Building a $6000 data science laptop for $2100 in 2022 - Part 1: choosing hardware

I've been doing clinical data science and engineering at Medical Intelligence One for about a year, which has given me the opportunity to test the limits of my current data science laptop.

Now it's time for an upgrade, and like any startup (or anybody who was raised on hand-me-downs and thrift stores), we're doing this lean and mean.

I started by looking for a computer that can reasonably handle all the pain I plan to put it through over the next couple of years. This search led me to some machines with incredible power and incredible price tags. The following looks like an ad, but it's actually part of the blog post, so take a look at its specs :).

This thing would be great to have, but I know from my many years of being a cheapskate that you're paying a lot for the brand and also for the bundle of things they're assembling for you.

So now we know what's possible. Let's figure out how to do it more cheaply.

Here are some key parts to consider:

- RAM

- Hard drive

- CPU

- GPU

- Motherboard

The motherboard usually doesn't make it into the marketing materials for a computer, and honestly I don't think you need to buy a laptop based on a specific motherboard. BUT... it helps to know that the motherboard is where you physically plug in the RAM and the hard drive, so it will determine how much you can upgrade a laptop. Generally the laptop's CPU is not something you can upgrade, so the CPU and motherboard are kind of a package deal. When you see a CPU with real fire power, it's probably sitting on a motherboard that can hold a lot of RAM and big, fast hard drives.

For those of you doing a lot of deep learning, the GPU is important because you're running a lot of operations in parallel. For the kind of work I'm doing, I need a strong CPU that can handle large mostly sequential operations. I'll still look for a good GPU, since I may end up doing some work with transformer language models, but it's not so critical for me in the near-term. Blockchain has caused the price of GPUs to go up significantly, so if it's not critical to have a good GPU, then put your money into the CPU and other components instead.

Key Point The CPU is your clue to a laptop with a lot of potential, even if the RAM and hard drive aren't that impressive. When you find a computer with a good CPU, start googling to see how much people have upgraded that specific model.

After poking around the local Best Buy and the internet a bit, I found a Lenovo Legion 5 which had been upgraded to 64gb RAM and 8TB SSD for $4,700. That caught my eye for three reasons:

1. It had the memory and storage of the $6,000 MacBook Pro we saw earlier

2. I had seen other Lenovo Legion 5 laptops selling for under $1,000

3. It can be relatively cheap to upgrade RAM and storage

Photo credit: https://www.theverge.com/22529429/lenovo-legion-5-pro-review-gaming-laptop-price-specs-performance-features

After a little digging, I found out that there is the Legion 5, and then there is the Legion 5 Pro. This is where the motherboard matters. While the Legion 5 can only be upgraded to 32gb (hence the lower prices I was seeing), the Pro can be upgraded to 64 gb.

Okay, now we're getting somewhere. Can I find a relatively cheap Legion 5 Pro and then upgrade my way to the laptop of my data science dreams? The answer is yes... at Walmart of all places.

I'll put the link here in case anyone wants to do the same thing I'm doing, but please know that I'm not getting paid for you to click on this. I wish I was, but I'm not.

Now for some speedy RAM and a solid solid state drive at Amazon, and we are good to GO! Again, these look like ads, but they're not. Just examples, I promise :).

The choice of hard drive matters. If you're going to read and write large datasets, you want something relatively fast. This laptop has slots for two M.2 SSD drives. The technical specs say it can go up to 1TB per slot, but in the fine print they note that "The system may support larger storage as the technology develops." Someone has already demonstrated it will work with 8TB, so I'll get a 4TB drive for one slot and leave the other empty with plans to upgrade to 8TB total if/when needed.

I decided to go for 64gb RAM (two 32gb cards) since that's what I foresee I'll need in the near-term. The technical specs say it can go up to 32gb, but again they note that "The system may support more memory as the technology develops," and we've seen an example where someone put 64gb into this model. This model only has two slots for RAM, and as far as I can tell there's nothing on the market right now bigger than a 32gb card, so we can't go any higher than 64gb total.

Okay, that's it! I've made my purchases, and it came out to (after tax):

$1512.42 for the laptop + $621.69 for the memory and storage upgrades = $2,134.11

Just by doing the upgrade myself, I'll save $2,600.

Does this compare with the $6,000 MacBook Pro? Not quite, but it's surprisingly close for the price, and it fits our needs. It's easy to get caught up in chasing some measure of performance, but at some point before you click "complete purchase," it's important to step back and recall why you are doing this. I guess that applies to a lot of things in life.

Next up, putting it all together into a data science workstation! I'll try to post some videos of the unboxing and the hardware and software installations if I can find the time.

For anyone reading this who may have a better grasp of computer hardware, please comment below if I've said anything misleading and I'll try to correct it.