ATX desktops are bad

[Mercutio]

Background:

I maintain about 60 PCs in a classroom setting for my day-job as an IT trainer. Of those PCs, 24 are in Enlight 7200 desktop cases. Another 12 are inCompucase 7106s while the rest are in cases visually similar to the Evercase e1205, but without the front panel.

These machines were built with either an 80mm (Evercase/Enlight) front intake fan or a 60mm output fan (Compucase) located immediately next to the PSU (ie close to the CPU).

Internal parts for all machines are the same: GA-7S748 motherboards with Barton AthlonXP 2500s, retail HSFs, 1GB RAM, Sapphire Radeon 7000 graphics (well... there are 12 with Leadtek FX5200s - someone else bought them and, no surprise, they're the PCs I have to work on the most) and HEC 350W PSUs.

HEC PSUs have an 80mm output fan and a vented "bottom" (the side that faces the PSU). In theory, hot air should be sucked away from the CPU.

That's the background.

Every once in a while I get complaints of flakey behavior on our classroom machines. Usually, I can't reproduce the problem, since I'm almost never told of the issue when it happens - my co-workers will just move students to another machine and carry on, and I maybe get an e-mail the next day.
My suspicion for a while has been that the problems are heat-related, as periodic checks of isolated machines would reveal temps that were north of 60C. I'd re-apply my silver compound and make sure fans were spinning at their normal 3600rpm.

Yesterday I finally took the time to study the issue in more detail, since I had representatives of every case in use, plus a couple machines that could easily be made hardware-identical in tower cases (a Foxconn TS-1 with an 80mm output fan under the PSU, and a Chenming 501AWBU-F with a 120mm fan beneath the PSU). To each of those cases, I installed the same 350W PSU and a Radeon 7000.

First thing I did was check CPU temps for all the classroom temps in my building, in their native location. This is what I found:
Enlight: low 54C mean 56.1C high 61C
Evercase: low 55C mean 60.3C high 66C
Compucase: Low 63C mean 67.3C high 75C (ack!)

I haven't paid much attention to CPU temps since I set those machines up in the first place, but those numbers seem pretty high to me.

My testing methodology: I put a representative machine in the same room, a room with its own thermostat set to 21.1C (70F). The PCs were all placed on tables at the I turned each machine on and let them all sit at a Windows XP desktop for a half hour, completely idle. I turned the machines off and checked the temperature from a BIOS report. I left the PCs off for another half hour while I changed conditions before re-testing.

For all machines, I removed the former thermal pad or compound and replaced it with an even coat of Arctic Silver 5. I also cleaned them all so they were free of dust. This also allowed me to check to make sure that the HSFs were identical (no copper bottoms or slugs. They all had the same aluminum construction and a visibly identical fan).

So, here's what I found out from my subjects.

Test 1: With no case fans, identical retail CPU fans, case sealed, no monitor on desktop cases
Enlight: 57C
Compucase: 69C <-- at idle!
Evercase: 62C
Foxconn: 48C
Chenming: 46C

Test 2: With standard case fans as described above, identical retail CPU fans, case sealed, no monitor on desktop cases
Enlight: 56C
Compucase: 67C
Evercase: 63C
Foxconn: 46C
Chenming 43C

Case fans don't seem to do a whole lot for the desktop machines, and it seems to me that having those CPUs in desktop enclosures isn't doing them any favors either. In fact, it seems to me that ATX desktop cases are really awful for the health of other components in a PC.

A flaw I see in my methodology is that I don't really have a control. I probably should've rigged up a board and PSU outside of a chassis, but it didn't occur to me until after I'd done my first reading, and I didn't have time yesterday to re-do it (and now one of my tower PCs is gone).

I have two Spire "BigRock II" HSFs here. If I had a few more, I'd test again with those instead of the retail HSFs, and I guess I could also do a test with some other PSU, but all the spares I have here are HEC units as well.

 

[MaxBurn]

I had a couple enlight 7237 towers (why the heck do I remember that and I can't remember the name of whom I am working with?) and I don't think the airflow was that good or bad, average case. Had a good vent to let in air on the front but the plastic cover looked way restrictive. I drilled out the holes on the front cover to open it up.

Is there something restricting the airflow on that compusa case like that? Take a look at how air gets into the case?

 

[mubs]

Case fans should help, provided air's being exhausted. A big problem in standard setups is that much of the hot air just keeps going round and round, gradually building up temperatures in the case. In this respect, in the absence of well-planned and executed airflow, it's better to have negative pressure in the case than positive; i.e., more air being sucked out than is coming in. If this is already the case, then two fans close by but not in the same plane, i.e., at right angles to each other, could be interfering with each other so badly that neither is able to do it's job properly. In such cases, a single fan would be better than two.

It's amazing how poor the aiflow is in most cases, even the ones that have multiple fans. I've modified my cases extensively so as to have good flow. Modding will definitely help the problem cases, but may not be practical in your situation with that many systems.

Of the three desktops, the Compucase is the big surprise; I notice that it has two grills in the top of the case. That should help exhaust some of the heat, but apparently not.

Note that Foxconn and chenming have fans front and rear that provides for a semblance of air flow. They also have grills/slots in the front for air intake. The desktops seem mostly restrictive with respect to ait intake. The easiest fix would be to remove one 3.5" or 5.25" bay cover. That should drop temps a bit, if you can live with the toothless grin. Maybe there's a market for decorative grill type bay covers!

I'd say that in general, tower styles are intrinsically better at air flow than desktops.

What is most worrisome is that you checked temps after the machines were idle for some time. Don't NT class Op Sys' "halt" the CPU? Recorded temps then are the lowest they'll be with power on. Run Prime 95 and check the temps again; some remedial action may be necessary.

 

[Buck]

Nice test and interesting findings Merc. Although it worries me, because two years ago I assembled four systems with Compucase cases. Recently, one of them has become flakey. All four have identical components and both sit in the same environment. They're used for recording audio from a telephone line.

 

[Mercutio]

Why all the desktops? The woman who owns the company I work for doesn't like tower cases.

Buck, are you talking about Compucase Desktop or Tower cases? I bought Compucase towers for as long as I could find them as part of my standard build and I don't EVER remember seeing idle CPU temps above 50C on Athlon XPs. I'd continue buying them if I could find them.

mubs, neither my Chenming nor Foxconn case had a front fan. Chenming (a PC I built yesterday) didn't come with one and the Foxconn's had been removed because there wasn't a place to plug in a 3rd 3-prong fan.

NT should be sending HALT instructions, yes.

The Compucase desktop has zero front intake of air. The front of the case is a solid piece of plastic. My case also does not have any holes in the top of the case nor an internal duct/baffle for airflow... but the invoice I have from when I bought those cases does say 7106. So nothing on the front, nothing on the top. On the other hand, I personally assembled all those machines, and I know that I took care to fold cables and wires out of the way. The area between the PSU and the optical drives is a bit of a mess but there was nothing I could do about that.

I'm genuinely alarmed by 55C+ idle temps, and I'm seeing them on all my machines.

 

[Buck]

Buck, are you talking about Compucase Desktop or Tower cases? I bought Compucase towers for as long as I could find them as part of my standard build and I don't EVER remember seeing idle CPU temps above 50C on Athlon XPs. I'd continue buying them if I could find them.

Compucase towers.

 

[jtr1962]

I've found that case fans don't do squat. In fact, I disconnect them in my machines as a matter of course since all they do is make noise and blow dust into the case.

Want to know what really helps? Get a Seasonic or other power supply with a 120 mm fan. They move a ton of air with very little noise. In fact, I put one in the new machine I'm building around the A7N8X-E Merc sent me since I had a positive experience with the one in my other system. After doing that I dropped the CPU fan to 6 volts and it still won't crack 65° C at full load (idle temps are in the low 50s at most).

Besides these recommendations, get the largest physical-size CPU heatsink which will fit on the motherboard, preferably one with high-fin density and made of copper like this one. I've seen similar ones for AMD processors as well. The idea is to have as much surface area and conduction as possible in the smallest space. If you're a quiet freak you can then undervolt the CPU and still have decent temps.

On the heat sink compound I've been there thanks to my thermoelectric (peltier) module hobby. Even the fancy ones only make about a 2°C difference provided the mating surfaces are fairly flat. For CPUs this is irrelevant although for thermoelectrics it matters if you're trying to get the lowest possible temps in whatever they're cooling. If the mating surfaces are uneven, they should be lapped first since fancy heat-sink compound will act as little more than a bandaid. BTW, I can get as little as 0.02°C/W per square inch of thermal impedance using standard thermal grease if I lap the mating surfaces properly (i.e. to within 0.0005" flatness). Not terribly hard to do, either. Get a piece of plate glass, put progressively finer grades of sandpaper on top of it, and move the item to be lapped in a circular motion while keeping the sandpaper wet. Going up to 2500 grit is usually fine. If you want to get fancy go up to 10000 girt and give a final polish for a mirror finish.

 

[mubs]

I've found that case fans don't do squat.

You ain't using them right.

 

[jtr1962]

I've found that case fans don't do squat.

You ain't using them right.

I'm just saying that my internal case temps generally stay less than 10°C above ambient without a case fan. A case fan might lower them a few degrees, but if your CPU is that close to the edge of stability that a few degrees make a difference then there are other issues. I consider cases fans a noisy nuisance and a dust collector. At the very least case makers should only use 120 mm case fans since they move the most air with the least noise.

As I've said several times, I hope using the case itself as a passive heat sink for the CPU, GPU, and power supply becomes the norm. A PC case certainly has the size to effective dissipate ~200 watts with proper external fins and no fans. Best of all, internally the case would be sealed, preventing dust ingress. With heat pipes it's possible to move heat from the CPU to the outside of the case.

 

[ddrueding]

I could've sworn I posted in this thread earlier, must have been eaten...

Use P180 cases on their side. They are beautiful, have the best airflow of any case I've ever seen, and are strong enough to hold a monitor on their side. Furthermore, they have a door with a lock across the front (good for public PCs).

 

[Mercutio]

It's not economically viable to replace the HSFs, PSUs or cases for every machine unless I can demonstrate a clear and immediate need. Even if it only took me 15 minutes to make a parts swap, that's still paying me for a day and a half of work + cost of whichever part x 60.

The larger issue to me is that ATX desktop cases in general just need to be avoided. Evercase, Enlight and Compucase are among the very small number of companies still making standard ATX mid-desktops. Modern PC parts just don't belong in them.

 

[P5-133XL]

I find this interesting. I have 9 machines at home built very similarly with tower Antec cases instead of desktop cases, Radeon 9600's instead of 7000 and one server being a Enlight 8950 case and one old generic tower case.

With Barton 2500's and retail fans the CPU temps under 100% folding load are universally between 50-53 degrees centigrade. Those temps for a cooling fanatic are totally unacceptable but they are way lower than the max acceptable for the chip and it doesn't bother me a bit. Now 3.2GHz P4's with retail fans are always between 60-65 degrees centigrade. Those P4 temps bother me more but but everyone knows that P4's are temp hogs, so I live with it. These machine have been running continiously for now going on 2 years and I have yet ever had to replace the Artic Silver that I used to replace the wax pad originally supplied with the CPU+fan's. Now note that I have had to replace 3 chipset fans on my Gigabyte 7400n2 motherboards....

First observation is that case fans do didly... One of the first things I did was remove all my case fans and found absolutely no effect of CPU temps. While the case fans did effectively drop the interior temps about 5 degrees, the CPU' temps just plain didn't budge.

The only case I have temp problems is my 8950 Enlight 9 bay server case. I have found that if I don't have a working 5 1/4 bay HD fan every alternating HD bay, the HD's overheat. Again, nothing really affects CPU temp -- This is a very annoying HD airflow problem: I destroyed serveral HD's before I figured out what was happening...

Now my standard PS is an Antec, but there have been variations (a replacement Foxcon on one machine and PC Power and cooling for my server case). The Foxconn with a 120mm fan really pointed out, to me, a design issue with standard ATX PS's: The intake is parallel to the output where in reallity it makes much more sense that it be at 90 degrees (i.e. vented on the bottom). Now on every Antec PS I have replaced the 80mm fan with a high volume, low noise Panflow and it really makes a difference in noise especially after removing all the case fans.

 

[CougTek]

I usually fix airflow problems with my enclosures by using my foot to create a large intake hole on the front panel. My CPU temps are very acceptable. For instance, after the foot operation, my 3GHz P4 is no more than 53C. That's with a Zalman 7700Al-Cu and inside a "modified" Sonata II enclosure.

 

[Adcadet]

I usually fix airflow problems with my enclosures by using my foot to create a large intake hole on the front panel. My CPU temps are very acceptable. For instance, after the foot operation, my 3GHz P4 is no more than 53C. That's with a Zalman 7700Al-Cu and inside a "modified" Sonata II enclosure.

Do you use a bare foot, sox, or some sort of shoe or boot?

 

[Adcadet]

My solution to temp issues for the past few years is a full tower case even though all my stuff would fit into a midtower. Since it just sits under my desk, the extra height isn't really noticed.

 

[CougTek]

Do you use a bare foot...

Man, you're sick in the head. Go get some medication.

 

[LOST6200]

Geeze, I think ti was a joke. Besied kicking the case seems rather ludicrous, so what kidn fo response did you expctc?