Nvidia Pascal GP104

Geforce GTX graphics card Nvidia 1070 undresses and teaches all the intricacies of its core Pascal GP104, which is exposed and is beautifully photographed. These are the first images that have reached the public of the core 314 mm2 at all if splendor, so if you’re a hardware enthusiast, grab your favorite drink, sit down, and get ready to drool relate.

Let’s make it clear, first, that all credit goes to Fritzchens Fritz has come to surprising extremes to get these amazing pictures of the die. He has a wonderful collection of images of dies with crystal clear quality of a wide range of graphics cards, the latest of which is the GP104, the model used to move Geforce GTX Nvidia 1070 and 1080 cards.

With Maxwell cores, predecessors of the current Pascal and who were responsible for moving the GTX 900 series, Nvidia introduced Maxwell Streaming Multiprocessor (SMM). The SMM is built on the strengths of the core Nvidia Kepler SM that was introduced with the GTX 600 and 700 series, which Nvidia called SMX. He also has rejected many unnecessary complexities that allowed the engine speeds offered more performance and higher clock. Pascal SM, in itself, is an evolution of SM Maxwell, a more streamlined and intelligent motor.


The physical layout of the core Pascal GP104 is grouped into four main divisions GPC engine, each of which occupies one fourth of the chip, being in turn divided into five sub SMs (Smart Shaders). To keep the beast well occupied GDDR5X there are eight segments of 32-bit memory that make up a ring around the periphery of the die.

Nvidia Pascal core block diagram GP 104

The SMs provision is almost identical to what we have seen in the GP100 core that moves the new accelerator Tesla P100, which is much larger by incorporating 3840 CUDA cores. Only within the GP100 each SM contains exactly half the number of CUDA cores, dispatch units and programmers warp the GP104. But in turn, there are twice SMs per GPC. So the main difference between the design of GP104 and GP102 is the Nvidia CUDA 64 cores are grouping in pairs formed by SM 128 CUDA cores each and, in turn, calling the older units 128 an SM. This, while the exact same proportion of units of issue, programmers warp and buffers instructions per core CUDA we’ve seen in the GP100 is maintained.

Get Latest Update, Enter your email address:

Delivered by FeedBurner

You might also likeclose