Hacker News Reader

Comment by justinlardinois | Hacker News Reader

I had a professor who liked to say things like "RAM is free" when talking about how the cost of computers has come down over the course of his life. I always rolled my eyes and thought yeah, it's basically free when you're an accomplished academic who brings in untold amounts of grant money and makes even more from consulting than you do from your university salary. But to a broke college student, anything that costs more than a few hours of minimum wage =/= free.

I really like this concept of cheap, all-in-one computers. We've had Arduinos and Raspberry Pis, but stuff like this really opens a lot of doors. $40 is a great price for a computer, but I would never consider using it for a project where its destruction is a legitimate risk (anything exposed to the elements, or liquid, or drones, for example), but with a $9 computer I would.

View on HN

I think you grossly misunderstood him. He's just saying that it's many orders of magnitude cheaper (more available) than in his days, if you go back a few decades you had to do serious bit wrangling in almost every application to get things to fit in memory, while those days no one cares your app uses 1kb more or less of RAM; he's probably not talking about how much specific RAM cards cost on your PC, or how much those dev boards cost.

On a side note, if you compare how much you pay for your college, directly or indirectly (including housing, books, food), $40 for an educational kit that you spend many hours on is almost free.

I understood him perfectly, and he had plenty of stories about limited resources back in the old days. One of my favorites was about the time he rewrote a routine his boss had implemented so it ran in O(log n) instead of O(n^2), only to be fired by his boss for one-upping him and then rehired by his boss's boss the next day. Overall his class was a great experience.

I just think that, especially when it comes to money, there's a disconnect between Fermi estimates and reality.

On one hand, I can sort of agree with your prof. I learned coding on a system that had 16 kB total memory. The laptop I'm using now has 16 GB RAM and 250 GB storage.

On the other hand, when I'm eyeballing a new environment I'm building in the cloud, one of the first things I'm asking myself is - how much total RAM? (because that's one major driver of cost)

It's all in the perspective.

Definitely all in the perspective, though of course the premium paid for more RAM does not necessarily represent an accurate extra cost incurred by the provider. Amount of available RAM is very often pretty far disconnected from the cost to provide the good or service and is just acting as a price tiering mechanism.

This is true for cloud services as well as other things like cell phones (eg. the 2*X GB version is often significantly more expensive than the X GB version, way more than the cost of the extra memory alone would warrant if you just considered the cost of the chips).

A multi user system that I bought in the mid 80's held a max of 4mb of memory and a max of 144mb hard drive. I think the memory was something like $4000 (in 80's dollars, that's close to $9000 today) for 2mb of memory. So that was just the cost of the memory. On that system by the way you could hook up 26 Wyse terminals. We had maybe 12 terminals hooked up with 2mb of memory and a 72mb hard drive. No graphics everything was pretty system. It ran accounting and job scheduling systems (some of which I actually wrote myself as a non-programmer). At the time I think dinner for two was about $35 for a nice place. Healthcare was $65 per month for the best possible plan. IIM I was a few years out of college at the time.

> We've had Arduinos and Raspberry Pis, but stuff like this really opens a lot of doors.

I think this is an inflexion point for IoT at amateur level. We have the CHIP for RPi-level devices, running full Linux, with WiFi and BT onboard.

We have particle.io and soon digistump.com for Arduino-level devices with WiFi and soon cellular onboard.

My prediction: the number of IoT projects I'll see next time at the Maker Faire will just explode.

I have a Photon (by particle.io), I've preordered a CHIP and an Oak (by digistump.com). Very excited to play with the new toys.

Power is still a problem for IoT.

In addition, this thing will not ship for anything close to $9 in the near future.

The BLE and WiFi bill of materials is more than that-it's why the BeagleBone Black and RaspberryPi don't include them.

I'm surprised there isn't more use of Power-over-Ethernet or similar concepts in projects. Ethernet doesn't need complicated configuration or crypto for basic security. It's not a good choice for general consumer products (how many people have cat6 running through their house?) but I imagine that won't slow down people assembling IoT projects.

PoE is on one hand (relatively, for ~$10 devices) expensive to do, with all the signaling and filtering needed, while on the other hand, not abundant in the environments where these projects are run.

I do have an 8-port-1Gb/s-switch with 4-PoE ports at home. It cost 5 times as much as the "plain" 8-port-1Gb/s with no PoE, and I'm the only person I know who has one at home.

Oh definitely. "Real" PoE is expensive and unhelpful but it doesn't have to be standard PoE. You can just steal some pairs from cat5 and still get some form of ethernet. There are even cheap splitters/injectors out there. You won't get the nifty auto-negotiation, filtering, or anything else but it'll work.

> Power is still a problem for IoT.

I'm not sure about that. If it's a stationary, long-term thing, just tap the main power conduit on whatever device you're hacking.

If it's mobile or short-term, Li-Ion cells are pretty awesome. For a few bucks, you could buy a Li-Ion cell, and a smart charger that will switch the load between battery and main power depending on voltage levels.

Many IoT concepts are remote or battery-powered. Most of the single-board computers on the market will drain a 9V battery in a couple of days.

Wireless power transmission will be the enabler that causes IoT to really advance.

ESP8266 is where the excitement is now. You can buy three ESP-12's on Aliexpress for the price of a CHIP and get free shipping to boot. No Linux, but you can program it now with the Arduino IDE. FCC approved 802.11 b/g/n built-in, SPI, UART, a few GPIOs, faster and more RAM / FLASH than Arduino, etc.

https://nurdspace.nl/ESP8266#Technical_Overview

I hope he wasn't a professor of computer science.

He was; it was a course on operating systems. As darkmighty noted, what said made perfect sense from a math perspective.

When you start comparing costs, RAM is extraordinarily cheap. Under $10CAD/GB. I do contract work primarily, and hardware that saves time (e.g. by reducing the time for compile cycles) pays for itself very very quickly. I have a client who likes to have me work on-site and has development machines with only 4GB of RAM. Last year, we were doing builds that took minutes on my personal beefy machine, and an hour or two on theirs. The extra RAM would have paid for itself the first day it showed up, and would have paid a dividend every following day...

Agreed in full. My development desktop, a repurposed Westmere PowerEdge tower, has 96GB of RAM in it. The value I realize from being able to load everything I care about, ever into RAM significantly outstrips a faster CPU (not that 16 logical cores at 2.8GHz is holding me up or anything). Turns out that ramdisks are super fast--who'd have thought?

Good to see likeminded people here on that subject. I used to push RAM drives, etc in late 90's to early 2000's for critical stuff. People said it was crazy but performance and security benefits were great. Being re-discovered in past 5 years or so in cloud industry in form of "RAM sleds," etc. As if RAM making stuff faster was a new thing. ;)

Another benefit of tons of RAM is in special-purpose systems using memory-safe runtime and GC. The copying GC's were relatively simple to implement for me a while ago. When RAM expanded, I had one design that just used it for GC instead of more storage. Lots of space not utilized in normal case but imagine a whole app/service GC'd while running on tiny kernel w/ good exception handling. Never crashed. Imagine a desktop where similarly all the system services were memory safe and GC'd with a dedicated piece of hardware doing pauseless, concurrent GC. So, a crashless, fast desktop with critical stuff stored in RAM. I'll take it.

Note: Oberon Systems (eg A2 Bluebottle) use a GC language and run very fast. That with modern features, HW acceleration, and user-mode drivers.

Holy crap. Just checked ebay. I didn't realize how cheaply you could get ridiculous equipment like that. There's an older PowerEdge with 48GB of RAM for around $600. Hmmm... the credit card is feeling warm in my pocket...

Be sure to remember electricity costs in your calculations.

It draws less than my old desktop with a Nehalem I7 and a Ti. And it's only on when I use it. =)

Helps that I got it for free when a startup went down, too.

"But to a broke college student, anything that costs more than a few hours of minimum wage =/= free."

With all due respect to a college student today there are so many things you can do to make money that your professor couldn't do back in his day. It doesn't even come close. And back in his day (if it's the same as my day) if you hustled (which I did for example) you never ever earned minimum wage. I did (as I am sure many people did) many things on the side in order to make money. And that was back when I have to say it was a bit more difficult than it is now.

Reply on news.ycombinator.com