It looks like !buildapc community isn’t super active so I apologize for posting here. Mods, let me know if I should post there instead.
I built my first PC when I was I think 10-11 years old. Built my next PC after that and then sort of moved toward pre-made HP/Dell/etc. My last PC’s mobo just gave out and I’m looking to replace the whole thing. I’ve read over the last few years that prefabs from HP/Dell/etc. have gone to shit and don’t really work like they used to. Since I’m looking to expand comfortably, I’ve been thinking of giving building my own again.
I remember when I was a young lad, that there were two big pain points when putting the rig together: motherboard alignment with the case (I shorted two mobos by having it touch the bare metal of the grounded case; not sure how that happened but it did) and CPU pin alignment so you don’t bend any pins when inserting into the socket.
Since it’s been several decades since my last build, what are some things I should be aware of? Things I should avoid?
For example, I only recently learned what M.2 SSD are. My desktop has (had) SATA 3.5" drives, only one of which is an SSD.
I’ll admit I am a bit overwhelmed by some of my choices. I’ve spent some time on pcpartpicker and feel very overwhelmed by some of the options. Most of my time is spent in code development (primarily containers and node). I am planning on installing Linux (Ubuntu, most likely) and I am hoping to tinker with some AI models, something I haven’t been able to do with my now broken desktop due to it’s age. For ML/AI, I know I’ll need some sort of GPU, knowing only that NVIDIA cards require closed-source drivers. While I fully support FOSS, I’m not a OSS purist and fully accept that using a closed source drivers for linux may not be avoidable. Happy to take recommendations on GPUs!
Since I also host a myriad of self hosted apps on my desktop, I know I’ll need to beef up my RAM (I usually go the max or at least plan for the max).
My main requirements:
- Intel i7 processor (I’ve tried i5s and they can’t keep up with what I code; I know i9s are the latest hotness but don’t think the price is worth it; I’ve also tried AMD processors before and had terrible luck. I’m willing to try them again but I’d need a GOOD recommendation)
- At least 3 SATA ports so that I can carry my drives over
- At least one M.2 port (I cannibalized a laptop I recycled recently and grabbed the 1TB M.2 card)
- On-board Ethernet/NIC (on-board wifi/bluetooth not required, but won’t complain if they have them)
- Support at least 32 GB of RAM
- GPU that can support some sort of ML/AI with DisplayPort (preferred)
Nice to haves:
- MoBo with front USB 3 ports but will accept USB 2 (C vs A doesn’t matter)
- On-board sound (I typically use headphones or bluetooth headset so I don’t need anything fancy. I mostly listen to music when I code and occasionally do video calls.)
I threw together this list: https://pcpartpicker.com/list/n6wVRK
It didn’t matter to me if it was in stock; just wanted a place to start. Advice is very much appreciated!
EDIT: WOW!! I am shocked and humbled by the great advice I’ve gotten here. And you’ve given me a boost in confidence in doing this myself. Thank you all and I’ll keep replying as I can.
Well, let’s see:
-
You no longer have to set jumpers to “master” or “slave” on your hard drives, both because we don’t put two drives on the same ribbon cable anymore and because the terminology is considered kinda offensive.
-
Speaking of jumpers, there’s a distinct lack of them on motherboards these days compared to the ones you’re familiar with: everything’s got to be configured in firmware instead.
-
There’s a thing called “plug 'n play” now, so you don’t have to worry about IRQ conflicts etc.
-
Make sure your power supply is “ATX”, not just “AT”. The computer has a soft on/off switch controlled through the motherboard now – the hard switch on the PSU itself can just normally stay on.
-
Cooling is a much bigger deal than it was last time you built a PC. CPUs require not just heat sinks now, but fans too! You’re even going to want some extra fans to cool the inside of the case instead of relying on the PSU fan to do it.
-
A lot more functionality is integrated onto motherboards these days, so you don’t need nearly as big a case or as many expansion slots as you used to. In fact, you could probably get by without any ISA slots at all!
While I love this list, it is more applicable to the turn of the century than a a decade ago. I was half expecting to see “ram no longer has to be installed in pairs” on the list.
ETA: Talking about EDO memory not dual channel
I think you may have misread OPs post. They haven’t built a PC since shirtly after they were 10-11, which was almost 30 years ago. So developments since the turn of the century are in fact relevant here, heh.
They haven’t built a PC since shirtly after they were 10-11…
Well, they’d need a much bigger shirt now than when they were 10 or 11.
I’ll see myself out.
LOL
Wait RAM doesn’t need to be installed in pair? I am an old apparently
-
AMD is the gold standard for general user PCs in the last 5+ years. Intel simply cannot compete at the same energy expenditure/performance. At the same/close price/performance, Intel either burn a small thermonuclear power plant to deliver comparable performance, or simply is worse compared to similar Ryzens
Ryzens are like aliens compared to what AMD used to be before them
So I’d go with them
As for the GPU, if you want to use Linux forget Nvidia
There is no need for a separate sound card now, it is built in.
GPUs these days use a whole lot of power. Ensure your power supply is specced appropriately.
And make sure it’s an actually good PSU too.
I know in gaming, possibly in other loads Nvidia 40 series, and especially 30 series love transient spikes which can easily exceed 2x the nominal power consumption. Make sure your PSU can handle those spikes both in terms of brevity, and current.
My 3090 is a light flickering machine. Kind of annoying tbh.
Get Nvidia GPU for AI, period.
Read the manual for the motherboard you want and make sure that the M2 slot supports NVMe rather than SATA. (Also, learn to tell NVMe from SATA chips.) M2 slots that are SATA usually share a SATA lane with the SATA connectors and if you populate the M2 slot you might lose a connector.
Another thing to read about is whether populating which M2 slot reduces the speed of one of the PCIe slots. Same reason (shared lanes) but with PCIe instead of SATA. These things should be spelled out next to the M2 connectors.
NVMe drives in Linux have /dev/nvme* designations not /dev/sd*.
lots of good advice here. I just want to restate: do yourself a favor and migrate your HDDs over to any solid state drive. Whether that means “classic” SSDs with a SATA-Port or M.2s is your prerogative, but in either case you’ll start wondering how you could ever stand that s pinning noise and the vibrations and the slow, slow data transfer.
For AI/ML workloads the VRAM is king
As you are starting out something older with lots of VRAM would be better than something faster with less VRAM for the same price.
The 4060 ti is a good baseline to compare against as it has a 16GB variant
“Minimum” VRAM for ML is around 10GB the more the better, less VRAM could be usable but with sacrefices with speed and quality.
If you like that stuff in couple of months, you could sell the GPU that you would buy and swap it with 4090 super
For AMD support is confusing as there is no official support for rocm (for mid range GPUs) on linux but someone said that it works.
There is new ZLUDA that enables running CUDA workloads on ROCm
https://www.xda-developers.com/nvidia-cuda-amd-zluda/
I don’t have enough info to reccomend AMD cards
I got ROCm to work on a 7800XT after fixing some Python errors. It was quite unstable though.
Some thoughts:
Ubuntu, most likely
I’d encourage you to take a look at Linux Mint, it alleviates some of the Ubuntu fuckiness. And if you want to join the “I use arch btw” crowd, maybe checkout EndeavourOS if you’re feeling more brave than just Ubuntu variants (which is built on arch, but makes barrier to entry a little easier).
i9s are the latest hotness but don’t think the price is worth it
Take a look at last generation to soften the blow to your wallet. E.g., instead of looking at a 14900k, look at 13 or even 12 series. In fact, this is a useful strategy all around if you’re being price conscious: go one gen older.
GPU that can support some sort of ML/AI with DisplayPort
Probably going to want to go with a discrete card, rather than just integrated. Other major consideration is going to be nvidia vs AMD, for which you’ll need to decide if CUDA should be part of your calculus or not. I’ll defer to any data science engineers that might wander through this post.
The rest of your reqs pretty much come as regular stock options when building a pc these days. Though another nicety for my latest builds, is multi-gig nics (though 2.5Gb was my ceiling, since you’ll also need the network gear to utilize it). Going multi-gig is nice for pushing around a fuckton of data between machines on my lan (including a NAS).
Very last thing that I’ve found helpful in my last 3 builds spanning 15 years: I use newegg for its reviews of items, specifically so I can search for the term “linux” in any given product’s reviews. Often times I can glean quick insight on how friendly (or not) hardware has been for other’s linux builds.
And I lied, I just remembered about another linux hardware resource: https://linux-hardware.org/?view=search
You can see other people that have built with given hardware. Just remember to do a scan too once your build is up to pay it forward.
Good luck, and remember to have fun!
Compared to those pain points building a modern PC should be a breeze. CPUs go in Zero Insertion Force sockets so as long as you remember to lift the little lever you won’t bend any pins. People don’t even wear static discharge wrist bands anymore (all though it couldn’t hurt) or worry about shorting things out. And power connectors only fit one way unlike the AT power connector.
Speaking of breeze your only pain point might be making sure you have enough air circulation for cooling all that gear.
Acronyms, initialisms, abbreviations, contractions, and other phrases which expand to something larger, that I’ve seen in this thread:
Fewer Letters More Letters NAS Network-Attached Storage NVMe Non-Volatile Memory Express interface for mass storage PCIe Peripheral Component Interconnect Express PSU Power Supply Unit RAID Redundant Array of Independent Disks for mass storage SATA Serial AT Attachment interface for mass storage SSD Solid State Drive mass storage ZFS Solaris/Linux filesystem focusing on data integrity
8 acronyms in this thread; the most compressed thread commented on today has 5 acronyms.
[Thread #529 for this sub, first seen 20th Feb 2024, 23:55] [FAQ] [Full list] [Contact] [Source code]
I went through the same process myself a couple years ago, first PC build in a while. The biggest shock for me was finding out hard drives (SSD, HHD, etc) were outdated: its all about NVMe cards which look like a stick of RAM and plug directly on the motherboard.
Unless you want a bunch of storage and modularity. The benefit to Sata is that it is much more flexible and Sata SSD’s are cheaper and can be put in a ZFS raid to increase maximum speeds.
I’ve gone back and forth on whether I need RAID locally. Giving up at least a third of your storage capacity (assuming RAID 5) for the off-chance that your hard drive dies in 3-4 years seems like a high price to pay. I had two drives fail in the lifespan of my current desktop. And I had enough warning from SMART that I could peel off the data before the drives bricked. I know I got lucky, but still…
If you’re practicing 3-2-1 backups then you probably don’t need to bother with RAID.
I can hear the mechanical keyboards clacking; Hear me out: If you’re not committed to a regular backup strategy, RAID can be a good way to protect yourself against a sudden hard drive failure, at which point you can do an “oh shit” backup and reconsider your life choices. RAID does nothing else beyond that. If your data gets corrupted, the wrong bits will happily be synced to the mirror drives. If you get ransomwared, congratulations you now have two copies of your inaccessible encrypted data.
Skip the RAID and set up backups. It can be as simple as an external drive that you plug in once a week and run rsync, or you can pay for a service like backblaze that has a client to handle things, or you can set up a NAS that receives a nightly backup from your PC and then pushes a copy up to something like B2 or S3 glacier.
Here is my next planned build. Also with linux in mind. Get yourself that GPU. It’ll get you there.
That GPU is
I for one would not purchase any Intel hardware as long as AMD is around. Not that they’re bad or anything, but AMD gives me much Kore “bang for the buck”. To future proof your rig, I strongly suggest you go with the latest socket (be it Intel or AMD, doesn’t matter) and make sure you get DDR5 RAM. PCI Gen 4, and then have at it.
Getting an 80 Plus Gold power supply is always nice too.
And then there’s the cooling. I see you went with a radiator and fan, but I strongly suggest getting some type of liquid cooling. The prices are not that bad anymore (unlike about 10 years ago, which was insane).
As for the board, you’ll get all kinds of different suggestions. Some people swear by Asus, I’d rather go with Gigabyte (love the Aorus line), so it’ll come down to brand trust at the end of the day.
As for the card, I hear a lot of crap given to Nvidia about being closed source, and I sort of agree that’s messed up, but ATI cards (while pretty good) are always a step behind Nvidia. Plus, most distros have them working out of the box.
It can be intimidating after so many years, but its way simpler than it was back then.
Good luck man, you got this, there’s nothing to fear but fear itself.
ATI cards (while pretty good) are always a step behind Nvidia.
Ok, you mean AMD. They bought ATI like 20 years ago now and that branding is long dead.
And AMD cards are hardly “a step behind” Nvidia. This is only true if you buy the 24GB top card of the series. Otherwise you’ll get comparable performance from AMD at a better value.
Plus, most distros have them working out of the box.
Unless you’re running a kernel <6.x then every distro will support AMD cards. And even then, you could always install the proprietary blobs from AMD and get full support on any distro. The kernel version only matters if you want to use the FOSS kernel drivers for the cards.
I agree that I could be wrong on the comparison. Maybe they are not that far behind, but guaranteed not at the same level when comparing apples to apples. I wish that wasn’t the case, but it still is.
when comparing apples to apples.
But this isn’t really easy to do, and impossible in some cases.
Historically, Nvidia has done better than AMD in gaming performance because there’s just so much game specific optimizations in the Nvidia drivers, whereas AMD didn’t.
On the other hand, AMD historically had better raw performance in scientific calculation tasks (pre-deeplearning trend).
Nvidia has had a stranglehold on the AI market entirely because of their CUDA dominance. But hopefully AMD has finally bucked that tend with their new ROCm release that is a drop-in replacement for CUDA (meaning you can just run CUDA compiled applications on AMD with no changes).
Also, AMD’s new MI300X AI processor is (supposedly) wiping the floor with Nvidia’s H100 cards. I say “supposedly” because I don’t have $50k USD to buy both cards and compare myself.
I have absolutely no counter for you on this one, as I’m jot aware of the highest level stuff between manufacturers. And it makes sense. Nvidia has been the goto manufacturer for gaming and developers usually improve their code based on what’s needed to run the best possible on Nvidia hardware. I’ll research Kore on this when I have a chance, this seems to he a very interesting topic. Thank you for pointing this out.
It is not exactly what you asked, but if it is customization and expand ability you are looking for, did you consider custom build cites like cyberpowerpc.com? You can select from wide variety of components, they build and test it for you, and make sure that everything is compatible and working. You do pay a bit premium for the assembly and tests but it is not that much, and you save your time and have a peace of mind that everything works.
Oooh, I had not considered that but thank you for the recommendation. The only thing I don’t like about these PCs is that they all have RGB lighting. I really don’t need this and I don’t get their appeal.
The only PC fan manufacture that has not gone RGB at all is a Noctua (premium). Their fans are poop brown and beige or black for consumer, grey for industrial, but are great in terms of noise to cool performance. If noise is important then there’s videos of people comparing fans so you can pick a tone that is subjectively best.
I enjoyed the days of one color LEDS. Couldn’t beat a Tron blue or The Matrix green.
The responsiveness between a hard drive and an SSD is night and day. NVMe is even faster but not noticeable unless you move a hell of a lot of data around. A motherboard having at least 1 M.2 NVMe slot is common, so installing the OS on it is an option. Hard drives have more storage per price, but unless space is significant factor I suggest using SSDs (also quieter than a spinning disk!). More info on storage formats in this video
Recent generations of motherboards use DDR5 RAM, which were very expensive on release. I think the price has come down but I am not up to date this generation. You may be able to save money making a DDR4 system but you’ll be stuck on a less supported platform.
AMD had like ~10 years of bad/power hungry processors and Intel stagnated, re-releasing 4-core processors over and over. AMD made a big comeback with their Ryzen series becoming best bang for buck, then even over taking Intel. I think it’s pretty even now.
If you don’t intend to game or do certain compute workloads then you can avoid buying a GPU. Integrated CPUs have come quite far (still low end compared to a dedicated GPU). Crypto mining, Covid and now AI has made the GPUs market expensive and boring. Nvidia has more higher-end cards, mid range is way more expensive for both and low end sucks ass. On Linux AMD GPUs drivers come with the OS, but Nvidia you have to get their proprietary drivers (Linux gaming has come a long way).
I was really hoping to cannibalize the 32 GBs of DDR3 RAM but I couldn’t find a MoBo that supports it anymore. Then I saw DDR5 is the latest!
I don’t really do any gaming. If I wasn’t going to tinker with AI, I’d just need a card for dual DisplayPort output. I can support HDMI but…I prefer DP
The 4070 TI will give you quite a few years out of it for sure. You could also completely forego the GPU and get a couple of CUDAs for a fraction of the cost. Just use the integrated graphics and you’re golden.
Are CUDAs something that I can select within pcpartpicker? Or is this like a cloud thing?
Are CUDAs something that I can select within pcpartpicker?
I’m not sure what they were trying to say, but there’s no such thing as “getting a couple of CUDA’s”.
CUDA is a framework that runs on Nvidia hardware. It’s the hardware that will have “CUDA cores” which are large amounts of low power processing units. AMD calls them “stream processors”.
I misspoke, and I apologize. I could not recall the term TPU, so I just went with the name of the protocol (CUDA). Nvidia has various TPU devices that use CUDA protocol (like the K80 for example). TPUs (Tensor Processing Units) are coprocessors designed to run some GPU intensive tasks without the expense of an actual GPU unit. They are not a one to one replacement, as they perform calculations in completely different ways.
I believe you would be well served by researching a bit and then making an informed decision on what to get (TPU, GPU or both).
You could also completely forego the GPU and get a couple of CUDAs for a fraction of the cost.
What is this sentence? How do you “get a couple of CUDA’s”?
Dude, you KNOW I’m talking about TPUs. The name escaped my mind at the moment. Sorry if my English is not to your royalty level. Are you really so hired that you have to make a party out of that? Ran out of credits on pornhub or something?
DDR5 has gone down dramatically compared to launch. You can get 64GB with a very fast bus for under 200 dollars now. At launch 32GB would easily set you back 250+. AMD has made a killing with Ryzen. Never mind the new naming convention that Intel came up with to make it even more complicated to choose the right CPU for your use cases, ridiculous. As for Nvidia GPU drivers, at the end of the day, they just work, regardless their proprietary drivers philosophy (which, again, I agree sucks). But if the OP is going to be doing AI development, machine learning and all that cool stuff, he’d be better served by getting a few CUDA TPUs. You can get those anywhere from 25 dollars to less than 100, and they come in all types (USB, PCI, M.2). https://coral.ai/products/#prototyping-products I have 1 USB Coral running the AI on my Frigate dicker for 16 cameras, and my CPU never reaches more than 12% while the TPU itself barely touches 20% utilization. You put 2 of those bad boys together, and the CPU would probably not even move from idle 🤣
Hold on a second, how come every time i look for TPUs i get a bunch of not-for-sale nvidia and Google cards, but this just exists out there and i never heard of it?
I found out about those about 6 months ago only, and it was by chance while going over the UnRaid forum for Frigate, so I decided to do some research. It took me almost 4 months to finally get my paws on one. They were seriously scarce back then, but have been available for a couple of month now. I only got mine finally at the end of November. They seem to be in an availability trend similar to Raspberry Pis.
getting a few CUDA TPUs
Those aren’t “CUDA” anything. CUDA is a parallel processing framework by Nvidia and for Nvidia’s cards.
Also, those devices are only good for inferencing smaller models for things like object detection. They aren’t good for developing AI models (in the sense of training). And they can’t run LLMs. Maybe you can run a smaller model under 4B, but those aren’t exactly great for accuracy.
At best you could hope for is to run a very small instruct model trained on very specific data (like robotic actions) that doesn’t need accuracy in the sense of “knowledge accuracy”.
And completely forgot any kind of generative image stuff.
Same reply. And you can add as many TPUs as you want to push it to whatever level you want. At 59 bucks a piece, they’ll blow any 4070 out of the water for the same or less cost. But to the OP, you don’t have to believe any of us. You’re in that field, I’m sure you can find the jnfo on if these would fit your needs or not.
And you can add as many TPUs as you want to push it to whatever level you want
No you can’t. You’re going to be limited by the number of PCI lanes. But putting that aside, those Coral TPUs don’t have any memory. Which means for each operation you need to shuffle the relevant data over the bus to the device for processing, and then back and forth again. You’re going to be doing this thousands of times per second (likely much more) and I can tell you from personal experience that running AI like is painfully slow (if you can get it to even work that way in the first place).
You’re talking about the equivalent of buying hundreds of dollars of groceries, and then getting everything home 10km away by walking with whatever you can put in your pockets, and then doing multiple trips.
What you’re suggesting can’t work.
Let’s get this out of the way. Not a single consumer grade board has more than 16 lanes on 1 PCI slot. With the exception of 2 or 3 very expensive new boards out there, you’ll be hard pressed to find a board with 3 slots giving you a total mas of 28 lanes (16+8+4). So, regardless of TPU or GPU that’s going to be your limit. GPUs are designed as general purpose processors that have to support millions of different applications and software. So while a GPU can run multiple functions at once, in order to do so, it must access registers or shared memory to read and store the intermediate calculation results. And since the GPU performs tons of parallel calculations on its thousands of ALUs, it also expends large amounts of energy in order to access memory, which in turn increases the footprint of the GPU. TPUs are application-specific integrated circuits (ASIC) designed specifically to handle the computational demands of machine learning and accelerate AI calculations and algorithms. They are created as a domain-specific architecture. What that means is that instead of designing a general purpose processor like a GPU or CPU, they were designed as a matrix processor that was specialized for neural network work loads. Since the TPU is a matrix processor instead of a general purpose processor, it removes the memory access problem that slows down GPUs and CPUs and requires them to use more processing power. Get your facts straight and read more before you try to send others on wild goose chases. As I said, the OP already works this field, it shouldn’t be hard for him to find the information and make an educated decision.
A lot of what you said is true.
Since the TPU is a matrix processor instead of a general purpose processor, it removes the memory access problem that slows down GPUs and CPUs and requires them to use more processing power.
Just no. Flat out no. Just so much wrong. How does the TPU process data? How does the data get there? It needs to be shuttled back and forth over the bus. Doing this for a 1080p image with of data several times a second is fine. An uncompressed 1080p image is about 8MB. Entirely manageable.
Edit: it’s not even 1080p, because the image would get resized to the input size. So again, 300x300x3 for the past model I could find.
/Edit
Look at this repo. You need to convert the models using the TFLite framework (Tensorflow Lite) which is designed for resource constrained edge devices. The max resolution for input size is 224x224x3. I would imagine it can’t handle anything larger.
https://github.com/jveitchmichaelis/edgetpu-yolo/tree/main/data
Now look at the official model zoo on the Google Coral website.
Not a single model is larger than 40MB. Whereas LLMs start at well over a big for even smaller (and inaccurate) models. The good ones start at about 4GB and I frequently run models at about 20GB. The size in parameters really makes a huge difference.
You likely/technically could run an LLM on a Coral, but you’re going to wait on the order of double-digit minutes for a basic response, of not way longer.
It’s just not going to happen.
OK mman, dont pop a vein over this. I’m a hobbyist, with some experience, but a hobbyist nonetheless. I’m speaking from personal experience, nothing else. You may well be right (and thanks for the links, they’re really good for me to learn even more).
I guess, at the end of the day, the OP will need to make an informed decision on what will work for him while adhering to his budget.
I’m glad to be here, because I can help people (at least some times) and learn at the same time.
I just hope the OP ends up with something that’ll fit his needs and budget. I will he adding a K80 to my rig soon, only because I can let go of 50 bucks and want to test it until it burns.
I wish you all a very nice weekend, and keep tweaking, its too Much fun.
