On 10/23/2014 11:21 AM, Manfred Mornhinweg wrote:
Thoughts on computer heatsinks and cooling
I am running 240 watt CPUs. 8 core @ 4.3 to 5 GHz. with 16 to 32 GB of RAM
Several are liquid cooled (closed circuit) and run in the 40C range
compared to the little CPUs of a few years ago that ran as high as 80C
plus with the little, noisy, high speed fans.
The video cards are on the order of 300 watts each.
The video cards have 2 to 3 fans with vapor phase cooling and the output
air is quite toasty.
7 fans per computer not counting video cards, CPU cooling and PS.
Fan locations: 120 mm: 3 front, 2 rear, 1 side, and 1 large on top.
There are 5 to 6 HDs with one SSD for the system drive in each computer.
These are going to be moved to an NAS
Typical HDs run 5W while these 500 GB SSDs are well less than 1W ~ 0.1W
It should be noted the heatsinks are about 1.25" X 1.25" and use exotic
heatsink compound (Arctic Silver sp?). This is the same compound often
used in high power SS finals, with a heatsink contact area several times
larger than what many SS finals use. The vapor phase cooling seemes to
work well, but becomes questionable above 200 watts
Power supplies are 850 Watts, but switcher are very efficient so there
is little heat output. The fan on the PS only runs occasionally. These
computers will not boot with 600 watt power supplies
These are in full size, steel tower cases @ about 30 to 40# each.
Complete computers are 50 to 60#
73
Roger (K8RI)
.
Marv,
After much consideration, I've abandoned the idea of using solid state
devices in an Envelope Modulated configuration.
Was this decision caused by the voltage-variable capacitances in them,
or any other reasons?
At present I wonder to what extent envelope modulation can be used
with FETs, without phase predistortion, before the unwanted phase
modulation causes too much IMD. Clearly it will be necessary to limit
envelope modulation to the higher parts of the envelope, and get
linearity in the smaller amplitude range through a different method.
I am still pursuing an Envelope Modulated architecture employing a tube
final. The test until will employ a 6146. The modulator is a standard
digital audio part operating at 200 KHz driving a bridge of power
FET's. The
4-pole output filter(s) are at 25 KHz. Designing a single filter
becomes
problematic at impedances greater than 1K or so due the self resonant
frequencies of available inductors. Therefore, four bridge rectifiers
driving four filters are seriesed on the output of the (switching
supply)
output transformer.
So far it looks like a good configuration. I was considering something
very similar, using a 4CX1500B. I have two of them in my junk box. But
no socket.
I can't imagine the resonance of inductors being a big problem. Using
ferrite cores, the Q should be low enough so they behave mostly as
resistors, at the frequencies where they would otherwise resonate.
Splitting up the HV supply into separate rectifier/filter groups and
putting them in series is actually the most common way to build a HV
switching supply.
> This also conveniently allows sourcing the screen at 1/4
of the plate supply.
But is it convenient to modulate the screen voltage too?
An auxiliary supply and LF amplifier is connected in
series with the screen source which may be controlled to envelope
correct
this final amp.
But that costs efficiency! What might be very reasonable is to keep
the screen voltage constant during mid/high amplitudes, modulating the
plate voltage, and in the low areas only freeze the plate voltage at a
fixed level and modulate the screen to get that part linear.
The grid can be left to self-bias through grid conduction and leak.
When/If this design becomes functional a legal limit version
utilizing a 4-400A will be attempted.
An interesting project!
I'm still contemplating a phase modulated bridge, similar to what's
employed
in the Broadcast Electronics 4MX series, for a solid state final.
The 4MX
employs ordinary power FET's since it only has to operate up to 1.8 Mhz.
(Much!) Better transistors will be required for HF.
Or just much smaller ones, using many in parallel!
> I plan a test unit using
ancient Siliconix VMP4's operating at 50 watts. Operating at 50 ohms
eases
the output coupling and filter requirements. To translate this to
the legal
limit level it may be best to pray for some appropriate GaN parts.
Don't be afraid of using many cheap small FETs in parallel. Using
individual source and gate resistors, they work fine as one big FET,
but with far better frequency range than if you actualy use a single
big FET.
The input
of such a device may be opto, or dielectricly, isolated at the
digital input
Optosiolation at RF isn't trivial. Dielectric or transformer isolation
is much easier.
and a simple 1:1 50 ohm transmission line transformer can isolate the RF
output from a direct line operated power supply.
Here you have me puzzled. How can you use a transmission line
transformer to provide DC isolation? That would need to be a
conventional transformer!
If you mean winding a ferrite core with 50 ohm coax cable, and then
using the center conductor as primary and the shield as secondary,
then that's a conventional transformer, not a transmission line
transformer!
> A series LC filter is
connected in the primary of this output transformer and a double
Pi-Network
on the secondary will provide output matching as well as harmonic
suppression.
That should work. Or you can place a parallel resonant circuit over
the secondary, and match by having both this circuit's cap, and that
of the series circuit in the primary, implemented as variable caps.
At legal limit power, the caps will necessarily be small vacuum
units.
That would be best. But air variables should work, too.
Envelope correction for this scheme will require a small (50nS /
Freq. or so) adjustable delay inserted into one side of the bridge
drive.
Here you caught me. I don't see why that's needed. Also I can't figure
out how much delay you mean. I suppose you mean "ns", not "nS", but
even so it makes no sense to me, since a time divided by a frequency
is a time squared! And if you mean 50ns divided by the frequency as a
dimensionless number, we end up with just some femtoseconds delay in
the HF range, which doesn't seem reasonable!
So, can you explain this bit?
Both of the above designs will be driven by a DSP based exciter,
capable of
AM, FM, CW & SSB, employing some standard Audio DSP parts, a few
Microchip
PICs, and an Analog Devices Quadrature Up-Converter which can output
directly
on frequency. The exciter portion of this design has been stable for
about 5
years and is under slow construction. A receive function is also
contemplated with a "conventional" front end, including somewhat less
than
standard IF gain, and a DSP baseband demodulator.
That sounds interesting, as you can put the required phase
corrections, and also the delay of the phase drive signal to
compensate for the long power supply delay, into the DSP. But it makes
this design unsuitable as an add-on amplifier. It can only be used
with the DSP radio. Still a very worthy project.
Operating designs of this type to "zero" output in SSB mode is
problematic.
I've considered adding a low level signal (say 35-40dB down) "out of
band" in
the tens of Hz region. This would be just a curious artifact on the
air and
be filtered out by anyone listening to the desired signal.
But it won't work!!! Even if you add such a "pilot tone", the envelope
of your SSB signal plus this tone _still_ crosses zero amplitude a few
hundred times every second! Such a pilot tone keeps the output from
staying near zero for a long time, but it absolutely doesn't avoid
zero crossings! So it doesn't help in any way.
Now if you can make the EER system work to -40dB, then you can simply
cut off any signal that's even lower! No ham operation requires more
than 40dB S/N ratio; in almost every operating condition there is
enough noise and QRM so that parts of the signal 40dB down can't be
heard; and any IMD arising from curring off what's below -40dB, is
irrelevant, specially if that cutting off is done smoothly.
The problem, of course, is getting the EER to work to -40dB. That will
be hard, I fear! I think it's a better approach to make the EER work
only to -20dB, or even just to -15dB, and control the output amplitude
in the smaller portions by screen modulation, or bias modulation,
while keeping the plate supply at the level corresponding to that -15
or -20dB level. The penalty in efficiency will be small, and things
will get much easier.
Just a thought on heatsinks. Check out the heatpipes from the Apple G5
(which are currently seeking their place in the landfill). They have
an inch
square base and are probably good for a couple hundred watts each.
That would need to be checked. In general I'm a bit cautious about
computer cooling components. There is too much snake oil being sold in
that field. For example, many people belief their computers need to
get rid of LOTS of heat, when they see PC power supplies rated at 600
watts, and CPUs rated at over 100 watts. But this isn't true, in
almost every case! My own computer does have a 550 watt rated supply,
a CPU rated at 130 watts "maximum design power", several hard disks,
two optical drives, loads of memory, and is stuffed full of expansion
cards. But the power consumption, measured at the AC input, is 27
watts! The CPU has an average dissipation of less than 5 watts. I
disconnected the CPU fan, to avoid its ugly whining noise, and the
CPU's heatsink gets barely lukewarm. The CPU temperature, as reported
by the BIOS, stays below 40°C. So, I fear that computer coolers rated
at hundreds of watts might actually not be what they promise!
I have been thinking a lot about a high efficiency amp using one of my
4CX1500B tubes, with a switching power supply. The good points: It
would be robust, I have the tube, no need to built a complex heat sink
assembly. The downsides: Some blower noise, even if less than when
working in low efficiency modes; 3 minutes warm-up delay; Need for
manual tuning, or a complex autotuning system; A constant base power
consumption of about 100 watts during all times, even RX, for the
filament, blower, and losses.
Specially this last thing, the significant power consumption during
standby times, is pretty much a show stopper for me. My idea of a high
efficiency amplifier just isn't compatible with 100 watts of power
waste during RX! So I guess the 4CX1500B is not for me... not even in
a high efficiency scheme. Maybe some instant-on tube. But I don't have
any suitable one, and buying tubes means spending more money than on
FETs...
Manfred
========================
Visit my hobby homepage!
http://ludens.cl
========================
_______________________________________________
Amps mailing list
Amps@contesting.com
http://lists.contesting.com/mailman/listinfo/amps
_______________________________________________
Amps mailing list
Amps@contesting.com
http://lists.contesting.com/mailman/listinfo/amps
|