macbook pro 15-inch Mid 2014 (Dual GPU), Nvdia GPU much slow then Intel GPU

Questions and Answers : Macintosh : macbook pro 15-inch Mid 2014 (Dual GPU), Nvdia GPU much slow then Intel GPU
Message board moderation

To post messages, you must log in.

AuthorMessage
pumbpeng

Send message
Joined: 23 Jan 07
Posts: 1
Credit: 1,844,088
RAC: 0
Message 1831879 - Posted: 22 Nov 2016, 10:50:45 UTC
Last modified: 22 Nov 2016, 10:51:17 UTC

I use MacBook Pro 15-inch mid 2014, which include two GPU, Intel Iris pro and GeForce GT 750M, both GPU in using, but the task on Nvidia seems much slow the Iris, both task computation size is about 183887 GFLOPs, but Nvidia take 2hours and Iris take only 1 hour.
ID: 1831879 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 1831889 - Posted: 22 Nov 2016, 11:44:28 UTC - in response to Message 1831879.  

Despite the hardware being comparable due to amount of FLOPs they churn out, the method in which they do calculations isn't comparable.

The Intel GPU does OpenCL calculations whereas the Nvidia GPU does CUDA. Both are comparable to each other but also very much apart from another.

CUDA can only be done on Nvidia GPUs, is only developed by Nvdia and is therefore the proprietary method of doing calculations on Nvidia GPUs.
OpenCL is open source, developed by a myriad of companies and volunteers, it can run on a lot of hardware, including on Nvidia hardware.

On top of that, Seti's tasks cannot be compared one on one with each other. There are tasks that run in mere seconds, but also that run in several hours - also depending on hardware.

The two tasks you ran, 07ja09af.13338.11115.15.42.30_0 and 07ja09af.13338.11115.15.42.217_1, may in name look like each other, but unless you looked up their angle range we don't know if they are the same.

07ja09af == the recording of this piece of data happened around 07 January 2009 on chunk af (there are 7 chunks per data set, aa to ag)
13338 == Process ID of the splitter.
11115 == Block number. The beginning of a chunk is 72, advances by about 409 for each subsequent group of Work Units.
15 == Telescope ID: 3 and 4 are beam zero of ALFA at 0 and 1 polarity, etc.
42 == splitter settings ID.
30 == Subband. The 0 band contains the 1.42 GHz. center frequency, 127 is the highest, 128 the lowest, and 255 is just below 0.
_0 == task 1 of 2 for this work unit. Each work unit is split into two equal parts which are sent out to two different computers, which do their own set of calculations and return a result. The result of both tasks is compared against each other (validated) and only when they agree with each other is credit given out.
_vlar == possible end on tasks, the acronym stands for Very Low Angle Range. These are angles recorded at the lower end of the telescope's beam and may cause problems on GPUs. These are very slow tasks.
_vhar == possible end on tasks, the acronym stands for Very High Angle Range. These are angles recorded at the higher end of the telescope's beam and shouldn't cause any problems on any hardware. They run very fast though.
ID: 1831889 · Report as offensive

Questions and Answers : Macintosh : macbook pro 15-inch Mid 2014 (Dual GPU), Nvdia GPU much slow then Intel GPU


 
©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.