Low FPS? Ailos' guide to having a better experience.

[Hamma]

Good stuff man nice work!

[basti]

GPUPhysix=1


That can actually help in case of CPU bound? :o
Cycles, can you give some intel here? Does PhysX just add some new fancy effects, or does it move some work from the CPU to the GPU?

[Ailos]

Originally Posted by basti

GPUPhysix=1


That can actually help in case of CPU bound? :o
Cycles, can you give some intel here? Does PhysX just add some new fancy effects, or does it move some work from the CPU to the GPU?

I don't know what it does, but I can personally vouch that it helps get an extra 5 FPS or so in really tight battles.

[Impaler]

I run an i5 2500k and a GTX 570, and I switch between CPU and GPU limited at the warpgate, so that combo is pretty even. In any battle, it becomes CPU limited. My brother has an i7 2700k and a GTX 670 and he's majority CPU limited all the time.

What I infer from the way the game performs is that it is not CPU or GPU limited, but memory bandwidth limited. This will still show up as CPU limited, so it's kinda hard to diagnose. One key symptom of this is the game being CPU bottlenecked while not fully stressing the cores - the cores are able to get the data out quicker than the memory bus can process it to the other components. A couple MMO engines I've played on previously (most notably Gamebryo: Warhammer Online & Rift) were notorious for being limited this way.

AMD's memory bandwidth is behind Intel's for the current generation by 20-30%, so an Intel processor will almost always run the game better than an AMD. A basic tree for performance (processor wise) would be:

Core 2 Duo/Quad = Phenom II = Llano
Trinity = Bulldozer FX = Nehalem (original i7s)
Piledriver FX
Sandy Bridge = Ivy Bridge

I think that better performance of the engine may be gained by using the fastest RAM clock speed & latency possible. I use 9-9-9 DDR3 @ 1600; at least in my case, I could go much higher.

[Ailos]

Originally Posted by Impaler

I run an i5 2500k and a GTX 570, and I switch between CPU and GPU limited at the warpgate, so that combo is pretty even. In any battle, it becomes CPU limited. My brother has an i7 2700k and a GTX 670 and he's majority CPU limited all the time.

What I infer from the way the game performs is that it is not CPU or GPU limited, but memory bandwidth limited. This will still show up as CPU limited, so it's kinda hard to diagnose. One key symptom of this is the game being CPU bottlenecked while not fully stressing the cores - the cores are able to get the data out quicker than the memory bus can process it to the other components. A couple MMO engines I've played on previously (most notably Gamebryo: Warhammer Online & Rift) were notorious for being limited this way.

AMD's memory bandwidth is behind Intel's for the current generation by 20-30%, so an Intel processor will almost always run the game better than an AMD. A basic tree for performance (processor wise) would be:

Core 2 Duo/Quad = Phenom II = Llano
Trinity = Bulldozer FX = Nehalem (original i7s)
Piledriver FX
Sandy Bridge = Ivy Bridge

I think that better performance of the engine may be gained by using the fastest RAM clock speed & latency possible. I use 9-9-9 DDR3 @ 1600; at least in my case, I could go much higher.

That is a very good point. I'll update the OP with that information in a little bit.

[CyclesMcHurtz]

Originally Posted by Ailos

That is a very good point. I'll update the OP with that information in a little bit.

It's an EXTREMELY important point. As a coder, it's come to the point where you have to treat some kinds of memory access the same way we treat hard disk or network access - the slowest of the slow.

When I first started in this industry (and this will show my age) I was working on a 65c02 processor, where memory was the only option. There was no cache. The SNES and Genesis both had faster processors, but still no cache. The first machine I worked on with a real cache system was PC's with the intel 386, but since the PC market didn't have them everywhere you couldn't rely on it.

Since I cut my teeth on systems where you (quite literally) counted every machine cycle I still cringe at even some of the basic C++ constructs (cough-cough-virtual-functions-cough-cough) that other coders take for granted.

[basti]

Oh, interresting.

How could I improve that for myself? Overclock my FSB?