a topic you would like me to write about)
Pretty much all of todays voltage regulator circuits in computer systems operates according to the PWM principle. With this article I would like to cover some questions I have had over the years. One is 'Why can't the power supply handle all voltage conversion?' and another is 'How can it be that a CPU draws over 200 amps while the maximum my power supply can output is below 100A?'
The answer to the first one is that there are so many different voltages needed in a computer system. Some are used for signaling and others for power and it would be very costly to create a general power supply to handle all of them. That is why there are voltage converters of various sizes in various places.
To answer the second question we need to understand the idea behind PWM, which stands for Pulse Width Modulation. A traditional linear voltage regulator has a high voltage as input and outputs a lower voltage. To achieve this it has to "burn" off the excessive voltage. This is very energy consuming already at small loads. The idea of a PWM converter is to switch the high voltage on and off quickly so the mean of the output is a lower voltage. As one would assume, this output voltage would be severely "spiky" and not very kind to the circuitry. This is where filtering comes into play that smoothens out the voltage.
To take a practical example; let's assume that we want to convert a 12V source to a lower voltage. If we switch the voltage on for one second, switch it off for one second and repeat this. The mean output voltage over time would be 6V. If we switch on the voltage for 1 second and leave it off for 11 seconds we would get a mean voltage of 1V. However, this voltage would not be healthy for any integrated circuit and in reality this switching is done at a much higher rate. Even though the switching rate is higher, we still need the filtering.
To illustrate how the filtering works, let's imagine that we want to water flowers but we only have access to a fire hose (or something similarly powerful). Those neat flowers would be killed instantly by the high pressure of the water. In a similar fashion to what was described above, we can switch the water on and off quickly, but that would not do the trick entirely. However, we can take an empty water tank and use it as a filter. The water tank would fill up slowly if we pulse the water into it and we can output non-pressurized water by drilling a hole in the bottom of the water tank.
This is a textual and highly simplified description of how a PWM works. I will shortly update the photo album with some illustration of this.
GOOC 2009 - EU final
posted by Robert "crotale" Kihlberg @ 2:38pm, Wednesday 29 April 2009.
After a pass in the GOOC Swedish final I was kindly invited to the European final in Prague, Czech Republic. Together with me I brought Henry de Kruijf, also known as ME4ME on the Internet. The time before the event was rather hectic with CeBIT, The Gathering and one exam. Luckily, we found a spot for some practicing two weeks before the event and we collected some equipment for it. Our main focus was to get to know the motherboard better than we had time to do during the Swedish final.
During our preparations we managed to run SuperPi at around 4990MHz and 3D at 4870MHz. This was pretty much what the CPU was capable of, so we felt we got pretty good control over the BIOS and settings in it. The second part of the testing was to try what a GTX260 was capable of. After some frustrating initial testings we realized we could run the card really cold and that voltage wasn't as necessary as we thought. In the end we managed to pass a couple of runs at 1000+ core and 2000+ shader at -130C, which was really satisfying.
Moving on to the GOOC event in Prague we were quite confident in our strategy. We had all insulation material already cut for both cards and the motherboard, so we pretty much just bolted everything together. Along with massman and pt1t from Belgium we were early with publishing results in SuperPi. We couldn't match their CPU clocks as they were running well over 5300MHz but thanks to nice efficiency we managed to control the second place with 5075MHz. We felt that the 3D part was where we had a really good chance to be successful, but it didn't start out well. The cards were working well up to 800/1600 but after that we got really strange lockups and stuttering which took us a good hour to figure out. One part of the problem was that one of the cards didn't really like cold as well as the first one but keeping it warmer than -100C fixed this issue. The second problem was related to 2D/3D-mode. Apparently, the slave card isn't force into 3D mode as the first one when this feature is enabled in RivaTuner. This made us set the 3D voltage of the first card and 2D voltage on the second. Then when both cards entered 3D mode, the second card would revert to it's original 3D-voltage and causing the test to crash. When we figured this out, the time was not on our side and while we managed to show some really impressive SM3/HDR scores, we did not manage to pass a complete 3DMark06 round. Looking back at it, taking the second place would have been easy if we just passed a single full round. Taking out massman and pt1t with their brutal CPU would have been tougher, but not impossible.
All in all, the event went smoothly and it was great fun meeting and talking to all other overclockers from all around Europe. I'm sure many of us will meet again.
The Gathering 2009
posted by Robert "crotale" Kihlberg @ 2:30pm, Tuesday 21 April 2009.
Last year I joined Marcus at ASUS Nordic for some overclocking adventures at The Gathering in Norway. The Gathering is Norway’s equivalence to Dreamhack in Sweden and has about 5000 visitors. This year we decided to join up again and give QuadSLI a try.
QuadSLI is created by using two GTX295 cards which each incorporates 2 PCBs with one GPU on each. The design of each PCB resembles the one of GTX260 as they both use 896MB of graphics memory. The core itself has more in common with GTX280 with its 240 shader units, but the clocks are the same as GTX260. To sum it up; 1.4 billion 65nm transistors on each PCB.
The first thing we did after smelling the fresh and “brand-new-smell” of the cards was to screw them completely apart. Due to space limitations and price savings, the GPUs have a limited voltage regulator which has a tight over current limit. Not unsurprisingly, the cards were modified to fix this issue. During the first two days we played around with settings for a single card (2 GPUs), testing temperatures, voltages and frequencies and during the final two days we ran QuadSLI. Running two of these cards with extreme cooling is a very hard and multi dimensional task. Not overly satisfied with the frequencies we were able to work with, we laid our hopes to efficiency. In the end we manages to conquer the world record in the 3DMark Vantage Extreme preset and the second place in both the High and Performance preset.
Here are some photos from the event.
