GPU rendering workstations: an unbiased article on the hardware you’ll need

Want to buy or build a workstation for GPU rendering but aren't sure what you should be looking for? Or perhaps you are, but don't know where to begin? Or maybe all the benchmarks and tech-talk online is a little overwhelming? Fear not and read on - we'll clue you in and keep it simple.

Here at Novatech, we understand that researching workstations is often a pretty dry, frontal lobotomy kind of experience. One that feels never-ending, with all the different available options, numbers and statistics piling up into some kind of bar chart labyrinth that you don't particularly want to spend days or weeks navigating. We get it - so sit back, let us moisten your dusty eyes and massage your frontal lobes, as we give you a no-BS breakdown of everything you need to know about the hardware which should be in your GPU rendering workstation.

Let's start with the most important thing to consider - which you might be surprised to find...

Is your GPU! Who'd have thought it?

Shockingly, it turns out that GPU rendering software, such as Redshift, Octane, and VRAY GPU, are very much GPU-based. But more than that, they often make use of multiple GPUs, meaning, generally speaking, the more you have the better your workstation will perform. 

As of right now, NVIDIA GeForce RTX cards will give you the best price:performance ratio, but most importantly, they use CUDA-cores, which are what you'll require in any of the three software we mentioned above. VRAY-RT however, does support OpenCL, so if you particularly wanted to use an AMD card here, then you could. But, and depending on your choice of render engine, you may require more VRAM, in which case you'll be heading towards the NVIDIA Quadro GPUs. As a minimum we would recommend dual Asus RTX 2080 Ti Turbos, which will give you a total of 22GB VRAM and 8704 CUDA cores.

So what about the CPU?

There are two key points to bear in mind for the CPU. Firstly, if you're not using the workstation for anything else, then it won't rely on it as much as others, so there's no need to grab a processor with more cores than you can count, because it won't be making great use of them. Your decision here depends on which software you're using, and what you require from it. For most of these applications, a higher single-core clock speed is what you'll benefit from the most, which generally means an Intel processor, rather than AMD. Unless you're on a budget, in which case AMD processors are usually cheaper, and, these days, AMD CPUs are definitely catching up with Intel, so you won't miss out too much.

The second point, and more important when it comes to multiple GPUs, is considering how many PCIe-Lanes it can support. Your GPU will connect to the motherboard via PCIe x16 ports, each of which will need x16 lanes from you CPU. Therefore, if you wanted to be able to run more than one GPU at full speed, you would need a CPU that supports more PCIe-Lanes. The AMD Threadripper CPUs for example, which can have up to 64 PCIe-Lanes, or Intel's i9 7900X which has 44, for example.

RAMming speed!

Or not. Of course, you will require RAM for your system to function, but frankly, despite what you might read elsewhere in hardcore PC-builder forums, you don't need to pay much attention to how fast that RAM is, or its latency for that matter (Mhz and CL). It's extremely fast by nature - even if you were just using the cheapest old stuff you could find. 

You're highly unlikely to notice the difference between say, 2400Mhz and 4000Mhz, in real world use, which means you'd just end up paying a premium for something you can never really appreciate. Honestly, if you can, then you're in the wrong career and should probably go get a job as the first ever human speed camera. 

In all fairness though, unlike some other types of workstations, there is some benefit to be had, so if money is of no object to you and you really want the best of the best, the cream of the RAM crop, then go for higher frequency (Mhz), and lower latency (CL).

But by and large, what you really want to focus on is how much you have, which is all dependent on the standard file size(s) you actively work on at any given moment, plus any other applications you might run in the background or alongside your project. Especially if those applications are other projects, and even if it's only a few tabs in Chrome for YouTube, Netflix, Spotify (all three if you're some kind of mutant with six eyes, six ears and three brains), or whatever your preferred form of procrastination might be.

We would recommend at least 64 GB, which will keep you covered on all fronts for the most part. If you wanted to play it super safe, or say, work with really large and/or multiple files at a time, then you could opt for 128 GB instead. But don't forget, in most cases your system will support up to four sticks of RAM (or even 8 depending on your motherboard), so you can always start lower and upgrade later. It's also worth noting though, that if you do either start with, or intend to upgrade to 128 GB at a later date, you'll need a motherboard with a chipset that supports it - anything upwards from the z390.

You may also have noticed that sticks of RAM tend to come as either a single, pair or two-pair kit, and might have heard the terms single, dual and quad module configurations thrown around. Much like speed, you don't really need to worry about these too much - the configurations are only really of importance when you build your own system or expect to upgrade your memory at a later date.

In this instance, I'm not going to give multiple options for RAM, since they're fairly the same wherever you look. The price range doesn't differ too much from brand to brand, and generally just going with something like Corsair's Vengeance modules is fine. Do however, consider ECC memory if you're working in an environment that cannot tolerate any kind of errors or failures, where losing data can be make or break. But bear in mind, if you do opt for ECC, then make sure that your CPU supports it - either an X-series Intel core (Xeon), or Threadripper.

Common sizes/modules available are:

2x/4x 8 GB, 2x/4x 16 GB, 2x/4x 32 GB

But where oh where to store everything?

M.2 NVMe, SSD, and HDD - in that order. Put your OS and applications on the NVMe storage, save your active and recent projects on the SSD, and keep all of your old or archived projects on a larger HDD. It's that simple.

M.2 NVMe is the fastest type of storage (and thus the most expensive), so is great for keeping small or hand-picked software and programmes on. It means your system will be fast to boot and open up applications.

SSD is the next fastest, but is much cheaper than NVMe, making it perfect for saving active projects that you want to open nice and promptly, so you can start work right away.

HDD, the slowest of the pack but by far the cheapest, is the perfect solution for storing older projects that you still may need to access but won't be in a hurry to jump into. That's not to say it'll take days to open a large file, not at all. It'll just take several seconds longer (maybe minutes for gargantuan files), rather than just a few seconds, as is the case with SSDs and NVMe.

When it comes to NVMe and SSD, the Samsung 970 Evo and 860 Evo, respectively, are some of the best around, though many other brands and models are just as good. As for HDD, SeaGate and Toshiba are well-know and reliable too. 

The ideal set up for most people is to have a 250-500 GB NVMe, as many TB of SSD storage you need for current projects, and anything upwards of 4 TB for an HDD, should you feel you wanted or needed one.

Storage is available in a variety of sizes, usually:

250 GB, 500 GB, 1 TB, 2 TB, 4 TB, 10 TB and so on, with prices scaling accordingly.

Other components to consider

If you've decided to build your own custom workstation, then there's a handful of other things to think about, namely: PSU, Motherboard and Cooling.

PSU is relatively straightforward - you just need something that can meet slightly more than your actual possible power consumption, usually a minimum of 10% over. You should also make sure it's got an efficiency rating of at least 80, and of course, that it fits your chosen case.

1000-2000W is the standard range for most higher-end GPU rendering workstations, but they more than cover the average power requirements. For a frame of reference, our HyperStation DLX-4R, a top-end workstation for GPU rendering, uses a 1600W PSU.

As for your motherboard, that'll be dependent on the other components you've chosen. You'll need to ensure you match up its chipset to your CPU, be it AMD or Intel, and you'll need to check it has enough DIMM slots to hold the amount of RAM modules you want to install. Plus you'll need to check that the case you have can house it by matching up its form factor (most commonly ATX). In Layman's terms, this essentially means a full-size tower. 

Equally, you'll need to check that the motherboard has the appropriate number of PCIe slots (all details that can (and should) be found in seller item descriptions/specifications) in order to mount the multiple GPUs you'll be using. If putting together a motherboard and all the associated components is a tad daunting, take a look at our Intel and AMD motherboard bundles - there may well be one that covers everything you need.

And finally, if the CPU doesn't come with a cooler by default, you'll need to consider getting a CPU cooler too. A fan with a heatsink is the most common type, and provided you're not building an absolute beast with a 64-core processor like AMD's latest Threadripper, then a mid-range product will be sufficient. Similarly to with the motherboard, the only other thing you'll need to double check is whether the CPU cooler is compatible with AMD or Intel CPUs. But often, they tend to be compatible with both.

That's all there is to it

With any luck, this post has proven of use to you and I can only hope I've managed to make a usually dry and dense subject a little lighter and more palatable. I would re-iterate though, that unlike some other types of workstations, GPU renderers are a bit trickier to get right if you're attempting to build one yourself, so be sure to know your software, and what that translates into for parts - the fine details really matter here.

But if you feel more confident about what it is you want from your workstation, even enough so to build your own, then follow the links below where you can find some pre-built systems as well as a custom configuration tool, where you can order your very own workstation - designed by you, for you. And of course, if you have any questions or concerns, don't hesitate to get in contact through our form below.

And as always, let us know what you think in the comments section down below. We love to hear your thoughts and feedback.