Message boards :
Number crunching :
raspberry pi 3 vs GPU, whats best?
Message board moderation
Author | Message |
---|---|
Anders Kihle Send message Joined: 27 Mar 16 Posts: 4 Credit: 162,948 RAC: 0 |
Hello, always want to help in rsearch and use computer power. Now i want to uppgrade to somthing better. Like a Clouster ore GPU`s RIG. But i dont know to much about this. So whats best to use if you think price and power output/cost: raspberry pi 3 ore GPU`S? Thanks fore any answere :) |
rob smith Send message Joined: 7 Mar 03 Posts: 22237 Credit: 416,307,556 RAC: 380 |
A raspberry pi cluster wins on the "fun" side, but, given that a GTX750 will complete a task in about half an hour compared to the day or more for a single pi it would have to be a large cluster to get anywhere near that sort of throughput. Cost wise a GTX750 casts about 100gbp, compared to the 30gbp for pi, and in energy terms the GTX750 uses about 50watts, and each pi about 4watts (one source I saw gave a figure of 12watts) Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I've starting calculating watt hours per task to determine which of devices are the most efficient. Which can be calculated two different ways. 1. daily watt hours/daily number of tasks or 2. ((watts * run time in minutes)/60)/number of concurrent tasks For the Raspberry Pi 3 I found someone stating it took ~25 hours to complete tasks. They didn't mention how many tasks they were running at once but it is easy to calculate for each possibility. Using method #2 If running 1 task (4w * 1500min)/60)/1 = 100Wh per MB task If running 2 task (4w * 1500min)/60)/2 = 50Wh per MB task If running 3 task (4w * 1500min)/60)/3 = 33.3Wh per MB task If running 4 task (4w * 1500min)/60)/4 = 25Wh per MB task Comparing that to my GTX 750ti FTW running 2 tasks at once in ~25 min & drawing up to 45w (according to GPUz). Running 2 task (45w * 25min)/60)/2 = 9.375Wh per MB task Even if I use the EVGA's TDP value for power usage Running 2 task (85w * 25min)/60)/2 = 17.7Wh per MB task For comparison some of my systems i5-4670K with a TDP of 84W running 4 MB tasks at once in ~1h each. (84w * 60min)/60)/4 = 21Wh per MB task Celeron J1900 with a TDP of 10W running 4 MB tasks at once in ~6h each. (10w * 360min)/60)/4 = 15Wh per MB task This tells me. Even if the Raspberry Pi 3 is running 4 tasks at once in ~25 hours it is less efficient than an i5-4670K. Also that my Celeron J1900 is more efficient than I thought it was. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Sidewinder Send message Joined: 15 Nov 09 Posts: 100 Credit: 79,432,465 RAC: 0 |
Rob and HAL are pretty much spot on. I see similar numbers with my crunchers vs my Pi's. I think the main issue is that the Pi is meant to be a low-cost, low power generic computer; not really specialized for this type of work. I bet there are some ARM designs that are optimized for computation (probably why researchers would even consider building large ARM clusters), but they may not be cheap and/or publicly available. There are some alternatives that do crunch faster than a Pi, but they cost more and use more power (examples, here and here). They also have their own issues with OSs/support/community. Pi does have a good community and wide use. One positive of the Pi though is the cost of cooling. They can be cooled passively for the most part which could save you money on cooling (my house gets quite hot from my dedicated crunchers). |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
A part of it might be a function of the architecture. From what I understand of the ARM architecture (which is limited), the memory subsystem, IO and cache orient themselves better to running one multithreaded process better than multiple single threaded ones. Probably future applications will take better advantage of what's available, at least once some general linuxey things changing lately settle down a bit. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Anders Kihle Send message Joined: 27 Mar 16 Posts: 4 Credit: 162,948 RAC: 0 |
Thank you all fore NICE answer. ps:Sorry my ENGLISH!!! |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Thank you all fore NICE answer. We try to help as best we can. Your English is much better than my Norwegian. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Anders Kihle Send message Joined: 27 Mar 16 Posts: 4 Credit: 162,948 RAC: 0 |
;) hehe |
OTS Send message Joined: 6 Jan 08 Posts: 369 Credit: 20,533,537 RAC: 0 |
Comparing that to my GTX 750ti FTW running 2 tasks at once in ~25 min & drawing up to 45w (according to GPUz). Interesting. It appears nvidia-smi is not very accurate, at least on watts on consumed. Nvidia-smi reports my 750ti is using only 28 watts but after reading what yours is drawing I checked it with both the watts reading on the UPS and a KillaWatt device. It is a little difficult to know for sure because as soon as you kill the GPU processes they restart, but both the Killawatt and the UPS report mine is drawing about 40 watts for two tasks which is in line with what you were saying. I wonder why the discrepancy between these two devices and nvidia-smi and if the nvidia-smi reported 95-99% utilization and temperature of 49C has any validity. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Comparing that to my GTX 750ti FTW running 2 tasks at once in ~25 min & drawing up to 45w (according to GPUz). I have noticed the GPU power consumption does seem to vary based on the tasks that are being processed. I just checked my system. GPUz & HWinfo are both telling me my 750 is running about 33w right now. Normally we figure about 80% of the TDP for power consumption while running SETI@home tasks. Perhaps consuming 25-45w is to be expected while running two tasks on a 750ti. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
OTS Send message Joined: 6 Jan 08 Posts: 369 Credit: 20,533,537 RAC: 0 |
Comparing that to my GTX 750ti FTW running 2 tasks at once in ~25 min & drawing up to 45w (according to GPUz). That makes sense, but I was wondering why nvidia-smi was reporting values so much less than the actual measured values with two independent separate hardware devices and if the temps and utilization figures as reported by nvidia-smi were off as well. I have never seen nvidia-smi report anywhere close to 35 watts let alone 40 watts. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Comparing that to my GTX 750ti FTW running 2 tasks at once in ~25 min & drawing up to 45w (according to GPUz). Perhaps that tool is only reporting power usage from the PCIe bus or maybe the PCIe power connector? SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
OTS Send message Joined: 6 Jan 08 Posts: 369 Credit: 20,533,537 RAC: 0 |
Comparing that to my GTX 750ti FTW running 2 tasks at once in ~25 min & drawing up to 45w (according to GPUz). Now that you mention it, I forgot about the fact when the GPU is working it also adds load to the CPU and nvidia-smi almost for certain does not show that load. Even if it doesn't report accurately, you can use it a a comparative tool which is the way I do use it. I guess we have beaten this horse into the ground. |
Mark Wyzenbeek Send message Joined: 28 Jun 99 Posts: 134 Credit: 6,203,079 RAC: 0 |
Could power supply efficiency be the difference? The difference being lost as heat in the power supply. The Universe is not only stranger than you imagine, it's stranger than you can imagine. SETI@home classic workunits 1,405 CPU time 57,318 hours |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
That makes sense, but I was wondering why nvidia-smi was reporting values so much less than the actual measured values with two independent separate hardware devices and if the temps and utilization figures as reported by nvidia-smi were off as well. I have never seen nvidia-smi report anywhere close to 35 watts let alone 40 watts. I found where nvidia-smi was hiding & the value it displays for power usage matches what I see in other monitoring software. I expect they are reading from the same source. My 750ti seems to want to run at 1345MHz. Perhaps that is related to the higher power usage I see? Even if I have abnormally high power usage for a 750ti it is still more efficient than several Rasberry Pi's & more cost effective. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13755 Credit: 208,696,464 RAC: 304 |
According to GPU-Z my main GTX 750Ti is generally around 65% of it's maximum TDP (60W) which works out to be around 40W. Running 2 WUs at a time with approx 90% GPU load at 1280.3MHz Grant Darwin NT |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14654 Credit: 200,643,578 RAC: 874 |
According to GPU-Z my main GTX 750Ti is generally around 65% of it's maximum TDP (60W) which works out to be around 40W. That seems to match mine. I have one host running with a kill-a-watt style meter, with different projects running on each of the major components: GTX 970, GTX 750Ti, Intel HD 4600, CPU - so I can see the power differential for each device by suspending the corresponding project. SETI runs on the 750Ti, two MB tasks. The wall power draw dropped by around 40W - 272W to 232W - when I suspended SETI. GPU-Z says it's drawing around 60%-65% TDP (fluctuating as I type), at a core clock of 1319.8 MHz. It's a "Golden Sample" card, running at factory settings - and on PCIe power only, no additional power cable required. |
EEVblog Send message Joined: 20 Apr 16 Posts: 20 Credit: 4,351,842 RAC: 0 |
My RPi2 only takes in the order of 2.5W running all 4 cores at 100% https://youtu.be/pcQaseUJeZI?t=22m24s [/quote] |
EEVblog Send message Joined: 20 Apr 16 Posts: 20 Credit: 4,351,842 RAC: 0 |
I'm looking at trying an array of Orange Pi One's: http://www.aliexpress.com/store/product/Orange-Pi-One-ubuntu-linux-and-android-mini-PC-Beyond-and-Compatible-with-Raspberry-Pi-2/1553371_32603308880.html Only $10 for a 4 core 1.2GHz ARM board compatible with the RPi2 |
EEVblog Send message Joined: 20 Apr 16 Posts: 20 Credit: 4,351,842 RAC: 0 |
Here is data from my RPi2 running 4 cores at 100% 180,000 seconds average per task, and as stated, 2.5W for 4 cores, so 0.625W per task. |
©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.