You’ve probably spent hours cabling up a new PC, loaded up the latest open-world role-playing game only to find that your shiny new high-end graphics card isn’t quite as powerful as you thought. Instead of silky smooth frame rates effortlessly clearing the 90 FPS mark, you’re stuck below 60 FPS — and that’s with all the settings turned down. It’s hard to think that the key to better performance is so diametrically opposed to the graphics card hype train, but there’s no getting around it: the graphics card chipset is the single most important part of the equation.
Chipsets are what provide the graphics capabilities that a video card offers. While the fans may be wowed by things such as cooling, memory overclocking, or custom exterior casings, the architectural design of a graphics processor cannot be changed. In today’s video, we explore the evolution of graphics processors and compare the key differentials between the many generations of video cards.
Architecture Sets the Foundation
The GPU architecture itself is the blueprint for the building, and this gives insight into how the “structure” will perform in terms of core efficiency, memory controllers, and compute units. The number of CUDA cores are predesigned into the chipset, and even with superior cooling or overclocking in the factory design that utilizes a particular chipset, it will never compare to the compute performance of a graphics card that was designed around a different chipset with only 896 CUDA cores vs the 2048 found here.
Your computer’s video card uses one or more GPUs to perform compute operations, and some of those will be Streaming Multiprocessor’s that can be used to compute your game in parallel, or specialized for things like ray tracing when supported by the game for enhanced lighting. Information about the specific details of this video card can be found in the the ‘the_specs’ section below. The NVIDIA GeForce GTX 1650 is a very power efficient video card with a budget price-point and is based on the NVIDIA GeForce GTX 1650 chipset utilizing NVIDIA’s latest-generation architecture, codenamed Turing.
This is especially important when looking at the memory architecture of the graphics card and the Chipset itself. The memory architecture of the Chipset essentially dictates the graphics card’s memory bus width and how fast it will be able to transfer data across said bus. For example, an 128-bit bus is no match for a 256-bit bus and, despite best efforts from manufacturers to hype the performance of the 128-bit bus, the facts are clear to anyone willing to do the maths: a wider bus equates to faster texture streaming and increased frame buffer performance.
Manufacturing Process Matters
The size of nodes has a significant impact on the performance and efficiency of nodes. Smaller nodes offer increased transistor density and power improvements. For example, a 7 nm chipset could potentially pack a greater number of compute units per square millimetre of real-estate than a 14 nm alternative, while also consuming less power.
PC parts can only get so new before the whole thing becomes less viable, but with the newer methods for building computers, there’s also room for increased clock speeds and better cooling. This means air cooling isn’t the handicapped, low-performance option it once was.
Power Efficiency Shapes Real Performance
It is common to hear the TDP of a CPU mentioned as a measure of efficiency, but the actual chipset is what determines how power is distributed and controlled around the silicon. This is a far better measure of true performance than the brief burst required to record a high Dhrystone or Cinebench score.
Most designs will go all out for huge burst score then drastically underperform as the temperatures rise. A few designs will actually keep up the performance over long periods of gaming, where this actually occurs. Power efficiency will also become a greater priority in smaller cases or laptops. Here a efficient design may allow for higher sustained boost clocks than an alternative design that prioritises peak score.
In addition to the support of multi-video output, the new chipset design for our multi-graphics video boards also features support for Voltage Regulation (VR) and Dynamic Voltage Scaling (DVxS). These power management features improve the overall stability of the system and allow for higher frame rates for the same graphics cards. In reality, the same spec cards can have drastically different frame rates.
Application-specific Architecture
We also threw some CPU-intensive tasks at the chips, such as photo editing and video encoding, as these are more suited to specialized encoding units, where the ZEN+-enriched X370 chipset really shines. But when it comes to games, it is all about raw parallel processing capability, and the Ryzen 5 generally comes out on top.
Professional applications will usually use compute shaders differently than game engines and as such, the performance improvements observed by chipster may not translate. Some chipsters may have dedicated silicon for video encoding and some have even been designed with support for machine learning (ML/AI) tasks and can see large gains in some applications while providing no benefits in others.
The Performance Ceiling
We are currently restricted to some degree by the underlying silicon design. As mentioned earlier, the problem of memory bandwidth is a major one and a large increase in core clock speed is unlikely to make any meaningful difference to memory bound applications that are already pushing the existing memory subsystem to its limits.
If an application is able to exhaust the system’s processor then increased boost clock speed will offer little benefits. The difference between 2 models based on the same chipset are very minimal, typically no more than a few percent. Also within the same model of chipset the variation between different cooling solutions and overclocking from the factory will be minimal.
The chipset is what will in the end determine how well a particular video card can hopefully perform. All else will determine how close that particular card will hopefully perform to those theoretical limits. Keep this in mind when deciding whether or not to purchase a particular video card. There are so many aspects of a video card that can be examined in depth. But if you’re planning on purchasing a video card, it’s foolish to place too much emphasis on anything outside the chipset.
