flightster

In all seriousness. I have set up several HPCs running on NVIDIA chips. They're running in my garage right now, making me about $4/hr. That is good but at the end of the day, they are commodity hardware chips like any other. Let’s look at AMD (worth 300B) for example. AMD also makes GPUs whose pure number-crunching capabilities rival those of NVIDIA. What does NVIDIA have that AMD lacks that justifies a ten times higher valuation? As of now, a proprietary software layer called CUDA. And just because most of the models driving the current AI hype cycle tend to rely on CUDA, doesn’t mean they always will.

In fact, I see stuff on Hacker News about how someone's hacked CUDA to run, with bugs, on AMD chips. So CUDA is not a true moat - it's a temporary moat. Does a temporary moat from a single proprietary driver justify a 10x valuation?

Now let’s compare NVIDIA to Apple. Is NVIDIA truly worth the NPV of all future Apple's cash flows? Think all future iPhone sales, their 30% app store tax, unbelievable balance sheet goodwill. The 20B check they receive from google every year in exchange for one line of the iOS Safari source code. Billions of high-income users in developed countries. An entire subculture of girls who will simply not reply to green texts.

What about Microsoft? Microsoft has an anticompetitive chokehold over B2B the likes of which the world has never seen. Our company, for example, couldn't leave Microsoft if we tried. The MSR team alone spends $50k monthly on cloud compute. We're paying Microsoft $600/mo for a machine I could build for $90. The entire financial ecosystem is built on the back of MSSQL and Excel. We're talking tens of trillions of USD flowing through Microsoft software every single day. European regulators look down upon anything non-Microsoft (with the occasional pity regulation favoring SAP).

And what has NVIDIA given us? The commodity hardware to train a bunch of chatbots that offer a 1.5x productivity boost to a few very specific sectors of the economy. Are they going to change anyone's life? No. Self-driving cars? At least a decade out. NVIDIA chips are nothing but super fast pocket calculators. They are not the step function for human computation that the market wants them to be.

And finally why am I right? Because I have experience working with NVIDIA hardware and software with my own two hands, unlike 99% of the institutional investor base who probably bases their opinions off of analyst drivel not unlike what you’re reading right now.

So I buy NVIDIA puts on a regular schedule because I think there's some serious tail bubble risk of NVIDIA bringing down the entire SPX. It's like health insurance premium without a deductible. And that's financial advice.

I've been tasked with creating an API to look up a customer based on a variety of fields. A non-exhaustive list of fields could look something like:

- customer_id - customer_first_name - customer_last_name - customer_loan_number - customer_phone_number - customer_email - customer_address - customer_date_of_birth - customer_zip_code

What I ultimately want is customer_id. But what I have is any combination of the subsequent fields. They might be all lowercase, they might not. There might be type-os. The phone number might have parentheses and dashes, it might not.

The address and zip are also wonky - for example customer_zip_code may look like XXXXX-YYYY, or it may just be XXXXX.

And example query would be: given customer_first_name and customer_last_name and customer_address, find the customer_id. The customer_last_name might be lowercase in the query and uppercase in the main database. The customer_address in the query might be "123 Main St" but in the database it might be "123 MAIN STREET".

And the data is in the 100m-1000m rows size, so this has to run reasonably quickly with that in mind. Can't just do a "select * where [x = y] and [p = q] and [t = s]". Would like to pre-load the big table in RAM, and then somehow look up the matches.

Because this almost definitely a common problem in our industry, I was wondering, is there some kind of good database already pre-optimized for this kind of fuzzy lookup?