Enlighten an AMD/ATI newbie

I have been using Nvidia cards since 2002, so I kinda know the pros and cons for gaming, video rendering, etc, with these cards.

I had recently sold my GTX 660Ti in hopes of upgrading to a GTX 970 but like some all pokemon lurking in the grass, a wild R9 390 appears!

With an identical price to the GTX 970, and on paper, seems to outshine the GTX 970, I have started to contemplate whether to switch to AMD (at least for this generation of cards).

But there are a couple of things I need to clear out that not even the wise Google can answer for me (or I haven't looked hard enough).

First, how is OpenCL compared to CUDA in rendering videos in After Effects or Premiere Pro.

And second, games that do use GPU Physx, since I am using an AMD card, will all physics now be CPU based?

Just a couple of things I want to clear out before going for the R9 390 by the end of the month...

Thanks!

OpenCL/OpenGL is much more widely adopted as apposed to CUDA. Strictly because it's proprietary and exclusive to Nvidia. and will ONLY work with Nvidia products. OpenCL/OpenGL works with everything, Intel, AMD, Apple, Linux, Windows and is very portable. mind you there's no guarantee that your code will work, effort must be put on your part to make something work. but to keep it short, OpenCL/OpenGL is much more open, CUDA isn't. also, Nvidia cards aren't good with OpenCL/OpenGL like AMD cards are. AMD has heavily invested onto OpenCL/OpenGL. AMD is very generous when it comes to alot of things. almost to the point where you can argue they are most likely losing money giving you the stuff they give you with their GPUs.

PhsyX is realistically a joke, if you want to get down to it. it hasn't been widely adopted. and is only available on Nvidia titles, it's not available anywhere else. and even some Nvidia titles today don't even have PhsyX on it. strictly because some developers have gotten better at making physics in game that they didn't need to use a Proprietary physics made by Nvidia that only works on their cards only. oh and i forgot to mention Nvidia has changed PhsyX quite a bit, most of the PhysX stuff is now implemented on "Gameworks" and all that stuff is CPU based as apposed to GPU.

If you have an AMD card however PhsyX will be run on your CPU but it will put a lot of taxing on the CPU. which as of now won't make sense anymore considering, DirectX 12 and Vulkan (OpenGL 4.4) are starting to put ALL the workload on the GPU. so you may not need a fast CPU anymore.

As for the 970 versus the 390.

they are neck and neck. almost identical in performance. BUT the main issue everyone tends to argue is that, the 970 has a VRAM issue. they all do. When you purchase a 970 it clearly states on the box 4GiB of VRAM on the box. that is true for the most part. BUT, 3.5GiBs of the VRAM has fast memory. the other 500mbs is slow as hell. so once you go above 3.5GiB of VRAM, the GPU will slow itself down. which let's be realistic, give it about a year, and games will start going above that. so to keep it short the 970 isn't good for the long run.

I do recommend you grab the 390. it's the overall better choice in the long run. PLUS you don't have to buy into Nvidia's proprietary non-sense. like G-Sync, Game-Stream, and Gameworks, (Which is bad for gaming as a whole) all Gameworks based titles have taken backlash for destroying everything that isn't a Maxwell based GPU. there has been people complaining how a 780ti was being beaten by 960. which is not supposed to happen at ALL.

Here's a good source on OpenGL/OpenCL vs CUDA
http://wiki.tiker.net/CudaVsOpenCL

2 Likes

I cannot say for sure on the open cl because I don't use it but it seems to be highly regarded and likeded from an efficiency but other will be much better able to tell you about that.

PhysX I can tell you about. In game that use physx as an underlying engine they are just fine and will not suffer very much if at all, personally I hate physx but the games I play that use it as thee main running gear fine run great. Where it becomes a problem is where physx is used as a fancy additional effect. Borderlands 2 is a perfect example where physx is used for the fluid on the ground that can be kicked and moved and rock too. Here that will, if enabled, be offloaded to the CPU. Mostly this will cause slowdowns. Depending on the level.of physx of course. If you are using an Intel thenphysx simulations will benefit from the increased single thread, it is worse on but both are noticeably worse than GPU native physx.

This would not be an issue if nvidia did not make the CPU side of physx so terrible. It is deliberately slow and drags things down.

Other things like hair works will also be pretty bad but at least for that you can limit the tessellation in the catalyst control center to help that a ton.

In terms of OpenCL depends on which version CS6 OpenCL isn't available for MPE however CC has much better support for both OpenCL as well as CUDA.

Physx is a old relic of Ageia which Nvidia bought which basically just add particles in certain games. Most physics get handled by CPU's however although you see it shifting more in the GPU too, but any PhysX title has specific Nvidia added eye candy. same goes for Hairworks for which AMD gas TressFX. I've played Borderlands with and without PhysX and well you'll only see it when it's side by side. and the fluid thing Zibob mentions you just won't see with an AMD card but as I said it you don't see it as a missing thing.