I'm new to overclocking (never tried it until the other night). Read through the clunk's, graysky, and wolfram beginner's guides and got a good idea of it. Mainly followed Clunk's, I was overclocking fine when it came to booting, temps, and stability tests, until I realized the system wasn't stable. There would be mouse jitters, small lags... and furthermore, there would be huge FPS drops in 3dmark.
Anyway, I am doing it on my 4-year-old build:
4x1GB G.SKILL DDR2-800 RAM F2-6400CL5D-2GBNQ (1.8-2.0V, 5-5-5-15)
Asus P5K-E WiFi AP mobo
EVGA GeForce 9800 GX2
COOLER MASTER Real Power Pro RS-750-ACAA-A1 750W Power Supply
Now this sounds like an entire noob question with a simple answer, but I have spent hours googling and couldn't find one. I set my BIOS according to clunk's guide. I set my ram timings manually to 5-5-5-15 and ram voltage to 1.8. I set my FSB RAM to a 1:1 ratio, for mine that is 333:667.
Here is the problem: When I overclock my FSB up to to 399 MHz, (DRAM 799 Mhz), I can benchmark in 3DMark and the FPS looks normal. But as soon as I bring the FSB up one more interval to 400 MHz (DRAM 801 MHz), the FPS drops severely in 3Dmark, and the final score is about cut in half. The CPU score raises, but the SM2.0 and SM3.0 scores severly drop. Also I notice some jittering in the mouse, lagging while trying to scroll down fast in a browser... which, according to what I've read, sounds just like basic symptoms of an unstable overclock.
I had started initially overclocking it far higher, to around 3.6 GHz, with boots, temps, and 8 hours of stress testing OK. I thought I had achieved a very successful overclock, but it was the 3DMark FPS that clued me that something was wrong.
Things I have tried:
I thought maybe it was the PCI-E bandwidth messing with the graphics card, although CPU-Z never said 1x instead of 16x. I tried setting the bandwidth to 100 and that did not help. I then set it back to auto, and tried increasing the RAM voltage to 1.9V. Did not help.
Before bringing the clock back down a notch, I went ahead and ran Memtest with the 400:801 set. I ran it overnight (about 7 hours) and had no errors.
I've heard of people getting to 3.6 and above with an E6750... Do I just need faster RAM? I know people clock their RAM faster than it's rated all the time, so I don't see how 1 Mhz over is causing the 3dmark score to immediately cut in half.
More things I have tried:
I tried changing the strap from Auto to 400, which brought the RAM down to 800 MHz instead of 801. Still had the choppiness. I then set the strap back to Auto, but this time adjusted the timings to 6-6-6-12 instead of 5-5-5-15 (rated).
So with the timings at CL6, went ahead and tried the 3Dmark benchmark again and it ran smoothly.
So now I figured lowering the timings would allow me to clock up the FSB, and in turn push the RAM clock above 800. I tried 420:841, and then ran the benchmark. Immediately I could see the huge performance drop, so I rebooted and tried pushing the RAM voltage to 1.9V, but that didn't help. Set it back to 1.8V, and brought the FSB down to 410:821, then 405:811, then 402:805, and finally 401:803. Every one of these had the same severely reduced performance in the benchmark.
I then set it back to 400:801 (while maintaining the CL6 timings), which worked earlier as I had mentioned... and oddly enough, I had the same poor performance issue on this run. I did not expect that at all, as it worked a couple of hours ago.
So to recap, with 5-5-5-15 or 6-6-6-18 timings, 399:799 runs the benchmark smoothly, and 400:801 or anything above cuts it in half.
I have no idea what is causing this!
I'm pretty sure I've covered everything in excruciating detail, but if I'm forgetting anything, please let me know. Thank you!
As you can see, only diminished performance on the CPU at 400 is the "Mandelbrot Parallelized." On the global benchmark, The Cache and Memory Performance, Memory Performance, and Video Performance drastically decrease. This is all with the 9800 GX2.
Bump.. any ideas? As soon as the RAM hits the rated speed or above, the graphics card performance is severely diminished. Has anyone experienced a similar problem where somehow overclocking the CPU or RAM to a certain point reduces the graphics card performance by an extremely large amount?
It's not completely clear to me from your post but you should lock the pci-e bus to 100 mhz. Definitely don't have it on auto.. It sounds like you originally have had it on auto but then tried 100? Just keep it locked to 100 to remove that variable.
If it's on auto it does also overclock the pci-e bus a small amount, eventually that could cause instability if it's high enough.
I had tried setting it to 100, but it did not fix the problem.
But, I have found something that might bring us closer to a solution:
I just figured out for sure it's not the RAM. I set the RAM divider to 800, and set the FSB back to stock speed of 333, making the RAM run at 889 (333:889). The performance issue does not happen. So the RAM can be overclocked. So this means the performance issue with the graphics card only happens when the FSB is clocked to 400 and above. FSB<399, no graphics issue. FSB>=400, graphics performance issues and jittery computer. Does that shed any light on the issue?
By FSB voltage, do you mean the CPU voltage or the NB voltage? I have kept them both to Auto the whole time, believing that they would overcompensate and not underfeed. CPU VCORE runs around 1.41 (i have to double check the hundredth decimal when I get home).
I'm not sure what NB voltage runs at while on Auto... is that one of the VIN numbers in HWmonitor?
It was the NB voltage the whole time. I set it to 1.4V manually, and the 3Dmark graphics card ran smoothly. After that I raised the FSB pretty high, to about 450, and got a BSOD. I jumped so high because before I noticed the graphics problem, I was passing stress tests around 3.6 GHz. My guess is without the NB voltage manually set, it probably never was really reaching that level. Because with the NB voltage set, as soon as I got in the 430 range, I needed to manually set vcore, as auto stopped working for that setting as well.
Anyway, I got my system stable at 440 with vcore set to 1.4850 if I can remember correctly, as I woke up to an orthos stress test with no errors. CoreTemp's max was in the high 60s, while HWMonitor's was 70 for both cores and RealTemp's 71. They always read higher than CoreTemp (and if anyone knows which one to trust the most, feel free to chime in). Thanks for all your help and input; I'm excited to see how much performance I can squeeze out of this thing. Now I'm gonna have to deal with figuring out when it's the CPU voltage, CPU PLL voltage, FSB termination voltage, NB voltage, or even if the RAM voltage needs to get just a tenth over rated for really high speeds.
I'm assuming that not getting the graphics error as I continue to OC doesn't rule out that I need to increase the NB voltage levels though, right?
It's funny, I had never been so happy to see a BSOD before. The graphics card problem was preventing me from pushing the chip to anything that required voltage tweaking (>400 FSB). So I hadn't gotten one BSOD since I started OCing this thing. They started to piss me off right away on my way to a stable 3.520 Ghz, though. :-)
Just remember that once you push your ram past it's rated speed it may require more voltage and/or looser timing. My last 775t board's NB heatsink would get so hot I couldn't touch it even with a 120mm fan pointed at it.