The Pascal microarchitecture, named after
Blaise Pascal, was announced in March 2014 as a successor to the
Maxwell microarchitecture. The first graphics cards from the series, the GeForce GTX 1080 and 1070, were announced on May 6, 2016, and were released several weeks later on May 27 and June 10, respectively. The architecture incorporates either 16 nm
FinFET (
TSMC) or 14 nm FinFET (
Samsung) technologies. Initially, chips were only produced in TSMC's 16 nm process, but later chips were made with Samsung's newer 14 nm process (GP107, GP108). New Features in GP10x: •
CUDA Compute Capability 6.0 (GP100 only), 6.1 (GP102, GP104, GP106, GP107, GP108) •
DisplayPort 1.4 (No
DSC) •
HDMI 2.0b • Fourth generation Delta Color Compression •
PureVideo Feature Set H hardware video decoding HEVC Main10 (10 bit), Main12 (12 bit) & VP9 hardware decoding (GM200 & GM204 did not support HEVC Main10/Main12 & VP9 hardware decoding) •
HDCP 2.2 support for 4K DRM protected content playback & streaming (Maxwell GM200 & GM204 lack HDCP 2.2 support, GM206 supports HDCP 2.2) •
NVENC HEVC Main10 10 bit hardware encoding (except GP108 which doesn't support NVENC) • GPU Boost 3.0 • Simultaneous Multi-Projection • HB SLI Bridge Technology • New memory controller with GDDR5X & GDDR5 support (GP102, GP104, GP106) • Dynamic load balancing scheduling system. This allows the scheduler to dynamically adjust the amount of the GPU assigned to multiple tasks, ensuring that the GPU remains saturated with work except when there is no more work that can safely be distributed. Nvidia therefore has safely enabled asynchronous compute in Pascal's driver. • Instruction-level preemption. In graphics tasks, the driver restricts this to pixel-level preemption because pixel tasks typically finish quickly and the overhead costs of doing pixel-level preemption are much lower than performing instruction-level preemption. Compute tasks get either thread-level or instruction-level preemption. Instruction-level preemption is useful because compute tasks can take long times to finish and there are no guarantees on when a compute task finishes, so the driver enables the very expensive instruction-level preemption for these tasks. •
Triple buffering implemented in the driver level. Nvidia calls this "Fast Sync". This has the GPU maintain three frame buffers per monitor. This results in the GPU continuously rendering frames, and the most recently completely rendered frame is sent to a monitor each time it needs one. This removes the initial delay that
double buffering with vsync causes, thereby preventing tearing. The costs are that more memory is consumed for the buffers and that the GPU will consume power drawing frames that might be wasted because two or more frames could possibly be drawn between the time a monitor is sent a frame and the time the same monitor needs to be sent another frame. In this case, the latest frame is picked, causing frames drawn after the previously displayed frame but before the frame that is picked to be completely wasted. This feature has been backported to Maxwell-based GPUs in driver version 372.70. Nvidia has announced that the Pascal GP100 GPU will feature four
High Bandwidth Memory stacks, allowing a total of 16 GB HBM2 on the highest-end models, 16 nm technology, Starting with
Windows 10 version 2004, support has been added for using a hardware graphics scheduler to reduce latency and improve performance, which requires a driver level of WDDM 2.7. == Products ==