NVIDIA’s next-generation flagship graphics processor, codenamed “GP100,” has moved to testing phase. This is done by a limited batch of completed chips are sent to NVIDIA from the foundry partner for testing and evaluation. The chips tripped speed-traps on changeover airports, on their way to NVIDIA. It is predicts by 3dcenter.org that the GP100, based on the company’s “Pascal” GPU architecture, will feature no less than 17 billion transistors, and will be built on the 16 nm FinFET+ node at TSMC. This GP100 will feature an HBM2 memory interface. And HBM2 allows you to cram up to 32 GB of memory. With the flagship product based on GP100 could feature about 16 GB of memory. Anyway NVIDIA’s design goal could be to squeeze out anywhere between 60-90% higher performance than the current-generation flagship GTX TITAN-X.
nVidia GP100
confirmed: Pascal architecture
sure: DirectX 12.1 in Hardware
sure: planned for the enthusiast / professional segment (successor to the GM200 chips)
allegedly: 17 billion transistors
pretty sure: 16ff + production by TSMC
pretty sure: Tape-out in June 2015
estimated: 4500-6000 shader units
confirmed: HBM2 memory interface with up to 32 GB HBM2 memory (probably a maximum of 16 GB for gamers variants)
suggesting 4096 bit DDR interface width (yields on 1000 MHz memory bandwidth of 1024 GB / sec)
suggesting Single Precision / Double Precision ratio of 2: 1
confirmed: NVLink feature for coupling of GPUs with CPUs in supercomputing applications (not active in Gamer variants)
Performance: estimated + 60-90% compared to GM200
Release: estimated Q2 / 2016, Gamer variants may only Q3 / 2016
Advertisement
