I’d recommend dedicating some of your budget to a motherboard-mounted m2 SSD - it’ll be much faster than your old HDD.
2 Likes
Hi there @stillyslalom
yep with you on that one but right now I want to get the mb,cpu,ram right then start upgrading the gpu and ssd. Thanks for the interest though. It would be the next thing I look at before the gpu. I’d like prices to come down a tad, I don’t want to pay the “mining tax”
You could drop the 5900X to a 5800X with minimal performance impact (except for heavy multithreaded workloads: 12 vs 8 cores) and apply the savings towards a m.2 SSD like this. It’ll give you something like 10-30x faster I/O performance than your old HDD, which will translate to significantly faster package load speeds for development workflows. I would prioritize a HDD → SSD upgrade over almost anything else.
2 Likes
Hi there,
thanks for the suggestion and I have added it to my list BUT I might need the extra cores. Here’s what I think I’m going to be looking at
So I could start with my old drive and lashings of memory THEN upgrade to the ssd. Sound sensible?
Since a lot of AMD-socket motherboards/chipsets support ECC RAM these days (at least “unofficially”, but you can find part compatibility lists on the net), personally I would aim for getting one of those instead of aiming for the highest clock speed in RAM. You can find lists online. I usually buy Asus and find it quite reliable.
As for the CPU, it really depends on your application. My rule of thumb would be getting the cheapest, eg 5600X, and spend the extra money on something sensible (books, bicycle parts, …), and use a high-end cloud server for anything intensive. Or an m2 SSD as recommended by @stillyslalom, will provide a much higher return in user experience.
1 Like
How do you get your code stored from your computer to running on the server?
Depends on the server. If I have access to a full Linux with ssh login and shell, then I usually set up a private repo on Gitlab, push to it when I am done working on the laptop, then pull on the server. This gives CI and allows cooperation with coauthors.
For some clusters it is more involved, but a variant of it usually works.
3 Likes
thanks @Tamas_Papp I chatted with some people and decided ( for 2021) that I would by an old HP Z840 with dual xeon 2699 v3 and 128gb ram. I wanted to investigate multithreading. I understand that the old chips don’t have the cache handling to be as fast as a ryzen but I’m a VERY happy bunny with what I got. Thanks for the advice.
1 Like
Regardless of how much memory you decide to buy initially (you may start with 128GB), it is important that your motherboard supports as much as possible, so that you can upgrade it. Then is a good idea to buy large modules to fill as few slots as possible.
If you are going to run complex simulations or load large amounts of data, memory can be very important. Lack of memory can prevent you from being able to perform certain calculations.
Lack of speed, on the other hand, will not prevent you from anything, it will just make it slower.
You might also need a good GPU. But keep in mind that many applications do not make use of them.
my initial goal is to look at NOT using BLAS to look into threaded coding. That’s why I picked dual xeon’s with 18 cores each. The HP z860 was designed from the ground up to be a desktop HPC and can support 2 tb of RAM. I am also looking at investigating MPI. I’ve been putting this off for about a year now. I did listen to everyone about the b550 mb/ryzen 5900 et al but decided that IF I can make the code work on 2014 tech then it should FLY on 2022 tech without recoding. The lads threw in a K5000 gpu which has a compute of 3 ( I think). Again my goal is to just GET CODING and have some fun. “IF” I architect this correctly then the hardware should not be a factor, I take my code and move it to new and it should improve, if NOT then I screwed up. As to the data set then it can get VERY big as it’s delivered in 25ms slices so that’s steaming pile of opportunity for this machine 
I suppose I MIGHT upgrade this machine but for now I’m fine with turnkey install of linux, EXCELLENT cooling, toolfree maintenance and lashings of support from the z860 forums.
Then maybe spending time on hardware choices is not the ideal use of your time — you could just start using a HPC cloud solution immediately, experiment, and decide later.
Don’t get me wrong; there is a valid use case for building and running your own hardware, but purely from an economic point of view its niche is shrinking. As a hobby it is of course fine 
2 Likes
I do use the Aws for many other things and it’s an excellent solution. Your point is a valid one and I respect your observations. I wanted to have a physical machine in my office which I can try things out on.