If you have tried GPU accelerated rendering and you think you can work fine within its limitations, this big-little machine can be your single box renderfarm solution.
It is far from “cheap”, but for a studio that can harness the 21.5K CUDA cores it packs, this “Baby Titan” supercomputer will certainly appear to be a worthy investment!
CG Creation and GPGPU
Despite the inertia within the CG industry re-adapting to the speed and flexibility parallel GPU computing (GPGPU) allows for, many other scientific disciplines have been using GPUs for computing complex simulation models for a long time now, in fields ranging from medical research, to stockmarket prediction algorithms and complex simulation math models.
My first experience with GPU accelerated rendering was with the 1st beta of Octane. Since then, many CG software houses have introduced their own version of GPU accelerated engines, including Chaosgroup’s VRay RT and nVidia’s iRay, a GPU accelerated engine related to MetalRay that is included with 3DS Max since v 2011.
Most CG artists are aware and curious about GPU rendering, trying to squeeze an extra GPU or two in their workstations. Multi-GPU solutions were popular for gaming the last decade, and fortunately most GPU accelerated software is fully utilizable by “gaming” GPUs which otherwise are severely underutilized in viewport acceleration.
With modern high-end desktop GPUs packing more and more VRam, dedicated GPGPU units like nVidia’s Tesla cards, are almost obsolete for the majority of CG artitsts, and most premium desktop motherboards for both AMD and intel CPUs allow for 2, 3 or more GPUs to be utilized – ofc with gaming in mind.
For quite some time now, the high-end X79 boards for intel’s SB-E CPUs are available with as many as 6 or 7 PCIe 16x slots, that can immediately can get wild thoughts in your mind. Unfortunately, these E-ATX or CEB sized boards have size limitations that cannot allow for more than 4x double width GPU cards to be plugged in without blocking every other slot. Unless you are willing to produce a custom case and use long, flexible PCIe 16x risers / ribbons, you won’t be able to mount more than 4x cards. This enough for the consumer profile of the multi-GPU user, which is a gamer interested in SLI or Crossfire X configurations that currently don’t allow for more than 4x cards anyways.
What if we want more than that for GPGPU?