Zalman TNN 500 AF based workstation (Q6600/NV6200LE/2 disks)

kcg · Post by **kcg** » Mon Mar 17, 2008 1:50 pm

Hello,

I've seen even JimX already rebuilt his Zalman workstation

Well I also decided to upgrade my former Zalman TNN 500 AF from AMD64 (winchester), Asus A8V Deluxe, 1GB RAM, NVidia FX 5200, Hitachi 5K100 80GB, Seagate 80GB to something more modern:

Board: Asus P5W64 WS Professional
CPU: Intel Q6600
RAM: ECC Kingston, planned amount 8GB was reduced to 6GB thanks to fact that 4GB kits were not available at the purchase time, so I'm running with 2GB kit now and waiting for 4GB which should be on the road
GPU: Asus EN6200LE card (NVidia 6200LE chip plus just 64MB RAM)
disks 2x Hitachi 7k200 160GB
disks enclosures 2x Scythe Quiet Drive 2.5"

Why this? This should be software development engineering workstation, so first of all I set myself on quad core CPU and then on ECC RAM. This means Intel and either 975X or X38 chipsets. The decision was hard, since I'd like to go modern X38 way, but it does not support LPT nor PS/2 (in Asus case) and also does have higher TDP than older i82975X and more importantly is not really tested (and recommended) by Zalman and pictures of Asus X38 boards does not look like it will run w/o removing of host-bridge heat-sink with connected heat-pipes in the case. At the end I decided to go more mature route by using 975X. I do have good experience with Asus, so this was obvious choice. I chosen P5WDG2 WS Pro (recommended by Zalman!) at the first, but since it was not available here for 2 months (I waited for it), then I decided to switch to P5W64 WS Pro which is nearly the same layout and so little risk that it will not work when P5WDG2 WS Pro is recommended board. I also decided not to wait for Yorkfield, but rather try Kensfield.i.e. time preassure. I've had a hard time deciding between 5k160 and 7k200 drives, but based on the feedback here I decided to give a try to 7200rpm disk again, although I do have bad experience with 3.5" seagate from my previous build (very noisy disk). The last to be decided was GPU, I decided based on power-consumption, first thought about going 7300LE/GS route, but then used Zalman recommendation and purchased one of remaining 6200LE cards which is perfectly supported by the case.

Few pictures are here:

Open case lying on the desk:

Drives and cables between them:

Detail of disks tack, unfortunately Scythe makes only screw holes on the sides of the enclosure and not on the bottom of it where they are supported by the case. So I needed to improvize a little bit and put drives vertically and use just one screw to fasten them to the drive backplate. And here is how it looks:

GPU cooling done in zalman way, since I've been able (thanks to supported low-end card) to use case capability to cool GPU:

I've also sticked host-bridge heat pipe to the motherboard's host-bridge although it does have it's own:

And here are two details of cable management:

...and cable management on the bottom of the case:

Now, few notes about experience with this build in comparison with previous build.

1. 2 7k200 in Scythe enclosures are less noisy in idle than single 5k100 lying on a foam on the bottom of the case (not enclosed), but they are more noisy during seek. I suspect that drive on the end of plate is this one which makes the seek noise more audible

2. Q6600 is _much_ more hot than previous Winchester 1.8GHz

3. 975X chipset is more hot than previous Via <something> I don't know exact number

4. ECC seems to really work, still needs to be tested for corrections, i.e. I need real proof

5. 7k200s are sweet, they are much faster than 5k100 and in addition I'm running mirroring of /home between both (OS is Solaris Express Developer Edition 1/08, filesystem is famous ZFS)

6. 6200LE is quite faster than FX 5200, even from the point of scrolling a lot of text in gnome-terminal

7. No need to use bulb to keep PSU running, which is a proof of higher power-consumption which does not make me that happy.

Temperatures of idle (in Linux which automatically underclock to 1.6GHz) are:

Host-bridge: 43C
CPU (Tcase): 43C
core0: 39C
core1: 38C
core2: 35C
core3: 38C

For a load I've compared running SuperPI (ie. systester in Linux) and compiling of http://www.mico.org project with make -j 4 which proofed to be better stress-test. I kept it running over night to warm up whole case and temperatures after this test were:

Host-bridge: 49C
CPU (Tcase): 68C
core0: 65C
core1: 62C
core2: 59C
core3: 61C

this is of course during the peak, since once I stopped compilation it goes down during a second to ~ 60 CPU. Ambient temperature was 21C

To be honest, I'm quite disappointed by the high CPU temperature, I'm not sure if Linux provides good value (but I quite trust this OS), but I've been used to around 45C for highest load on Winchester. The problem with Q6600 is that under highest load it seems to run just few Cs from it's maximal Tcase temperature and I know that during the summer ambient temperature in my workroom will be 10C higher which probably means that CPU will be overheated. I'm also maximally surprised by the fact that temperatures of independent cores are lower than Tcase. I think this is not possible, but cannot decide what's going wrong here. Any idea?

Thanks!

BillTodd · Post by **BillTodd** » Mon Mar 17, 2008 2:11 pm

I'm also maximally surprised by the fact that temperatures of independent cores are lower than Tcase. I think this is not possible, but cannot decide what's going wrong here. Any idea?

Assuming the monitors are calibrated correctly (i.e. they are reading the correct temperature) then the only reason for the case being at a higher temperature than the cores would be if there was another heat source in the chip (cache perhaps?) running at an even higher temperature.

(I'm not familiar with the quad core's internals so this might be complete B'lox)

fjf · Post by **fjf** » Mon Mar 17, 2008 2:30 pm

It is well known that modern intel cpus don't give real temps. Maybe they don't want us to know.

I like the case a lot, but I know it is very expensive. The question is: is it worth it?. Is it so much silent that a solo with a ninja and a good rear 120 mm-600 rpm fan, a suspended hard drive and fanless video card?.

BillTodd · Post by **BillTodd** » Mon Mar 17, 2008 3:01 pm

is it worth it?.

Who knows?

But, it IS the way all PC's should be made

kcg · Post by **kcg** » Mon Mar 17, 2008 3:18 pm

fjf wrote:I like the case a lot, but I know it is very expensive. The question is: is it worth it?. Is it so much silent that a solo with a ninja and a good rear 120 mm-600 rpm fan, a suspended hard drive and fanless video card?.

I cannot compare, since I don't have solo nor p182 with ninja here. I've bought just average EuroCASE with Fortron 400W for the previous build components and I'm using it as a test/backup machine now. And it's _noisy_. But yes, I think that Ninja with Scythe 8dB 12cm fan (don't remember exact part number) and ditto for rear fan might be really interesting comparison, but don't know where to get one for doing it.

Karel

JimX · Post by **JimX** » Tue Mar 18, 2008 12:55 am

kcg wrote:
fjf wrote:I like the case a lot, but I know it is very expensive. The question is: is it worth it?. Is it so much silent that a solo with a ninja and a good rear 120 mm-600 rpm fan, a suspended hard drive and fanless video card?.
I cannot compare, since I don't have solo nor p182 with ninja here. I've bought just average EuroCASE with Fortron 400W for the previous build components and I'm using it as a test/backup machine now. And it's _noisy_. But yes, I think that Ninja with Scythe 8dB 12cm fan (don't remember exact part number) and ditto for rear fan might be really interesting comparison, but don't know where to get one for doing it.

Karel

I do have a P182 and when comparing with no HDDs in the dead of night, the P182 is very quiet while the Zalman is silent. With HDDs, it's just a matter of how far you're willing to go in either case. I'm not willing to go far enough in neither!

JimX · Post by **JimX** » Tue Mar 18, 2008 1:11 am

Karel, great WS!

A few questions:

It is well documented in SPCR's copper Ninja review that pressure makes a difference in temps. Is the CPU HS tight enough?

Did you put new thermal paste between the CPU block heatpipes? I cleaned them and used AS5 when I rebuilt mine.

Did you use the rear thermal blocks? A 100+ watt processor needs them.

What is the temp of the CPU block and the top CPU blocks under load?

I could recheck all temps if you want, I use a cheap RadioShack infrared thermometer, which works very well. We could at least draw some conclusions by extrapolation.

I would also remove the cap from the northbridge and stick the block on the fins.

kcg · Post by **kcg** » Tue Mar 18, 2008 1:45 pm

JimX wrote:Karel, great WS!

A few questions:

1. It is well documented in SPCR's copper Ninja review that pressure makes a difference in temps. Is the CPU HS tight enough?

2. Did you put new thermal paste between the CPU block heatpipes? I cleaned them and used AS5 when I rebuilt mine.

3. Did you use the rear thermal blocks? A 100+ watt processor needs them.

4. What is the temp of the CPU block and the top CPU blocks under load?

I could recheck all temps if you want, I use a cheap RadioShack infrared thermometer, which works very well. We could at least draw some conclusions by extrapolation.

I would also remove the cap from the northbridge and stick the block on the fins.

Thanks Jim!

ad 1) I think it is, in fact I've tightened it again few days ago
ad 2) Sure! I also cleaned them from the old paste and applied the new, but here I probably made my first mistake and applied the past which I've bough 3 years ago for Winchester assembly. It's still OK, but I'm afraid it's not the most high-tech paste. FYI: it provides 0.4 W/m.K heat conductivity which might not be comparable with top-class PC pastes.
ad 3) no, yes, this is my second mistake. I'm using SLACR Q6600 which is considered to be 95W TDP, even Q6700 is 95W so I've considered rear-mount thermal blocks as a disutility.
ad 4) I really don't know, since I don't have any inframeter here.

Anyway, I've just updated my BIOS and it seems core's temepratures are in Linux still lower that thermal diode temp. That was my first todo item, now I know thermal registers are not misconfigured. The second item might be to disassemble, add rear-mount thermal blocks, change to use better thermal paste and assemble again. I'll think about it, hard to know if this is really needed if I don't hit Tcase max even under the highest load for hours. Let see what happen during the summer. At the worst case I might end with switching to Yorkfield which seems to be ~20W under load cooler.

BTW, are those thermal-blocks really that efficient?

Thanks for the hints!
Karel

kcg · Post by **kcg** » Tue Mar 18, 2008 1:51 pm

JimX wrote: I would also remove the cap from the northbridge and stick the block on the fins.

Forgotten to reply to this hint, this is also option, but at the current stage I would not risk it because of possible RMA. The second issue might be that Zalman recommends IIRC just to cool 10W heat source by this heat-pipe, while MCH 82975 (aka 975X) is marked with 13.5W TDP. FYI, 82X38 (aka X38) is 26.5W TDP and 82P35 (aka P35) is 16W TDP. This is also the reason why our beloved case will not be that useful in the future thanks to unability to cool the host-bridge. Well, things might move forward with come of Nehalem.

Cheers!
Karel

JimX · Post by **JimX** » Tue Mar 18, 2008 5:04 pm

Karel, when I installed the RTMs behind the CPU, I didn't notice much of a difference (2 degrees Celsius, if memory serves me right), but it was a ~75W one. I believe that putting them behind the CPU also allows a more even pressure without the board flexing at all.

I did notice a more significant drop in HS temperature from the 2 RTMs I installed behind the old 925 northbridge (~4c), and the case was warm right behind them. Which means that we may probably use the case through another upgrade (if PSU specs don't change much by then).

I do not have RTMs installed now, the P35 on the Abit feels quite cool. And the same goes for the P5E (X38) in my P182. So even though the TDP may be higher, the various heatpipe solutions implemented actually work to dissipate the heat more efficiently.