Was testing today VR view using a Vive Cosmos headset, a GTX 1070 GPU, Ryzen 2600X CPU and 64GB RAM.
Works but it’s not really fluid, a bit of nauseating shutters.
Tried to change various setting but didn’t find better performance maintaining the same visual quality.
Since the GPU was just 50% busy wanted to ask if with a more recent and powerful (was thing a RX 6800XT) will really improve things.
You can tune volume rendering quality settings in SlicerVirtualReality module settings to ensure that volume rendering updates are fast enough (using any GPU) while the head is moving, of course at the cost temporarily lowering the image quality .
A significantly better GPU is expected to provide significantly better rendering speed for volume rendering, so if the dynamic quality adjustment does not give you satisfactory results then it may make sense to upgrade. In general I would recommend NVidia GeForce GPUs for volume rendering (NVidia for software compatibility and driver quality; and GeForce series for cost effectiveness and software compatibility).
The fact my 1070 was not saturated was wondering me, or a faster GPU will be anyway get better results even if not saturated.
Also I’m not using very big models, having total VRAM>model size will allow better compatibility with AMD GPUs.
I tend to prefer AMD hw for next rig due better price/performance (6900 XT at €1700 vs 3090 at €3700 here).
Do you have performance issues with visualizing models or volumes?
If you visualize volumes, what is the volume size in voxels?
With an AMD GPU there is a higher chance that you will run into problems, but it could be a good option if you are willing to take the risk and/or you are ready to do some low-level OpenGL debugging and contribute fixes to VTK.
I visualize volumes, testing on ircad 12, 512x512x260
Not able to debug OpenGL, reading the issues was having the impression AMD has problems in case of big volumes using a lot of VRAM or using Adaptive quality.
It is not a small volume, so it is not surprising that a several-year-old GPU struggles to render it at full resolution in stereo at 60fps.
Most “large volume” rendering issues start at texture sizes above 512, so probably you would not run into them even if you use an AMD GPU.
Not an expert but 16 bit grayscale is meaning 68MB image… Not that much to me. Wrong calculations probably.
Buy 3090. I bought 6900XT and regret very hard =).
I ran into data sets I can visualize with volume rendering on 1070, but not 6900XT and some other data sets vice versa. 3090 opened them all.
But there is some rule? Getting problems only with big files, or using Adaptive resolution, or simply not track able to any specific feature so is not predictable.
I find it unpredictable.
I have a file with size around 4Gb I cannot volume render at all with 6900XT, but works fine with 1070 or CPU. I also have a file at 11Gb, which I can volume render with 6900XT but it shows wrong stiched. The 11Gb file can not be volume rendered using 1070.
Volume rendering 11Gb file, the speed is slow when using Adaptive resolution with 3090.
If you have 3090, you really shouldn’t need to use Adaptive. Normal should give you better performance. Another thing to check is to see if you are really set to use the discrete GPU to for 3D rendering (In case your CPU has an internal GPU as well). In windows 10 this is handled by the Graphics Settings hardware-accelerated GPU scheduling. Make sure SlicerrealApp.exe is chosen to use the 3090.
You mean 4GB on the file system, or in VRAM (how you check it?). As example I’m testing using also the Ircadb 1.12 image if you have time to have a test on it.
Do you mean volume rendering without virtual reality?
Is it possible to post an anonymized copy of this volume ? I’m eager to test it with my 6700XT without virtual reality.
I’ll check on Monday on the Ircadb file. The file size is 4Gb, I don’t know VRAM.
Hi! Yes, I’m volume rendering without virtual reality. 6900XT is not very predictable in my hands, I’m not handy at all though. The specific data belongs to other people, I’ll try to find similar data which belongs to me and send to you.
Thanks. Btw you are on Windows or Linux.
I’m on windows 10. Maybe windows operative system comes into play as well…
Discovered this opened issue on VTK
No assignee after one year, just a pity 20% of the market is ignored (considering i.e. all the Apple have AMD GPU).
Also seems AMD subpar OpenGL support (and essentially switching their effort on Vulkan) play a role, not good for VTK due a migration to Vulkan seems not to be happening fast
Was reading a post where AMD states his Windows OpenGL drivers are “workstation” drivers, so will be curios to know if there is someone able to test on AMD workstation GPU to check if the problems are still there.
Anyway interersted to know if we can track down the limitations on AMD GPUs to lower than 4GB files.
Hi! I have tried the Ircadb files. Patient CT and VTKs. I used 6900XT, it runs very smoothly. I used normal, adaptive and maximum methods, all three ran smoothly. I tried same files in 3090, there is no difference.
Will be possible for you to test on this file
This time a 1004x1024x1018 volume, 2GB size.