Seventh generation and beyond

In early 2010, NVIDIA's Fermi architecture was introduced for GPGPU computing. The new features it brought were the following:

• True HW cache hierarchy
• Error-correcting Code (ECC)
• Unified memory address space
• Concurrent kernel execution
• Better double-precision performance
• Dual warp schedulers

The GTX 480 was launched with 480 CUDA cores: 15 streaming multiprocessors, each with 32 CUDA cores.

An AMD equivalent card to the GTX 480 was the Radeon HD 5870. The HD 5000, 6000, and 7000 families were launched between 2010 and 2012. AMD's Radeon HD 5000 family had a particular advantage over NVIDIA's products at that time.

AMD announced the first generation accelerated processing units (APUs), Llano for high-performance and Brazos for low-power devices, in January 2011. These 64-bit microprocessors were designed to function as a CPU and GPU on a single die. Thus, APUs were deemed to be more powerful than Intel's integrated graphics.

NVIDIA grew dominant in the HPC and self-driving car industry with its GPUs. From 2010-2017, it launched the GTX 500, 600, 700, 800, 900, and 10 series of GPUs.

In 2016, AMD launched a new open source initiative called GPUOpen. Within this initiative, ROCm (Radeon Open Compute Platform) was introduced. This was a fresh new start for AMD. For the first time, an open source alternative to CUDA had arrived! Like CUDA's NVCC compiler driver, ROCm released the open source HCC compiler. During its launch, AMD highlighted the open source advantage in one of its early presentations:

"Through GPUOpen, we can notice the significance of an open source model, making the compiler much more programmer friendly to allow better debugging!"

In addition to HCC, ROCm also includes HIP, which can be very useful for converting CUDA code into portable C++. HIP can be used for GPU computing on both NVIDIA and AMD GPUs. OpenCL is also a part of the ROCm stack.

In 2017, AMD launched its Vega series of GPUs. In 2018, NVIDIA launched its new RTX line of GPUs with tensor cores specialized for deep learning (a subset of machine learning).

In 2019, AMD announced its new Vega 20 GPU, namely the Radeon VII.

So, we can note how the competition between NVIDIA and ATI continues to evolve; it now continues between NVIDIA and AMD, since the latter acquired ATI in 2006.

Let's now look deeper into NVIDIA's computing platforms in the next section.