Single Status Update
I know I could go to a random tech site and ask this, but I feel an answer from here would be more personal.
How does a Nvidia Geforce 6200 256MB compare to a ATI radeaon 9600XT 128MB? I found with the 6200 it allowed me to play doom 3 on max settings, but other games I'm getting shit performance. I thought that because the VRAM was higher, the 6200 would be better, but I'm having doubts. I heard that some 6200's have 'TurboCache'. From what I know, the card actually only has a small amount of RAM and uses system RAM for most of it's processing.
With that, it might not matter. I may be getting some 512MB card (no idea on brand/model), but the price will be free, so can't complain.
- Show previous comments 12 more
I always thought the GPU only processes as much information as the CPU can give it, and bottlenecking referred to two things: overall performance being hindered by a slower CPU where CPU performance is relevant, and the GPU's ability to calculate the amount of instructions that it's capable of not being put to full use thanks to a slower CPU.
Goes the GPU read directly off the system memory, negating the need for the CPU? I think I might be out of touch with the way hardware is connected to each other on modern motherboard designs. Or I'm forgetting something which might well have gone over my head even when I kept up to date with the latest hardware.
I always thought the GPU only processes as much information as the CPU can give it
Depends on what kind of GPU processing we're talking about. If it involved a lot of data being moved from the CPU to the GPU or even from main memory to the GPU (e.g. very large textures that change all the time or geometry that can't be computed entirely on the GPU) then yeah, you'll have a bottleneck. The opposite scenario are GPGPU numerical applications, where code + data are preloaded into the card's memory and all processing runs independently. Video games are more like the former scenario, but there are exceptions, especially now with CUDA and GPGPU/physics. Running a game like Trine that requires PhysX on a system that has no CUDA-compatible hardware will be MUCH slower, as a lot of processing will have to be done in software. This alone justifies getting at least a CUDA-capable card, even if you have a single-core CPU. There would be no sense in "keeping things back" to the level of, say, an nVidia 6200.
Goes the GPU read directly off the system memory, negating the need for the CPU?
Only some "static" things like textures get cached or preloaded into video RAM, while geometry (the scene's description) is usually actively pushed by the CPU, at least in usual OpenGL/Direct3D game design. But there are exceptions, and with current GPGPU architectures, a great deal of it could be done entirely by the card's hardware, eliminating the CPU-card bottleneck.
Again, even with the above restrictions, if a more powerful card manages to process a scene of a given complexity in 13 ms instead of 13.1 ms, all other things (CPU, RAM) being equal, that's still an improvement. Of course, if the CPU or memory bus itself become bottlenecks, then that limits the benefits of a faster card for most applications. Pure GPGPU applications obviously don't care what CPU or memory bus you have.
Okay, I've narrowed down my cheap graphics card purchase to 2 cards. I've been reading reviews on both the cards, but I don't understand a lot of what they say. Also the numbers used to mark the cards confuse me. So I'm asking for help to choose between these two cards:
GeForce GT 430 1Gb
Sapphire HD 6450 2Gb
I've been leaning towards an Nvidia card since I read they have better drivers.
Edit: looking at that comparison page, it seems the nvidia card is better, but then again, I'm not sure what all these terms and specs mean.
- Show next comments 3 more