[Chimera-users] Hardware for VR of volume data

[Tom Goddard]

Hi Bernard,

  VR is not going to handle large volumes no matter what hardware you buy.   To avoid causing nausea current VR headsets use 90 frames/second graphics updating.  That is in two eyes, so the effective graphics frame rate needed is 180 frames/second.  By comparison on a flat panel display it is routine to show a very large volume at 10 frames per second and this causes no problem.  So VR requires about 20 times the rendering speed of visualization on a conventional display.

  So here are some ideas of what you can expect.  Some recent 3d light microscopy I was looking at in VR is 4K by 2K by 31 z planes with 3 channels — and it rendered fast enough with VR on a GTX 1080 and also on a Quadro P6000.  Comparing to your 2K cubed example my data was equivalent to about 2k x 2k x 186, or about 1/10th the size.  I haven’t benchmarked volume rendering so I don’t actually know at what size graphics is too slow — I should test this.  But I suspect this is close to the limit.  For molecular structure display, 100,000 atoms with very coarsely triangulated spheres works in VR but is near the limit.  For surface and triangle rendering the limit is about 2-5 million triangles depending on which high-end graphics card before frame drops start causing flicker.  For molecules showing 2.4 million atoms of HIV capsid 3j3q is fine on a desktop display but is not going to work in VR.

  With that in mind I would recommend a GTX 1080 Ti graphics card with a high-end CPU, maybe 64 Gbytes of memory, definitely get a high-end SSD drive or two as the data transfer rates of about 3 Gbytes/second are very handy for large data compared to the 100 Mbytes/second of spinning drives — the whole system would be about $4-5k.  You could spend more but you won’t get any better performance.  The main VR performance bottleneck is going to be in the software if you use ChimeraX.  Unfortunately it does color blending and mapping data values to colors and cropping on the CPU instead of the GPU which is painfully slow, tens of seconds to adjust brightness on the above-mentioned 3-channel light microscopy.  I expect to put the blending and color lookup on the GPU in ChimeraX over the next 12 months.

	Tom


> On Jul 10, 2018, at 12:48 PM, Heymann, Bernard (NIH/NIAMS) [E]  wrote:
> 
> Hi
> 
> I'm interested in acquiring a PC with capabilities to use ChimeraX for 3D VR visualization of large volumes (2k cube). What is the best configuration (CPU, GPU, VR hardware) for around a $5-10k budget?
> 
> Bernard
> 
> 
> 
> 
> 
> _______________________________________________
> Chimera-users mailing list: Chimera-users at cgl.ucsf.edu
> Manage subscription: http://plato.cgl.ucsf.edu/mailman/listinfo/chimera-users
>