Hi,
This may sound like a very stupid question - however, would this phone:
LeTV LeEco Le Max 2 which has he Snapdragon 820 be able to run NPE powered apps? It runs Android M or eUI 5.8. I see no reason it should not but I just wanted to double check that NPE doesn't have to be activated or anything devices etc.
Thank you for your time,
Kind regards,
Dan
Opinions expressed in the content posted here are the personal opinions of the original authors, and do not necessarily reflect those of Qualcomm Incorporated or its subsidiaries (“Qualcomm”). The content is provided for informational purposes only and is not meant to be an endorsement or representation by Qualcomm or any other party. This site may also provide links or references to non-Qualcomm sites and resources. Qualcomm makes no representations, warranties, or other commitments whatsoever about any non-Qualcomm sites or third-party resources that may be referenced, accessible from, or linked to this site.
As long as you include the snpe-release and setup your app correctly, it should run just fine.
Thank you! So there is a similar phone with the 652 SoC...that should work as well I assume? So as long as we release SNPE in our app any SNPE supported SoC should run our app at improved performance?
If you go to the documentation included in the SDK, there is a chart of all supported chips.
The 625 has CPU and GPU support officially (I haven't personally tried these platforms though).
Can anyone clarify a more specific series of questions.
1) How does sNPE / NPE relate to Androids NNAPI? Is this a vendor specific integration which eventually will become an NNAPI device implementation, or is it something Qualcomm made entirely for their own devices in parallel?
2) I assume the Qualcomm DLC special model format will be redundant in the face of TF Lite's flatbuffers
3) Will these older chips that arent the coming 845 flag ship such as the 820 be supported in the newest Android and gain acceleartion from the NNAPI?
The marketing says: The Qualcomm® Snapdragon™ Neural Processing Engine (NPE) SDK for artificial intelligence (AI) is designed to help developers run one or more neural network models trained in Caffe/Caffe2 or TensorFlow on Snapdragon mobile platforms, whether that is the CPU, GPU or DSP.
I assume that means that simply being supported by NPE SDK does not itself guarantee sufficient performance for complex Computer Vision tasks like real time Object Detection on video and additional OpenCV processing?
Considering the way Vendors, older handsets and Android updates go, there is a history of lower end devices simply not getting that attention or making the cut.
4) Can anyone actually quantify the performance difference of say the 820 with sNPE vs normal CPU? Is there any negative affects of simply offloading this to the GPU under sNPE SDK, e.g. heat and battery drain as we currently see when running any of the normal TF Mobile (not TF Lite) demos on Android or iPhone? Is there any example demos which show the kind of performance benefit on low end chips, and would that performance be expected this year in Android 8+?
5) If I was to integrate the Qualcomm sNPE SDK into an Android app would it become redundant when TF Lite is more stable, or would theyre be a situation where a developer would want both code paths to be in the app?
Sorry for all the questions but its very unclear what is happening here going forward, and I get the feeling that the only safe bet would be to stick to a Flag Ship device with the 845, but I would like to be wrong since they aren't available yet and will be very expensive on release.
Thanks!
Can anyone clarify a more specific series of questions.
1) How does sNPE / NPE relate to Androids NNAPI? Is this a vendor specific integration which eventually will become an NNAPI device implementation, or is it something Qualcomm made entirely for their own devices in parallel?
2) I assume the Qualcomm DLC special model format will be redundant in the face of TF Lite's flatbuffers
3) Will these older chips that arent the coming 845 flag ship such as the 820 be supported in the newest Android and gain acceleartion from the NNAPI?
The marketing says: The Qualcomm® Snapdragon™ Neural Processing Engine (NPE) SDK for artificial intelligence (AI) is designed to help developers run one or more neural network models trained in Caffe/Caffe2 or TensorFlow on Snapdragon mobile platforms, whether that is the CPU, GPU or DSP.
I assume that means that simply being supported by NPE SDK does not itself guarantee sufficient performance for complex Computer Vision tasks like real time Object Detection on video and additional OpenCV processing?
Considering the way Vendors, older handsets and Android updates go, there is a history of lower end devices simply not getting that attention or making the cut.
4) Can anyone actually quantify the performance difference of say the 820 with sNPE vs normal CPU? Is there any negative affects of simply offloading this to the GPU under sNPE SDK, e.g. heat and battery drain as we currently see when running any of the normal TF Mobile (not TF Lite) demos on Android or iPhone? Is there any example demos which show the kind of performance benefit on low end chips, and would that performance be expected this year in Android 8+?
5) If I was to integrate the Qualcomm sNPE SDK into an Android app would it become redundant when TF Lite is more stable, or would theyre be a situation where a developer would want both code paths to be in the app?
Sorry for all the questions but its very unclear what is happening here going forward, and I get the feeling that the only safe bet would be to stick to a Flag Ship device with the 845, but I would like to be wrong since they aren't available yet and will be very expensive on release.
Thanks!
sorry the form failed, then double posted, now i cant delete my duplicate message, it says:
Qualcomm Technologies, Inc. requires you to have a validated account to access this area. Please log in or create an account.
Sigh...
The NN API is exciting. It should allow developers to connect low level hardware to a unified API.
However, SNPE existed before the NN API. It acts as an optimizer and manager for neural nets. I would appreciate if Qualcomm would add NN API support for the sole reason of optimizing neural nets for more than just SnapDragon chips.
So there is no relationship between the two and support for NNAPI by Qualcomms chips and its implementation by phone vendors will be dependant on their desire to do so?
From what I can see here:
https://www.qualcomm.com/news/onq/2018/02/01/how-can-snapdragon-845s-new...
Its the 845 that has the special DSP for low power high performance inference, and it supports Tensorflow Lite (which I assume will be through NNAPI) and also their own Qualcomm NPE SDK.
But, I assume that means that those cheaper devices with chips like 820 may only get accelerated performance with the NPE SDK?
Is anyone at Qualcomm able to comment?
It would be good to know that developing now with the Qualcomm NPE SDK on existing devices means that later those same devices will likely to be supported by say Androids NNAPI and while the development tools and enviroment will change the investment in hardware wont be wasted.
The alternative seems to be that I should wait for 845, or just focus on the latest TF Lite support which includes iOS.
4) Can anyone actually quantify the performance difference of say the 820 with sNPE vs normal CPU? Is there any negative affects of simply offloading this to the GPU under sNPE SDK, e.g. heat and battery drain as we currently see when running any of the normal TF Mobile (not TF Lite) demos on Android or iPhone? Is there any example demos which show the kind of performance benefit on low end chips, and would that performance be expected this year in Android 8+?
hi, here is the Nexus5x 7.1.1, cpu 808, son of google
inception_v3.dlc:
sNPE 1.13 CPU: success
sNPE 1.13 GPU: Not Support
Tensorflow Mobile & Lite: Success
Custom Model DLC:
sNPE 1.13 CPU: crash (can't get libqti-perfd-client.so, but inception is ok, why??!! )
sNPE 1.13 GPU: Not Support
Tensorflow Mobile & Lite: Success
=================================
second model is 835, android 7.x
inception_v3.dlc:
sNPE 1.13 CPU: success
sNPE 1.13 GPU: success
Tensorflow Mobile & Lite: Success
Custom Model DLC:
sNPE 1.13 CPU: crash (can't get libqti-perfd-client.so, too)
sNPE 1.13 GPU: Success
Tensorflow Mobile & Lite: Success