What is the reason? Why dual GPUs, once the pinnacle of technology, were “killed”?

OUWA – What is the reason? Why dual GPUs, once the pinnacle of technology, were “killed”? Perhaps many of you already know about setting up multiple GPUs for gaming and work, but before that, there was even a time when graphics cards containing multiple GPUs existed. For example, in 1997, the Dynamic Pictures Oxygen 402 was one of the first products to have not two but four GPU chips on a single board.

Companies like 3dfx brought dual GPUs to prominence and the position was later taken by ATI (later AMD) and Nvidia, which continued to produce dual GPUs until the last years of the last decade. Sadly, you won’t find any dual GPUs for end users anymore. What exactly makes the once revolutionary dual GPU formula a thing of the past? Let’s find out with Sforum in this article.

The dual GPU design results in high latency and poor efficiency

With the graphics quality of games improving rapidly after 2000, developers began to opt for deferred rendering. Unlike the outdated forward rendering technique, deferred rendering is more effective at handling scenes with multiple light sources and has less impact on the graphics card without affecting image quality too much. It did this by dividing the rendering process into several stages and using information from previous stages to avoid any unnecessary rendering.

The key point between deferred rendering and multi-GPU or dual-GPU designs is that the final frame depends on its previous rendering stages so data related to those stages must be transferred from this GPU to another GPU. Initially, the PCIe interface was used to transfer that data, and eventually, SLI and Crossfire connectors were added, but even they did not have the necessary bandwidth.

Another issue raised is the inefficient use of VRAM. Since both GPUs only have access to their own VRAM, each VRAM must store the same set of textures to allow both GPUs to render the same scene. That means a dual graphics card with 8GB of VRAM is only really 4GB effective.

Both of these problems exist in dual GPUs. Even though both chips have the same PCB, there still exists a communication gap between the chips and their VRAM. This gap can be reduced with enough effort, at least on graphics cards that have both chips on the same PCB. However, due to lack of motivation, Nvidia and AMD did not continue to develop.

Performance issue

Among the most notable problems that come with using multiple GPUs or dual GPUs is stuttering. This is due to the way both GPUs divide the workload of the rendering frames.

With the release of DirectX12, multi-GPU support has really improved. However, this cannot change the situation because the number of people using dual cards or cards with dual GPUs is actually too small, and it is not worth the effort for game developers to optimize them, so dual GPUs cannot be used. promote inherent strength.

Rapid innovation in single-chip GPUs

In 2014, Nvidia released its last dual GPU – the Titan Z for $3000. This is a dream card for many people with 5 TFLOPS of FP32 computing speed and 336 GB/s memory bandwidth. Just four years later, in 2018, the RTX 2080 Ti was launched at less than half the price of the Titan Z with 13.45 TFLOPS of FP32 and 616 GB/s memory bandwidth, while also having iconic features. networks like DLSS and ray tracing.

With single-chip GPUs improving so quickly, investing such a ridiculous amount of money in dual GPUs may no longer make sense, so periodically upgrading the GPU is a better option. Therefore, even if GPU manufacturers create dual GPUs, only truly passionate PC gamers will buy them and certainly sales will not be high nor enough to encourage game developers to optimize. optimized for such graphics cards.

Dual GPUs are no longer relevant

While dual-chip GPUs were once competitively priced and offered increased performance, this soon became a race to see which company could make the most expensive and power-hungry GPUs. In its prime, dual GPUs delivered much better performance for the money. For example, the GTX 295, a dual GPU with an MSRP of $499, offers better performance than two SLI GTX 260 cards with a total price of $900.

But Nvidia’s Titan Z in 2014 was nothing more than hubris. With a price tag of $3000 and less performance than two GTX 780s running SLI, each only costing $650, it was a complete failure. AMD’s R9 380×2 is a similar story; it requires four 8-pin connectors and a 580W TDP, it is bound to fail.

Although these graphics cards have attracted attention, hardly anyone buys them. The only real advantage of dual GPUs is the space they take up, because instead of using two separate GPUs with 2 PCIe slots, dual GPU graphics cards only need a single slot. However, nowadays even setting up 2 graphics cards is no longer popular in the gaming world.

Will dual GPUs return?

In fact, dual GPUs are still being produced today. AMD’s Radeon Pro Vega II Duo used in Apple’s Mac Pro is an example. Setting up multiple GPUs still exists, but it is only used on graphics computer systems, scientific research, developing AI models… As for the gaming world, is it really a good dual GPU design? Multi-graphics card setups are pretty much dead and it’s unlikely to return, and there doesn’t seem to be a reason for AMD or NVIDIA to make them.