Contribute
Register

Can you have AMD RX 470 *AND* an NVIDIA GTX9xx?

Status
Not open for further replies.
Joined
Dec 24, 2015
Messages
55
Motherboard
ASrock X99 Extreme4
CPU
i7-5820K
Graphics
GTX 960, R9 280X & RX 580
Mac
  1. iMac
  2. MacBook
  3. MacBook Pro
Classic Mac
  1. Apple
  2. Power Mac
Mobile Phone
  1. iOS
Will two different types of video cards cause conflicts? I have currently have an NVIDIA GTX 960, and have two more PCIe16 slots open. I want to fill one of those slots with a Radeon RX 470.

I'm not going to try to bind them together with SLI or anything like that. Both of them will be doing nothing but running the GPU, with only the GTX960 connected to the monitor.

I'm guessing there should be no problem doing this, since a lot of Macs have built in AMD cards, and people add in NVIDIA cards all the time, right?
 
Should work. You won't get any OpenGL acceleration from the RX 470 unless you attach it to your display though.

Crap. That's all I wanted the RX470 for was to run OpenCL GPU computing. So I guess the NVIDIA GTX960 will spin free and GPU compute while the RX 470 does both. Will both be able to run OpenCL (the NVIDIA was running CUDA, but my project switched to OpenCL all around).
 
OpenCL should work on both cards regardless of attached displays, and the RX 470 should be a lot faster than the Nvidia card. It's up to the application to decide which GPU is used.

The project I am running will normally use all of the GPUs you allow, but I'm not sure if I have ever seen anyone use different types (except I used OpenCL on the iGPU at the same time as the NVIDIA used CUDA--the former was a mistake because it sucked up all of the memory bandwidth from the CPU and was so slow it was counterproductive).

And I hadn't considered using the AMD RX 470 until yesterday, when I saw that they key to performance on this project under OpenCL is double-precision FLOPS, which the AMD somehow does better than the NVIDIA Pascal series, which can't even run on Macs yet.

Thanks a ton for your advice so far!
 
Last edited:
For double precision computing, older AMD hardware such as a R9 280X should be even faster thanks to its high FP64 ratio: http://www.geeks3d.com/20140305/amd-radeon-and-nvidia-geforce-fp32-fp64-gflops-table-computing/

On the Nvidia side, the old Titan / Titan Black should be a nice choice. The original Titan even runs on Apple's stock drivers (as does the AMD 280X).

Both of those are a little out of my price range. I took another look at the RX 480 though, and it might be worth bumping up a level for that, except that from what I've read the RX 470 is nice and stable with a minor tweak on a kext, while the 480 is not very stable.
 
The R9 280X doesn't cost much more than $100 on eBay. But the Titan / Titan Black is very rare, which makes it expensive.

The RX 480 has also become stable during the last months, but since you said you need FP64 performance I'd opt for its ancestors (-> R9 280, 280X), which is 100% OOB in OS X and should be a lot faster.
 
The R9 280X doesn't cost much more than $100 on eBay. But the Titan / Titan Black is very rare, which makes it expensive.

The RX 480 has also become stable during the last months, but since you said you need FP64 performance I'd opt for its ancestors (-> R9 280, 280X), which is 100% OOB in OS X and should be a lot faster.

I hope I can find it for that price range. The benchmarks I see say 280X gets 1024 GFLOPs and the RX 480 gets 323.

I can see why the Titan/Titan X are so rare. For most GPU computing applications (to include machine vision) they seem to be indomitable, which is why I am so surprised by the stellar double-precision floating point capabilities on the AMD side, and the disparity between that and their overall benchmark comparisons with the NVIDIA cards.
 
Yes, the architecture of those Tahiti chips is quite strange due to its high FP32/FP64 ratio. FP64 is of very little for normal consumer application / games, so you usually don't waste any die area for functions like that. That's what Nvidia did with their Kepler cards, they have a 1:24 FP64 ratio (compared to 1:4 in Tahiti).

AMD chose to do so, which puts the GPUs into the same league as other professional video cards, making it very strong in FP64 compute but (comparably) weak in gaming or other consumer stuff.
 
Status
Not open for further replies.
Back
Top