CPU for "heavy" programmers discussion

[cristoferoswald]

I've searched quite a lot about this, but it's hard to find some sort of benchmarks for programming applications and workloads, so I think this may be a good discussion.

 

So, first, I want to make some things clear, I know that for pure programming, all you need is a text editor that can be ran in a potato (I'm in this area for some years), but I'm not talking about this kind of software development. I'm talking about having to run a 8GB ram usage for a single IDE, with debugging (that can REALLY slow down the execution), or training some simple neural networks (for the sake of college work) that can take hours to train (at least in my 6th gen i5 laptop that I have now), try to get some games done in Unity or whatever engine, running a VM for Android while using Android Studio or, in the end, just trying to compile a 50-file C++ project in the least amount of time so you can fix that null pointers as quick as possible.

You get the point here, programming itself is not demanding, but getting the job done IS.

 

The question is, what are your experiences with the most recent desktop CPUs for these kind of workloads?

 

From what I saw until now, probably the Ryzens are the best for general balance, and working with compiled languages use better the higher core counts, but there are some applications that hard to know what they best use, specially the most "higher level" ones, Unity and IDEs for example, wich totally depends on how they are made (and I have little clue of). Another side that is quite hard to figure out is interpreted and/or browser-based languages, because the way they use the CPU is not as simples as "prefers higher clock speeds" or "prefers more core count" and quite honestly, there is no way to measure these things but to hear from people's opinion.

 

Obviously, I am not considering the part where the program you are making may be better for a CPU or another. A simple single thread C program will obviously benefit from pure IPC and a multi-thread simulation will benefit from core counts.

 

And for the true linux users that just use VIM and GCC, I know that there is benchmark for this, but that's not the only thing I do, for you guys, I have another question. Have you experienced any kind of difference between different CPUs among distros? I am a linux user, and would be very happy to hear your opinion too.

 

My last "disclaimer" is that I know that some kind of the workloads I said depend on the data drive too (SSD or HDD), specially the large ones like engines and IDEs, so I'm trying to make this a CPU (and maybe RAM related) discussion only.

[Enderman]

9900K

More than enough cores for almost every task, and the best core performance available.

The high core performance makes up for fewer threads when compared to higher core count CPUs.

So even though a $10000 xeon with 28 cores will outperform it in certain multithreaded tasks the 9900K will beat it in the huge majority of programming and compiling related loads.

[SupaKomputa]

9900k is the king of ipc right now.

but ryzen 3000 is on the horizon, and amd clearly demonstrate that the future (mid range) cpu will have at least the same ipc at intel's highest end.

core for core and clock for clock it matches the performance in cinebench.

by getting a 2700x it will paved the way for the new cpu without changing the motherboard.

[tarfeef101]

Here's my experiences:

6600K - Worst I put it through was android studio with a VM running at the same time, and it was completely seamless. Simple apps, nothing as complex as say, fortnite mobile, but it handled that perfectly

 

7700K - Recompiling windows apps, testing DX11/12 games, compiling on g++/jvm all ran well. Even could  virtualize a hadoop cluster without becoming sluggish.

 

Chromebook + server using either of the above, + 1080ti: Text editing on the chromebook is great, and having a separate server (especially with a jupyterhub to do neural net stuff in with CUDA support) is a godsend. Take the heavy lifting away from my laptop, and keeps me productive. I love having a server and would never go back. 

[cristoferoswald]

I get your points, and definitely the 9900k is the king right now, but I think that it is pretty unfair to just trow away the price numbers in this comparison.

 

A i9 9900k is ~$500 and a r7 2700x is ~$300, that's a 40% diference. Looking at pure single and multi thread performance, the i9 is just 20% better. We should consider the mid-ranges here too.

 

For example, let's say I want to build a $1250 PC for general programming. There are more things to consider when taking the hole build in account. The 9900k is not just $500, it needs a high (and maybe premium) mobo and a very decent cooler, taking its price per dollar why lower.

 

I think I forgot to mention that I'm not taking this in a utopic world where money is infinite, we need to consider all things, including price.

[brob]

The higher IPC of Intel cores generally makes them significantly better performers with programming type workloads. Core counts do matter, but I think the new Coffee Lake Refresh cpu, (e.g., i7-9700), offer decent choices. In my experience most compilers and linkers are or are mostly single threaded.

 

Often overlooked is storage performance. An NVMe ssd can do an awful lot to improve large compiles and builds since a significant amount of the work consists of opening, reading and writing many files.