You want to get the best GPU for your money. That’s natural, because your graphics card is probably expensive, and you want your money to be well-spent. But how do you know what to look for? What GPU specs should you look at? What do the numbers mean?
In this article, I will explain what certain key GPU specifications mean, and roughly how they translate into actual in-game or program performance.
Important GPU Specs
GPU Core Clock
This is how many clock cycles your GPU’s cores can accomplish per second. Basically, a clock cycle is when the transistors of your GPU open and close. More cycles in the same period of time means faster calculations. This, in turn, results in more FPS in games, faster data processing, faster rendering, smoother encoding, and so on.
In games and renders, this especially affects performance for light/shadow calculations. Both modern AAA games and rendering software (like Cinema 4D and Blender) do lots of calculations relating to the bouncing of light. But as the graphics card is also just generally handling all output of images to the monitor, the faster it can work, the better for you.
Core Count and Core Type
As mentioned above, the cores of the GPU are the parts that handle the instructions and return the data that should be displayed. So, in addition to higher raw speeds, more performance can result from having more cores to handle more tasks (or ‘instructions’) simultaneously. Whether achieved through higher speeds, more cores, or both, the target result in the same: a faster rendered frame. And beyond count, some companies offer different types of cores that are specialized for different tasks. Nvidia, for instance, splits their cores up in different types: CUDA, Tensor and raytracing cores.
CUDA cores are Nvidia’s ‘normal’ cores. These are parallel processing cores that can receive algorithms written in programming languages like C and C++. Since these are the ‘basic’ cores, they are used for almost every GPU task, and more CUDA cores almost always translates directly into additional performance.
Tensor cores are cores that are faster for AI and data science purposes. This could also mean faster frames, with Nvidia’s DLSS (Deep Learning Super Sampling) technology, which renders a game at a low resolution and then scales it up. But unless you use DLSS or you are using your GPU to run a neural network, more Tensor Cores usually doesn’t mean more performance—which is why these cores are more common on Nvidia’s workstation graphics cards than they are on Nvidia’s consumer-grade/gaming graphics cards.
Raytracing cores are cores designed to perform raytracing (the kind of ‘light bouncing’ work mentioned earlier) fast and efficient. But once again, unless you enable special raytracing options or generally go heavy with lighting effects, having more of these often doesn’t immediately translate to noticeably higher performance. When those circumstance are in play, though, the performance jump can be big.
Video Memory (VRAM)
Next, we’ll cover a very important specification: GPU memory. This is lightning-fast, short-term memory directly on a graphics card. We’ve covered this topic in some depth on this blog previously, but in brief: the GPU uses VRAM to store textures, meshes, shaders, and other data it needs to render a frame. If the GPU memory is full, it must store those things on the system RAM instead. System RAM, while faster than long-term storage on a hard drive, is slower than VRAM and physically further away from the GPU, slowing down your frame generation.
If you have more video memory, you can set textures and detail levels higher without as much impact on frame rates, since there is more room to store them. Similarly, if you are rendering a 3D scene in, for instance, Cinema 4D with a large amount of VRAM, you can manipulate your project and render it out faster; this is because more of the scene can fit into the immediately accessible memory of your GPU at once.
Very large amounts of memory can have those benefits, but the most important thing about VRAM is simply having enough, so pay attention to memory requirements provided by game developers, software developers, and reviews/benchmarks.
Memory Bandwidth and Memory Clock
These two specs have much to do with each other. Your GPU has, as just discussed, memory (usually called VRAM). The speed of this memory is defined by its bandwidth and clock. The more data that can be received, the faster your GPU can load (or move) scenes, textures, and other elements.
Bandwidth is the literal throughput width of the communication channel, but clock speed tells you how fast one single operation is. Both have an impact on the performance. With a higher bandwidth, more data can be sent in each operation; with a higher clock speed, more total operations can be done in shorter spans of time. So, obviously, the best possible scenario would be both moving a lot of data at once and moving it quickly. Recent VRAM types like HBM3 and GDDR6X accomplish this.
Overall, more bandwidth and/or more clock speed results in faster loading, as well as a prevention of frame dips at moments where loading is happening in the background (like in some open-world games).
TMUs and ROPs
Rarely, Texture Mapping Units and Render Output Units are mentioned. You need to know little about such things, since you can’t compare them between different architectures (the way chips are built). This means that these specs are only relevant when comparing GPUs based on the same architecture, which is relatively uncommon for a normal person making a build plan. However, I will explain them in short:
A TMU (Texture Mapping Unit) is a processor that must resize and rotate bitmaps of 3D meshes. More TMUs = faster rendering, but the effect can only be compared through benchmarks by knowledgeable reviewers (for the reason stated above).
An ROP (Render Output Pipeline) is another component that processes pixel values before drawing them on your screen. More ROPs = faster image drawing . . . but once again, this effect can only be accurately measured by expert benchmarks.
I hope you’ve found this overview of GPU specs helpful! The next step for figuring out what matters when picking a graphics card would be looking at a lot of reviews and benchmarks, since they can give you a better image of what tends to matter for real-world performance. (Or, for a bit of a shortcut, you can always take a look at the GPU recommendations in our main build chart, where a large amount of the research has been done for you.)
Also, if you enjoyed this tour, you may want to check out my previous article that takes a similar look at CPU specs. But what do you think? Did I miss any vital GPU specs? Do you have any other questions? You can let us know in the comments below.