Skulltrail

Posted on March 19, 2008 by tyrou.
Categories: CeBit, Overclocking.

For the CeBIT show we had on Cooler Master’s behalf at the CeBIT, we decided to set up a Skulltrail system. Skull…what ? Skulltrail ! That’s the codename for the newest Intel delirium. Let’s consider a few specs:

  • A workstation mainboard using the 5400 chipset,with four PCI-Ex 16x slots supporting both Crossfire and SLI, four FB-DIMM slots (yes, that’s server memory), and of course, two LGA771 sockets for the CPUs. Note that even though they’re LGA771 for the sockets, the mounting holes for the cooling solution are LGA775 style, allowing to use any common Core2 heatsink, good idea from the Intel designers. Last but not least, a BIOS which boasts almost every usual overclocking setting, except for FSB:RAM ratio. The board complies with the E-ATX form factor, meaning it won’t fit any case. and you’d better have a top notch power supply, because it requires a 24 pin plus two 8 pins connectors (one for each CPU). Overall design is good, and looks to be the fruit of a good brainstorming to fit all that stuff on a single PCB, especially for the VRMs, made for high loads, even the memory one. (We’ll see later how to mod them !)
  • Special CPUs, called X9775. They’re Core2 based Xeons, with fully unlocked multipliers, factory setting is 8×400 = 3.2GHz. So that’s the latest generation of Intel processors, 45nm, FSB 1600 and for sure, quad cores. Throw a pair on the mainboard, and that makes a octo-core home system !
  • Memory: that’s the bad point of the system, you’re forced to use FB-DIMM. It’s DDR2 based, but with ECC, and a completely different way to address it from the chipset (won’t enter into the details). It’s not that expensive, less than DDR3, but slow, and there’s no sticks available rated over DDR2-800, combine that with the automatic ratio selection, and you have a rather bad memory sub-system for overclocking.

Now the presentation is over, we can go further on funny things… let’s clock that stuff! Good news, it overclocks just like a “common” QX9770, meaning I can reach around 4.5GHz on aircooling, at the CeBIT, we went up to 5.26GHz using liquid nitrogen (limited by poor memory which wouldn’t overclock any further). At these speeds, it runs really fast, best way to test it is to run some Cinebench, 9.5 version takes as low as 5 seconds to complete… my good old quadcore took 12!

We didn’t have that much time to run every bench we wanted, but new tests are scheduled for next weeks.

For those interested, we couldn’t find any published volt mod for this board, so we created one. it allows up to ~1.77V for the CPUs, higher values make the board really overheating, and you don’t want to try, trust me. (I got my finger burnt holding it). Following pictures show the mod for one CPU VRM, but it is exactly the same for the other processor, and it’s easy to copy it. Voltage regulation is based on an ADP3189 chip, so you want to solder a 10K VR between the resistor shown on the picture and the ground.

More pictures and benchmark results will come quickly, but here are a few, and they can be used as a reference for the volt mod.

This SimpleViewer gallery requires Macromedia Flash. Please open this post in your browser or get Macromedia Flash here.
This is a WPSimpleViewerGallery

How to spend a (geek) weekend.

Posted on February 18, 2008 by tyrou.
Categories: Overclocking.

It’s Friday, and you’re on the road going home… you’re phone rings, that’s a SMS. Is it for a party ? … NO ! It’s just a thousand times better: Fedex telling you your parcel has been delivered !

Half an hour later, you jump out of the car, run to the door, and finally get you hands on that darn package you’ve been expecting for days and days. Time to open it, grab a drink, and run to the computer desk. That box contains everything you need for a perfect geek weekend alone at home: a new cpu and a DDR3 memory kit.

… fast forward … parts are now getting installed, and it’s only 7PM

7:15, moment of truth: will it boot up ?

7:16, breathing restarts, it works !

After a boring hour reinstalling Windows and some drivers, I’m now ready for the real fun: discovering that new platform, and trying to overclock it properly.

9PM, first SuperPI run… score is in the 11s, that’s correct for an aircooled rig, and gives a good start.

11PM, time for a break… get a pizza out of the freezer, stick it in the oven. While waiting, open the windows in the computer room, it’s 0°C outdoor, it would be stupid to miss that nice cold air which could do good on the cpu temperature.

12PM, pizza has disappeared, and it’s time to get serious… run SuperPi as fast as possible as many times as it needs to get a good score.

… fast forward … it’s now 4AM, it’s real cold in the room, and that has been 22 hours since wake-up time. Oh, yes, I forgot the most important: Pi time is now below 10s, and has even been better than my previous personal best. Knowing that former best score was achieved using a cascade cooler which draws 5kW of power, and that the new one just happened with a simple heatsink…

Saturday.

It’s 8:30, and the alarm rings. Time to wake up, good clocks need time, and a weekend only lasts for two days, that means you shouldn’t lose time sleeping too much.

Day was spent doing almost the same thing, but with a new target: getting a 9.5s time. After a long benchmarking run, it finally occurred, as I saw times were not improving a lot with the following runs, I gave up on aircooling systems, and searched for a more powerful cooler.

In the past, I’ve been building a lot of phase-change coolers, but most of them are now either stocked as spare parts in the garage, or need to be repaired. Not speaking of the cascade cooler, which needs a week to make it run for an hour :D

Finally, I found one single stage phase change cooler (technically, the same thing as a basic refrigerator) still working… not that good, but working at last. I brought it back into the room, and installed it on the benching rig. After a lot of adjustments, it started to work almost as I expected…

That cooler was not as great as I thought, and needed some re-work, it took a few hours to make it run properly.

The evening was spent trying to get good scores with the cold cpu, and doing some 3DMark tests, so no improvements on the SuperPi time.

Sunday.

You don’t get any record with an usual stock mainboard… except if it’s a Foxconn X48 ! … but as it’s a pre-release one, it still needs some re-work, which were sent from the manufacturer, via a good friend, Shamino. I asked my volt modder, and we changed some things on the board, making it run better.

It’s now time for real benching: we have great hardware, a modded top notch mainboard, and some cold. After a long evening, it finally happened, we were able to break in the 8.xxx s for SuperPi 1M ! That was the main target for the weekend, and we did it.

SuperPI score

The sun now sets, and in a few hours, those two short days will end. Find it a loss of time ? Who cares. Weekends are for funny things, and for the geek I am, that’s real fun.

In now two weeks, we’ll be doing that for a whole week. The CeBIT will be all about that kind of things for us, and that was a good start for our preparation.

[/geek]