Dear All,
Could I ask it is possible to scale the frequencies base on MALI T628 GPU?
I just want to see the performance differences when I varied the frequencies.
And the other question is I saw an OpenCL guide instruction for Samsung Chromebook,
but also can I use that user space driver and settings for my HP chromebook 11?
I tried many times to compile and run several sample applications which are included in MALI OpenCL SDK, but failed.
So if frequency scaling is supported and HP chromebook 11 does not support for OpenCL programming,
then I am thinking to buy new samsung chromebook 2.
Thanks a lot in advance,
Woojung Myung
Oh I am so sorry, I have one more question which is can I use OpenCL's device fission functionality based on Mali GPU like T628?
Hi Woojung Myung,
Many thanks for your questions - I'll do my best to answer.
Regarding changing the GPU frequency, it should be possible to do this typically through a command like the following...
echo frequency > /sys/class/misc/mali0/device/clock
Valid values for 'frequency' can be found by looking at the following file...
cat /sys/class/misc/mali0/device/available_frequencies
You might also want to disable dynamic voltage and frequency scaling as follows...
echo off > /sys/class/misc/mali0/device/dvfs
I don't believe we provide support for the HP Chromebook 11 despite it being based on the same SoC as the Samsung Chromebook 1. I will double-check this for you. If not, all three Samsung Chromebooks (the Chromebook 1 and both versions of the Chromebook 2) are good candidates for running OpenCL on Mali. The guide you referenced should get you up and running relatively easily.
Regarding your question about device fission, I don't believe this functionality is included in the OpenCL 1.1 driver. As you may know, device fission is a part of the OpenCL 1.2 specification, and sometimes implemented as an extension to CL 1.1. It's worth mentioning that with the Mali-T628 MP6 - as included in both models of the Chromebook 2 - you have 6 cores arranged into 2 core groups, 4 + 2. These appear as separate platforms within the GPU device. This does give you some manual control to split the workload amongst the 6 available cores as you need a separate context for each core group.
I hope that helps... let me know if you have any further questions on this,
Regards, Tim
Dear timhar01
Really thanks for your detailed explanation about those questions
Probably I need to change my target system to Chromebook 2.
Btw, you mentioned "you have 6 cores arranged into 2 core groups, 4 + 2. These appear as separate platforms within the GPU device."
this sentence means if I want I can use 2 max compute units or 4 max compute units whether which GPU platform I choose, right?
Thanks a lot, Tim
Btw, you mentioned "you have 6 cores arranged into 2 core groups, 4 + 2. These appear as separate platforms within the GPU device." this sentence means if I want I can use 2 max compute units or 4 max compute units whether which GPU platform I choose, right?
Yes, that's right. It's not as fine gained an approach as device fission, but it gives you some options.
I have checked about the HP Chromebook 11. It should certainly be possible to use this like the Samsung Chromebook with only a few small changes to the instructions. However it's not something we have tested fully yet and I don't currently have a timescale for when we will publish a guide to include this device. Bear in mind that the HP Chromebook - like the Chromebook 1 - has 4 GPU cores in a single core group. For the 2-core group, 6-core GPU you would need either of the Chromebook 2 models.