On GPU/CPU Hegemony

Summary

• Next 5 years in processor leadership.
• Strategy topic, not valuation.
• Community input and discussion.

In an earlier discussion I mentioned the idea of "CPU/GPU Hegemony" in light of NVIDIA’s recent purchase of ARM.   I have to say none of the articles coming out on SA are quite getting to the heart of what I perceive about this deal.

My belief is that while CPUs are and will always be important, and will never go away, GPUs surpassed CPUs in potential and performance some time ago. 

GPUs do not claim the breadth of usability of CPUs, but they have shown their potential under very heavy work loads in Artificial Intelligence, machine learning, and high performance computing in general.  With the advent of more utilitarian AI (not AGI or Artificial General Intelligence, like the Borg in Deep Space9) for image recognition and language understanding, commercial and research enterprises are really making a go of incorporating this technology into problem solving.  And that sort of cracks the door on what could be. 

Artificial intelligence has always been an exciting and interesting place to try and invest, to identify the alpha, the difference makers.   All of the largest Clouds or Cloud Service Providers (Google, Alibaba, Amazon, Microsoft Azure et al) have incorporated these products and are building services to address this nascent market.

With NVIDIA's proposed purchase of CPU Intellectual Property (IP) provider ARM from Softbank, the game is changing.

In the past ARM CPUs haven't been considered on par with the powerful x86 CPUs AMD and Intel build.   But recent executions by Amazon with Gravitron (https://aws.amazon.com/ec2/graviton/) and Fujitsu's A64FX (https://www.fujitsu.com/global/products/computing/servers/supercomputer/a64fx/) which is the heart of the world's fastest supercomputer, show that ARM CPUs have a lot of potential.

In the end, the company that can wrangle CPU + GPU assets for the best customer experience is the one that can dominate the market. My sense is it will not take too many years to figure out.

There are now potentially three companies with similar assets at the center of this discussion.  The question is which one prevails and why? Are there any others who should be considered such as Apple or QCOM, or perhaps a startup?  

It's a little unconventional, but the focus is blog is which company provides the best risk/reward for investors over the next 5-10 years.

This isn’t about valuation.

It is about assets, strategy and execution -- the core to I what I think are long term hold investments.

Off the top of my head, I’ve listed strengths and weaknesses in no particular order for each of these companies. I’m trying to be intellectually honest here, not cheer leading. Interestingly they all have roots in Personal Computers, and they each have obviously different strengths and weaknesses. 

One of these companies will shape the next era of computing:

I’m sure I’ve missed some attributes, and I’m sure some will disagree with my list, so I welcome all comments.

Please weigh in with your views:

• Who has the stronger assets?
• What weaknesses could be detrimental or lethal?
• What are your thoughts on strategy?
• Ultimately who prevails?

===========

Intel – Pros

• CPU market leadership
• X86 Installed Base
• Customer relationships
• Legacy software compatibility
• FPGA and ASIC assets
• Solid engineering
• Mobileye
• Highly Profitable
• Healthy R&D spending
• Patent Portfolio/IP

Intel – Cons

• Management appears rudderless
• Manufacturing (leaning negative, certainly not the positive it’s been in past)
• OneAPI – Trying to establish a new programming language with no products
• Untested in dGPU
• AI is a big question mark, no traction despite large investments
• Not a software company

===========

AMD – Pros

• Ryzen/CPU architecture
• CPU/GPU integration experience (semi custom)
• Fabless
• CPU Team
• Infinity fabric
• dGPU legacy
• Loyal consumer customer
• Patent Portfolio/IP
• Lisa Su executing well in CPU
• Exascale wins
• Roadmaps
• Nimble
• Data Center/server traction

AMD – Cons

• R&D focus on CPU
• GPU Software strategy
• GPU footprint outside Gaming/GPGPU
• AI experience
• Not a software company
• R&D budget limited
• Financials - most challenged of the 3

===========

NVIDIA – Pros

• Ownership of leading Parallel Compute Platform
• GPU market leadership
• NVIDIA Research Team
• R&D Budget
• Patent Portfolio/IP
• Fabless
• CUDA API
• Management Team
• Vision
• Systems architecture experience
• GPGPU footprint
• NVIDIA Marketing
• Acquisition history – regulatory and cultural issues managed with aplomb on Mellanox
• Free Cash Flow
• Establishment of growing verticals (Automotive, Medical, HPC, Drones, AVs, Robotics, smart cities)
• Software focused
• ARM Team, business as usual
• Tegra SoC business
• Networking and DPU

NVIDIA – Cons

• ARM is a different business model than selling chips
• ARM integration
• Regulatory issues
• ARM China
• Perceived or potential conflicts of interest with licensees
• Challenge of pushing AI solutions through IP license (vs owning yourself)
• Competing with customers
• Too many projects risk losing focus