New Alix board for 2013
-
Nice. :)
I'm certainly prepared to believe those measurements, assuming the psu holds 12V well enough. ;)How many Watts do you think are lost in the PSU, it's efficiency?
Given the small variation in power, 5W max, perhaps the supplied psu is still highly efficient at 3.36W.
Pure speculation time: I would expect to see perhaps 0.25-0.5W loss in the PSU in which case I would hope an AC side power meter should read 4W (if it doesn't display fractions of a Watt).User phil.davis could tell us a lot about the power consumption of the old Alix board since a lot of his sites are solar powered.
Steve
-
Yeah I was using a bench power supply and double-checked voltage and current with a couple of multimeters.
The PSU I normally use, and did the AC measurements with, is a 60W FSP one which I also use to supply some other network kit (I've also measured with some 12W PSUs with the same results IIRC). It's efficiency level V so >87% average efficiency, although the points for that average are measured at 25, 50, 75 and 100 percent load, so it may not be that efficient at <10% load.
Either way, you're likely correct the AC meter should be displaying 4W. The resolution isn't ideal for measuring this low TBF, but it's ballpark accurate at least. Even being less than half a Watt out and clipping the decimals rather than rounding up could explain why it displays 3.
-
I ran iperf again, this time using two computers connected to the board on individual port, all running 1000baseT, iperf server on a windows 8.1 box, and client on macbook laptop running os x 10.7.5, only running squid on pfsense 2.1, here are the results:
–----------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 64.0 KByte (default)[ 4] local 192.168.10.11 port 5001 connected with 192.168.20.11 port 49272
[ 4] 0.0-10.0 sec 539 MBytes 452 Mbits/secUsing netio:
NETIO - Network Throughput Benchmark, Version 1.32
(C) 1997-2012 Kai Uwe RommelUDP server listening.
TCP server listening.
UDP connection established …
Receiving from client, packet size 1k ... 21.98 MByte/s
Sending to client, packet size 1k ... 184.75 MByte/s
Receiving from client, packet size 2k ... 4.16 MByte/s
Sending to client, packet size 2k ... 263.60 MByte/s
Receiving from client, packet size 4k ... 0 Byte/s
Sending to client, packet size 4k ... 428.63 MByte/s
Receiving from client, packet size 8k ... 403.75 KByte/s
Sending to client, packet size 8k ... 567.76 MByte/s
Receiving from client, packet size 16k ... 203.54 KByte/s
Sending to client, packet size 16k ... 746.44 MByte/s
Receiving from client, packet size 32k ... 0 Byte/s
Sending to client, packet size 32k ... 913.70 MByte/s
Done.TCP connection established …
Receiving from client, packet size 1k ... 28.91 MByte/s
Sending to client, packet size 1k ... 30.98 MByte/s
Receiving from client, packet size 2k ... 23.72 MByte/s
Sending to client, packet size 2k ... 23.23 MByte/s
Receiving from client, packet size 4k ... 33.43 MByte/s
Sending to client, packet size 4k ... 43.70 MByte/s
Receiving from client, packet size 8k ... 23.25 MByte/s
Sending to client, packet size 8k ... 46.61 MByte/s
Receiving from client, packet size 16k ... 31.16 MByte/s
Sending to client, packet size 16k ... 47.44 MByte/s
Receiving from client, packet size 32k ... 14.18 MByte/s
Sending to client, packet size 32k ... 47.83 MByte/s
Done.Looks like I need something other than the crappy Macbook to test with :-\
-
Nice! :)
If you run 'top -SH' on the apu board what does the cpu usage look like during that test?Not sure how netio is measuring that udp speed but most of those numbers are far higher than you could get down a gigabit connection so it looks like it's buffering somewhere.
Steve
-
It's the UDP transmit speed, but of course most of that won't make it on to the wire, so it's essentially irrelevant. The receive speeds are what you'd want to look at, but it seems something is up with the testing as they're zero, or close to it, on most of the runs.
Edit: I've just tried the UDP benchmark myself and it never reported above roughly 113MB/s, so it seems it works a bit differently than I thought, and something's up with the above benchmark.
-
FYI I've just finished doing iperf testing on an older alix 2d2.
Sadly I did not take power usage measurements while testing.Full results at: http://forum.pfsense.org/index.php/topic,70911.0.html
Short version, okay for up to 50 Mbit, can do up to 95 Mbit but you're wringing the nuts off there.
-
How are you testing that? A throughput test is what's needed, iperf running on two separate machine not on the pfSense box.
326Mbps seems disappointingly slow. :-\Steve
We ran a throughput test on a very similar box (same cpu, same NICs), and were disappointed.
Our APU only recently arrived.
-
First impressions (running IPFire):
http://www.tuxone.ch/2013/12/alix-nachfolger-im-test.html
-
i heard this new alix apu gets pretty hot, almost 81.5 degrees
-
Since it relies on the enclosure for cooling, what case was that in?
Steve
-
in the default case it heats up that much, atleast thats what the developer told me himself and that it might be fixed in the next redesign so consider that temperature it would make it useless in hot countries like mine where the summer goes upto 50 degrees
-
Hmm, seems very close to the 90 degrees maximum rating.
Steve
-
That seems really high for a heatsinked low-TDP processor like this. How is it transferring heat to the case exactly, e.g. a really thick thermal pad?
-
in the default case it heats up that much, atleast thats what the developer told me himself and that it might be fixed in the next redesign so consider that temperature it would make it useless in hot countries like mine where the summer goes upto 50 degrees
You can find below a new enclosure designed for PC Engines APU by Calexium.
The thermal dissipation is better than closed small cases from PC Engines. There is also HDD fixation for up to 2 HDD.
http://store.calexium.com/en/boitiers/324-pc-engines-alix-2d3-2d13-or-openvox-ipc100-110-120-case-with-hdd-wifi-black.html -
the other thing is that the processor is on the bottom so most of the heat is towards a wall or the ground based on where its placed so in countries like mine thats another issue as the summer temps here r 50 degrees so the ground is much warmer than the rest of the house. it would be better if there was a fan, even low speed would be better than nothing and the processor on top rather than bottom
-
I'm sure that extensive testing has been done by pc-engines during development. I don't believe for a second that they didn't think about keeping the CPU cool enough. Have we actually seen any heat related failure? Shutdowns? Reduced speed?
That's using cases that are just sheet aluminium. If even a small amount of finning were added I'm sure it could run cooler for use in a high temperature environment.
Aluminium is amazingly good at conducting heat so the fact that the CPU is in contact with the bottom may not make all that much difference.Steve
-
Those temps just seem off to me - were the die directly contacting the aluminium case I wouldn't expect it to get nearly that hot. Does anyone know how the production platform is mounted? Those temps suggest the die could be floating i.e. only conduction through PCB, or a thick thermal pad which don't have great conductivity.
The new board uses barely more power than the 2D13, and that barely gets lukewarm to the touch; my GigE switch gets warmer. Even the Geode CPU itself without heatsinking only gets warm to the touch, although we're talking about different packages of course.
-
That seems really high for a heatsinked low-TDP processor like this. How is it transferring heat to the case exactly, e.g. a really thick thermal pad?
PC Engines web site states "using a 3 mm alu heat spreader". Without having seen pre-production units, I understand this as using the 3mm spreader to bridge the gap between the CPU die and the enclosure. This seems supported by the visible thermal grease residue on CPU and south bridge die in the pictures.
These high temperatures do not at all seem implausible, let alone surprising, to me. As a reference point, I am aware of this test. Featuring the AMD G-T40E (which is the lower-TDP version of the two which were considered for the the PC Engines APU), the Jetway test unit approached 80° C under CPU load. And this in, at least judging from a quick glance, an enclosure that seems to have been specifically designed for thermal dissipation (unlike the Alix enclosures). In contrast, the cooling solution from PC Engines looks like an attempt to shoehorn the new APU thermals into cases that were not designed for passive conductive cooling to begin with.
That's not to say they can't do it successfully and won't be able to deliver a solid & stable (in the long term) solution - but employing the T-40_N_ in a fanless build using the existing Alix enclosures always seemed quite a stretch to me.
-
The Jetway board itself is using more power than the ALIX is stated to though, which means extra heat to expel.
My reasoning is, the new board uses only a couple of Watts more than the existing one, at least when idle, so it seems unlikely the case would go from barely lukewarm to too hot to touch. Hence my theory of something being less-than-ideal with the thermal path from die to enclosure.
Has anyone with the new board observed the enclosure getting hot? Bear in mind that since the heat still needs to get through the enclosure to the air I'd expect similar enclosure temperatures whether the CPU was well bonded to the enclosure or not, but if there's a large temperature delta between the die and enclosure (e.g. CPU is 80+ but case is merely warm), then things could improve with a better thermal path between the two.
-
How are you measuring the temperature? Are you sure it's accurate? Even close?
Steve
