https://www.reuters.com/technology/apple-says-it-uses-no-nvidia-gpus-train-its-ai-models-2024-07-29/
Nvidia is the clear leader, but Apple’s choice of Google chips improves their competitive position. Competition in AI chips heats up.
" Including the chips made by Google, Amazon.com (AMZN.O), opens new tab and other cloud computing companies, Nvidia commands roughly 80% of the market."
No comment from Apple but it also states that you have to use Google’s cloud to get access to their chips so Apple isn’t buying the chips but leasing services from Google. Also Google does buy NVDIA chips so it’s not really clear if it is true that Apple doesn’t use NVDA chips.
Google Tensor (TPU) vs. Nvidia Graphic (GPU) and then there is the Tesla Dojo and the upcoming Tesla HW-5 or AI-5.
Non of the above manufacture chips, they hire fabs. Many of their chips are ARM based.
As I have note before, neural network based AI is a Brute Force technology that mimics the brain. I found a relevant article:
How does that fit in with technology that mimics the brain? Intelligence uses energy to reverse entrophy…
Tracing oxygen consumption, the brain accounts for about 20% of the body’s energy consumption, despite only representing 2 percent of its weight.
How much energy do we expend using our brains? | Bond University | Gold Coast, Queensland, Australia.
The Captain
Captain your post is very interesting. Up to the reversing of entropy. That never happens. Energy can be focused on doing work but escapes as entropy. There is no other definition in engineering for entropy.
That doesn’t really get into TPU vs GPU but talks mostly about power. Although that is important it has nothing to do with what Paul brought up. Paul’s post is important because everyone is trying to get from under NVDA and the GPU which NVDA has basically cornered. Everyone I talk to, that has any idea about AI says that GPU’s are the only viable option now. But something will come in and disrupt sooner or later. I don’t think it will be Intel that disrupts though it could be AMD. I have heard of other pre-ipo companies that are working on chips to disrupt NVDA but nothing has actually shown up.
I have no doubt that Apple would use TPU’s but maybe for inferencing. Has anyone heard of any chip that has a chance of taking on NVDA and the lead they have taken?
IBM and Qualcomm could get interesting. Deep enough pockets to get a good product.
Thank you!
Globally it never happens but locally it does provided you use energy from outside the system, like from the Sun.
In general, entropy increases as order decreases. This means that as a system becomes more disordered, its entropy increases. This is because there are more ways for the particles in a disordered system to be arranged compared to a highly ordered system.
To create intelligence you have to organize the brain’s or the data center’s neural networks, as the case ma be, with requires energy.
The Captain
I had to find out how Google’s method differed from Nvidia’s. Irrelevant to the layman but I’m a data processing geek and a high tech investor. 
Tesla’s Dojo. Musk has not touted it so as not to antagonize Nvidia, an important Tesla supplier, at last for now.
About leads, I don’t have a clue.
The Captain
AMD Q2 Earnings: Key Updates About AI
https://seekingalpha.com/article/4708588-amd-q2-earnings-key-updates-about-ai
AMD Q2 Analysis: With Data Centers Dominating, It’s Ever Closer To Nvidia
The Captain
Maybe you could explain how Google’s method differs from NVIDIA’s. That would be interesting.
Ok Here is a video that explains TPU VS GPU. It’s short but gives a basic understanding of what is happening. It very well could be that Google doesn’t sell the TPU ever but they could still build out a nice business with a moat.
Yes but that defeats the definition of entropy.
Simple and less scalable.
TPU does not have the drivers. We had someone come in here and explain NVDA v AMD and INTC GPUs. Unreal and Unity demand generations of drivers. NVDA has constantly kept up with that. AMD and Intel do not have the driver for older releases.
Without the gaming industry TPU and other GPU are not entering the most profitable part of the market. Or one of the major profit centers.
Huh? The Captain was correct. On the scale of the universe entropy is increasing, but locally entropy can be decreasing.
Quick example: Life on Earth is an example of decreasing entropy. But life on Earth requires energy from the sun, which is an example of increasing entropy.
No laws of thermodynamics are violated.
Thank you!
The Captain
Quite easy from a macro perspective. Very difficult from a micro perspective. I’ll take the easy route and link to some micro perspective.
CPU → GPU → TPU
I started programming in machine language meaning addressing the CPU directly. The CPU could to just one task at a time
CPUs got more complex trying to do more than one thing at a time, it was the route Intel took, CISC, Complex Instruction Set Computer. The problem with CISC is that some of the functionality was wasted. ARM took the opposite route, RISC, Reduced Instruction Set Computer. Both had the same limitation, still one thing at a time. The obvious solution was to do more than one thing at a time. Operating systems developed things like Paging which was to work on several tasks on different pages loaded into memory as needed. This allowed Multitasking but that was still one thing at a time but the computers had become fast enough that it was not noticeable to the users and efficiency increased.
Vacuum tubes (valves in British English) gave way to transistors which introduced Solid state computers. Solid state gave way to computers on a chip. This technology allowed many CPUs to exist side by side next to each other, opening the route to parallel computing, working at different parts of the problem at the same time. The challenge was how to split the problem into parts that could be worked on in parallel. This was solved by software (and I don’t have a clue how).
As technology shrank chip size, it became possible to increase parallelism but not all jobs have the same hardware requirements, some need more memory, some more compute power. This is how
arrived. Each designed for a specific task. Some sharp programmers found they could be used for more than just the intended tasks.
How does a GPU work?
GPUs work by using a method called parallel processing, where multiple processors handle separate parts of a single task. A GPU will also have its own RAM to store the data it is processing. This RAM is designed specifically to hold the large amounts of information coming into the GPU for highly intensive graphics use cases.
Introduction to Cloud TPU
Tensor Processing Units (TPUs) are Google’s custom-developed application-specific integrated circuits (ASICs) used to accelerate machine learning workloads. For more detailed information about TPU hardware, see System Architecture. Cloud TPU is a web service that makes TPUs available as scalable computing resources on Google Cloud.
In this page:
Some of the technology was used to break the Geman Enigma code
Systolic arrays were first used in Colossus, which was an early computer used to break German Lorenz ciphers during World War II.[1] Due to the classified nature of Colossus, they were independently invented or rediscovered by H. T. Kung and Charles Leiserson who described arrays for many dense linear algebra computations (matrix product, solving systems of linear equations, LU decomposition, etc.) for banded matrices.
The Captain
Thanks Captain. Here is something Google put out on their TPU’s and gives a better understanding of why Apple went with it.
Machine learning models trained in the cloud increasingly need to run inferencing “at the edge”—that is, on devices that operate on the edge of the Internet of Things (IoT). These devices include sensors and other smart devices that gather real-time data, make intelligent decisions, and then take action or communicate their information to other devices or the cloud.
Because such devices must operate on limited power (including battery power), Google designed the Edge TPU coprocessor to accelerate ML inferencing on low-power devices. An individual Edge TPU can perform 4 trillion operations per second (4 TOPS), using only 2 watts of power—in other words, you get 2 TOPS per watt. For example, the Edge TPU can execute state-of-the-art mobile vision models such as MobileNet V2 at almost 400 frames per second, and in a power efficient manner.
This low-power ML accelerator augments Cloud TPU and Cloud IoT to provide an end-to-end (cloud-to-edge, hardware + software) infrastructure that facilitates your AI-based solutions.
The Edge TPU is available for your own prototyping and production devices in several form-factors, including a single-board computer, a system-on-module, a PCIe/M.2 card, and a surface-mounted module. For more information about the Edge TPU and all available products, visit coral.ai.
Edit: I am sorry it was the same thing you posted.
Twice a good thing is twice as good! 
Think of Teslas as EVs driving on the Edge. Tesla’s HW-x, or AI-5 are inference computers developed with low power ARM technology while Dojo is the machine learning computer at the Core.
What would be really interesting is benchmarking Google’s TPUs vs. Nvidia’s GPUs vs. Tesla’s HW-5.
The Captain
That is not the definition of entropy. Entropy happens at random rates but it is not focused as to decrease. If the Captain means that the system offered is more efficient that is so possible but is not about entropy. It is entropy happening more so later than now.
en·tro·py
/ˈentrəpē/
noun
Physics
a thermodynamic quantity representing the unavailability of a system’s thermal energy for conversion into mechanical work, often interpreted as the degree of disorder or randomness in the system.
“the second law of thermodynamics says that entropy always increases with time”
lack of order or predictability; gradual decline into disorder.
“a marketplace where entropy reigns supreme”
Leap, you’re wrong. Entropy always increasing only is valid for an ISOLATED SYSTEM. Entropy can, and does, decrease for systems in certain situations (such as those with energy inflows), and this is NOT a violation of the second law or physics in general. Stop digging this hole you are in.