Bypass capacitor arrays and the AGP bus

[maluckey]

I tried this on the OC forum, and got no hits, so maybe here is the place.

For those who tinker.........Abit KR7A-133 (KT266a).

Does anyone think it would be worth the time to set up an array of bypass capacitors in the AGP signal path, seeing as how the AGP craps out on this board way before the PCI bus or the CPU? By worth the time, I mean will it help, or is the AGP archetecture/design such that it will not help.

The PCI Bus, the RAM, and the HDD's are all able to handle up to 200 Mhz FSB without issues (tested Prime95 for 13 hours, no errors @ 200 Mhz FSB).

Just pondering........

 

[Matthias99]

Dude... it's time to get a motherboard that has an AGP lock. Like, say, any board manufactured in the last year and built for overclocking. They even make boards now that run 200Mhz at *spec*! (gasp)

I seriously doubt you can do much to help it. Even if you did, I seriously doubt it would be enough to make the AGP bus run 50% faster than spec.

 

[maluckey]

I'll ignore the first comment and just add that it already is almost 40 percent above stock data rate, when I game at 175 Mhz FSB (350 Mhz Data rate versus 266 Mhz data rate stock). This is without any mods. Besides, I have three mobos already. This is something to do with spare time (which I have lots of).

To cut to the chase, is that you have doubts that it will get me much, as far as cleaning the signal. I'll chock that up to one "no" as far as would it be worth the effort.

 

[CTho9305]

So get an oscilloscope and look at the voltage across the existing caps. If it is fluctuating a lot more at higher frequencies, then maybe more capacitance will help. If it isn't, then that isn't the source of your problem.

 

[maluckey]

The signal is degraded, of that there is no doubt. The ripple is visibly greater at 348 Mhz data rate as compared to 266 Mhz Data rate, when we hooked it up at the shop to the O-scope. I have thought of using a couple of caps in parallel (as filters), mainly .1 microfarads, coupled with a .01 microfarads to attack the noise at a couple of different frequencies. This should help to ensure that I get a broader spectrum of rolloff, as compared to a single cap.

I have done this to a couple of auidiophile amplifiers to clean up the noise, and had good success so far. I'm just unsure about the architecture of the AGP bus itself, and whether or not this will work enough to merit the mod. I guess that maybe I'll just solder it up, then run some benches with intense 3d apps to see if it's any better. The cost is under 15 dollars total, so it's not breaking the bank, and I'll kill some time till the wife gets home.

 

[Matthias99]

This is something to do with spare time (which I have lots of).

See, that's the vital difference.

It wouldn't be worth it to me to even try, since I have somewhat of a scarcity of free time. If the only problem you're having is signal degradation, this may help. However, there are other factors to consider -- the interfaces on the graphics card/northbridge itself, for instance, may not be able to cycle their registers that fast, or there may be too much noise/crosstalk in the circuits on the card. In those cases there would be essentially nothing you could do, except maybe better cooling for the northbridge, or underclocking the graphics card.

I also wasn't paying attention to what was in your sig -- clearly you're the type that likes trying to make things like this actually work, whereas I'm the type that would rather take what I can get with minimal effort or spend a few bucks for something that works.

 

[CTho9305]

Well, here is my stupid capacitor experiment from freshman year. Link to the pics page is at the bottom. This pic and this one are a very quick summary .

 

[maluckey]

A lot of good thoughts, and a kindered spirit too! I patch together discarded, or unwanted things on a weekly basis. Sometimes cars, sometimes washing machines, and when I'm lucky, TV's and computers.

I hadn't considered the cross-talk from the video card. It might be that the card can't push the signal fast enough, no matter how clear the bus signal is. Hmmmmmnnnnn........Maybe it's time to work on the card first, then the AGP bus. then the signal would be about as clean as can be. The reason I haven't thought about the card being a weak link, is because no matter what driving value I use for it, it showed no difference. Perhaps that was because it was at it's OC limit at 324/616 Core/RAM, and the headroom was nil.

Thanks for the responses thus far!

Mark