Exploration into Overclocked AGP Graphics — THG.RU

Introduction

It has been a short period of time since we released the shocking information about Intel’s i820 being inferior the old yet nimble 440BX chipset (Showdown at 133MHz FSB — Part 2, The Real McCoy) that could run a 133MHz FSB setting. After heavy evaluation of each chipsets’ performance, the conclusion was that the 440BX was the best solution when using a Coppermine processor for the performance and price as long as you had no issues running the AGP and BX chip at higher than specified speeds. The biggest issue involved here if you are that serious is the fact that graphics cards may not like running at such a high AGP speed (89MHz vs. 66MHz). This is what stemmed this particular article. What graphics cards could boot at this speed? Which cards could run reliably? Would it cause the temperatures to increase on the card itself? What would the gains be? All these questions needed answers so we’ve run a bunch of tests to see who made the cut, and who didn’t.

The Questions

So there are a few obvious questions we needed to answer before pushing the 133MHz FSB setting on the 440BX motherboard. First things first, you have to have a reliable 440BX that can run the 133MHz FSB setting and memory that can do this as well, preferably CAS latency 2 memory. Once you have those two issues taken care of, you are now faced with which video card to choose. Currently, we know that the reference GeForce 256 has no issues running at the overclock setting of 89MHz but this doesn’t mean any flavor of GeForce 256 will as board designs vary. We also don’t know how other graphics chipsets will react. We need to find out not only if they’ll work but if they can handle the stress for an extended period of time if we’re to deem them stable.

The Testing Methodology

Testing each card would be done on a P3 800/133, Asus P3B-F, PC133 CAS 2 (and 3) memory, Netgear FA-310TX network card and WD Expert 418000 HDD. Each card would be tested by booting it, loading the OS, and then finally, running Quake3 for 30 minutes while connected to a server full of bots. Any crashes or failures would be noted and idle/working temperatures would be tracked to see how the added buss speed affected heat, if at all. Obviously if the card failed to even boot, no tests would be run and the card would automatically fail the test. It is possible that a card may boot, get into the OS yet fail to start Quake3 or run it with stability. These tests were run on CAS2 and CAS3 latency memory to see a comparison of the performance and stability of each. Technically, the memory timing shouldn’t matter to the video card but it’s good to check regardless.

The Test Subjects

With the performance parts in mind, I chose graphics cards that would most likely be used on this caliber of a system. I also tried various brands if the particular chipset was shipping from a range of companies. Here is the list of the cards tested:

With 15 cards total, less than half passed the full test. During the testing, basically all the boards failed during the boot process and not during the testing itself. If the board made it as far as the OS, it ran just fine. The most surprising part is that a given chipset didn’t always pass across the board. Variations in the design proved that some companies had designed the boards to deal with a greater AGP speed tolerance (or were just plain lucky). This doesn’t mean that the companies that failed obviously made a inferior design but didn’t (couldn’t) design the board to reach such high speeds. Remember, we’re talking about running the AGP 33% faster than normal. Let’s take a look at who did what.

Benchmark Setup

The Results — 133/100MHz Bus Quake 3 Arena Results

To give you an idea of what type of performance gain you’ll see from the various bus speeds, I’ve tested each chipset that was successful at 133MHz bus with PC100 CAS2 and PC133 CAS2/3 in Quake 3 Arena in Normal mode.

Here you can see that the jump from PC100 to PC133 is a solid gain for most but not all cards. In the case of the TNT2 Ultra it appears that the system is so fast that it actually reaches the video card’s fill-rate peak at this low resolution as the increased system performance barely helps the framerate. The step from CAS3 to CAS2 is minor but for the high performance enthusiasts, this still is an attractive gain.

The Results — Heat Measurements

I decided to monitor the heat of each card while idle and while running Quake 3 Arena over an extended period of time at each of the bus settings (133 and 100 MHz bus). The measurement was made with the infrared thermometer on the backside of the video card’s graphics chip. This typically gives the best measurement. Let’s take a look at the results.

For most of the cards, the change from 100 to 133MHz bus was minimal if none at all. Obviously working temperatures went up from the temperature when idle but the bus speed had no affect on the heat level of each card. Looking at the various temperatures of the leading cards is a bit interesting. That Voodoo3 3500 is a scorcher and is bar far a the best video card to double as a heater in the winter.

Afterthoughts

On a performance level, a 133MHz bus speed on the BX platform performs very well and shows a solid improvement on almost every single card. The CAS latency doesn’t seem to make such a huge difference but it may be worth looking into if you feel that you need every single % squeezed out of your system. We also noticed that the heat differences from running a card at the two bus speeds was basically nothing so you need not worry about heat issues if you take the 133MHz plunge.

This situation is interesting for two kinds of people. Die hard overclockers may want to try upping the bus speed with their Pentium III 550E or 600E processor to reach 733 or 800 MHz if they are lucky, thus benefiting from the overclocked processor as well as from the superior performance of BX at 133 MHz FSB. People who don’t really want to overclock their CPUs, but who don’t mind that their graphics card is running at 89 MHz AGP might want to plug an official Coppermine processor for 133 MHz FSB (ending with ‘EB’) into their BX-board. In this case there is still the benefit of the high performance that BX offers at 133 MHz FSB, which is superior to any platform that officially supports this FSB-clock.

After testing all the boards I had handy, we’ve learned that not only does the 3D-chip matter but also the board vendor design itself comes into play as we’ve seen the same 3D-chip fail and pass running at 133MHz bus. By no means is this a complete list of graphics cards available or that will work but it should give you an idea of which popular cards should work in case you wish to switch over immediately to the fastest platform setup available right now. We also cannot tell you if the 3D-cards might suffer long term damage from running at an AGP-clock that’s 33% beyond spec.