Show HN: PC Builder AI (pcbuilderai.com)
Hey HN!
You know when you wanna build a new PC, and need to watch a bunch of videos to find the best parts, and then tweak your shopping cart to fit your budget? Well, this new app can help you.
It recommends the best hardware for your usage (gaming or work), and considers your budget. It's really helpful, even when you got expertise on hardware. Check it out, and feel free to suggest stuff! Hope you like it!
83 comments
[ 4.0 ms ] story [ 166 ms ] threadDo you verify that all the parts you suggest are compatible ?
Are you able to complement with actual build instructions ?
Generating a part list compatible with pcpartpicker.com could fix that easily if not.
> Are you able to complement with actual build instructions ?
It's almost easier to build a PC today than a 100 piece LEGO kit. Otherwise, most components come with manuals for how to assemble them together with the other pieces too, like the motherboard comes with instructions on where things go.
I've walked 3 different under 21 kids through a build since start of the pandemic... only because my hand/eye coordination for some bits (panel headers, etc) isn't so great combined with low light visibility.
Slightly recent example is newly launched CPU architectures using contemporary sockets, where you have to insert a supported CPU first, update BIOS and only then could the motherboard support newly launched CPU. But if you just have the one, new CPU, it might not be able to boot.
I put I wanted the computer to cost $8000, and it generated components that cost $5340 together (1000 + 500 + 2100 + 700 + 90 + 250 + 500 + 200) but still the UI showed the correct prices for the individual items, but the total still showed as "$8000" although adding them together is nowhere near that.
(1) Parts that total $9,980, but claims it totals $33,950 (2) Recommendations for parts that are both years out of date, and vastly inferior to other options (CPU = AMD Ryzen Threadripper 3990X, GPU = RTX3090)
So I'm guessing the AI data isn't trained enough and is going off some poor metrics like sorting by price, perhaps recommending multiples without indicating (was it asking me to double up on RAM or Stroage?)
newegg has something similar -works pretty well
https://www.newegg.com/tools/custom-pc-builder/
I need a relatively high-performance i9-based desktop PC (for ML and toying with LLMs) with on-site warranty, so building my own PC would not work for me.
What is the backing AI/data source? I tried using ChatGPT for this, but given it's cutoff is September 2021 it doesn't really give you the most relevant hardware.
It told me to buy a 3950X.
I think this is a good tool, but not a great tool. Some refinement is needed, but the API is good, if sparse.
When I think of an AI I'd imagine some more personalisation, and there's so much that goes into building a PC but the only input here is the price.
What about appearance? Opaque causes, transparent cases, white vs black PCBs, RGB? Preferences around the prioritisation of certain specs (more storage?). Then there's the many form factors.
Even for just a "gaming" build - are you going for blockbuster graphics, or more CPU-intensive simulation genres.
- seems like the market data about components is out of date, so I wouldn't actually trust this.
- pcpartbuilder is helpful in large part because there are links to actually buy the components
- really what I'd want to do is describe what I want to do with the computer, then have it suggest things -- not just select from a dropdown. That would actually be helpful; there's a learning curve to stay abreast of the latest hardware things and really what I would want is "tell me what's the good stuff to buy" like you might find if you google "best workstation build for ai" or watch a Linus tech tips video
- I'd want to take a look at the list of components and edit them and have it recalibrate, like "actually I think this case looks really cool" and then have it automatically readjust components based on for example things that fit in the case
- I'm guessing this is using a LLM based on the out-of-date market data. For the above, I think you would really just need a robust set of apis about pc part data and then use LLMs to function call into the apis. It seems like pcpartpicker is the group with this data, not sure if they expose it to other developers
We will add links soon. We have this feature already, it's just not available because the prices don't match (the AI price is higher than the actual Amazon price, for instance)
Your feedback is much appreciated and very very helpful, thank you so much.
How does PCPartPicker get pricing data from Amazon, NewEgg, Walmart,etc.
PC Builder AI is more of a fun project that might help by giving you direction of what to search
What I would do is classify each item to the best item year (similar to FB's best device year), find benchmark scores for representative sets of item-years, and use that as a 'oracle' to guide the solver.
Changing to a $550 build keeps the same cpu but changes its price from $100 to $170, adds a $20 cooler, and drops Ram from 16gb to 8gb.
Sorry for the negativity but algorithm needs work.
* Ryzen 5950x — a 16-core Zen 3 CPU for gaming in 2023?
* RTX 3090 for 1500 — you can get a 4090 FE for about that price
* an X570E motherboard which generally doesn’t make a lot of sense for gaming
* a 1200W PSU which is wildly overspecced for this build
Now, there are ways to make the ai more useful, and that would be to have a conversation with it about what you want to do on the computer. Then that makes the ai use apparent and reasonable.
Have a conversation with the AI about the PC you want, and challenge it on things like "can I see the equivalent PC but with AMD instead of Intel" and things like that. I think that would have extremely good value. And for prices, you might be better off scraping amazon (or whatever site makes the most sense), instead of whatever stale data you have. Prices for PC parts fluctuate so much that I don't know that it make sense to try and store/cache it locally.