In the realm of high-performance computing, the bandwidth limitations imposed by GPUs and the challenges related to location in centralized systems present formidable hurdles. However, by embracing decentralized solutions such as edge computing, distributed paradigms, and optimized data movement strategies, the potential to alleviate these constraints and unlock enhanced computational power becomes a promising reality.
I became the Marketing Committee Director (if you're into the DAO titles) of a company that was on a quest to mitigate bandwidth constraints and optimize HPC software by exploring its decentralized solution - in the summer of 2023.
In other words (If you're more of a traditionalist capitalist) I was the Cheif Marketing Officer of HPChain - a decentralized GPU network for high performance computing power. Also something I like to refer to as providing data that is "the oil of the internet".
Compounding in numbers at unprecedented historical rates - ChatGPT pushed forth a new era of computationally driven capability use cases. It drove the need for GPUs - graphics processing units. caused a crash in NVDIA. By the way, GPU cards are the preferred hardware for AI processing purposes (instead of CPUs, Central Processing Units) because AI computations heavily rely on parallel processing, which GPUs enable better. GPUs have higher throughput, better memory bandwidth, and better cost-performance than CPUs for deep learning computational needs.
As a result - NVIDIA is basically now the standard on AI workload processing being the main provider of GPUs. Sadly Intel, the one you actually have probably heard of, fell behind in the chip maker race. I wrote about that more here ~The recommendation engines, natural language processing, and generative AI large language models like ChatGPT all are being accelerated on Nvidia platforms
Nvidia's Form 10-Q [PDF] showed how sales in China grew faster than elsewhere, going from $1.15 billion in Q2 2022 to $4.08 billion in Q2 2023.
Curiously, revenue from the tiny city state of Singapore grew even faster – from $562 million in 2022 to $2.7 billion in 2023. Singapore, the country-city-state I flew back to this year to begin working on progressing decentralization of the data spreading through these chips. It's a directly positive feedback loop, more chips = more data. More data = more chips. The opposite, is also true. For now, we can watch as people scramble to Intel for scrappy alternatives admist the NVIDIA GPU shortages. Hey, at least Intel is getting some piece of the pie after their historical flop!
We have innovative companies, like the Singaporean company I worked with (see video above) creating decentralized solutions for enabling access to AI globally. Currently as it stands today, the proximity to a data center plays a critical role in high-performance cloud computing due to latency (the delay between sending a request and receiving a response). In other words, you'd need a data center under 100 kilometers away to create a standardized performance.
The solution for progress would be
Harnessing idle computing power
Creating sustainable and smaller data center frameworks
Onboarding a global community to utilize the product
The last one makes me laugh. A lot. Because it's the actual way to create anything decentralized, a distribution of power (and physical GPU power in this case!) yet somehow this obvious goal gets shafted to the side as a later care, once the 5 million dollar seed fundriaise is first accomplished.
Ah yes, I'll focus on the thing that would make this company a reality, the thing that money itself can't buy, once I have received money from others to make costly mistakes with. Because for whatever reason, I'll still focus on the product and marketing only. And leave community as an afterthought.
I'm excited for life ahead. And the explosion of community-driven commerce. What happens when we actually put community first? And create a product people will actually use.
We have all the solutions in the world already. Clean water. Ending world hunger. Free electricity. We don't have people using these things though. The user drives the market. You're analyzing the market but have no users. (Now I'm just ranting at all the techbros I've had the honor of working with to no avail).
My purpose for writing this blog was to share how the AI world is still limited by the physical, as it stands today on this brisk end-of-November night.
Speak again later,
Dr. Zee
Comentários