Behind AMD’s Polaris and NVIDA’s Pascal Face-Off



The AMD-NVIDIA GPU competition is getting heavy

We’ve already discussed how Advanced Micro Devices (AMD) has the first mover advantage in HBM (high bandwidth memory) technology with its Polaris GPU (graphic processing unit). But now NVIDIA (NVDA) is introducing its own HBM 2 technology, Pascal GPU, to compete.

Behind AMD's Polaris and NVIDA's Pascal Face-Off

Article continues below advertisement

AMD’s Polaris

In 2016, AMD will launch its next-generation Polaris, which should bring significant improvements in performance, power-efficiency, and features. The key advancement in Polaris is the use of Samsung (SSNLF) and Global Foundries’ 14nm FinFET (fin-shaped field effect transistor) technology.

With its HBM 2 technology, Polaris should also significantly improve performance per watt. AMD has already demonstrated power efficiency by putting Polaris against Maxwell Nvidia GTX 950, with Polaris delivering a similar performance using 86-watt power compared to Maxwell’s 140 watts. This makes Polaris suitable for notebooks and light-weight laptops. Additionally, Polaris will have a CPU (central processing unit) feature of instruction pre-fetch, which improves performance by reducing stalls and wait times.

NVIDIA’s Pascal GPU 

But Polaris will now face tough competition from NVIDIA’s Pascal GPU. Like Polaris, Pascal also has HBM2, but it is being built on TSMC’s (TSM) 16nm FinFET technology. Moreover, Pascal uses NVIDIA’s NVLink, a high-bandwidth communications channel that moves data on the node 5–12x faster than PCIe Gen3 x16 and consumes 3x less energy.

Specifically, Pascal will use 16FP mixed-precision computing, which has lower accuracy than the standard 32FP, making it unsuitable for modern gaming applications. But 16FP should deliver strong power efficiency, making it more suitable for mobile devices.

NVIDIA plans to launch Pascal in June 2016. Notably, the VanEck Vectors Semiconductor ETF (SMH) has exposure to the entire semiconductor supply chain. It has ~13.1% exposure in TSM and 3.7% in NVDA.

Pascal in trouble

It has also been reported that NVIDIA may unveil Pascal at the 2016 GTC (GPU Technology Conference) on April 4–7, 2016. However, the site states that the company is facing technical difficulties in bringing Pascal up to 16nm FinFET speed. If this problem is not resolved, the launch may be delayed.

Now let’s look at AMD’s most talked about technology, Zen Core, which may be equipped with Polaris GPU.


More From Market Realist