Dear Uber, Could you inform us of the technical specification of your servers? I for one am very interested if the game is optimized for a certain amount of cores, or if more is always better. Also I'm very curious on the processor types/Ram amounts (taking there are multiple testing servers/servers for different amounts of players) if the server uses graphic cards for processing data and in general, how it works as a backbone for the PA game. Thank you, Kite.
I was wondering about this too, because the servers seem to die after a 3 player match reaches 20k units in total (massive slowdown). My personal guess is that they've got a lot of server computers that are occupied with games and released as they end, and do not run on a single superserver like for example Tranquility (not fair to compare because CCP has a fetish for this kind of stuff) http://www.pcgamer.com/eve-online-1/ I might think that the servers are running the Intel Quad E3-1200 series, with a 64gb+ RAM support, which are very commonly used in computer on other server hosing companies at least. Again, this is just wild guessing from my side, it would be interesting to see if Uber have any raw data for us, I like this kind o'stuff. ¨~ Chris
Lies... the soup doesn't like lies... I'm not a soup... http://bit.ly/15pRSQS <-- I was not amused ¨~ Chris
http://forrestthewoods.com/the-tech-of-planetary-annihilation-chronocam/ [17] Dedicated and elastic refer to a subset of always-on servers, and dynamically started servers to handle any load above this, respectively.
Currently the sim part of the game, server-side, is single-threaded. This is the main reason why 'large' games tend to experience slow-down as more and more units are being simulated. Everything else, I think, performs quite well. Modders have got custom servers running on high spec dedicated and virtual machines, and the sim speed has always suffered. Multi-threading that kind of code is no doubt all kinds of nasty, but hopefully Uber will be able to deliver improvements in that area. Maybe it could be optimised and multi-threading it will not be necessary.. Otherwise, I guess, we'll be waiting for faster chips!
What we know so far is: Also the other stuff that has been pointed out here. I would check out Amazon Web Services for actual specs. Uber's servers hardware requirements are the same as the server that shipping with PA. With one minor difference number of instances one for us.
multicore sim should have been the absolute first thing they should have worked on imo - from day 1. Maybe Uber could track down that guy who made multithreaded spring
I don't see the problem with multi-threading really. Haven't seen a computer in years with less than QuadCore, and my two last computers have been HexaCore. Also, the Intel Xeon comes with 12 core, and the Nvidia Tesla technology has support for over 240 cores. http://www.nvidia.com/object/what-is-gpu-computing.html of course I can't expect all people to run computers with intel extreme models, 32gb+ RAM and TITAN graphics like I do, but I hope you see my point, the general computer has become quite powerful in terms of processing power, even the laptop. ¨~ Chris
It's not about the hardware - you are right in that it's very commonplace now. The difficulty is with writing the software, and how complex it is to split something up in a way that can efficiently run separately, when in fact there is often a large amount of common information that is needed. If you do it wrong, you can easily cause invalid results to the simulation (because two threads made a calculation without taking into account the other), and it can actually make performance worse in some cases. That's why you'll commonly find the multi-threading in a program divided along easily separated code - for example, Sup Com had a rendering thread, simulation thread, audio thread and a few other misc thread - each of these don't require much info (or more specifically, back-and-forth communication) from the others to function. Trying to split those into multiple threads is far, far harder. As for GPU computing, anything other than highly repetitive simple tasks are more trouble than they are worth at the moment. A GPU core is not at all equal to a CPU one.
ooh, that was interesting, thanks! I'm quite into the hardware bit and personally think I'm a bit over average when it comes to knowledge in that subject, but programming and such is something I don't have the slightest clue about. This clears things up a bit ^^ ¨~ Chris
