didn't Apple used to make an Xserve? - Page 2

Actually, they (the machines) are not the same. The 1750's were single power supplies and single drives while every 2600 and 2650 had redundant power supplies, multiple drives and a RAID card. Otherwise, the machines had the same CPU's. We also did not measure the power going into the machines as we were only concerned about the heat generated not the power consumption.

Again, my statements towards the beginning of this thread were in relation to the heat generated by a 1U versus a MultipleU.

People are attempting to clarify and slightly ammend your description of the physics involved. They are pointing out the difference between heat and temperature. While a smaller box may run hotter. This isn't synonymous with it generating more heat.

I TOTALLY understand it...having an EE degree I was simply pointing out that a 1U Dell Xeon machine does produce more heat then a 2U or 4U Dell Xeon machine with the same CPU's. This is not based on conjecture or guessing or pencil and paper figures...it is based on real world tests...nothing more...nothing less

Then how are you measuring that heat? How many BTUs do the 2U and 1U boxes pump out per hour for instance?

I may be showing my ignorance here, but if you are measuring the air output from the exhaust you are not measuring the heat generated by the CPU's but rather the heat transfered to the air that is coming out of the enclosure. If both computers have the same heat sink and same fan then the larger enclosure would still probably have a lower heat at the exhaust vent because there is more room for air in the enclosure that is receiving heat from the heat sinks. To measure the heat generated by the CPU's you need to get a temperature reading from the CPU to be accurate.

I wish we really know emphatically the exact number. I was talking about this with one of my guys today that has a similar Physics background to mine and we were wondering the same. Unforunately, Dell's Rack Advisor still has not been updated with the 1750's. We were actually looking to see what the power consumption was on all the machines...

It still baffles us that a Dell 2600 Dual 3.06 with internal RAID with 2 drives, an additional RAID card for an external RAID chassis and redundant power supplies is producing more heat then a Dell 1750 with Dual 3.06's and a single hard drive and single power supply!!!

BTW, the heatsinks are WAY different!

The heatsinks are NOT the same. Maybe I can find the time tomorrow to snap some pictures. We were talking about checking the CPU temps on the machines...this is monitored, we just did not have the time to check them.

Also, the 1750's fans SEEM to push more air volume...although they may be moving the same CFM in a much smaller area.

I'm confused. I also have an EE degree, but perhaps I don't remember it as well as I thought: I was under the impression that the law of conservation of energy imply that the heat produced by a computer per unit time must be directly proportional to its energy consumption.

So surely measuring the power consumption of the server would be the best, most reliable method of determining its heat output, since even a very noisy server with lots of blinking lights still produces heat as its primary form of energy output.

If you try to measure the temperature of the processor or the air coming out of the vents, then you may miss other places where heat is escaping, such as radiation from the casing.

It doesn't really matter how a server distributes the processor heat, in the end, for a given processor at a given load the exact same amount of heat will be produced and this heat will end up warming the air in the room, regardless of how many heatsinks, fans, casings, etc it passes through first (after all, heatsinks and fans are basically just ways to transfer heat to the air).

It may be that a server with a lot more fans produces more heat because of the fans own motors, which cannot run completely without friction. In which case you may be right that the smaller servers (which typically need more fans) may be running hotter. But then in that case you could find out that quite easily by measuring their power input, as was suggested.

You are correct that temperature is dependent on size, however heat is not. A smaller case will get hotter than a larger case given the same amount of heating. However I presume that it is the temperature of your server room that you are concerned about, not the individual servers themselves. A small server gets hotter because it only has half as much case area to absorb heat. However, that also means it only has half as much area to radiate heat from, which means that it will heat the room exactly as much as a server that is half the temperature, but twice as big.

So if I were you, I'd worry about ventilation, and not how big your server boxes are.

Socrates

I may be showing my ignorance here, but if you are measuring the air output from the exhaust you are not measuring the heat generated by the CPU's but rather the heat transfered to the air that is coming out of the enclosure.

Well it's not like the heat just stays in the computer. It has to go somewhere. Probably the best measurement would be putting the comp in a room by itself and measuring the increase in air temperature.

As Socrates so wisely pointed out the larger server has a larger surface area, as well as internal volume to both absorb and radiate the heat through means other than the exhaust vent. The heat is still going into the room, but not directly through the exhaust which I assume is the point that the temperature was being measured.

shawk is right. Pixar has intense interest in G5 Xserves.

Heat can be transfered 3 ways.
1. Radiation
2. Convection through air or liquid ( all that air pumped over the heat sinks)
3. Conductance through solid material (like from the CPU to the heat sink)

Basically most parts of the computer is heated up through radiation and conductance spreading the heat from a handful of really hot components and then the airflow will cool of the whole thing.

The simplest way to measure heat is to measure the power comsumption by having a meter on the power cord. A computer basically is a machine with 0% "efficency" 100% end up as heat and (allmost) 0% as "work".

And please use Watts as opposed to such arcane units as BTU

We are measuring the actual ROOM temperature...which is what is puzzling. During our spare time we may try and figure out what is going on Cause you know that funny thing about energy...

That was all we were concerend about (cooling)...however, the 1750's and the 2600's are the only servers with U320 built-in controllers. The 2650's are still U180. Being DB servers disk IO is the true limiting factor. Now imagine the space 6 2600's take up vs 6 1750's. We also need room to grow this cluster to 32 nodes...size does start to matter.

You are correct. We do not have the tools to measure this however.

Nothing arcane about BTUs at all.

But I wasn't suggesting what unit of measurement to use, I was trying to find out how they were taking the measurement. Comparing the heat output of a lightly used CPU in a 2U box against a 1U box that's heavily loaded with computaionally intensive tasks using the "lick and touch" measurement technique isn't going to tell you much beyond "one's warmer'. But warmer does not mean "produces more heat" which is the point folks were trying to get across.

Measuring BTUs would give you a rough idea of how the CPUs in question compare in terms of actual heat output in this instance, but since the amount of heat generated by a CPU is related to how much electricity it uses which will vary according to computational load it's not too surprising that an under utilized system will actually produce less heat than one that is heavily utilized all other things being equal (which in this case, they're not).

Go watch the dial spin in the electricity meter. Duh.

i don't want to repeat myself but it seems as if i was - and still am - in a good position to win the bet on 90nm chips as well as my prediction for g5 xserves at the end of february

funny that every answer i got in that old thread was refusal

I agree 8) I know some other tidbits that should come to pass in a couple months and really change the IT department's outlook on G5 servers...

The dial moves too fast

Except IBM never said anything about 1U servers, they talked about blades. They also have avaialble in the same form factor dual XEON 2.8GHz units, which I suspect dissipate quite a bit more heat than a pair of 130nm 1.6GHz G5s.

michael

Hi Bluesigns;

I do hope that Apple doesn't abandon the 1U market. It's completely possible to build such a machine with todays chips, but I'm worried that Apple will be concerned about the noise the unit develops. This may lead them to dismiss the form factor, even though that is the hottest part of the market

On the otherhand I would not be completely opposed to 3U iether if they resultant machine Is truely something ot crow about. That is if the delievered a low cost machine with 4 to 6 processors in a 3U case then the equation may change a bit. A lot of things come into play, so it is hard to say if Apple would get the 3U unit right. By right I mean a system that would offer better peformance than three 1U machines working together. The 3U unit of course would have to be delivered at a lower cost than the three 1U units.

I'm trying to keep an open mind with respect to what Apple could release. There is nothing to keep them from releaseing both a 1U and a 3U unit. In fact a good argument may be made that they should do just that. There is also the wild card of the 750VX (or what ever it is called these days), that is a 32 bit implementation. A quite 32 bit machine that could double present performance of the 1U machine would be recieved very well by some markets. A 32 bit machine would be short lived though as the whole market will move to 64 bit rather quickly - possibly faster that Apple and some of the PC manufactures may suspect.

Time will Tell

Dave