So reducing the memory bandwidth, whether that be by reducing the RAM/QPI multiplier(s) and/or the base clock, and/or by widening timings, would reduce the heat produced there, and reducing any voltages associated with memory (i.e. IOH, ICH/IOH) would help significantly.Were it me, I'd start with a very aggressive CPU cooling solution (water with a high-end CPU block and at least a 360mm x 120mm radiator), maybe including an additional external thermal sensor, and I'd reconsider how hard I pushed the memory bandwidth limits.  I don't think the risk of cooking a $1000 CPU is worth pushing memory bandwidth to extremes...added memory bandwidth produces relatively modest performance increases at CPU speeds below 4.5GHz.  And remember that Intel's design spec for the EE chips is only for 1333 MHz RAM.CheersBob ScottColonel, USAF (ret)ATP IMEL Gulfstream II-III-IV-VColorado Springs, CO

I agree.  A cheap sealed cooling system like the Corsair H70 is the ticket for me.  It doesn't need to be so hard.  But as long as people are asking questions - well that is a good thing. IMO jja

awesome piece of information Bob. Thanks a lot. I also overclocked my former AMD setup (my first OC experience) trying to max out memory speed, I know that I'm not stressing my IMC that much with a single card, but with my new I5 I'll relax memory OC as much as I can.two questions please:1.- would reducing QPI freq, and memory speed & latencies help get a better CPU overclock?2.- how slow do you think memory can be before it affects performance in FSX? I know it might be hard to tell. I'll be testing that myself .PS: Mike, hope you overcome the difficulties soon and you are back simming ASAP

Dazz; Mike's setup is pushing the limits in several ways--2 GHz RAM at tight CAS 7 timings plus two monster video cards being pushed by an i7 quad CPU overclocked more than 33% to 4.3 GHz. And I don't know what voltages, particularly IOH, he was using. Hopefully he can fill us in. But in a moderately-overclocked single GPU system with RAM at 1600MHz or below, I doubt you can get close enough to the wall to worry about this. With extremely aggressive BCLK settings (200+) I do begin to worry about the memory controller. Reducing memory timings or the memory clock multiplier is likely to be necessary when overclocking a CPU that has a locked CPU multiplier, because the only way to overclock is to push up the base clock (BCLK), and the memory freq is directly proportional to that. So if an upward adjustment to BCLK ends up pushing the RAM past its limits at the stock multiplier, you can slow the RAM back down either by dropping the multipler (a very coarse adjustment because mem mult is only adjustable in increments of 2) or by loosening the timings. Or you might try dropping the mult and then tightening the timings at the lower mult so the bandwidth drop isn't quite so big. For example, if you push BCLK to 190 MHz with a locked CPU mult of 21, you'll get a CPU freq of 3.99 GHz, but at the stock memory multiplier of 10 your memory will be trying to run at 1900 MHz (10 x 190). You can drop to a multiplier of 8, which will give you a memory freq of 1520 MHz, and if you're using 1600MHz RAM, possibly gain some of the lost performance back by trying to tighten the timings. As far as how far you can back off memory performance before affecting FSX...it depends. On a well-balanced high-performance system that is making good use of the available memory bandwidth, any drop will have some effect. But in a system where there's a CPU or GPU processing bottleneck, some of that available memory bandwidth may be going to waste anyway, so backing down the RAM won't have much of an effect until you get down to where you start slowing down the bottlenecked subsystem. It's going to be both situation and system-dependent. One good effect of slowing down the RAM is that you might be able to drop the DIMM voltage at the lower speed, making your RAM run cooler (and presumably last longer). In my simming PC, I have preset various speed/voltage/timing configs in the BIOS for 3.33 GHz(stock speed) with RAM at 1333 MHz, 4.0 GHz w/ RAM at 1600MHz, 4.4GHz with RAM @ 1600 MHz, 4.4 GHz with RAM at 1708 MHz, and 4.55 GHz with RAM at 1708 MHz. When I am doing maintenance and other routine stuff, I run the machine at stock speeds with the RAM derated to 1333 MHz. Most of my simming is at 4.4GHz with the RAM at it's rated 1600 MHz/CAS 6, and if I'm running a really complex FSX profile (i.e. PMDG B747 into NY area in bad wx) I'll use 4.55 and 1708 CAS 6. Based on Mike's experience, I wouldn't likely push my RAM at 1708 MHz if I ever added a second video card. And to Mike...sorry about your dead system. The problem with being on the forefront is that sometimes you get to be John Glenn, and sometimes you get to be Gus Grissom. At least in this case the only thing burned up was an expensive CPU. But thanks for sharing your (painful) experience so we can all discover where the limits are.RegardsBob ScottColonel, USAF (ret)ATP IMEL Gulfstream II-III-IV-VColorado Springs, CO

As far as how far you can back off memory performance before affecting FSX...it depends. On a well-balanced high-performance system that is making good use of the available memory bandwidth, any drop will have some effect. But in a system where there's a CPU or GPU processing bottleneck, some of that available memory bandwidth may be going to waste anyway, so backing down the RAM won't have much of an effect until you get down to where you start slowing down the bottlenecked subsystem. It's going to be both situation and system-dependent.

that sort of explains my tests results with my former AMD CPU limited set up. I'll be repeating them with my new intel I5 750.thanks again Bob

Just thought I would add to this thread, instead of starting my own, by sharing my experience over the last few days. I also bought SLI EVGA GTX 480s to run my 3 26" monitors, and everything was running fine in FSX (a lot better than the 5870 eyefinity) when I BSOD'ed. However, a few hours later my PC made a large crackling sound and immediately shut off. While turning on the PC a second time, large sparks flew out the back of the power supply! I was running a 1200w Thermaltake. I took the PC to a local repair place and they diagnosed only the PSU went bad, as the PC started with a new PSU and the post test was successful. Does this rule out any additional damage? No components on the motherboard looked fried. Do any experts here believe the PSU went bad, or something worse is at play here?My specs:QX9770 @ 3.6 Ghz (9x400)EVGA 780iG.Skill 4GB DDR2-1066

Just thought I would add to this thread, instead of starting my own, by sharing my experience over the last few days. I also bought SLI EVGA GTX 480s to run my 3 26" monitors, and everything was running fine in FSX (a lot better than the 5870 eyefinity) when I BSOD'ed. However, a few hours later my PC made a large crackling sound and immediately shut off. While turning on the PC a second time, large sparks flew out the back of the power supply! I was running a 1200w Thermaltake. I took the PC to a local repair place and they diagnosed only the PSU went bad, as the PC started with a new PSU and the post test was successful. Does this rule out any additional damage? No components on the motherboard looked fried. Do any experts here believe the PSU went bad, or something worse is at play here?My specs:QX9770 @ 3.6 Ghz (9x400)EVGA 780iG.Skill 4GB DDR2-1066

You're probably OK if it POSTs and boots OK. But I'll tell ya, I'd sure want to know how a 1200W PSU got overloaded to the point of a meltdown before I fired it up again in that config.Was the Thermaltake a single rail PSU? If not, is it possible that one or both of those cards was over or near the rated current for the rail it was plugged into? Some of the multi-rail PSUs have two (or three) high-current rails and a lower-rated one that usually powers the mobo and accessory pigtails. I highly recommend a quality single rail PSU, and if not, I'd definitely want to know what the current ratings are for each pigtail coming out of the box, if I had to call the company and talk to their engineering guys to do it.At the very least, I'd be watching the 12v rail voltage(s) like a hawk when I fired it up again.You're the fourth person I've heard of in the past two weeks that has had some kind of "fire in the hole" running two 480s in SLI. I wonder why...RegardsBob ScottColonel, USAF (ret)ATP IMEL Gulfstream II-III-IV-VColorado Springs, COEdit: I just looked up a Thermaltake W0133RU 1200W PSU. It's a four-rail box :excl:, with 2 x 20A (one is connected to the mobo pigtails) and 2 x 36A 12v rails. If you were using that PSU and had one of the GPUs plugged into the 20A instead of a 36A rail, that might explain the fireworks, especially if those are peak, rather than continuous, current ratings... My advice: get yourself a good single-rail PSU.

Hey Bob, Thanks for the reply. It actually had 4 12V rails (http://www.newegg.com/Product/Product.aspx?Item=N82E16817153054). Pretty much the only thing in the system, apart from the GPU's, is the QX9770 with the mild overclock. I figured that the system would top out at 800-1000w max. It was unfortunately two years old, but I doubt old age was the cause. Are there any tips for when I receive my new PSU to ensure that I do not end up with another decimated PSU? =P

Hey Bob, Thanks for the reply. It actually had 4 12V rails (http://www.newegg.com/Product/Product.aspx?Item=N82E16817153054). Pretty much the only thing in the system, apart from the GPU's, is the QX9770 with the mild overclock. I figured that the system would top out at 800-1000w max. It was unfortunately two years old, but I doubt old age was the cause. Are there any tips for when I receive my new PSU to ensure that I do not end up with another decimated PSU? =P

See the edit below my sig in the post above. If you're going to use the same PSU (I'd reconsider), make sure you're plugging the GPUs into the two 36A rails, and not one 20A and one 36A.Don't underestimate the combined power draw from other system components...mobo, HDDs, fans, etc. And the total system load isn't the critical issue...it's how much of the current load is connected to each rail. That PSU is essentially four smaller PSUs in one box, rated at 1200W total "maximum" power...I read "maximum" as weaselwording for peak power, not continuous. So I don't think you have nearly the current capacity that the 1200W number on the box might suggest.Contrast that to a PC Power and Cooling 950W Silencer II--it's rated to supply a continuous 83.4A on a single common 12v rail, so there's no issue of how to split the load amongst a set of cheaper, low-to-medium current busses.CheersBob ScottColonel, USAF (ret)ATP IMEL Gulfstream II-III-IV-VColorado Springs, CO