Besides how many GB of VRAM your graphics card has, there are many GPU specs you should know about to make the right decision when purchasing one. This guide explains the most important GPU specs to help you compare graphics cards.
FYI: buying a new graphics card? Check out our graphics card buyer’s guide.
GPU Die
Although we use the terms GPU and graphics card synonymously, they’re technically not the same thing. A typical graphics card has a GPU die inside it, soldered to the PCB, with memory modules around it. It is then encased inside a heatsink, shroud, and fans to create what we call a graphics card. Companies like AMD and Nvidia use a few different GPU dies every generation to create multiple SKUs, then tweaking and cut down the specs of the dies as needed.
GPU dies are usually named in such a way that the lower the number in the name on the spec sheet, the larger and more powerful the die is. For instance, Nvidia’s AD102 GPU is used in its flagship RTX 4090 while the RTX 4080 and RTX 4080 Super use the AD103 GPU. Newer GPUs usually use smaller manufacturing processes, leading to more power being packed in the same physical area.
GPU Cores
The compute units on a GPU are what people normally call “cores.” This nomenclature is misleading, as the cores on a GPU are not like physical cores on a CPU. GPUs have thousands of small cores or processing elements grouped into clusters that are then grouped into compute units. These compute units are what every GPU manufacturer calls by different names – compute units (AMD), Xe-cores (Intel), and stream multiprocessors (Nvidia).
There has been much marketing around CUDA cores. These are actually the processing elements that make up compute units. Hence, don’t mistake them for anything resembling cores on a CPU. Besides general-purpose processing elements like CUDA cores, GPUs can have specialized ones, such as Ray Tracing cores and Tensor cores. (AMD and Intel may use other names.)
Faster GPUs will have more CUDA cores (Nvidia) or streaming processors (AMD) than slower models, but you can’t compare these specs across manufacturers and generations. These will only help when comparing different models of graphics cards from the same generation from the same manufacturer.
Tip: hearing a lot about the AI ambitions of Nvidia? Learn how the AI hype is hurting PC hardware and gamers.
Memory: VRAM and Bandwidth
GPU VRAM is something almost every PC user is familiar with. VRAM or Video RAM is the memory on a graphics card that is available for accommodating the framebuffer, texture information, and other graphical information. If your in-game settings and video resolution use all the VRAM on your GPU, you may experience GPU artifacting and performance slowdowns.
There are other critical memory specs besides the size. For instance, the memory type (GDDR6, GDDR6X, etc.) decides how fast and power efficient your GPU memory is. The memory bus width (384-bit, 256-bit, 128-bit, etc.) determines how fast data can transfer between the GPU and the memory. The memory speed and bus width determine the memory bandwidth (in GB/s) of a graphics card.
FYI: confused between graphics cards? Check out the GPUs to avoid.
Clock Speeds
Just like a CPU, a GPU also has a clock speed – for both the GPU core and the memory. The core clock is the speed at which the GPU can process instructions, and the memory clock is the speed at which the memory (VRAM) can send or receive data to or from the GPU. Generally, the higher the core clock and memory clock of a graphics card, the higher the performance.
The memory clock also has a role to play in determining the effective memory speed (in Gbps) of a GPU. This metric, combined with the bus width, allows you to calculate the maximum bandwidth of a GPU. You can even change the core clock and memory clock, which is the same as overclocking your graphics card.
Total Graphics Power (TGP)
Just like a CPU has TDP (Thermal Design Power) indicating its maximum power draw, a GPU has TGP (Total Graphics Power). The TGP denotes the maximum amount of power a graphics card can consume during regular use and is measured in watts. For instance, the RTX 4090 is rated at 450W, meaning that’s the maximum amount of power it generally draws. Momentary spikes can push the power draw beyond the TGP, but those last for micro-seconds.
The TGP is not just an indicator of how power-hungry your graphics card will be. It also tells you about the GPU temperature you can expect when running games or other GPU-intensive tasks. A budget or low-end GPU that doesn’t draw more than 150W may not get as hot as a more power-hungry graphics card.
Good to know: did you know that you can undervolt your GPU to save power?
FP32 Performance
FP32 – or floating-point single precision – is a mathematical way of measuring the theoretical performance of a GPU. Although it doesn’t encompass everything that goes into performance, it can be useful when comparing different GPUs, even across generations. It indicates how many floating-point operations a GPU can perform in a second and is measured in TFLOPS, where 1 TFLOP is equal to one trillion operations per second.
For example, if you compare the new RTX 4070 Super with, say, the RTX 2070 Super, the 4070 Super has 35.48 TFLOPS against the 2070 Super’s 9.06 TFLOPS. Even though this number seems to indicate an increase of four times more, the 4070 Super is only around twice as powerful as the 2070 Super. Hence, FP32 isn’t comprehensive, but it’s still a helpful metric when you need a general idea about the performance of a GPU.
Other Specifications
While the specs shown above are the most critical when comparing GPUs, there are some other specs that can help you when making a buying decision.
For instance, the PCIe generation that your GPU supports translates to a higher theoretical bandwidth. This may not affect your gaming performance today, but could have an impact in the future. Another important aspect is the support for graphical APIs like DirectX, Vulkan, and OpenGL. Almost all modern GPUs would support each of these, so this may only be of help when comparing a really old GPU with a modern one.
Lastly, the software features supported by your graphics card can make or break your decision. Nvidia’s DLSS, AMD’s FSR, and Intel’s XeSS upscaling technologies differ in their performance and game support, so check which upscaling tech (and which version) your GPU supports before buying.
Some software features are hardware-limited, such as Nvidia’s Frame Generation technology that boosts framerates in games. Currently, only the RTX 4000 series GPUs support it. Ensure that you’re not missing out on key features due to the GPU generation you’re investing in.
FYI: don’t know what’s inside your PC? Learn how to check your PC specs on Windows.
There are tons of GPU specifications you can get into once you dive into the process. Always check real-world reviews and benchmarks by reputed outlets and creators to form your decision. Check the performance numbers of your shortlisted GPUs in the games and applications you want to run. Refer to Reddit posts, YouTube benchmarks, and reviews by trusted online publications. There are always multiple things to consider when building a gaming PC, so take your time, and don’t rush into anything.
Image credit: Unsplash
Our latest tutorials delivered straight to your inbox
