Cost of acquisition + performance per W for newer single board computers vs. desktop and dedicated GPUs.

Message boards : Number crunching : Cost of acquisition + performance per W for newer single board computers vs. desktop and dedicated GPUs.
Message board moderation

To post messages, you must log in.

AuthorMessage
NON

Send message
Joined: 22 May 99
Posts: 2
Credit: 2,677,060
RAC: 1,549
United States
Message 1911996 - Posted: 9 Jan 2018, 7:51:26 UTC

Apologies if this has been covered before recently but the last through comparison I found was for the now severely aging RPi3 and 3-4 year old GPUs/CiSCs.
Has anyone done a cost vs. performance analysis of any of the more recent 8 core, 64 bit arm SBCs vs current GPUs/CISC setups? Compared to a Raspberry Pi 3 the newer generation of SBC's seem to get about %240 the performance with the same upfront cost ($35ish) but with a 2-3W higher power consumption (about 25% increase on average compared to my Rpi3's)
ID: 1911996 · Report as offensive     Reply Quote
spitfire_mk_2
Avatar

Send message
Joined: 14 Apr 00
Posts: 557
Credit: 24,889,153
RAC: 8,211
United States
Message 1912100 - Posted: 10 Jan 2018, 11:04:28 UTC - in response to Message 1911996.  

Link examples of these newer SBCs.
ID: 1912100 · Report as offensive     Reply Quote
NON

Send message
Joined: 22 May 99
Posts: 2
Credit: 2,677,060
RAC: 1,549
United States
Message 1916450 - Posted: 2 Feb 2018, 6:13:15 UTC - in response to Message 1912100.  

I totally missed that I got a reply to this post for a good month :)
The most direct comparison to the Rpi3 that conforms to the stats I posted is the Nanopi Fire 3, http://nanopi.io/nanopi-Fire3.html .Same price, double the cores working roughly 20% faster per core on my systems. Then there's the Odroid Xu4 http://www.hardkernel.com/main/products/prdt_info.php?g_code=G143452239825 which is $60 but has 4 cores working at near 210% the speed of a RPI3 core and the remaining 4 working at the aforementioned 120% improvement for about the same gain per dollar. Then you have what seems about a million boards going for 20$ using the same A53 cores as the Rpi3 but clocked higher.
ID: 1916450 · Report as offensive     Reply Quote
Profile David Anderson (not *that* DA) Project Donor
Avatar

Send message
Joined: 5 Dec 09
Posts: 172
Credit: 48,240,460
RAC: 24,523
United States
Message 1917889 - Posted: 9 Feb 2018, 16:58:48 UTC - in response to Message 1916450.  

Just a single datapoint here. I thought the sbc concept was
interesting. (Not using rasbperry pi for seti: have a couple
doing record-keeping/control though).

I acquired an odroid-xu4 (16GiB eMMC disk, 4 core,2GiB ram, 2GhZ clock) and
finished the setup a day or so ago.
Preliminary results suggest that it will accumulate about
700-900 cobblestones per day total
across the 4 cores. One built-in 40mm fan
cooling the cpu.

Preliminary results indicate that Ubuntu-Mate 16.04
sees the 4 cores as 8 (some threading bios
setting) but (again, preliminary) it appears that
running more than 4 seti threads is not helpful.

A single core takes about 8 hours for each current
task it gets.

Power use is under 40 watts at 5V.

I had been running an old HP laptop (68 watts, 1 AMD core)
as an experiment (it took 4-5 hours per task)
but it was big/heavy at 17-inches and its screen and keyboard
were failing so I've junked it.
ID: 1917889 · Report as offensive     Reply Quote
Profile David Anderson (not *that* DA) Project Donor
Avatar

Send message
Joined: 5 Dec 09
Posts: 172
Credit: 48,240,460
RAC: 24,523
United States
Message 1917891 - Posted: 9 Feb 2018, 17:16:16 UTC

I decided that I should measure power instead
of just quoting the max power the ac-dc
converter can deliver.

Running seti (four tasks) its power
use bounces around between 11W
and 12.x watts.

I'll call it 12W.
ID: 1917891 · Report as offensive     Reply Quote
Profile David Anderson (not *that* DA) Project Donor
Avatar

Send message
Joined: 5 Dec 09
Posts: 172
Credit: 48,240,460
RAC: 24,523
United States
Message 1917892 - Posted: 9 Feb 2018, 17:20:31 UTC - in response to Message 1917891.  

I'm measuring the (12 W) power use at the wall plug
with a Brand Electronics Digital Power Meter.
ID: 1917892 · Report as offensive     Reply Quote

Message boards : Number crunching : Cost of acquisition + performance per W for newer single board computers vs. desktop and dedicated GPUs.


 
©2018 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.