The Qualcomm video today is doing my head in. I lack the technical knowledge or experience but I get obsessed with finding answers so I keep researching to get a result, which on this subject is to me like trying to read a book in an unknown foreign language.
On the Qualcomms website there are numerous new videos showing what the new Snapdragon 8 can do and the features are remarkably similar to Akida’s. If it’s not us then it‘s unfortunately a good competitor as they already have a large market share and have sold 2Billion AI products to date.
This is their webpage which has the video’s and describes their latest Snapdragon 8 gen 2.
Learn about the advantages of AI on-device, such as improved user privacy and enhanced application performance. Unlock the power of mobile AI today.
www.qualcomm.com
The article I have pasted below is from Qualcomm dated 17/11/22 and discusses the Hexagon AI processor which in researching goes back many years so it’s not new technology but likely improved year on year
Traditionally, laptop performance has been measured by CPU and GPU, but on-device AI processing is now a critical third measure. The integrated Qualcomm Hexagon AI processor in
Snapdragon compute platforms offloads compute intensive AI tasks from the CPU and GPU to dramatically increase performance on the device and deliver new user experiences. In fact, we're excited to share that with the pre-release Procyon AI Inference Benchmark, results show the
Hexagon AI processor scores up to 5x faster than the competitor's CPU and up to 2.5x faster than the competitor's GPU.
We're excited to share that with the pre-release Procyon AI Inference Benchmark, results show the Hexagon AI processor scores up to 5X faster than the competitor's CPU and up to 2.5X faster than the competitor's GPU.
www.qualcomm.com
This article below is so far the best I have found and discusses both neural networks and transformers however our Akida 2.0 which was possibly going to contain the transformers has not been made yet so that rules that out from being on the new snapdragon 8 gen 2.
Qualcomm Snapdragon 8 Gen 2 Delivers More AI For Mobile
Jim McGregor
Contributor
Tirias Research
Contributor Group
0
New! Click On The Conversation Bubble To Join The Conversation Got it!
Nov 16, 2022,04:41am EST
https://policies.google.com/privacy
Listen to article6 minutes
The Qualcomm Snapdragon 8 gen 2 SoC for flagship smartphones
QUALCOMM
The Snapdragon Tech Summit is a multi-day event that showcases the latest mobile technology Qualcomm has to offer. This is the second year that Qualcomm has held simultaneous events in China and Hawaii, as well as streaming the keynote addresses. Day 1 of the Snapdragon Tech Summit kicked off with the introduction of the latest smartphone system-on-chip (SoC) for smartphones – the Snapdragon 8 Gen 2. As expected, it delivers improvements in performance and efficiency for camera, connectivity, gaming, sound, and security. But the biggest punch comes from the use of artificial intelligence (AI) in just about every area. The company went so far as to call it “purpose built for AI.”
Qualcomm uses all of the Snapdragon SoC’s processing elements for AI processing and calls the combination of these processing elements the “AI engine.” Among the enhancements incorporated into the AI engine was a dedicated power plane and a doubling of the tensor processing cores within the Hexagon processor. The result is a 4.35x improvement in performance and an equally impressive 60% improvement in performance per watt efficiency.
Qualcomm also added support for transformer neural network models which are critical for applications like natural language processing (NLP) for speech-to-text and text-to-speech translation. The Hexagon can splice the neural NLP model into smaller elements to run on micro tiles allowing for more efficient use of the processing cores.
Additionally, Qualcomm added support for Int4 data structures. In many cases, this lower precision data structure can be used by neural network models like computational photography image enhancement without a noticeable loss of accuracy while improving speed and power efficiency. The end result is faster and more efficient processing of neural network models.
With the Snapdragon 8 Gen 2, Qualcomm is also introducing the Hexagon Direct Link – a physical link between the Hexagon processor and the other AI processing cores for fast, real-time processing. According to the company, this will enable the first cognitive processing by the image signal processor (ISP) in the Snapdragon 8 Gen 2.
@Diogenese
I saw Qualcomm put out a white paper trying to standardise 8 bits whereas PVDM did a blog recently saying 4 bits is enough. I THINK the part I bolded above where Qualcomm added support for Int4 data structures means they added support for 4 bits which is contrary to their own white paper and I THINK supports what PVDM neural networks plan is?
That‘s all I’ve got for dots to support Akida bring in Snapdragon 8 Gen 2 but contradictory info re transformer NN as we haven’t made them yet.
Hope it makes sense.
As
@Adam said earlier tonight: I might be better off, rather than trying to do all this dot joining to just wait for the announcement. Wise words!
There are ways to sleuth into this. 
