GPU stream processors.

[matthenge]

I came across this term "GPU stream processors" on http://yourultimatepc.com/ when reading about how to pick parts for a build I am doing and I just want to know more about them and how they affect the performance of GPUs.

Thanks

 

[Icaraeus]

Like nVidia's "CUDA Cores," AMD's Stream Processors are a more central component of the GPU. Stream Processors and CUDA Cores are not linearly comparable due to vast architectural differences, but can be thought of as similar when it comes to the primary function of each component. Stream Processing is focused intensely on parallelism of datasets to ensure efficient processing when performing tasks that are better-suited for parallel processing. These tasks can include aspects of audio processing (stream processors are used in DSP-equipped devices -- digital signal processors

Each GPU can contain hundreds to thousands of Stream Processors. Architecture changes in a fashion that makes cross-generation comparisons often non-linear, but generally speaking (within a generation), more Stream Processors will equate more raw compute power from the GPU. The processors are responsible for processing all the data that is fed into and out of the GPU, performing game graphics calculations that are resolved visually to the end-user. An example of something a Stream Processors might do would include rendering scenery in-game, drawing character models, or resolving complex lighting and shading within an environment.

 

[TJ Hooker]

You don't really need to concern yourself with stream processors though when choosing a part, as looking at raw GPU specs isn't a very good way to compare performance. Just look at benchmarks.

 

[Icaraeus]

You don't really need to concern yourself with stream processors though when choosing a part, as looking at raw GPU specs isn't a very good way to compare performance. Just look at benchmarks.

It is a good way to compare performance when you study *all* of the graphics card specifications, not just amount of CUDA/stream processors

 

[TJ Hooker]

I don't see why, unless you're specifically looking at two cards from the same series/lineup. You can't really estimate performance based by looking at cores/frequency/memory bandwidth for a single GPU, and you can't compare those specs from one GPU to the next unless they share the same microarchitecture.

If you want to understand how a GPU will perform at a given task, the best thing to do is to look at benchmarks of that task e.g. gaming benchmarks. I don't see how knowing how many stream processors the GPU has would add to that understanding. Maybe looking at VRAM amount can give you an idea if one card is more future proof than another, but trying to extrapolate and say which card will perform better in games in the future is always a crapshoot.

Edit: OK, there's one case where knowing the specs would be important, and that's when dealing with some of the sneaky cards Nvidia and AMD have put out recently where they have two cards with different specs but the same name. You'd want to look at number of cores etc. just to make sure you know which version you're getting.

 

[Icaraeus]

Maybe looking at VRAM amount can give you an idea if one card is more future proof than another, but trying to extrapolate and say which card will perform better in games in the future is always a crapshoot.

Nothing is future proof, and amount of video memory is only a very small part of a graphics chip. You could have 100GB of video memory, it won't mean anything if the memory bandwidth is 30-40GB/s like DDR system memory (see GT 1030 w/ DDR4)

 

[TJ Hooker]

Maybe looking at VRAM amount can give you an idea if one card is more future proof than another, but trying to extrapolate and say which card will perform better in games in the future is always a crapshoot.

Nothing is future proof, and amount of video memory is only a very small part of a graphics chip. You could have 100GB of video memory, it won't mean anything if the memory bandwidth is 30-40GB/s like DDR system memory (see GT 1030 w/ DDR4)

Yes, but if something like that (e.g. very low memory bandwidth) were the case, it would already be apparent in existing benchmarks. I.e. it's pretty obvious a GT 1030 isn't much of a gaming card today, let alone a year from now. I was referring more to cards to a case where you're comparing two cards that perform similarly today (e.g. 1060 3GB vs 570 4GB, or RX 580 4GB vs 8GB). I realize nothing is future proof, but some cards do age better than others.