Building an Affordable "Skulltrail" System

[Mike]

Northbridge heat problem

One big problem right now is the MCH (Northbridge) overheating...

I am thinking this is the main problem, as all other temperatures are quite moderate. I have a fan on the heatsink, but even still touching the heatsink it is extremely hot. What coolers are available to keep this at a reasonable temperature? I have looked around but am not too sure what would be the best solution. Should I put liquid cooling on this? Or would a third party heatsink/fan solve the problem?

How would you go about taking the stock heatsink off the board?

Hi,

Any 40mm fan will fit it the included mounting bracket. I've just ordered one too. Additionally, I've got my coolers (Thermalright Ultra 120 Extreme with add-on fans) oriented in the same direction as the front and rear fans for a continuous air stream corridor which crosses right over the northbridge. You can also try "lapping" the heatsink. Google for "lapping heatsink" to get an idea. Basically it means sanding down the base of your heatsink for better contact. There's even a kit available if you can't be bothered to shop for the materials yourself (http://www.frozencpu.com/products/2888/lag-01/FrozenCPU_Deluxe_Heatsink_Lapping_Kit.html).

I've read reports of over 5 degrees improvement (for CPUs) after lapping.

Mike

 

[Mike]

Memory

I admit I'm not much of an overclocker to begin with (I just overclock the CPU and be done with it), but I wouldn't be too concerned with latency when using four modules. It does increase it a bit, but I'd be hard-pressed to find someone who'd notice it in real-world scenarios.

The best thing would be to just use 2x2GB, but if you need 8GB, the latency hit is going to be moot.

Well, I ended up going with the 4x1GB. I had no problem at all with a FSB of 400 and changing the memory latency to 4-4-4-12 as advocated by Kingston. I ran the same thing at FSB 425 without issue until the memory would overheat after about 20 minutes and the machine would shut down. That was without a RAM fan. I've just invested in a bridge of fans to mount over the RAM. I'm hoping that that will allow me to stay somewhere closer to 425. The Xeon 5420's had seemingly no trouble at all at FSB 425. Just in case, I've also ordered a fan for the northbridge.

 

[Mike]

Coolers and Cases Don't Always Mix

Hi Mike:

I chose the Zalman cooler for that article because it's the same model that Intel sent with Skulltrail for our review, so I assumed they liked it for a reason. The Thermalright cooler is better though, I'll admit, and we'll be using the TRUE Black version going forward for both our Core 2 and Core i7 content.

When it comes down to it, if you had to return the Zalman's, I'm not sure it'd be worth the time, but I do know that the Thermalright's seem to be better-respected. Bill here recommends them far above the Zalman's.

As a precautionary note to those that may follow. Don't combine the Thermalright Ultra 120 Extreme with the Thermaltake Armor case having a side panel intake fan. The added dimension of the fan pushes against the top of the coolers, causing the motherboard to dip just enough to contact the backside of the case and short circuit. Doh! I didn't even realize that that was the cause until I tried installing the replacement board (thank you Intel) and paid *extra* special attention to the whole process.

 

[Rob Williams]

Good to hear things are going well with your build, for the most part. That is one big downside of chassis' with huge fans in the door... not good at all. Oddly enough, my mom of all people is using a Tt Armor with that exact cooler, but only on a single-CPU board, and she doesn't have an issue there. I guess the locations of the CPUS on dual-socket motherboard complicates things a little more.

As for the fans... I'm glad I'm not alone with overheating DIMMs. It's a little ridiculous to be honest... at stock speeds, RAM shouldn't overheat like that. We are not dealing with DDR2-1200 speeds here, but a modest DDR2-800. Intel should have stuck to regular DDR2 there, without question.

 

[Mike]

Side Case Fan

Good to hear things are going well with your build, for the most part. That is one big downside of chassis' with huge fans in the door... not good at all. Oddly enough, my mom of all people is using a Tt Armor with that exact cooler, but only on a single-CPU board, and she doesn't have an issue there. I guess the locations of the CPUS on dual-socket motherboard complicates things a little more.

Yup, the heatpipes on the cooler over the western CPU get pushed down by the fan framing. The cooler over the eastern CPU fits between the framing and is not a problem (though might interfere with the fan blades if not for the fact that the western cooler were preventing the fan from being lowered all the way).

As a quick work around, I mounted the fan to the exterior wall of the case, rather than the interior wall. It now looks like something belonging to the Borg out of a Star Trek film, but it works. If I find the time I may try to play with different positioning of the cooler to see if I can get it all to fit.

Depending on how high the Corsair CMXAF1 fans stand above the RAM, I may need to keep the side fan outside anyway I'll find out soon enough.

 

[Rob Williams]

Is it just me, or is it a lot of work to keep such a beast cool? When we first learned of Skulltrail, we knew that extra cooling consideration would have to be taken in, but man, this is ridiculous! I haven't had our Skulltrail rig hooked up for a few weeks, but winters almost here, so I'm almost considering hauling it back out!

 

[Mike]

Skulltrail Cooling

Is it just me, or is it a lot of work to keep such a beast cool? When we first learned of Skulltrail, we knew that extra cooling consideration would have to be taken in, but man, this is ridiculous! I haven't had our Skulltrail rig hooked up for a few weeks, but winters almost here, so I'm almost considering hauling it back out!

Heatsinks:
- CPU x 2
- Northbridge
- Southbridge
- Voltage Regulator Circuit x 2
- RAM heat spreaders x 4

Board Fans:
- CPU x 2
- Northbridge
- Southbridge
- RAM x 3 (single unit of 3 fans)

Case Fans:
- Front intake
- Side intake
- Top exhaust
- PSU exhaust
- Upper rear exhaust
- Main rear exhaust

Since I'm having so much fun, I may venture down the road to liquid cooling next

 

[Rob Williams]

Hah, that's a lot of cooling. Liquid cooling would interest me greatly, but it would take quite a bit of effort. As long as time permits, I think it'd be a great choice though. Quieter and cooler... can't beat it.

 

[Unregistered]

I've been in contact with Intel over a few issues I'm experiencing with my Skulltrail build. I'm using the same case as Rob (Tj10), but with an added 120mm side fan.

In short order, I've been having problems getting the machine "Prime blend test" stable at stock speed (using version 25.7). When using Windows XP, the system would freeze anywhere between 3-6 hours into the test. No rounding error reported, just a straight up system hang.

After days of tweaking I discovered that increasing the VCore on both CPUs to 1.2875 (which the lowest manually selectable option) would allow the test to reach 9 hours whereupon Prime itself crashed to desktop. I subsequently discovered that the latter crash is actually related to a memory allocation bug in the program and not a hardware flaw, the freezing however, does seem hardware related since the Vcore increase appeared to have a direct affect on the system's ability to run the test.

However, to further complicated things, two days ago, I tried running the test in Vista (x32), with all voltages reset to their defaults but with the CPUs overclocked to 3.4ghz and this time it got to 20 hours before I manually aborted it...very strange.

At the risk of straying from the topic of the last few posts, one fundamental issue I did encounter was, as others on the have discovered, with the Northbridge. Active cooling should have been deemed essential by Intel, not optional. I initially had spontaneous reboots within only 30 minutes of running the stress test and I'm nigh on certain they were down to the Northbridge overheating. Without some form of active cooling, either via a 40mm fan and/or pronounced cross flow from the CPU coolers, the Northbridge will overheat under load (frequently topping 80 degrees), regardless of whether one is overclocking or overvolting.

I'd recommend a dedicated 40mm x 20mm fan with airflow of no less than 6cfm, this should be positioned to blow down onto the heat sink and not draw air off it.. Moreover, If you're using the onboard header, be sure to disable system fan control in the BIOS otherwise it prevents the fan from ever running at full speed, even when it needs to!!!

Throughout all this testing I've tried 2 different boards, two different sets of memory (from different vendors) and 3 different PSUs. Pretty much the same symptoms were manifest with each and every combination.

If anyone reading and/or contributing to this thread has further advice or has had similar experiences with their setup, please let me know as I don't feel I've quite solved this mystery yet.

Spec is

D5400XS

2x Qx9775 CPUs (cooled by Zalman 9700-nTs - I have the front cooler - for CPU 1 - positioned with the fan facing downwards pulling cool air upwards from the front case fan and the rear cooler - for CPU 0 - with the fan facing forwards pulling air from the middle of the case backwards towards the rear case fan, found this gave me the lowest temps overall.

8GB Crucial PC6400 FB-DIMMS (2 kits - CT2KIT25672AF80E)
4x WD Velociraptors
RocketRAID 4320
Soundblaster X-Fi Elite Pro
2x BFG GTX 280 OCX Cards

 

[Toledo_Speedo]

THANK GOD!!!!!!!!!! A dedicated SkullTrail Thread with active MODS! I Just RMA'd my D5400XS board because it had 2 dead USB ports and I am waiting to get it back from Intel. I didn't put all the components together in case more than just USB's were bad but here is my set up:

CM Stacker 830, SkullTrail Mobo, Dual Xeon 5420's, Velociraptor 150HDD & a GeFORCE 8800GTS GPU.

I am not using it for gaming just Multi-Threaded apps related to Electronic Data Discovery. From what it looks like I shouldn't have any trouble with the E5420's even though they are not on Intel's supported CPU list.

 

[Toledo_Speedo]

You got it... they want to sell more higher-end processors. Anything from their 5xxx line-up will work fine. Just make sure it's LGA771 and DP (Dual-Processor). The E5420 is one of those.

http://www.intel.com/products/processor/xeon5000/specifications.htm?iid=products_xeon5000+tab_specs

Sorry to post again. Rob, are you sure that any 5000 series Xeon will work with the SkullTrail? I have two Xeon 5050's with the 65nm technology on my S5000PAL Server Board and I would think that they would be incompatible with the D5400XS.

 

[Rob Williams]

Unregistered: That's too bad about the instabilities. One thing you might want to try is setting the voltages to 1.30v and see if things improve. Those CPUs are designed to operate with up to 1.35v, I believe. The CPUs shouldn't need that much, but it's still a safe spot to keep it and might avoid potential instability down the road.

The biggest cause of instability in general is the overall lack of voltage (or temperatures), so I'd raise some of the voltages if possible... even the RAM, since sometimes, configurations with more than four banks like to see extra voltage. It's definitely something that will vary from system to system.

Also, did you happen to track temperatures during these stress tests?

Toledo_Speedo: Welcome to the forums

That CPU should work fine... it's still LGA771, and designed for dual-processor motherboards. I'm not sure why you'd want to, though. Even a budget 45nm Quad-Core would be a nice improvement. Two would literally obliterate it, though. Not to mention that two such CPUs would feel out-of-place in such an expensive motherboard.

 

[Toledo_Speedo]

Toledo_Speedo: Welcome to the forums

That CPU should work fine... it's still LGA771, and designed for dual-processor motherboards. I'm not sure why you'd want to, though. Even a budget 45nm Quad-Core would be a nice improvement. Two would literally obliterate it, though. Not to mention that two such CPUs would feel out-of-place in such an expensive motherboard.

When I was using Tech Support from Intel to trouble shoot my USB issues the tech wanted me to replace components to make sure the board was bad. I told him I could drop in some Xeon 5050's and he said they wouldn't work because of the 65nm tech compared to the 45nm. He also tried to tell me that my E5420's weren't supported but he had trouble keeping his story together on why the E5420 was any different then a supported X5450.

I can't wait to get this thing together, i figure I can cut my workload in half by going from 1 Core to 8!

 

[Toledo_Speedo]

WOW! Intel has great support; I RMA'd my board on 11-13 and they shipped a new on to me this morning! I can't wait to put this beast together!

 

[Rob Williams]

I told him I could drop in some Xeon 5050's and he said they wouldn't work because of the 65nm tech compared to the 45nm.

The 65nm vs. 45nm argument doesn't make sense. The motherboard doesn't care what process the processor is based on. It cares whether or not the CPU is going to fit into the socket, and that it can sufficiently power it. I did notice something that could be a potential issue, though:

http://www.intel.com/products/desktop/motherboards/D5400XS/D5400XS-overview.htm

On the specs page, you can see that the lowest supported FSB is 1066MHz. The Xeon 5050 has a 667MHz FSB, so if there is one limiter, that's going to be it. So in all honesty, I wouldn't try it. The worst that would probably happen is a non-bootable system, but there's no sense in taking a chance. Far too much at risk.

Glad to see Intel reacted so fast on the RMA though... good stuff.

 

[Unregistered]

Unregistered: That's too bad about the instabilities. One thing you might want to try is setting the voltages to 1.30v and see if things improve. Those CPUs are designed to operate with up to 1.35v, I believe. The CPUs shouldn't need that much, but it's still a safe spot to keep it and might avoid potential instability down the road.

The biggest cause of instability in general is the overall lack of voltage (or temperatures), so I'd raise some of the voltages if possible... even the RAM, since sometimes, configurations with more than four banks like to see extra voltage. It's definitely something that will vary from system to system.

Also, did you happen to track temperatures during these stress tests?

.

Yes, I did make a careful note of all the temperatures. With the CPU coolers in the positions mentioned and a fan on the northbridge, the highest MCH temp I saw was 65 degrees (idle was around 59). According to Core temp 99.3, the CPUs never reached in exess of 57 degrees on any one core. I also used Realtemp which indicated that the "distance to TJMAX" was always above 27 degrees on all cores.

Intel is insisting it could be the software despite that fact that I don't have any issues when running each CPU seperately and that a VCore increase made an obvious difference when running them together.

They have also stated that they don't endorse running the CPUs at 1.2875 since it is outside the reccommended specification, which really isn't very helpful, especially when considering there are YouTube videos of their in house tech staff overclocking and overvolting the platform like crazy.

There are other Skulltrail owners that have had similar issues when loading all 8 cores (at least 4 cases documented on forums) though when I asked if they'd encountered any issues that were similar to mine and if they could run the test themselves, they fell silent and just asked me to use a different application.

No system I've ever built has had problems running Prime at stock speeds and voltages and even though Vista seems happy enough running it thus far, I'd really like to establish why XP is not and why the VCore increase makes the difference it does.

 

[Rob Williams]

Hmm. That's odd. I used Skulltrail as my main machine for a few months (granted, it wasn't a full-time PC) and I never experienced an issue like that. The worst issue I had was general heat. Two cards in SLI + poor overall cooling was one of the reasons I ended up tearing it apart. I didn't want to deal with anymore.

Still, these instabilities are ridiculous. As for 1.2875v being outside of spec... I don't really see how. I could almost swear that they were designed to go up to 1.35v, though it's very difficult to find that exact information. If you are talking about the E5420's though, I think that much voltage is higher than it needs to be, at stock speeds.

I am not sure if you mentioned it already ( I can't see it ), but did you stress other parts of the system to see if they are all stable (memory, mainly)? Also, did you check the temperatures for those specifically? FB-DIMM's get ridiculously hot.

 

[Unregistered]

Still, these instabilities are ridiculous. As for 1.2875v being outside of spec... I don't really see how. I could almost swear that they were designed to go up to 1.35v, though it's very difficult to find that exact information. If you are talking about the E5420's though, I think that much voltage is higher than it needs to be, at stock speeds.

I am not sure if you mentioned it already ( I can't see it ), but did you stress other parts of the system to see if they are all stable (memory, mainly)? Also, did you check the temperatures for those specifically? FB-DIMM's get ridiculously hot.

Yes, ran memtest, no errors throughout multiple passes. Crucial FB-DIMMS reach 75 degrees under maximum load, Kingstons are lower due to their enhanced heat spreaders (around 67 degrees tops). Same problem with both. They are cooled with the Corsair 3x 40mm fan bracket. The Prime blend test stresses memory as much as almost anything,

As for voltage, I totally agree (and I'm talking about the QX9775s), it even says on the box they they're rated max is 1.35v, though their default is 1.212v, that's what Intel believe theu should be run at - at least, that's what they're telling me.

I just can't work out why the system is stable for 20 hours of the test in Vista (even when slightly overclocked and with all voltages at their defaults). Yet in XP, the CPUs need a higher VCore to be stable even when at stock speed. I'm begninng to wonder if the VCore actually made a difference or was just a lucky conincidence and that something else in XP might be causing this issue.

Perhaps I'll try a clean install, or just stick with Vista.

Thing is, I ran the test in XP twice and both times it reached 9 hours with the Vcore raised whereas before, the system always froze within 4.

Can you recall any specific issues you had with freezing/random rebooting etc, and what you did to overcome them?

 

[Mike]

RAM Cooling

...
Board Fans:
- CPU x 2
- Northbridge
- Southbridge
- RAM x 3 (single unit of 3 fans)
...

Wow! That RAM fan unit made a HUGE difference. From mid 90's C down to mid 60's C for the Kingston's with the built in heat pipes (HyperX).

Unfortunately, the northbridge fan (40mm x 10mm) was not enough to control the heat at FSB 425. I tried with fan pointing up and pointing down. Temps rise to 70 C then blue screen. I think my next attempt will have to be to replace the heatsink, else remove it, sand it down and put it back using Arctic Silver. Hopefullly that will take it down the 5 degrees or so I need to stay operational.

What, if anything, has anyone done to control the heat of the northbridge? Would a 40mm x 20mm (as opposed to 10mm) make much difference?

 

[Unregistered]

Wow! That RAM fan unit made a HUGE difference. From mid 90's C down to mid 60's C for the Kingston's with the built in heat pipes (HyperX).

Unfortunately, the northbridge fan (40mm x 10mm) was not enough to control the heat at FSB 425. I tried with fan pointing up and pointing down. Temps rise to 70 C then blue screen. I think my next attempt will have to be to replace the heatsink, else remove it, sand it down and put it back using Arctic Silver. Hopefullly that will take it down the 5 degrees or so I need to stay operational.

What, if anything, has anyone done to control the heat of the northbridge? Would a 40mm x 20mm (as opposed to 10mm) make much difference?

If you get hold of two Zalman 9700s (either the Nvidia edition or standard) and have the front one with the fan facing downard and the rear one with the fan facing forwards and also a 40x20mm <6.25 CFM fan blowing onto the northbridge, this should keep it well under 70 deg, at least it does seem to work for TJ09/TJ10 cases with all 4 120mm chassis fans installed. You will find that the fins on the left side of the front Zalman 9700 will just kiss the northbridge with the 40x20mm fan, but not enough to prevent installation.


What is your case and how many chassis fans do you have hooked up?