Here's asummary of my similar experience. My issues were a lttle complicated. Hopefully this wordy diatribe will offer up something of help...
My GTX670 is the EVGA SuperClocked 4GB w/ Backplane (NVIDIA driver version 126.96.36.1990), It looks like a reference card. Not sure if brand or model makes much difference with our board since it follows the NVIDIA reference design for the most part, but it is generally touchy with these Keplers. My BIOS is v0503.Fixing Posting and GPU Issue
After putting my beast together I also had problems getting my board to post. Not seeing any 64GB kits out there I had purchased two 32GB Vengeance RAM kits. I assumed that they would play well together. At the time I was clueless as to the cause of the posting issue. But by swapping RAM sticks in positions1 & 2. I was able to get posting to happen; but then my monitors only came to life maybe 2 out of the next 10 attempts and then not at all. My monitors kept complaining of no HDMI signal. I put the GTX670 in various PCIe slots with no better results. EVGA helped me. The problem there turned out to be the fact I had tightened my Dynatron CPU coolers too tight. Evidently the excess tightness was warping the CPU seats on the board just enough to cause PCIe anomalies. So I loosened them up to the point that they weren't pressuring the CPUs at all; but not so much that there was play between the CPUs and seats. I then tugged very gently on the CPUs to equalize pressures and re-tightened the coolers a small amount. My system posted, booted into Windows, and my monitors worked!
After the scramble to get it to post, my GPU had ended up in slot 3, so I moved it to slot 5 where I had planned to put it for better cooling distribution and because it doesn't switch down to 8x based on the presence of other cards in even numbered slots. Further tests confirmed that everything was still working.
Now at this point I was thinking that my problems were only due to CPU seat warping causing issues with the PCIe and RAM channels. I hadn't realized that I had also been dealing with some intermittent problems in the RAM itself.Further Problems
Some days later as I worked to optimize my system and apps, I began to experience corruption in the OS, apps, and files. It wasn't nearly as stable as I'd have liked. But I was so anxious to see how this beast performed by bench-marking and crunching video & 3D. After a few tests Photoshop got corrupted somehow and required a re-install. 3DsMax wasn't working at all and no number of applied patches fixed the issues totally. I was suspecting RAM problems but wasn't certain. Occasionally Window was getting corrupted and I had to revert to earlier system states and/or an older image. Something was just not right. As I continued to try work things on a software settings level, I downloaded some benchmark/diagnostic programs.
Sandra crashed when I tried to view memory and 3DMark11 would run but wouldn't post results up to their website. FutureMark said that the non-posting was due to bad RAM or a bad boot drive. Piriform Speccy indicated just half of my RAM capacity in slots 1 and 2. When I looked in BIOS I got the same capacity readings. So I swapped sticks1 and 2 again. Surprisingly it booted (considering this was the config I had when it wouldn't post originally). BIOS indicated that it had all of its capacity back; but strangely enough Speccy still listed only half capacity on those slots. I tried 3DMark11 again and still no results posted.
So I created a MemTest boot disk and ran it overnight to really give it a workout and to get an overall view of RAM health. In the morning I was stupid and forgot that I had turned off my monitors. So I thought that it had crashed. Of course when I pushed the boot button nothing happened. So I shut it off using the back switch. It booted and I realized my monitors were off. I was able to check system logs to see if MemTest had saved a log, but as I explored Windows crashed. Then it wouldn't even boot. Attempts at reverting to earlier states did nothing and it BSOD'd. Crap! At that point I suspected I had pushed any marginal RAM into failure and that I most likely had corrupted Windows beyond repair.Stable Machine
Before wiping my boot drive and reinstalling the OS/apps, I knew I needed to verify the RAM as being a cause of instability, corruption, and my 3DsMax issues. Using my MemtTest boot disk I was able to test each RAM stick one by one without Windows, and as I was doing so I called Corsair. They told me that there were two potential issues: 1) Mixed 32GB kits, and/or 2) bad RAM. He explained that when mixing kits there are very likely compatibility issues at the advertised speed/frequency/voltage settings. That's because there's a lot of variation between sticks. And stable faster kits are assembled through rigorous compatibility testing/selection. Different kits are never tested/verified to work together at those high levels. He indicated that I should I should have purchased a single 64GB kit. His go forward suggestion was to first finish my MemTests. Any bad RAM sticks would be replaced by Corsair with a new kit. And if no RAM was found to be marginal or bad, I could fix compatibility issues by lowering the common denominator. That would mean upping the voltage and/or reducing speed/frequency. It turns out that I had 2 very bad sticks in one kit and one fairly bad one in the other. I was in luck - they were obligated to send replacements for both of my kits. But they were courteous enough to replace the two 32s with a more pricey single 64GB kit instead. I wiped my boot drive and reinstalled everything. I let BIOS auto-detect voltage and latency, but had to manually up the speed to the rated 1600. That solved all of my problems. Things are now stable. It's now a great machine. Most of it's snappiness is due to launching of apps from the SSD boot drive. The processor clock speed of my e5-2630s is high average. Switching between apps and my browser I noticed that focus takes longer to come to the browser than with a lot of single CPU units - there must be a lot of handoffs in such a complicated system; but heavy duty processing tasks are way faster.As for the GTX670 GPU...
3DsMax works (with viewports set to Direct-X 10). 3DMark11 is primarily driven by the awesome GTX670 and works well. Adobe Apps work well, including GPU acceleration of Premiere's "Mercury Playback Engine" that relies on heavily on technologies like CUDA in the card. I just needed to hack it's list of accepted GPUs to be able to select it in the preferences. And simultaneous multitasking/rendering is acceptably efficient with the 12 cores and 64GB RAM it's got. I'm assuming that After Effects uses the GTX670 as the primary processor when I enable "Use OpenGL Renderer" in "Render Settings". I've noticed that it's a touch faster than without it selected; but also that particular layers in my complicated composites don't show up in the rendered output. So some further research and experimentation with app performance settings might be in order if I think that I need the flexibility.My Advice
Not knowing what you've done or not and what you know or don't know... I don't wish to insult anyone's intelligence. Given that this board seems haunted at times or at least hormonal, this is what I would do taking each subsequent step if posting doesn't happen:
- Clear RTC RAM. Make sure jumper ends up in correct position
- Check and verify that CPU coolers are minimally tight
- Reduce PCIe occupation to only the GPU. Try GPU in odd-numbered slot 5 or 7 (both are 16x, driven by CPU2, and not subject to speed switching complications)
- Try GPU in odd-numbered slot 1 or 3 (both 16x/8x, driven by CPU1, and subject to speed switching complications)
- Reduce RAM to a single stick in slot DIMM_A1 (the leftmost slot for for CPU1)
- Swap alternate RAM sticks into DIMM_A1
- Swap GPU into the alternate PCIe slot (5 or 7)
- Reduce to most minimal config: only one CPU, one stick of RAM, and one PCIe card (GPU).
Single CPU must be in seat CPU1 (upper right), Single RAM stick must be in slot DIMM_A1 , and GPU must be in odd-numbered PCIe slot 1 or 3 (where it's driven by CPU1).
- Try swapping in other RAM sticks into DIMM_A1 and moving GPU to the other odd-numbered slot
If you get it to post at most minimal config, then populate with additional RAM (from one kit). Begin by making a 2x config - add DIMM_B1 and test it, Then make a 4x config by adding both DIMM_C1 & DIMM_D1 and test. If that works, try adding the second CPU.
Good luck!Edited by BlackenedTush - 2/25/13 at 5:00pm