Best hardware for a replacement desktop?

Paul Pathiakis pathiaki2 at
Tue Jul 30 02:21:07 UTC 2019


Just not into asking for justification just saying it's nonsense?

I use a very simple 'at the wall' draw meter.  Over time, when things used to consume a LOT more power.... CPUs could draw 250W, DDR3, etc  As you added higher end graphics cards, they would increase the draw further..... now, if you've ever done benchmarking, you'd find that you, typically, throw out the first 3 runs.  Why?  Heat yields increased resistance.  Increasing memory, a higher end graphics card, a higher speed clock on the same micro-architecture, etc Things are in flux until the machine comes up to full operating temperature.

So, depending on the number of peripherals.... maybe a second graphics card if you want dual monitors, you have some incredible protein folding software that requires gobs of memory.... or a filesystem of ZFS that loves more memory, etc.  Add in more memory... more draw...

Do you see a pattern with what I'm saying?

Today, we see that high-energy efficiency is being pushed everywhere.... the CPU ( TDP dropping from 300 to 180 to 140 to 100 to 65 ), the GPU ( TDP dropping from 300W to 250W to 170W) , the memory (in case you didn't notice the voltage requirements dropping from 1.75 to 1.5 to 1.35 over the last 5-7 years....)
So, when I know I could potentially plug in second graphics card, fill all my memory slots, overclock a CPU (for only temporary reasons... ), my observations of the last 10 years or so, even through all these changes, was showing a high of about 40-45% over the base to where it is now, about 30-35% of the base.

And, again, despite all of this, I still won't overspend for something that makes no sense and I will, appropriately, size a PSU to my baseline needs while leaving a good amount of headroom for expansion.

I also consider the 80-Plus standards that have been implemented.  I really don't want to hit more than 80-85% of my PSU's upper bound as that seems to be why the standard was created.... drop off of the efficiency of power delivery.  Although the higher standards of gold, platinum and titanium tend to push that envelope of efficiency upwards to 90%  The efficiency of most modern PSUs hit a peak of efficiency around 60% (that may be outdated).  You would really like to have you machine drawing around there.  However, as I said, I tend to allow for enough headroom to plug-in additional peripherals if needed. 

If you read between the lines of the above, you can figure out why you DON'T oversize if unnecessary.  If you only draw 300W (there is a doubling factor of AC/DC conversion - I haven't checked if that's still viable as I don't have time to check all these factors), why buy a 1200W PSU?  Only $50 more?  If you want to spend that money, sure, whatever.  If you want to be at the peak of efficiency and be 'green', you can be at your 'most efficient' by being at the peak of your power curve that the PSU delivers.... so, at 1200W... you'd need to be drawing what?  720W ??!!

(so... what's that?  3x1070ti, 4x8GB of RAM, some ridiculous high-end Intel chip that sucks down 300W at TDP...  Yeah, if you want to spend $5000 on a machine, then, by all means, spend that $50 on that 1200W PSU)

This was what was meant when someone replied that they have different load curves.  (and, no, I really didn't want to have to write this letter....)

Oh... and yeah... I let my little meter calculate my draw after looking at the specs and building the machine.... I'm fairly close.  Also, there are a TON of PSU calculators out there and a LOT of articles about why you don't over-size.


    On Monday, July 29, 2019, 7:49:42 PM EDT, Ralf Mardorf via freebsd-questions <freebsd-questions at> wrote:  
 On Tue, 23 Jul 2019 11:27:03 +0000 (UTC), Paul Pathiakis wrote:
>Take that and add 33%.

On Tue, 23 Jul 2019 17:31:26 +0200, I wrote:
>I would be careful when calculating real power consumption. What makes
>you think that adding 33% is a good value?

On Mon, 29 Jul 2019 14:30:10 +0200, Matthias Fechner wrote:
>If you under or over 50% you will lose efficiency.

Again, oversized is way better, since calculating a realistic power
consumption is virtually impossible and it's also virtually impossible
to know the optimal workload of a power supply. Actually you only could
use specifications provided by the vendors. In the end you get a rough
guess on the values.

What are 33% or 50% of rough estimated values? The result is mere
freebsd-questions at mailing list
To unsubscribe, send any mail to "freebsd-questions-unsubscribe at"

More information about the freebsd-questions mailing list