RAM overclocking: How can I tell I'm going too far?

[clicknext]

I've overclocked the 5.0ns RAM on my vid card which was at 200, to 250. There are no stability or artifacting issues, having looped the 3Dmark2001 vertex shader test 500 times without visible artifacts. While working, the RAM does get quite hot to the touch. Almost at the point where it stings a bit when I keep my finger on it for a while, but not quite. A lot hotter than idle temp anyway. So can my RAM take it as long as its stable and there are no visible problems, or do I need ramsinks?

 

[Cerb]

The RAM should be hot to the touch regardless. Run 3Dmark while you're sleeping or off to work in loop mode. When you wake up or get back, wait for it to get to the point-sprites test (unless it shows artifacts sooner). If none are there, you're good to go...however, I can't say RAMsinks would be a bad idea.

 

[vss1980]

I dont trust the point sprites test as a good indicator of graphics artifacts from overclocking. I have seen the use of different drivers have a bigger effect on the quality of that test than overclocking artifacts on some cards.

Also, dont just settle for using one application for finding visual artifacts such as 3DMark. For example, I have personally been able to overclock my card and have a artifact free run through 3DMark2001 while UT2003 shows not just pixel popping artifacts but even errors in polygon and texture placement.

There have been a couple of programs in the past that did artifact testing, but their accuracy always varied a bit. One that comes to mind was called artifact tester and you may want to search for it and give it a go.

 

[chizow]

Yah, I agree that 3dmark2k1 isn't the best test for stressing your OC. Its pretty dated compared to whats out there in 3d graphics today. UT2k3 is a much better stress test b/c it'll tax your cpu and vid card to a much greater extent, causing any artifacts or crashes to arise much sooner.

My biggest issue for OCing my video card has been with case temps, and it takes a while for them to build up. If you can manage to keep your case cool with good airflow and active cooling, it'll definitely keep your OC more stable. I'm not a huge fan of ramsinks, but I suppose they couldn't hurt. Just don't go spending $30 on expensive ramsinks for that extra 2% increase in performance from OCing your RAM a tiny bit higher.

Chiz

 

[lobadobadingdong]

5.0ns ram 5nsX60(gets you to microseconds 1micros=1Mhz)=300MHz Maximum speed of this chip according to the manufacturer.
Go to school and learn frequency calculations in relation to time shift. Makes life easy.

 

[Jeff7]

Originally posted by: vss1980
I dont trust the point sprites test as a good indicator of graphics artifacts from overclocking. I have seen the use of different drivers have a bigger effect on the quality of that test than overclocking artifacts on some cards.

Agreed - the point sprites test always gives flickering dots and speckles at default speeds; has done so on any card I've tried - nVidia's Geforce2 GTS and Pro, Geforce4 Ti4200, and ATi's Radeon.
I run a loop using the high-poly test in 3dmark2001, The Lobby game, and the Advanced Pixel Shader test at high resolution and quality settings.
If you have a card capable of pixel and vertex shading, look for a demo called Codecreatures; for some reason it's not available at the company's website anymore, don't know why they stopped - too much bandwidth maybe. That thing will drag anything below a Radeon 9700 to a dead crawl. My Geforce4 Ti4200 running 300/567 only managed something like 7fps on average - that was at 1024x768, 32bit, WITHOUT antialiasing. It's really an intensive benchmark; something like that will strain your card and should force any flaws out into the open.

Just got a screenshot. Be warned, the file is around 1.2MB. The settings it used were: 1280x1024, 32bpp; 4xFSAA, anisotropic level 2. My videocard managed what looked like .5-1 fps with those settings.
The Polygon counts are obviously off; they weren't updating for some reason. The Polys/sec should be more like 1.8+ Mio (million I assume that means), and the Polys/frame should be above 250,000.
Link.
A very good demo; the water effects later in it are very impressive as well. If any games are based on this engine, an NV30 will probably be needed. Of course, by the time games of this level come out, we'll probably have NV40's.


Oh, the original issue - ramsinks. I did mine the cheap way - cut up 2 old 486 CPU heatsinks, and attached them with Arctic Silver II and epoxy. They work just fine.