Let's Talk about 9.0 Pie On Pixel 3

[mustang7757]

even though Google not blowing people away by specs what their chips can do will

 

[anon(10092459)]

Here's a link I've been reading
I dont think it's going to cloud with Google.
https://www.fastcompany.com/90247454...in-your-pocket

Good read and I definitely agree in the immediate future with what they are saying. The "right now" of it is getting more on device cramed into that Pixel as possible. I'll continue my thought on the next post.

 

[mustang7757]

Good read and I definitely agree in the immediate future with what they are saying. The "right now" of it is getting more on device cramed into that Pixel as possible. I'll continue my thought on the next post.

Like I said they not cramming more ram just to say hey we got 8gb ram on board they are making their chips more smarter ,efficient,AI faster based... more then the others which is where it really counts in my opinion.

 

[mustang7757]

I agree they def. could be. On my Pixel 2 XL while they worked I still see the iOS gestures as better. Much smooth transitions on what I do on the phone.

I agree , as you know I have quite a few phones and Google gestures need improvements ,can I deal with it yes but would like them to improve it to give a better experience. Sometimes if you haven't used gestures from other phones you don't t see what's missing.

 

[anon(10092459)]

even though Google not blowing people away by specs what their chips can do will

Continued.. I think this is where things will get interesting. The QCOM issue and what to do long-term there. The reason I'm anchoring on the cloud is that's where GA lives and in order to get that all the way on device, I don't know that QCOM can make the long haul and address that kind of computational appetite. Need some really big guns for processining power, power management, etc.

Custom silicone has to come into play at some point. There's a slide from a presentation recently on the larger Titan chip (here) that's titled "Why make our own" and on there it says because there was "No existing solution" for what they needed/wanted to develop for their servers. Custom was the way to go.

 

[anon(10092459)]

Like I said they not cramming more ram just to say hey we got 8gb ram on board they are making their chips more smarter ,efficient,AI faster based... more then the others which is where it really counts in my opinion.

Completely agree!

 

[mustang7757]

Continued.. I think this is where things will get interesting. The QCOM issue and what to do long-term there. The reason I'm anchoring on the cloud is that's where GA lives and in order to get that all the way on device, I don't know that QCOM can make the long haul and address that kind of computational appetite. Need some really big guns for processining power, power management, etc.

Custom silicone has to come into play at some point. There's a slide from a presentation recently on the larger Titan chip (here) that's titled "Why make our own" and on there it says because there was "No existing solution" for what they needed/wanted to develop for their servers. Custom was the way to go.

I think cloud can be a little slower for function on demand, I think they need few chips small to communicate between them to be effective

 

[anon(10092459)]

I think cloud can be a little slower for function on demand, I think they need few chips small to communicate between them to be effective

Agree about the latency. I am all about local, chip powered on device AI. Just need QCOM to keep up with the vision for now. The 845 is good for now, for sure. But man, the speed with which all this going. Wavenet alone takes so much power just for the voice/speech capabilities.

AlphaGo needed 1202 CPUs and 176 GPUs to win that GO competition. How do you ever even dream to get that kind of computational need down to a cell phone??

 

[mustang7757]

Agree about the latency. I am all about local, chip powered on device AI. Just need QCOM to keep up with the vision for now. The 845 is good for now, for sure. But man, the speed with which all this going. Wavenet alone takes so much power just for the voice/speech capabilities.

AlphaGo needed 1202 CPUs and 176 GPUs to win that GO competition. How do you ever even dream to get that kind of computational need down to a cell phone??

Yes because it's the beginning stages of what's to come, I'm fascinated by this that all this can come to a phone. I'm a auto technician my specialty is diagnostic/trouble shoot problem with cars . You know how fast the engine computer,trans computer,abs computer,body control computer communicate with each other as you drive your car ? It's unbelievable they all communicate through networking called CAN communication area network ..or how many times the computers are self checking for faults or problems crazy stuff almost 25 years doing this . But for phone is even more amazing.

 

[Mike Dee]

Yes because it's the beginning stages of what's to come, I'm fascinated by this that all this can come to a phone. I'm a auto technician my specialty is diagnostic/trouble shoot problem with cars . You know how fast the engine computer,trans computer,abs computer,body control computer communicate with each other as you drive your car ? It's unbelievable they all communicate through networking called CAN communication area network ..or how many times the computers are self checking for faults or problems crazy stuff almost 25 years doing this . But for phone is even more amazing.

My odometer mysteriously added 25,000 miles to my car....is that fixable with an update?

 

[mustang7757]

My odometer mysteriously added 25,000 miles to my car....is that fixable with an update?

Easy fix
The problem between steering wheel and the seat

 

[anon(10092459)]

Yes because it's the beginning stages of what's to come, I'm fascinated by this that all this can come to a phone. I'm a auto technician my specialty is diagnostic/trouble shoot problem with cars . You know how fast the engine computer,trans computer,abs computer,body control computer communicate with each other as you drive your car ? It's unbelievable they all communicate through networking called CAN communication area network ..or how many times the computers are self checking for faults or problems crazy stuff almost 25 years doing this . But for phone is even more amazing.

man, I hear ya! It's exciting stuff for sure. I absolutely am looking forward to how much more on device AI they can get in. Just turn more and more of the operational parts of the device over to the neural net!

 

[mustang7757]

man, I hear ya! It's exciting stuff for sure. I absolutely am looking forward to how much more on device AI they can get in. Just turn more and more of the operational parts of the device over to the neural net!

It's the future, who wins in AI is going to be successful. 3 companies I see Google/Apple/Samsung
Rumors Samsung going big in 2019 with AI ..I think Google will lead the way

 

[anon(10092459)]

I think cloud can be a little slower for function on demand, I think they need few chips small to communicate between them to be effective

Alright... So this has had me thinking since we brought it up .. The future AND if Google will eventually do custom silicone considering the Titan M they just put in the Pixel 3.

So going back and looking at what they've changed since Sundar took over in 2016, one of the things is they started building custom silicone for their servers specific for the AI/ML. Tensor Processing Unit (TPU) was custom silicone built and fitted on the servers in 2016. 2017 they build the next generation called Cloud TPU, faster and more advanced. TPU v3 is in beta..

https://cloud.google.com/tpu/

I think longer term, since Google is doing custom silicone for the AI/ML in the servers, we get a smaller version TPU very soon.

 

[mustang7757]

Alright... So this has had me thinking since we brought it up .. The futute AND if Google will eventually do custom silicone considering the Titan M the just put in the Pixel 3.

So going back and looking at what they've change since Sindar took over in 2016, one of the things is they started building custom silicone for the servers specific for the AI/ML. Tensor Processing Unit (TPU) was custom silicone built and fitted on the servers in 2016. 2017 they build the next generation called Cloud TPU, faster and more advanced. TPU v3 is in beta..

https://cloud.google.com/tpu/

I think longer term, since Google is doing custom silicone for the AI/ML in the servers, we get a smaller version TPU very soon.

I like to see that in action