CUDA

Neutrino has stated that the game will be CPU limited. Will the servers be able to use CUDA to do calculations on nvidia graphics cards to support more players/planets/units. Also will the game client be able to unload some of the workload to the graphics card?

Sounds like a lot of work, for only a few CUDA cards in servers.

Actually the topic of moving some of the math to the graphics card came up with OpenCL. And yeah it turns out that it's too much work, and there is going to be too much hardware skew where you have no idea what you can throw at any given user's machine.

Assuming that most of the servers, despite neutrino's prediction, will be an ordinary gaming PCs, GPGPU (CUDA/OpenCL/DirectCompute) acceleration makes a lot of sense - even powerful CPUs like i7 3960X is likely not enough to handle a massive interplanetary war. However, neutrino mentioned that simulation is multithreaded, but not massively parallel to take advantage from GPGPU power. Developing massively parallel simulation algorithm is probably far beyond the scope of project, since it will require gameplay to rely much more on physical simulation and avoid scripting, maybe even to the point of modeling unit's internal structure as particles and physical force fields, some of which are controllable.

I seriously doubt most servers will be normal gaming PCs as I expect most servers to be located in data centers. We'll see though, I could be wrong.

In this regard, I hope the server will also be available on linux. Sorry If it's already been said elsewhere

Considering you said that you're expecting most of them to be runned by community and unless we suddenly became able to afford a server itself (around $4000 for mid-powered one) and a data center colocation (around $200 per month), that prediction just cannot be true. Buying a gaming PC to use it daily can be reasonable, but buying same-priced device and paying monthly subscription fee 13 times more expensive than WoW to run just one game is out of question for the majority. If ordinary gaming PC is not be able to handle game server, we will have 4-5 hosts, and, as a side effects:

• Little to no modding - no server for mod will make it unusable, so modding will be limited to server owners or ones who can convince them to use their mod.
• No mod stacking - server owners will try to keep additional setup packages as easy and small as possible, so we'll not see big, game-transforming mod packages like BO full set + 4DC + TM + TVg in SCFA.
• Possible reduce of playerbase - sitting in spectate mode and waiting for the next match to join game is not very motivating.

It is pretty unrealistic that normal gaming machines cannot run the server for normal games.
It works just fine for SupCom (sure SupCom is p2p, but that just means that every player is a server in itself) and to my understanding PA aims to offer more units than SupCom at the same performance-cost.
So normal games will probably be run on the machines that the players have at home.
Why should I ever rent a server if I can easily host the 1v1 I want to play on my own machine?

You are missing my point. Most of these servers are LESS powerful than a real gaming machine.

Your pricing is way off on renting servers. Regardless we will run enough servers if the community doesn't.

There is nothing stopping you from running a server. I want people to run servers.

Of course. Likely it will be the most performant platform for servers.

I had asked this question and a question about SLI in a PM back before the kickstarter met its initial funding goal. The reply I got was (paraphrased)
"Not planning to right now, suggest you power more monitors instead"

So why do you expect less powerful servers to make up most of the game's servers?
I fail to see the advantage that arises from using such servers? Sure a small server is payable, but it gives me no advantage to just using my own machine, which I have already payed for.
The only reason I can see is that a dedicated server will not just ragequit and kill the game for all players involved. Which might be an important reason or not, I am not sure. Depends on how well people can behave.

Xeon phi pci-express card (2000$) embeds 60 1ghz cores and are used in hpc stampede.
Afaik they are to be addressed via ip (the card has it's own linux kernel) . This simple pci-x card has 1Tflop power. Don't if pa engine could take advantage of it

I break the servers into several classes.

Big game servers - these are the industrial strength servers that will need to have a lot of cores and live in data centers

small game servers - games of up to maybe 6-8 people that can be run on "normal" game boxes. A lot of people use their old machine to run dedicated servers so they don't usually have the best hardware.

In neither of these cases do I expect these machines to have a decent videocard.

Don't get me wrong, I am not arguing in favor of CUDA. I fully agree that CUDA makes no sense for the server since a server might even have no graphics card at all.
I just don't understand why you think that most servers will be in datacenters. Most games will probably end up in the category of 8 or less players, so most games are part of your 2nd category. And that's the "no datacenter" category.

In datacenters :
You usually provision virtual machines on top of big boxes, which gives you the ability to scale out based on your requirements. A 4 cores vm can be easily switched to 8 cores. More over infrastructure is monitored, there's redundency. You do not have to manage all this it. Should you need a bigger system your vm can be transparently moved to another box. All this comes at a lower price than if you had to manage all this by yourself.

Running the server on my own gaming machine at home is still cheaper, since it creates no additional costs

What about the Intel GPU inside the new cpu's like the Intel i3570k? Don't the newer servers dont have a GPU on their CPU?

I suppose it depends on the ratio between games we host and servers people put up. Also it depends on what kind of game mode is the most popular. I think it would be great if the community ran most of the servers.

Regardless there is no good argument for supporting cuda in either case.