Hacker News new | ask | show | jobs
by michaelgiba 782 days ago
I actually went down nearly this exact same rabbit hole recently. Specifically I was curious why, given all of the demand for GPUs, compute not been commoditized and resold through public markets. This also led me to the Enron's ideas before their collapse :)

One thing I found surprising is how hard it is to discover prices of this compute time in today's world as compared to traditional commodities like oil. A factor could be a difference in regulation, for example in the US I believe there are requirements for reporting energy trades between producers and consumers.

side note: I made this minimal website for tracking these prices which shows some of the disparity: https://computeindex.michaelgiba.com/

6 comments

I am not sure the actual use of compute is fungible enough as usually you'd need data to travel with the compute then (or somehow connect to it). If it were purely GPU cycles with a little bit of programming it could probably be a lot more commodity like.
Fungible is a spherical economic simplification. Some of our theoretically most fungible resources like electricity or bandwidth or water are not fungible in most markets. Even crude oil is not actually fungible (example sweet versus sour).

There is plenty of demand for pure compute. However no market has been created and the large compute providers have no incentive to commoditise compute (and are smart enough to recognise the value by creating their various lock-ins).

Lots of things are fully fungible (e.g., money, fuels of the same specs. Water is fungible but the supply network is not.).

Demand for compute has created a market but not necessarily a futures market or an exchange/CLOB like thing. The latter needs more than just supply and demand.

"Fully" is anything but fully. It is an unrealisable ideal.

Money is not fungible across currencies or across time or in different locations or a wide variety of different characteristics. Even 1USD at the same time in the same place can have have different values depending on how/where it is stored ($1 on credit card, $1 in USDT, $1 note at a place that doesn't accept cash, $1 silver coin, $1 in Argentina, $1 promised, etcetera).

Even "fuels of the same specs" are not really fungible. Only approximately on some markets. Russian oil? Fungible is an ideal and markets are designed to try and get closer to the ideal but markets and specs are nowhere near perfect.

Two dollars in the same situation are fully fungible between each other - if they are in different systems, or even gold, silver, etc. they are not in the same situation or even money or dollars (but something of a dollar value). For example, dollars in a checking account are fully fungible, you would not know which specific dollar you get if you transfer one.

That situationally swapping between fungible things can be blocked etc. doesn't mean that they thing itself isn't fungible.

I am not sure either. Although maybe it just comes down to how the purchased compute is “delivered”
Perhaps. It might also collapse to basically an energy market if it were fully fungible.

Market design is really tough, but it might be something worth thinking about here.

I suppose one complication is that not all flops are equal, computation is not only "how many kW/h did this math take" but "how much memory and at what bandwidth"
kWh, not per hour
thanks :)
There are definitely thresholds you reach that make it difficult to generalize. large cluster setups can differ significantly and actual usability of the clusters makes a big difference too.

However I guess there are some differences like these even in real commodities like oil based on the blend/refinement/producer/geography.

I work at a startup that does something like that. It's called Nunet and it's product is to build a descentralized compute economy. It's a spinoff from singularitynet so the main focus is machine learning and web3, but the overall goal is to have a compute agnostic platform of descentralized systems that commoditize latent compute power.

For every computer job you need four things, compute, data, storage and algorithms. The goal is to have an economy where everyone provides what they have in exchange for either tokens or access to something they don't have.

Currently the focus is the use case of descentralized SPO, but for instance there is already research in integrating the platform with kubernetes.

You can learn more at NuNet.io with the white paper.

> compute not been commoditized and resold through public markets.

I'm not very good at language, but I take it cloud computing (and e.g. spot markets) don't fit this bill, right?

I could imagine a company that abstracts over the cloud providers that can combine the spot markets etc somehow, but then I don't think the cloud providers would be happy with that because they want exclusive contracts.

There’s a company called Archera doing this - they’re doing well, cash flowing.

https://archera.ai/

However, I think it’s 1. absolutely not insurance, and they’ll have to stop using that word when the lawyers get involved and 2. best understood as an alternative AWS plan to GIs and SPs. So essentially there’s a combination of discount/term/guaranteed capacity that the market wants and Amazon isn’t supplying, and Archera are supplying that synthetically. Cool business!

This sort of business model is generally known as "cost engineering". It can work, to an extent, when you're small enough to kind of fly under the radar. Once you grow enough to start impacting margins for the big players that actually control the resources then they become highly motivated to crush you.
That would certainly be an obstacle.

Hypothecally it could be beneficial to other smaller providers. I think it all comes down to how the purchased compute would end up being used by purchasers

Do you mean something like TensorDock? https://www.tensordock.com/host
Sort of, gpulist.ai as well