Message boards :
Number crunching :
Energy Efficiency of Arm Based sysetms over x86 or GPU based systems
Message board moderation
Author | Message |
---|---|
mavrrick Send message Joined: 12 Apr 00 Posts: 17 Credit: 1,894,993 RAC: 4 |
The thought just crossed my mind. Has anyone take the time to collect some numbers to compare power efficiency between ARM systems and x86 with GPU's. I am sure the analysis would take some time but it would be interesting to see. I know one of the big reasons I stopped crunching all the time was because of heat and power usage with my systems. If I could lean on my ouya or maybe even a ODROID-XU from Hardkernal.com I would be much more inclined to get back into crunching full time. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
What kind of platforms do you want to compare? Desktop to desktop, mobile to desktop, or mobile to mobile? SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
mavrrick Send message Joined: 12 Apr 00 Posts: 17 Credit: 1,894,993 RAC: 4 |
I am talking about a comparison between lets say a ARM based Qualcomm Snapdragon 800 to a typical Wintel platform. The concept being, if I could get a ODroid-XU with a Quadcore ARM A15 CPU's and compare it to a Core i5. What would the numbers be like when calculating performance with power efficiency. There is no question which is actually faster per work unit. But i am really after efficieny for how much power per WU. If the ODROID-XU could Complete 4 WU in 10 hours but used 1/10th the power of a compareable Intel platform to do the same work. I am asking because I noticed my old Galaxy Note 1 (a snapdragon S3) is running around 1 WU every 10 hours give or take a little bit. That just amazes me. I was expecting it to take longer actually. I know it isn't using much power to do that work to This |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I am talking about a comparison between lets say a ARM based Qualcomm Snapdragon 800 to a typical Wintel platform. Recently I have been wanting to compare the performance per watt of my Galaxy S4. However being my phone and only running when plugged in I haven't been able to get a good idea of its processing ability. Synthetic benchmarks could give an idea, but not a direct comparison for SETI@home use. When I last did a PPW comparison of my home machines my old 1.5GHz Pentium M notebook was the best performer. I haven't generated enough data to give a value for my i5-4670 systems yet tho. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Performance per watt. Like my refrigerant cooled rig? It's about the least green rig on the planet. Consumes about 500 watts of compressor power in addition to the 600 watts or so used by the actual cruncher. In addition to the unkown blend of R based refrigerents running in it's veins. NOT a green machine. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
shizaru Send message Joined: 14 Jun 04 Posts: 1130 Credit: 1,967,904 RAC: 0 |
I am asking because I noticed my old Galaxy Note 1 (a snapdragon S3) is running around 1 WU every 10 hours give or take a little bit. Just a couple quick notes: - That must be what we call a "shortie". A "normal" WU would probably take double that on an old S3 proc. - In a nutshell I'm guessing that no, ARM wouldn't and will not be more efficient at SETI per Watt. But there's a catch, a next-gen Tegra (CUDA capable) will probably blur that line for good? Sometime later this year? Maybe next year for SETI... Edit: Forgot the most important bit! ARM apps are essentially still in BETA making any comparison... well it's basically too soon. |
David S Send message Joined: 4 Oct 99 Posts: 18352 Credit: 27,761,924 RAC: 12 |
I am asking because I noticed my old Galaxy Note 1 (a snapdragon S3) is running around 1 WU every 10 hours give or take a little bit. You certainly don't want to compare anything to the current 7.27 Beta app, which is taking over 150 hours on fairly fast devices. David Sitting on my butt while others boldly go, Waiting for a message from a small furry creature from Alpha Centauri. |
jackyman38 Send message Joined: 6 Oct 04 Posts: 28 Credit: 125,912 RAC: 0 |
The current betas are debug version so it take a lot longer. With previous non-debug version it average 24-25 hours per task (depends on the work unit size and type ofc). I have 2 Android TV stick with RK3188 running fulltime on beta. They hardly uses over 5w with all 4 cores running. They are no comparison to PC+GPU at the moment. Also GPU on these devices are not use mainly due to lacking of OpenCL support or lacking source code available. It will be interesting to build a arm-cuda version when more NVidia chips are being use but market share may not worth the effort required to port for Arm-NVidia. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
My Galaxy S4 was running VLAR's 4 at a time in 6-9 hours on the v7.21 apps. I have held off on running more on it until I can get another high power wall charger to use at work. As it is consumes enough power to run the battery down when just connected to a 500mA USB port. I do have some 1 & 1.5A USB wall power sources, but my phone only sees them as standard 500mA USB ports. No chip in them to provide USB info I expect. So it will not switch to high power charging mode. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
David S Send message Joined: 4 Oct 99 Posts: 18352 Credit: 27,761,924 RAC: 12 |
My Galaxy S4 was running VLAR's 4 at a time in 6-9 hours on the v7.21 apps. As I mentioned recently, my Note 2 uses more than 500mA running just 2 at a time, and even a car cord can't keep the battery charged if it's running the GPS display. David Sitting on my butt while others boldly go, Waiting for a message from a small furry creature from Alpha Centauri. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
My Galaxy S4 was running VLAR's 4 at a time in 6-9 hours on the v7.21 apps. Do you happen to know the rated output of the car charger, or do you have an issue like I do? Where the Samsung device will only pull 500mA unless there is an IC in the charger to tell it "I can provide more power". I use an app BatteryBot. When I on plugged into a 500mA port I see "Charging(USB)" & using the provided charger or a high powered charge port from my PC I see "Charging(AC)". When using either of my 3 car chargers the phone indicates "Charging(USB)". Despite their rated outputs of 1.5-2A. Considering the power consumption is >2.5w but <10w the devices do seem to have a significant PPW. I did notice that when running 2 or 4 tasks and using CPUz that the phone processors were not running at a constant speed. Fluctuating and normally running around 1280MHz instead of 1.9GHz. So there may need to be some tweaks to see the full power of these processors under Android. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
I have held off on running more on it until I can get another high power wall charger to use at work. As it is consumes enough power to run the battery down when just connected to a 500mA USB port. The USB cable apparently is important, I believe they have something special done to the data cables to enable a higher current. Claggy |
jackyman38 Send message Joined: 6 Oct 04 Posts: 28 Credit: 125,912 RAC: 0 |
The S4 has a Exynos 5 Octa 5410 CPU which is 4x Cortex-A15 + 4x A7 (big.LITTLE). Most devices won't have that kind of performance at the moment. I guess the computation would be done on the A15 cores as it is processing intensive. The charging power is most likely to do with the cable. The charging standard is a mess at the moment however there is a way to increase current on some devices by shorting the 2 data wires on the phone side. I custom made my charging cable they can all deliver 800ma when the 2 data pins are shorted on the connector even when I connect it to a USB2.0 port on my PC. You can give that a try but no guarantee that it will work for your particular device. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I have held off on running more on it until I can get another high power wall charger to use at work. As it is consumes enough power to run the battery down when just connected to a 500mA USB port. It would make sense to spec out wire slightly heavier. In not even thinking about that I have been using my old phone microusb cables, from '06-'07, for higher powered charging on my PC. I will have to make an USB shunt and do some power tests at some point with my various cables. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
The S4 has a Exynos 5 Octa 5410 CPU which is 4x Cortex-A15 + 4x A7 (big.LITTLE). Most devices won't have that kind of performance at the moment. I guess the computation would be done on the A15 cores as it is processing intensive. My S4 is the Qualcomm Snapdragon 600 - 1.9GHz quad-core variant. There is also a 3rd variant with a Qualcomm Snapdragon 800 - 2.3GHz. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
David S Send message Joined: 4 Oct 99 Posts: 18352 Credit: 27,761,924 RAC: 12 |
As I mentioned recently, my Note 2 uses more than 500mA running just 2 at a time, and even a car cord can't keep the battery charged if it's running the GPS display. No, no idea what its rating is. I'm not even sure where I got it. At the 12V end, it's hard wired, not a USB jack to plug the cable into. It might have come with my bluetooth speakerphone (which I never use any more). I'll check for that app. David Sitting on my butt while others boldly go, Waiting for a message from a small furry creature from Alpha Centauri. |
jackyman38 Send message Joined: 6 Oct 04 Posts: 28 Credit: 125,912 RAC: 0 |
Made the cable with 26 AWG silicone wires, works great. https://plus.google.com/u/0/photos/115666642611145843502/albums/5988969450085969633/5988969449085937602?pid=5988969449085937602&oid=115666642611145843502 |
David S Send message Joined: 4 Oct 99 Posts: 18352 Credit: 27,761,924 RAC: 12 |
As I mentioned recently, my Note 2 uses more than 500mA running just 2 at a time, and even a car cord can't keep the battery charged if it's running the GPS display. I just installed BatteryBot. It shows charging AC using the AC charger I have at work. It also shows Boinc is only using 2% of the battery. However, all the uses it lists add up to 103% (rounding, maybe?). I'll keep an eye on it. And I'll see what it says on the car charger. David Sitting on my butt while others boldly go, Waiting for a message from a small furry creature from Alpha Centauri. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I just discovered that my Galaxy S4 has a power usage rating on the inside cover under the battery 3.8v 1300mA or 4.94w. Currently I am assuming the rating is the max device consumption. Much like the rating labels other electrical equipment. So full CPU load with screen at full brightness. While crunching I am able to turn the screen off so actual power consumption should be less but more than 2.5w I expect. Even at 5w that is only about 44KW/h a year. Doing 12-16 tasks a day it seems like it would be highly efficient crunching. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
shizaru Send message Joined: 14 Jun 04 Posts: 1130 Credit: 1,967,904 RAC: 0 |
Even at 5w that is only about 44KW/h a year. Doing 12-16 tasks a day it seems like it would be highly efficient crunching. Whoa! Boy was I wrong. My 12w GPU can do only 8... I'll get me coat... |
©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.