Message boards :
Number crunching :
Pine A64+ 2GB running 24/7
Message board moderation
Author | Message |
---|---|
Defense Engineer Send message Joined: 11 Jun 99 Posts: 15 Credit: 7,204,997 RAC: 0 |
In case anyone is curious, here is a https://www.pine64.org/ running SETI@Home 24/7 for the past couple of months: http://setiathome.berkeley.edu/show_host_detail.php?hostid=8135422 I did not overclock it. However, I did put heatsinks on it and a tabletop fan circulating air across it. Otherwise, it's just sitting at room temperature. Which right now in Khon Kaen, Thailand is about 28C during the day and 25C at night. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
With that kind of processing throughput it would take 12 of those to equal my AMD 4200+ dual core CPU and at $30x12 a way over priced. |
JWNoctis Send message Joined: 6 Nov 07 Posts: 20 Credit: 329,409 RAC: 0 |
My phone with a Snapdragon 800 would probably run faster, but definitely would not run with all four cores at the same time without some bad heat and throttling. Meanwhile I think I'd probably try a similar setup with RPi 3 soon...It sounded fun. |
qbit Send message Joined: 19 Sep 04 Posts: 630 Credit: 6,868,528 RAC: 0 |
With that kind of processing throughput it would take 12 of those to equal my AMD 4200+ dual core CPU and at $30x12 a way over priced. Is it overpriced? I mean, you can't just get an CPU and run it, you need a complete computer for that. And that means that you need to find a computer for $360 or less to match those tiny computers. And don't forget that these are much more power efficient also. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
In case anyone is curious, here is a https://www.pine64.org/ running SETI@Home 24/7 for the past couple of months: CPU type unfilled in host page for some reason.What CPU this device has? Seems ARM family but what exactly? And what estimate of consuming power? SETI apps news We're not gonna fight them. We're gonna transcend them. |
qbit Send message Joined: 19 Sep 04 Posts: 630 Credit: 6,868,528 RAC: 0 |
Raistmer, there's a link in the opening post;-) However, seems to be a Cortex A53, 1.2 Ghz Quad-Core ARM. |
JWNoctis Send message Joined: 6 Nov 07 Posts: 20 Credit: 329,409 RAC: 0 |
With that kind of processing throughput it would take 12 of those to equal my AMD 4200+ dual core CPU and at $30x12 a way over priced. You can probably get more powerful combinations of second-hand and leftover parts for less, but the point of power consumption still stands. And being new and almost entirely solid-state meant they should also last considerably longer than years-old hard drives and motherboard caps. Wouldn't make much noise or take too much space to shove around, either. Meanwhile they had this neat thing linked on their site: http://climbers.net/sbc/bargain-pine-a64-cluster/ |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Thanks. So, ARMv8-A Interesting to compare with modern low-end smartphone I currently testing, Samsung Galaxy J1 mini (ARMv7). Its price was higher than 15$, more like 64$. App running with APR :1.77 GFLOPS This Pine rinning with APR 1.58 GFLOPS (both are 4-cores). BLC VLAR 0.010995 completion time for J1 mini:110-111ks BLC VLAR 0.006303 completion time for Pine: 80-86ks (~6h faster) used set of optimized functions: J1 mini: vfp_GetPowerSpectrum 0.001373 0.00000 vfp_ChirpData 0.052066 0.00000 fftwf_transpose 0.022859 0.00000 opt NEON folding 0.014100 0.00000 Pine: vfp_GetPowerSpectrum 0.001141 0.00000 v_ChirpData 0.081789 0.00000 fftwf_transpose 0.024457 0.00000 opt NEON folding 0.003022 0.00000 For comparison, x86 based (IvyBridge i5) device functions: v_avxGetPowerSpectrum 0.000051 0.00000 avx_ChirpData_a 0.002272 0.00000 v_avxTranspose4x16ntw 0.001394 0.00000 JS AVX_a folding 0.000302 0.00000 At least in part of Chirp (quite time consuming area for CPU apps, with trigonometry calls) app used on Pine has big optimization potential (compare J1 mini vs Pine chirp and optimized NEON folding timings). It can be made even more faster. SETI apps news We're not gonna fight them. We're gonna transcend them. |
JWNoctis Send message Joined: 6 Nov 07 Posts: 20 Credit: 329,409 RAC: 0 |
In the mean time this looked even sweeter: http://climbers.net/sbc/40-core-arm-cluster-nanopc-t3/ Seems like floating point and NEON SIMD units are standard on Cortex-A53 too...Wonder how difficult would it be to port existing Android/ARMv7 apps over to that, and maybe optimize them further. And just how much throughput 8 of those would have on a single board sharing bus to 1GB of memory. Or 40 of those on 5 boards - They seemed to have a $35 version with the same chip too. Wonder how cost-effective is that. |
Defense Engineer Send message Joined: 11 Jun 99 Posts: 15 Credit: 7,204,997 RAC: 0 |
Glad to see this sparked some conversation. If anyone has anything they want me to check on this setup for them, please let me know. I can even put it on a power meter with or without the table fan if anyone really wants to know exact energy consumption data. |
Tom Rinehart Send message Joined: 12 Dec 01 Posts: 113 Credit: 13,255,975 RAC: 6 |
I've been running an ODROID C2 with the SETI app compiled for ARM64. I'm getting about 2 GFLOP. It is not using the NEON routines. It is just a straight compile. https://setiathome.berkeley.edu/show_host_detail.php?hostid=8169546 I sent the app I compiled to Eric for the Beta site, but haven't heard from him. |
rob smith Send message Joined: 7 Mar 03 Posts: 22190 Credit: 416,307,556 RAC: 380 |
For some it is as much about the challenge of running SETI on unusual hardware and not about the credits - To such people I raise my hat, both when they succeed and fail with the words "Congratulations on pushing the boundaries". Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
Marco Franceschini Send message Joined: 4 Jul 01 Posts: 54 Credit: 69,877,354 RAC: 135 |
Hi all, this is my "little" Orange Pi Lite Allwinner H3 SoC based running 24/7 (with auxiliary little fan above since H3 generates high heat). https://setiathome.berkeley.edu/show_host_detail.php?hostid=8092096 Marco. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Sometimes few letters required due to letters lost time to time. SETI apps news We're not gonna fight them. We're gonna transcend them. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Glad to see this sparked some conversation. If anyone has anything they want me to check on this setup for them, please let me know. If you have the time that would be great. I've been looking that some of these devices for a while, but their actual efficiency when when running SETI@home work is typically worse than a desktop. I normally calculate watt hours per task and compare efficiency that way. My host 5837483 with an i5-4670K runs 4 MB tasks at once in about an hour while using between 97-102w So if I use 102w and 60min run time. (102w * 60min)/60)/4 = 25.5Wh per MB task Before I added a GPU to my low power J1900 host. It would run 4 MB CPU tasks in about 6 hours while using 15-16w. (16w * 360min)/60)/4 = 24Wh per MB task Numbers I have from a Raspberry Pi2. To run 4 MB tasks took about 50 hours while using 2.5w. (2.5w * 3000min)/60)/4 = 31.25Wh per MB task Since the site lists that a 10w(5v@2A) PSU is needed for the Pine64+ I have a feeling it may have power usage in the 4-5w range. While that is low with task run times of about 35 hours that would be 35-43Wh per MB task. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Defense Engineer Send message Joined: 11 Jun 99 Posts: 15 Credit: 7,204,997 RAC: 0 |
If you have the time that would be great. I'm at work right now. But when I get home, I'll power it down, move it to a power meter, restart everything, and let it run for a few hours. Once power consumption stabilizes at a constant running temperature, I'll post an average reading from the meter. It will be just the Pine A64+ It will not have any loads from fans on it. The fan will be on a separate circuit. |
JWNoctis Send message Joined: 6 Nov 07 Posts: 20 Credit: 329,409 RAC: 0 |
My laptop's i7-3632QM currently completes MB and guppi tasks both on average of about 95 minutes on 27W, that makes it...(27w*95min)/60/4=10.69Wh per task. Laptops are efficient. Though I'd throw in maybe 50% overhead on other components like display, hard drive, and a madly spinning fan. Meanwhile, my phone's Snapdragon 800 is eating up a half-new 3200mAh/12.16Wh lithium battery in a bit more than 2h30min, while taking an average of 8.64 hours per recent full-sized MB/guppi task. That makes it...(12.16wh/2.5h*8.64h)/4=10.50wh. And probably would be even more efficient if those were all on armv7 apps. Definitely a lot of potential for optimization here for those devices! |
Shaggie76 Send message Joined: 9 Oct 09 Posts: 282 Credit: 271,858,118 RAC: 196 |
I was curious so I did back of the envelope calculation for a modern GPU. My 1070's do at maybe 4 tasks every 25 minutes (2 tasks concurrently, 10 minutes for Arecibo, 15 minutes for GBA); so call it 6.25 minutes/task. My triple-card boxes are about 400-440 Watts, so at most 147W/card including CPU-overhead. 146W * 6.25m / 60 = 15Wh per task as a conservative estimate. 50% more power and nearly 4 times faster. |
JWNoctis Send message Joined: 6 Nov 07 Posts: 20 Credit: 329,409 RAC: 0 |
I was curious so I did back of the envelope calculation for a modern GPU. It's probably in fact about the same power and nearly 4 times faster. My ~10Wh/task on that laptop was without overhead added and I'd expect a bit more than 10W of power draw from everything else estimated from battery monitor. Though it's hard to be sure without a power meter. |
Defense Engineer Send message Joined: 11 Jun 99 Posts: 15 Credit: 7,204,997 RAC: 0 |
If you have the time that would be great. Ok. I let it run overnight and it is now pulling down 5.7 Watts according to the meter. https://flic.kr/p/QNCWq2 |
©2024 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.