Oculus PCs require a high end GPU with a reasonable GPU and a powerful CPU. The goal of this post is to inform the user how to build a a oculus ready PC. In this post, you will:
1. Learn little about GPUs and my recommendations for one for oculus.
2. And, order a bunch of GPUs and build your own oculus ready PC.
What is a GPU?
A GPU is a special purpose processor (as opposed to general purpose CPU), that can perform large number of highly parallel operations very quickly. It does so by employing a large number of processors (or cores), that can perform operations that are naturally parallel such as image manipulation, matrix operations, in parallel on these large number of cores. As an example, the latest nvidia GPU, Titan-X has about 3000 cores. GPUs are connected to the CPU and memory using the PCI bus, and modern GPU software does a good job of hiding this latency by cleverly overlapping communication between GPU and main memory with computation over GPUs.
The most popular top of the line GPUs today are Titan-X and Titan-Z with 3000 and 5000 cores.Titan-X is actually a single GPU with 3000 cores and 12 GB of memory. Titan-Z has 5000 cores split in two GPUs internally with about 6 GB each. I personally prefer the Titan-X because of the large amounts of memory per GPU. You can buy these off newegg or bestbuy. For the current line of oculus hardware a lower end GPU like GTX980 should also suffice. I prefer getting them from Amazon because of their excellent 30 day return policy. The full list of recommended NVIDIA GPUs by oculus are:
- GeForce GTX 780 Ti
- GeForce GTX 970
- GeForce GTX 980
- GeForce GTX 980 Ti
- Quadro K6000
- Quadro M6000
- GeForce GTX TITAN Z
- GeForce GTX TITAN
- GeForce GTX TITAN X
- GeForce GTX TITAN Black
Existing pre-built gaming machines with GPUs
The rule of thumb in building/buying PCs is to build a machine yourself unless you want to build a commodity off-the shelf sub 1000$ computer. If you want a high end PC, it is always economical to build. This holds true for machines with GPUs such as high end gaming systems or deep learning systems. It will always be worth the hassle financially to order and build a machine. As an example, branded high end gaming boxes cost about ~5K$ with nvidia GPUs. You can easily build a state of the art high end machine with a Titan-X for around ~2K$.
As I mentioned earlier, GPUs attach to the CPUs over the PCI bus. The PCI bus support in motherboard is referred in terms of lanes. A single GPU needs about 16 lanes to operate at full bandwidth. Also, to attach a GPU you need a 16x slot. You cannot attach a modern GPU to a 8x slot. However, a 16x slot can run at 8x speeds. The newest Intel processors Haswell-E, support a maximum of 28 or 40 PCI lanes. The Skylake-E support has not been announced yet and is perhaps not expected until late 2016-early 2017. However, the configuration I describe below will work with Broadwell-E configuration which is supposed to be released in few months since Broadwell-E also uses the LGA 2011 v3 socket. Also, please note that perhaps you can build a cheaper PC than this one.
Building your own machine or ordering GPUs for existing machines
In order to support the latest GPUs you need to make sure two things:
Your computer motherboard has PCI 3.0 16x slots. You need as many slots as the number of GPUs you wish to run on a single machine. (See the motherboard for wide PCI slots)
If your machine already has a PCI 3.0 slot and you have at least 16 lane PCI support. If you want to build a new machine, I recommend the following configuration as of today (early 2016). I recommend getting them from Amazon for their return policy (no need to RMA for the first 30 days). However, you can shop the CPU from a local microcenter shop and get a deep discount (and the 3% sales tax).
– An Intel 40 lane CPU (The latest Haswell-E processor Intel 5930K is a good entry level CPU)
– One (or two) nvidia Titan-X GPUs- X99 motherboard LGA 2011 v3 (Asus Sabertooth X99 – a stable motherboard for workstation grade reliability with gaming motherboard specs)
– Other parts to complete your build DDR4 DRAM (GSkill 2400), water cooling (Cooler Master 280L), power supply (EVGA 1000W), a computer case (Corsair 540), a boot/primary disk (samsung Evo 1TB), a disk for large datasets (WD 3 TB). Finally, here is a gorgeous monitor to use with this setup LG34UM95.
Note that the above configuration is a balanced system, and all components work well with one another. If you switch out, anything here make sure the parts are compatible. E.g. a X99 motherboard does not support DDR3 DRAM. This machine can easily run two GPUs at 16x speeds since it has 40 lane CPU-motherboard support. Furthermore, you can also use this same configuration an order a Broadwell-E CPU instead when they become available (instead of the Haswell-E used in this post).
The built PC!!
