hannsl

Your statement is inaccurate. The CAS latency increases in terms of the number of cycles, but each cycle becomes shorter in duration, so the total latency decreases. Of course threading is not a silver bullet, but we both know that it may be a useful tool to increase performance if the software is designed with threading in mind. I'm not a game developer so I don't really understand the limitations that apis like Directx impose in terms of threading, but I can't imagine that arma3's threading model is optimal.

please ignore

I just had a look at that video and it seems like the detail level changes when the chat is opened. Perhaps that is why the fps changes. Can you confirm this?

Fair enough, but I think people may have been getting the impression from your posts that memory bandwidth is the underlying problem, which I don't think is the case.

By the way, here's a interesting read on reddit where dayz's creator discussed the new engine they're working on. Clearly the 32bit memory limitation is an issue because according to him a move to 64bits has helped them with performance. This points to the fact that the file mapping api they're using is inefficient. Obviously rewriting the engine for 64bits is not possible at this point. http://www.reddit.com/r/Games/comments/27uf74/rgames_rdayz_ama_with_rocket_dayz_is_moving_to_a/ ---------- Post added at 01:21 ---------- Previous post was at 01:12 ---------- Again, you're making assumptions. I have a degree in computer science from one of the best universities in the world. I was trying to explain to you, that a lot the information you provide in your posts is common knowledge to software developers and doesn't need to be stated. We already know that memory allocation in this game is inefficient, because it uses a file mapping api to get around the limitations of 32bit memory. However, a custom memory allocator won't get around these issues because it will only apply to memory that is allocated within the 32bit region. Bohemia is already aware of these issues as you can see from that discussion. I stand by my claim that the only way to get good performance is for them to rewrite the engine.

Firstly, I think you should make less assumptions about other people's backgrounds. I'm a software developer with 8 years of experience in writing commercial software. I think you also posted a reply to another developer, which also assumed that he had no knowledge of software development. Secondly, of course the visibility setting will affect the number of calls to memory, that was not my point. My point was that focusing on memory allocators and those types of low-level optimizations will not lead to the massive gains we need. I can get roughly 75 FPS when I lower the visibility setting to it lowest value, whereas I get 40 FPS when it is set to Ultra. Obviously that proves that the visibility setting itself has a large impact on performance, without taking ai and other processing into account. How much of an improvement do you expect by making low level optimizations? Do you honestly think that you would be able to learn something about the engine that the BIS developers don't already know about? Clearly they have profiled it to death and done most of the optimizations that don't require rewriting of large portions of the engine. Otherwise they would be completely incompetent. Clearly the only way for us to get the performance improvements we desire is for them to rewrite the engine.

In my opinion, it's fairly obvious where the performance bottleneck lies in the singleplayer campaign. The discussion around memory bandwidth and custom memory allocators is a waste of time, because the gains to be had in this area are probably marginal. Everybody knows the visibility settings (in particular of objects) have a greater impact on performance than anything else. Obviously the engine is processing a whole lot more information when we increase the visibility. Maybe the high memory bandwidth requirement is a clue that this processing is data intensive, but the fact remains that Bohemia needs to find a more efficient way of doing this processing or to do it in a different way. I'm not a game developer, so I can't offer much insight into what the exact problem might be.