Bakc to Fusion and GPU+Vram Question

Hello

Since i finished studying, i never used Fusion again, but now i saw that there is a free version of Fusion and so i began to use it again - and i still like it! And now i can finish a project thats pendig since i studied.
But now my question. I want to upscale my system a bit with a new graphic card. It is an older workstation with an Xeon-Processor and 24GB RAM, but the graphic card is an quite small Quadro.
Now my questions if these two GPUs are ok and which of the two is best for Fusion
The AMD Radeon R9 290X with 8GB DDR5 VRAM
or the Nvidia GTX970 with 4GB DDR5 VRAM
I found nothing about Fusion and Radeon GPUs so i ask here.

Best regards from Germany
Asta

Welcome back!

I would go for the Nvidia card, but get one with more ram. A slightly slower card with more ram will benefit you if you plan on doing a lot of 3D compositing, otherwise, both cards will work quite well.

There has been some buzz about NVIDIA GTX 970 and memory access. However I remember from bench-marking Fusion with the GTX 970 that it rocked price/performance. So I guess it has to do more with realtime performance. Don't take my word for granted, though.

I personally can't comment on Radeon, but would be also interested to hear experiences - especially with the insane amount of VRAM AMD offers.

What about dual GTX 970 or 980? Does Fusion benefits from 2?


Sent from my iPhone using Tapatalk

More RAM? That cheap-as-chips 970 has an much RAM as a K5000, and at 40% higher bandwidth, so I can't imagine that's the bottleneck. There are limits for image sizes in OpenCL that are probably going to prevent you from doing huge memory chunks anyway, and unless you are doing HFR (and I don't mean measly 48fps), you can load huge images from RAM so fast that you should be fine with swapping out to system memory. Given a $4500 upgrade budget and choice between 1) K6000 (12GB) with 32GB of system memory and 2) K4200 (4GB) and 192 GB of system memory, I'd choose option 2, at least for Fusion.

I do love that you can get the 980m for a laptop with 8GB of GDDR5. Good times! Now if only you could find it in a non-Optimus configuration for cheap...

I agree with Blazej about the AMD prices for RAM, the W7100 is less than $700 and has 8GB. That's superb. I wonder how it compares speed-wise in Fusion for things like OpenCL, OpenGL, stereoscopic viewing, high pixel rates, and such. What compatibility issues would there be? If Nvidia is dragging feet on OpenCL support anyway, what is the issues, and if there are any, should BMD specify that support for one brand is mandatory? I can compare a K5000 to a W7000 (though they are not really comparable on cost) but I don't have a W7100 or a K4200 to compare, and those are current cards. Who's up for some testing?

EDIT: If you Shift+RMB on the memory counter in the lower right, you will get the option to List OpenCL Device Info. Check for cl_device_image2d_max_height and cl_device_image2d_max_width. I'm seeing 2^15 on Nvidia cards and 2^14 on AMD. What that would mean for say, 16-bit RGBA is >2GB for a max image on the AMD card and >8GB for a max image on the Nvidia. AMD's 16K limit is a bit low for my taste, I could see 4K work wanting to have 16K plates for BG's and such, so that would mean you'd hit the OpenCL limit long before you hit the memory limit on say a W7100.

For pro graphics, video, 3D, always go with Nvidia before AMD GPU's, it's just less issues with drivers and such. And I have no issues with AMD for gaming, just not for pro use in this area.

Personally I replaced a Quadro 4000 with dual GTX580s a couple of years back and I dropped those for a GTX970 4Gb just a few weeks back, it's been absolutely awesome. Faster than my dual GTX580s (where applicable, Cycles, VrayRT etc.) but at 1/3rd the power/heat. Now, there has been some whining about the shared memory controller for the last 500Mb's on the GTX970, but this, this is something that will affect gaming in QHD and 4k, but as the VRAM is still 4x the speed of your system RAM, it's just very unlikely to affect this type of use.

The only question is if it's worth waiting for the release of the GTX970 with 8 Gb - and, oh yeah, I would say so. VRAM is the only static bottleneck for GPU's, like in rendering, if your scene doesn't fit in RAM, you're screwed.

I'm not sure what the official word on this is from BMD, but Fusion supports OpenCL 1.2, which AMD and Intel support, but Nvidia does not. So If you wanted to use OpenCL 1.2 and were running an Nvidia card, it wouldn't work. Fusion doesn't allow you to have multiple OpenCL devices, so you can't say "run all the OCL 1.1 on the Nvidia GPU, and all the OCL 1.2 on the Intel GPU".

Other than the OpenCL 1.2 support, though, I would say Nvidia has pretty good drivers relative to the competition.

So conclusion is that Radeon cards are absolutely OK for Fusion considering fully supported of OpenCL? Meanwhile nVidia have (or had) some issues with v.1.2
There is nothing about Cuda cores, which Radeon doesn't have

I'm setting my new PC now for some comp job along with general 3d software and going to make a choice for RX580 / 590 which have a really good price for 8gb version comparing to a "green" 1660 for example.

Previous post was from 2015. Since then, OpenCL 1.2 is supported on most newer generations of nVidia cards and OpenCL 2.0 is rolling out.

CUDA is nVidia brand name for general compute cores in GPU, the same cores can be targeted either by nVidia own proprietary CUDA code/toolkit or with OpenCL. AMD doesn't have (at least I haven't heard) proprietary toolkit and uses only OpenCL. And then there is Metal, which Apple is developing, which is a hybrid of graphics and general compute and is supported by gpus from both AMD and nVidia.

Hendrik Proosa,
Yeah, I've noticed that this thread is quite old, and since I faced pretty similar question, i would like to finalize this topic: both brands of video chips are absolutely ok for GPU rendering in Fusion, right?

Hendrik Proosa,
Yeah, I've noticed that this thread is quite old, and since I faced pretty similar question, i would like to finalize this topic: both brands of video chips are absolutely ok for GPU rendering in Fusion, right?

Assuming there is no bugs in Fusion or in the GPU drivers and hardware.