According to AMD, the new W9100 will be girded by 32 GB of fast GDDR5 memory. Diving into the computational weeds, AMD’s engineers have stated that the W9100 will use a 512-bit memory interface and have a 320 GB/s memory bandwidth. While those might seem like just a couple numbers, what they actually translate to is the ability for W9100 users to work with extremely larger assemblies or blocks of data in real time without any processing lag.
When it comes to supporting simulation, the W9100 doesn’t falter. To backstop its mammoth 32 GB of memory, the W9100 can also deliver 5.24 teraflops (TFLOPS) of peak single-precision floating-point performance and 2.62 TFLOPS of peak double-precision floating-point performance. Again, sorting through what that means can be a bit of a chore, but the W9100’s single-precision floating-point performance means that simulations and rendering can be done extremely quickly.
If, however, engineers find themselves with the need for even greater numerical accuracy, like, say, when they’re working with computation fluid dynamics simulations, structural mechanics or aerodynamics, then double-precision performance can be unleashed to ensure that calculations are correct. So, is 2.62 TFLOPS am impressive score? Well, let’s just say that no graphics card in the past has ever breached the 2.0 TFLOPS mark when it comes to double-precision calculations.
Beyond its raw computing power, the W9100 can also support up to six 4K screens, making it an ideal card for immersive virtual reality and augmented reality, not to mention the visualization of simulations.
AMD says that its W9100 will be available beginning Q2 2016. The 32 GB flavor of the W9100 will retail for $4,999, and its older, 16 GB counterpart will cost $3,999.