Monday, October 3rd 2016
NVIDIA GP107 GPU Built on Samsung 14 nm Node?
Could the upcoming "GP107" ASIC by NVIDIA be its first on the 14 nanometer silicon fabrication process? That's what 3DCenter.org uncovered. Informed sources tell the German tech-site that the GP107 could be the first GPU built by NVIDIA's partnership with Samsung Electronics, after it emerged that the Korean silicon giant could manufacture certain GeForce "Pascal" GPUs on its 14 nm LPP (low-power plus) node. There's also talk of NVIDIA optical-shrinking its existing GeForce Pascal chips to 14 nm, built by Samsung.
The GP107 silicon will power two known mid-range desktop SKUs slated for launch later this month, the GeForce GTX 1050 Ti, and the GTX 1050. Bound for mid-October, the GTX 1050 Ti features 768 CUDA cores, 48 TMUs, 32 ROPs, and a 128-bit GDDR5 memory interface holding 4 GB of memory; with GPU clocks above the 1.50 GHz mark. The GTX 1050, on the other hand, could launch in late-October, featuring 640 CUDA cores, 40 TMUs, 32 ROPs, and 2 GB of GDDR5 memory across the chip's 128-bit memory interface, with the possibility of custom-design 4 GB SKUs. NVIDIA is targeting the $150 and $120 price-points with these SKUs. The company could also work on mobile SKUs based on the chip.
Source:
3DCenter.org
The GP107 silicon will power two known mid-range desktop SKUs slated for launch later this month, the GeForce GTX 1050 Ti, and the GTX 1050. Bound for mid-October, the GTX 1050 Ti features 768 CUDA cores, 48 TMUs, 32 ROPs, and a 128-bit GDDR5 memory interface holding 4 GB of memory; with GPU clocks above the 1.50 GHz mark. The GTX 1050, on the other hand, could launch in late-October, featuring 640 CUDA cores, 40 TMUs, 32 ROPs, and 2 GB of GDDR5 memory across the chip's 128-bit memory interface, with the possibility of custom-design 4 GB SKUs. NVIDIA is targeting the $150 and $120 price-points with these SKUs. The company could also work on mobile SKUs based on the chip.
12 Comments on NVIDIA GP107 GPU Built on Samsung 14 nm Node?
GM200 -> Titan, 980/Ti,970,960,950
GM300 -> Titan, 1080, 1070, 1060
GM400 -> 1050 and a rumored refresh of Pascwell
This whole generation is built out of fresh cores, yet someone will find something to b@#ch about.
WCCF is this way >>>
GCN 1.0, 1.1, 1.2, 1.3
Fundamentally, architecture is still more or less the same as it was in 2011
But other optimizations are performed
1_16nm FinFet node
2_Architectural Optimizations for higher clock
3_Improved Asynchronous compute
4_Fourth generation Delta Color Compression
I think Pascal architecture is not just a simple Die-Shrink of the Maxwell