Building an Affordable "Skulltrail" System

[Mike]

Fan Speed

If you are having that rough of a time in finding it, I'd be willing to bet it's not there. I looked at the screenshots from our review and can't find reference to it either :-/

Make sure that you have upgraded to the most recent BIOS. There have been at least two updates since the article was first written. I'm running version 1149, and it's there. I can't say exactly where at the moment since I'm halfway into a multi day process and cannot reboot. But, it's there.

Mike

 

[Toledo_Speedo]

Thanks Mike! I upgraded to BIOS v1175 and there it was! All speeds are @ 100% and it is running MUCH cooler!

Edit: I am running a Prime95 blend test as we speak @ 3.21GHz and no temps have hit 65C! Before turning off the fan speed control all temps would hit 75C+ except the southbridge.

Maybe 3.25GHz is possible.......

 

[balth_dire]

).

Conclusion: CPU/Memory interface (i.e. MCH a.k.a NorthBridge) not stable!
In fact, I suspect that even your 400 FSB won't make it through 24 hours of Prime95 blend, mine didn't.
There is a forum posting at HardForum from a guy who narrowed it down to the exact FFT size that causes the failure (3072k): http://hardforum.com/showthread.php?t=1362510


Mike

Hello, sorry to dig this little issue out of the archives again but there appears to be some confusion.

As you'll note I was the poster of the message linked to above on HardOCP and I believe I made mention of it some pages back on this particular thread (though anonymously). I have also raised the case on Mersenne's own forums, where a couple of other users have also encountered it.

To clear things up, the 3072K FFT crash is NOT caused by unstable hardware/overclocking/overheating or any physical fault with your system, it is a bug in the SOFTWARE that is somehow caused by Prime not allocating enough or mismanaging memory between all 8 CPU cores.

If you try running a custom test in Prime95 (goto options/torture test/select custom) and set the "memory to use" value to under 1550mb, you'll pass 3072K FFT everytime, and proceed onto 3584k and then 4096k, the latter being the final test in the program before it loops back to 1024k.

Up until Prime95 v25.6, the blend test was, by default, set to use 1750MB of memory, in 25.7, this was reduced to 1600mb, the reasons behind this are unclear, though they may well have something to do with the issue in question, the value remains the same in 25.8, the latest version.

Sometimes, if you start the blend test in version 25.7 (using the default "memory to use" option of 1600mb) it will pass 3072K FFT without crashing, but then it may crash at 3584K, or possibly 4096K, though occasionally, all the tests will pass without the crash at this setting (hence why some are reporting success) it all relates to how the memory is being allocated to each core during these final three tests in the sequence.


I've found the only consistent way of getting it to pass these three final tests is as stated, to run a custom test and set to use 1550mb or under (remember, this only applies to those running 8 core machines).

In any case, I'd advise using version 25.7/25.8, since these work best, even at their default settings.

There is no question that this is a bug since I have replicated the problem on two different Skulltrail boards, with two different sets of memory (from Kingston and Crucial, using both 2 and 4 sticks in each case) at overclocked/overvolted and stock settings, with 2 different PSUs and ample cooling etc. The problem was identical in every case - a crash to desktop at 3072K.

Moreover, I've just tested on a newely built Core I7 system (using a 965 chip, so again, 8 cores as far as Prime is concerned) and guess what....the EXACT SAME ISSUE!

Running the 3072K, 3584K or 4096K tests on their own results in the program crashing to desktop within seconds, unless "memory to use" is set to under 1550mb.

 

[Mike]

FFT Crash on Skulltrail

...
To clear things up, the 3072K FFT crash is NOT caused by unstable hardware/overclocking/overheating or any physical fault with your system, it is a bug in the SOFTWARE that is somehow caused by Prime not allocating enough or mismanaging memory between all 8 CPU cores.

If you try running a custom test in Prime95 (goto options/torture test/select custom) and set the "memory to use" value to under 1550mb, you'll pass 3072K FFT everytime, and proceed onto 3584k and then 4096k, the latter being the final test in the program before it loops back to 1024k.

...

I've found the only consistent way of getting it to pass these three final tests is as stated, to run a custom test and set to use 1550mb or under (remember, this only applies to those running 8 core machines).

In any case, I'd advise using version 25.7/25.8, since these work best, even at their default settings.
...

Excellent news. Thanks so much for following up. Based on so many other posts across multiple forums, I "knew" that my setup should have been able to do better, and I wanted to push the limits. Yet, I was getting shut down time and time again. Now you've given me the incentive to go through it all once more in pursuit of the perfect overclock!

'Gotta love the web. Complete strangers helping each other out at every turn. There's just something so right about that!

 

[Toledo_Speedo]

That is good news, if I would have known that I probably could have ignored a couple failed blend tests.

Here is something else that is odd: I am totally stable @ 3.21GHz but the NorthBridge was still hitting 65C+ so to see if I could cool it down I dropped all the voltages back to default (CPU, FSB & MCH). I was going to start bumping up from there until it stabilized but it is running fine, all temps are under 60C!

Is it normal to be able to run a .71GHz OC (2.5GHz Stock - 3.21GHz OC'd) without a voltage increase?

 

[Mike]

Overclock

...
Here is something else that is odd: I am totally stable @ 3.21GHz but the NorthBridge was still hitting 65C+ so to see if I could cool it down I dropped all the voltages back to default (CPU, FSB & MCH). I was going to start bumping up from there until it stabilized but it is running fine, all temps are under 60C!

Is it normal to be able to run a .71GHz OC (2.5GHz Stock - 3.21GHz OC'd) without a voltage increase?

Toledo,

Have you tried verifying your overclock with Intel Burn Test? http://downloads.guru3d.com/IntelBurnTest-v1.6-download-2047.html

IBT is a truely amazing piece of software. After wasting DAYS of stability testing with Prime95, I was able to get a solid overclcok in just a couple of hours! I would regularly go for hours at a time with Prime95 (over 12 hours some times) before failing. I even went over 24 hours on small FFT, which made me suspect my memory was the problem. Yet, memtest86+ went over 24 hours without issue. So, I was stumped.

With IBT the same settings would fail in under 5 minutes! I was able to very quickly identify exactly what was causing the instability (my memory could only achieve 4-5-4-12 @2.04v as opposed to the advertised 4-4-4-12 @2.00v). Once I got that under control, I worked on the FSB. Again, FSB settings that would run for hours and hours without issue in Prime95 would fail by the 8th iteration of IBT (about 5 minutes!).

I have the same CPUs as you. But, I had to raise my FSB to 1.30v, my MCH to 1.35v, my CPUs to 1.325v (I think) and my memory to 2.04v. At that rate I was able to pass 15 iterations of IBT (triple the recommended minimum) at a 3.11Ghz overclock. Those settings then breezed through Prime95 Custom (1550 as recommened in thread above) for over 18 hours before I decided to kill it.

 

[Rob Williams]

IBT is a truely amazing piece of software. After wasting DAYS of stability testing with Prime95, I was able to get a solid overclcok in just a couple of hours! I would regularly go for hours at a time with Prime95 (over 12 hours some times) before failing. I even went over 24 hours on small FFT, which made me suspect my memory was the problem. Yet, memtest86+ went over 24 hours without issue. So, I was stumped.

So IBT uses all eight cores without an issue on Skulltrail? That's great to hear, because currently, it won't use all eight threads on a Core i7 processor. It's frustrating, heh.

I agree, though. IBT for the win:

http://forums.techgage.com/showthread.php?t=4376

 

[Mike]

Don't need dual 8 pin power supply

This thread is becoming somewhat obsolete. But, just as a reference for any that follow. You don't actually need a dual 8 pin connector from the PSU, thus opening up the selection to include many more options.

With a single 12v rail PSU, you can use a "Y" splitter to convert a single 8 pin connector into a dual 8 pin connector. Alternatively, if the PSU has a single 8 pin and a single 4 pin, then a 4 pin to 8 pin adapter can be used to convert the 4 pin connector into the second 8 pin required by the board.

Also, I read report by a respected overclocker of
"2x5430's @3.2ghz, dual 8800gtx's in SLI and 2x raptor150's. Along with your normal DVD, and a slew 5x120mm fans" running on a PCP&C 750w Silencer (using 4pin->8pin adapter) which are now going for just over $100.

Big savings compared to the $400 alternatives Too late for my first attempt, since I bought the PCP&C 1000w when I did my build.

But, since I just bought another D5400XS/2xQX9775/2xZalmans/4x2GB FB-DIMM/fans/Lian Li case all for $1650 off of eBay (Whoohoo!). I'll throw the 1000w into that and buy a 750w Silencer for the box with the Xeon's. Add a small hard drive and a cheapo graphics card, and I've got myself another ultra powerful server! I smell a 4.4Ghz overclock coming

Now if you'll excuse me, I have to try and explain all this to my wife (Yikes!)...

 

[Rob Williams]

With a single 12v rail PSU, you can use a "Y" splitter to convert a single 8 pin connector into a dual 8 pin connector. Alternatively, if the PSU has a single 8 pin and a single 4 pin, then a 4 pin to 8 pin adapter can be used to convert the 4 pin connector into the second 8 pin required by the board.

Oh wow, that's a great idea! Can't believe I didn't even consider that in the past. Great tip!

Grats on the second build, as well. Good luck hiding out in your own house ;-)

 

[Toledo_Speedo]

Now that core i7 has been released I have been able to find new QX9775's for $500 and since I have so much wrapped up in my build already I bought one. My E5420's were running stable @ 3.21GHz but I have been able to push my QX9775 to 3.85GHz and stay stable. I am waiting to catch another one on eBay for cheap but I wanted to touch base and see if anyone has OC tips specific to Skulltrail?

 

[Kougar]

I've seen some Core i7 965's reaching $800 just because the 975 is around the corner and happens to be a D0 stepping. Gotta love the progress of technology... when you're not one of those being run over by it.

As far as overclocking Skulltrail goes, only thing I can say is be sure to keep the chipset cool. I wouldn't know anything else specific as it's a modified server motherboard, but the chipset gets hotter than most supposedly.

 

[Mike]

Skulltrail overclock

My second Skulltrail is using 2xQX9775. On air I got the overclock to 3.8Ghz stable through 10 iterations of Intel Burn Test (IBT).

Here's what I ended up with for the rig using the QX9775's:
vCore 1.35 @ 9.5 multiplier
vFSB 1.425 @ 400
vMCH Default
vDIMM 1.84 @ 5-5-5-18 4x2GB

I've since upgraded to liquid cooling on both CPUs (d-tek fuzion v2), the northbridge (danger den mpc-x38) and the FB-DIMMs (mips ramfreezer 4 - this was a @#$%# to customize and install for FB-DIMMs).

Whereas I could not post at 4.0Ghz on air (10x400FSB), I have no trouble posting all the way through 4.4Ghz (11x400FSB) on liquid. However, I have not been able to get through a single iteration of IBT

I keep getting a blue screen and my HWMonitor temps show a whopping 95C for "Remote 1" under each of the two "Analog Device ADT7490" nodes. Does anyone know what this is measuring? I want to add cooling for that component, but I don't know what it is!

- MyMachine

- Analog Device ADT7490​

- Remote 1​

- Local​

- Remote 2​

I don't have a waterblock on the southbridge since I only have a cheap $25 graphics card that never gets used (this is a server machine). Could the high readings be from the southbridge? It feels warm to the touch at load, but not "hot". The MOSFETs around the CPUs feel "hot"! Could the HWMonitor temps be for the MOSFETs?

I don't mind paying the $75 for a southbridge block if that is the problem. But, I don't want to throw money away if it won't help anything. If it's the MOSFETs, what options do I have?

Thanks,

Mike

 

[Greg King]

Hey Mike,

I can't speak too close to this platform as I have only seen it in person when with Intel. Rob however should talk your ear off so you might want to wait to hear from him.

One of the advantages of going the water route is that your cooling capacity is raised significantly (you know this). I do not know which component is getting too hot for comfort but out of the two that you mentioned, I would almost guarantee that it's the mosfets. When you removed your air cooling, you took away the air that flows across the heat sinks on the 'fets that came off of your CPU heatsink. Try rigging up a few 120mm fans (or whatever you have laying around for that matter) and direct their flow onto the mosfets. It's going to be ghetto as hell but if your stability improves, you ought to be able to work something up that isn't so unsightly.

When I had my last water system, I had to rig up a 120mm to keep the CPU's surrounding areas cool otherwise I would have been in the same boat that you are in now.

Good luck and post back you're experiences.

 

[Rob Williams]

Cooling is the biggest thing with Skulltrail. In talking to one of the performance nuts at Intel a few weeks ago, who's currently running Skulltrail at home, he told me he's running CoolIT's Skulltrail edition cooler. Nice, but the thing is damned expensive. Far too expensive. At that point, I'd rather buy a workstation board and a couple of the i7 Xeon's that came out not long ago.

Mike: If you can touch the Southbridge when pushing the CPUs hard and you still have your fingers afterwards, I'm going to jump to conclusions and say it's not the Southbridge. If the MOSFETs are extremely hot, on the other hand (50C+), you might consider getting some small heatsinks to place on each one, if you can find some that will fit. To be honest, that probably won't do much, but it might be worth a shot.

You might also want to consider somehow getting a fan in where the MOSFETs are and cool them that way. If they are too hot to touch, that's not a good thing.

Greg is right about the watercooling though. The Southbridge might not be the issue (and I'm almost guaranteeing it's not... the Southbridge handle's I/O, not the FSB), but adding water to it might help spread out the temperatures a little bit. I know when I added water to an ASUS board that offered a waterblock on the Northbridge, overclocking didn't improve, but temperatures did for both the Northbridge and the PWM area, even though that area was a good three or four inches away.

If all else fails, you could always just move to Canada and live outdoors.

 

[Mike]

Skulltrail/HWMonitor: southbridge or MOSFET?

Cooling is the biggest thing with Skulltrail. In talking to one of the performance nuts at Intel a few weeks ago, who's currently running Skulltrail at home, he told me he's running CoolIT's Skulltrail edition cooler. Nice, but the thing is damned expensive. Far too expensive.

If you're referring to the MTEC version, then yeah, that's just crazy. If you're referring to the liquid version (PURE ST), that's less than what I've already put into my liquid setup, thanks in large part to the $100+ MIPS RamFreezer 4. But, on the plus side, my FB-DIMMs are around 50C under IntelBurnTest full load

At that point, I'd rather buy a workstation board and a couple of the i7 Xeon's that came out not long ago.

Yeah, if I started over again, that's the way I'd go. I may yet try to unload my two Skulltrail systems to buy one dual socket i7 Xeon rig

Mike: If you can touch the Southbridge when pushing the CPUs hard and you still have your fingers afterwards, I'm going to jump to conclusions and say it's not the Southbridge. If the MOSFETs are extremely hot, on the other hand (50C+), you might consider getting some small heatsinks to place on each one, if you can find some that will fit. To be honest, that probably won't do much, but it might be worth a shot.

You might also want to consider somehow getting a fan in where the MOSFETs are and cool them that way. If they are too hot to touch, that's not a good thing.

The MOSFETs are not "extremely" hot, but they are definately hotter than all other components. I can hold my fingers to them for a few seconds. I tried using an 80mm fan that I had, and the dual 50mm OCZ ram fan, but they didn't change the temp readings.

I've been doing a lot of searching, and I think that I've found a pretty big clue. I came accross a build thread where the guy showed a big closeup of his board (too close to tell which part of the board he was showing) and right there in the photo was a chip with the label "ADT7490" on it! Now I've just got to look for that chip on my board

The only problem is that he may have removed stock heat sinks to expose the chip since he was adding liquid nitrogen. I've found a lot of data dumps too that mention this chip very near the words "SMBus" and "2ch". So, now I'm thinking that it's related to the the memory. I'll look for the "ADT7490" chips near there first. Since I'm using the MIPS RamFreezer, I no longer have the ram fan there. That might explain the rise in temps for "Remote 1".

If all else fails, you could always just move to Canada and live outdoors.

That's where I came from Still trying to convince my American wife to move back to Montreal with me!

Thanks for the comments guys. I really appreciate it.

 

[Rob Williams]

Mike: Yeah, that's the Skulltrail cooler I was talking about... I think it goes for about $600. Far too expensive. It'd be way cheaper to get the new Xeon setup like I said. They run hot too, but you don't feel like you need a $600 CPU cooler either. I'd actually be interested in seeing how those new dual-socket i7 rigs do for regular use... might be an interesting follow-up to our Skulltrail article.

Since I had my Skulltrail board handy, I hauled it out, dusted it off, and found two ADT7490 chips on the board... one around each socket. According to the website for the chip (URL), it's for the fan controller, so I don't think it's really that important for you to look at.

I think if any area on your board is too hot to touch, that'd be the part to focus on. The MOSFETs are likely too hot, even if you can hold your finger there for a few seconds. They can run warm to begin with, but I'm not sure that warm.

Ahh, Montreal... the city I've always meant to go partying in, but never got around to ;-)

 

[Kougar]

Got a link to the photo with the chip in it I could see?

You also might try checking for Speedfan profiles of your motherboard... someone may have already gone through the trouble of renaming the sensors and saving a profile to their database. Considering the motherboard in question it's likely, actually.

 

[Mike]

Skulltrail MOSFET

... might be an interesting follow-up to our Skulltrail article.

That I would love to see. I'm a little afraid to find out how much more powerful it is than what I've got, for probably less money

Since I had my Skulltrail board handy, I hauled it out, dusted it off, and found two ADT7490 chips on the board... one around each socket. According to the website for the chip (URL), it's for the fan controller, so I don't think it's really that important for you to look at.

Yep, now that I've got home, I've done the same thing. Your link is actually quite useful. It would appear to be the fan controller for the CPU. The local is probably for the chip itself, which is not hot to the touch. Remote 2 is probably for the CPU, which is also well under control. I'm going to make a leap and conclude that the Remote 1 is for the MOSFET serving that CPU.

I think if any area on your board is too hot to touch, that'd be the part to focus on. The MOSFETs are likely too hot, even if you can hold your finger there for a few seconds. They can run warm to begin with, but I'm not sure that warm.

Now that I've tested again tonight, I'll revise my earlier statement. When IBT is running, and Remote 1 is showing as 95C, the MOSFETs are very very hot. I burned my thumb .

I tried holding an 80mm fan directly over the MOSFETs, but it did not seem to make any difference in the temps. I tried it as push and as pull. Any difference was only two or three degrees, which may only have been the natural fluctuation. There's no room for a larger fan due to the lliquid tubing. As it is my finger got caught in the fan blades and broke two blades off (doh!)

The MOSFET heat sink is arched, so I can't just pile on another heat sink on top of it, unless I find one that fits perfectly upside down such that its blades slide between the existing blades. I may have room to attach a tall thin piece of metal to the side of the heatsink instead.

I read a blog post somewhere about a guy who pulled off the heatsinks to install Koolance MOSFET waterblocks, only to discover that the surrounding chips were taller than the MOSFETs, so he had to place a thin piece of copper on top of the MOSFETs to bring them to the same height in order to lay the waterblock evenly accross. I had to do the same thing to get my FB-DIMMs to work with the RamFreezer (controller chip in center of card is higher than the DRAMs) and let me tell you; it was a huge pain in the butt. May be my best option though.

Thanks for the detective work.

Ahh, Montreal... the city I've always meant to go partying in, but never got around to ;-)

Definately worth your while

Mike

 

[Mike]

Skulltrail MOSFET

Got a link to the photo with the chip in it I could see?

You also might try checking for Speedfan profiles of your motherboard... someone may have already gone through the trouble of renaming the sensors and saving a profile to their database. Considering the motherboard in question it's likely, actually.

Post #76 of this thread, second photo down, top right corner of photo:
http://www.xtremesystems.org/forums/showthread.php?t=180267&highlight=skulltrail&page=4

As Rob pointed out, the chip is in the immedate perimeter of the CPU. In this case, it is right below the rightmost CPU on the board.

I could not find any SpeedFan profile for Skulltrail or D5400Xs. But, thanks for the suggestion. I am definately going to focus on trying to cool those MOSFETs.

Mike

 

[Kougar]

Ah, Rob to the rescue! Glad to see ya both tracked down the chips. Thanks for the image link.

I'm sorry to hear ya didn't find any Speedfan profiles, that's surprising. Anyway, regarding the chips ya could probably slap some Enzotech BMR-C1 or C1L chips on them. Perhaps the same with the mosfets depending on your setup, the L is the low-profile version. These things are expensive but I use them on my watercooled GPU's memory and I'm quite happy with them.