This post assumes that you have a GPU miner running NVIDIA 1070’s or higher (AMD’s are currently not supported by Otoy’s OctaneRender which powers the RNDR network, but will be in the future). It also assumes you have a setup similar to this:
- CPU: Intel Celeron
- RAM: 4GB
- HD: 32GB USB Drive
- Motherboard: Any board with at least one(1) 16x PCIe lane and five(5) PCIe 1x slots (more on this later)
- PSU: Enough to handle all of your equipment at full power and some headroom
- GPUs: NVIDIA 1070 or higher
Bottom Line Up Front (BLUF)
You will need to upgrade your RAM, Hard Drive and possibly some of your risers. The Minimum recommended setup is:
- RAM: 16GB Minimum, 32GB Recommended.
- HD: 50GB
- GPUs: Don’t mix and match, it hurts your performance
- Risers: We want the most data throughput possible see below
- OS: Windows (for now)
GPU matching matters when it comes to the maximum scene (project) size. The card with the lowest amount of VRAM will limit the size of the projects you’re able to render. A card’s onboard RAM affects how large and complex of a scene can be rendered, so having even just one card with less VRAM could limit your whole system. You could disable that card when rendering bigger scenes, but that defeats the whole purpose of that additional card.
The video card selection is the driving factor for performance in OctaneRender and hence RNDR. There are two aspects of a video card that impact render capabilities: the raw speed of the GPU itself and the amount of memory on the card. Video memory will limit how large and complex of scenes can be rendered, speed will determine how fast it can render each frame.
Puget Systems did a really great analysis on mixing GPUs of different types if you want to see some numbers.
The number of cores does not have significant impact on OctaneRender performance, but the clock speeds do have a slight impact. Although all of the computing is done on the GPUs, the CPU is loading data from the scene and coordinating between the GPUs. All these things overtime have a measurable impact, but may not be worth spending the money on an upgrade just yet. Again Puget Systems has done a great job testing and verifying this.
Your motherboard will be a limiting factor in the total number of GPU’s you can connect, but there are options to expand even more. It is always a desire to get as many PCIe 16x slots, but the costs increase greatly as you start running into server boards. When looking to get the most bang for your buck using existing motherboards, there are ways to squeeze the most OctaneBench score out of your existing system. We’ll go into build comparisons and ways to expand your system in a future post.
Mining risers are not recommended as it slows down the transfer of scene files to your GPUs VRAM, but for most mining builds there is no option. You’ll take a 5-10% hit in your OctaneBench score running off x1 risers, but in our tests its better to run six GPUs on 1x vs three GPUs on 16x.
PCIe 1.x – 250MB/s per lane (16x == 16 lanes)
PCIe 2.0 – 500MB/s per lane (thus 16x PCIe 2.0 is 8GB/s)
PCIe 3.0 – 1 GB/s per lane
Check and Benchmark
OctaneBench® allows you to benchmark your GPU using OctaneRender. This provides a level playing field by making sure that everybody uses the same version and the same scenes and settings. Without these constraints, benchmark results can vary a lot and can’t be compared.
You can head on over to Otoy’s Website and download OctaneBench. Once you run it, you can compare your results to other systems. Best thing to do is to turn off all overclocking and make tweaks until OctaneBench crashes or unable to complete its tests. At that point tune it back and you’ve got a stable system ready for the RNDR network.