NVIDIA launches Pascal
In the previous part of the series, we learned that deep learning drove NVIDIA’s (NVDA) revenue in fiscal 1Q17. The company recently launched its flagship Pascal GPU (graphics processing unit) architecture, further boosting its deep learning, gaming, and virtual reality segments.
The Pascal launch came as a surprise, as it wasn’t expected to happen so soon. Advanced Micro Devices has yet to launch its competing technology Polaris in fiscal 2Q16. Several speculations have been running in the Market regarding Pascal versus Polaris. Let’s look at Polaris and the features that make it a flagship product.
How does NVIDIA leverage Pascal architecture?
Instead of developing a new chip for every application, NVIDIA develops a single architecture and leverages it across multiple segments. The company launched its Pascal architecture, and on this architecture it built the GTX 1080 and 1070 GPUs for gaming and virtual reality and the P100 and DGX1 for deep learning.
The GTX 1080 and 1070 Founders editions will be available for sale on May 27, 2016, and June 10, 2016, respectively.
Pascal climbs two generations
NVIDIA’s GPU segment has been using 28nm (nanometer) process nodes since 2012, whereas microprocessor makers such as Intel (INTC) have reached 14nm process nodes. NVIDIA built Pascal on TSMC’s (TSM) 16nm FinFET (fin-shaped field effect transistor) technology, which significantly increases performance and reduces power consumption while providing cost efficiencies in manufacturing.
NVIDIA’s P100 product houses 15 billion transistors on a single chip due to the 16nm process node. The GeForce GTX 1080 Pascal-based GPU is a mellowed version of the P100. It houses 7.2 billion transistors and 2,560 CUDA cores, compared to the currently used GTX 960 GPU, which houses 5.2 billion transistors and 2,048 CUDA cores.
Moreover, GTX 1080 has a computing speed of 9TFLOPS (tera floating-point operations per second) compared to GTX 980’s 5TFLOPS. There was speculation that Pascal would be NVIDIA’s first GPU to house HBM 2 (High Bandwidth Memory). However, the original Pascal houses Micron’s (MU) GDDR5X (double data rate type five synchronous graphics random access memory).
The GTX 1080 delivers twice the speed and performance of its Maxwell-based predecessor Titan X and is priced 40% lower. We’ll look at the prices of Pascal GPUs in the next part of the series.
You can get exposure to the companies associated with NVIDIA’s Pascal architecture by investing in the VanEck Vectors Semiconductor ETF (SMH). SMH has 2.2% exposure to MU, 4.4% to NVDA, and 12.2% to TSM.