GTX 1080 ARISES
A Second Form of Pascal The era of the Maxwell GPU is officially over, and now comes the days of the Pascal GPU. But unlike the Pascal GPU that was unveiled during GTC this year, the GeForce variant of this Pascal GPU isn’t built on the foundations of NVIDIA’s allnew powerful HBM2 format.
We bet you’re wondering what the Founders Edition moniker stands for. It’s actually just NVIDIA’s way of saying ‘reference design’, as well as being a graphics card that NVIDIA themselves will offer to consumers directly, rather than just granting the rights to its manufacturing partners.
Aside from that fact, the GTX 1080 Founders Edition is still very much using the new 16nm FinFET process, GP104 Pascal GPU architecture, but instead of HBM2, the card uses the recently introduced GDDR5X memory format. More specifically, it comes equipped with 8GB of GDRR5X memory that’s been clocked at 10,000MHz, and a core GPU clock of 1,607MHz that’s been factory overclocked to 1,734MHz. Couple that with a 256-bit memory bus and an effective bandwidth of 320GB/s, what you have is a gamer-oriented card that is set to roll through graphically-demanding 4K titles, along with the recently reincarnated medium of Virtual Reality (VR). Additionally, the card also has 2,560 CUDA cores.
NVIDIA has stated that Pascal is more power efficient and more graphically driven than the GTX 980 Ti and the TITAN X. To give you an idea as to just how powerful this card really is (before we jump into the benchmarks and performance numbers): the GTX 1080 Founders Edition is a 180W card that runs off a single 8-pin PCIe power port, which ran our games at 4K resolution without a hitch. Both the GTX 980 Ti and the TITAN X are both 250W that run on one 8-pin and one 6-pin PCIe port, and we witnessed both cards struggling with some effort to actually maintain a decent framerate at the same resolution level.
How the GTX 1080 Founders Edition is able to maintain its breakneck performance basically boils down to its four Graphics Processing Clusters (GPCs). These GPCs consist of 20 Streaming Multiprocessors (SM), and each SM is fused together with a Polymorph Engine to create something known as a Texture Processing Cluster (TPC). Long story short: it’s these TPCs that help the card perform tasks, such as tessellations and perspective correction.