NVIDIA has been rumored to be working with an upcoming Pascal graphics processing unit. Recently, The NVIDIA GP104 has been spotted revealing that it might feature GDDR5X. On the other hand, the GP100 has is seen in transit from TSMC's fabrication plants to Nvidia's testing facilities in India.
According to TweakTown, it seems that NVIDIA is already playing around with its next-gen Pascal GPUs. This is due to the new listing spotted on a shipping manifest from Zauba. The NVIDIA's upcoming GP104 will be the mid-range part, just like the GM204 which give rise in the GeForce GTX 980.
The new GP104 GPU arrives in a 37.5 x 37.5mm BPA package that is basically smaller than the GM204 which arrived in 40 x 40mm. Additionally it is armed with more pins than the GM204, with 2152 vs 1745, which will be thanks to the 16nm FinFET process.
The GP104-based card will use GDDR5X, while the higher-end offerings will be powered by HBM2. This will make the GP104 and cards under that much inexpensive compared to the much higher prices HBM2 technology on the enthusiast products.
Furthermore, TechFrag added that the shipping manifests also revealed that the GP104 graphics card could be powered by a completely new liquid cooled solution. This is an indication of the Nvidia's plans for adopting it as a stock feature on its future GPUs.
Aside from the NVIDIA GP104 Pascal GPU, there is another chipset that the company is working on. As a matter of fact it has been spotted on NVIDIA'a testing facility in India.
WCCFtech reported that the chipset is named GP100 GPU, the largest and most powerful of Nvidia's Pascal graphics chips. This chipset features a Pascal microarchitecture and DirectX 12 feature level 12_1 or higher. The Pascal GP100 is the successor to the GM200 GPU found in the GTX Titan X and GTX 980 Ti. It has a built on the 16FF+ manufacturing process from TSMC.
The NVIDIA chip allegedly has a total of 17 billion transistors, more than twice that of GM200. It also features four HBM2 stacks, for a total of 16GB of VRAM for the consumer variant and 32GB for the professional variant. The GP100 will sport a 4096bit memory interface and it will feature NVLink and support for Mixed Precision FP16 compute tasks at twice the rate of FP32. Its 2016 release, estimated to be out during the Summer break.
Additionally, HPCwire reported that NVIDIA Pascal GPUs are coming to Automotive Supercomputer. It is known to everyone that NVIDIA is enthusiastic to bring elements of its effective GPGPU ecosystem with the HPC community and handover these to the much wider, and more profitable, consumer space. At SC15, CEO Jen-Hsun Huang referred to machine learning as HPC's first customer killer app. Google, Microsoft, and Facebook have all made major machine learning in-roads in recent months and GPUs are a key element in all of these projects.
With the NVIDIA GP104 and GP100 seen in Zauba and NVIDIA testing facilities respectively, these chipsets could really be out shortly. The company, however, still hasn't given any confirmation regarding the aforementioned reports. Hopefully, more details about the GPUs will be confirmed soon.
