DIR Return Create A Forum - Home
---------------------------------------------------------
ComputerGrapes
HTML https://computergrapes.createaforum.com
---------------------------------------------------------
*****************************************************
DIR Return to: Volta
*****************************************************
#Post#: 5--------------------------------------------------
Rumors Roundup 8/29/2017
DIR By: ComputerGrapes
Date: August 28, 2017, 2:59 pm
---------------------------------------------------------
Sources:
HTML https://www.pcgamesn.com/nvidia/nvidia-volta-gpu-specifications/
HTML https://www.nvidia.com/en-us/data-center/volta-gpu-architecture/
HTML http://wccftech.com/nvidia-geforce-20-volta-graphics-card-q3-2017/
HTML http://www.anandtech.com/show/11367/nvidia-volta-unveiled-gv100-gpu-and-tesla-v100-accelerator-
announced
HTML https://arstechnica.com/gadgets/2017/05/nvidia-tesla-v100-gpu-details/
HTML http://www.trustedreviews.com/news/nvidia-volta-specs-release-date-technology-rumours-2952823
HTML http://www.techradar.com/news/nvidia-is-in-no-rush-to-counter-amds-vega-with-next-gen-volta-graphics-cards/
Volta is the name of the microarchitecture that will be the
successor to Pascal. TSMC has announced that they will start
production on the GPUs in late 2017, leaving many to believe
that it will be released early 2018. Earlier rumors said that it
will be released as early as Q3 2017, but NVidia said they
aren't planning to release it that soon, implying that VEGA was
not powerful enough to dethrone the 1080TI.
The first Volta cards have already been released, but it is not
a consumer card. The card that has been released is the Tesla
V100. It comes ready with HBM2 memory, and the GV100 GPU. Memory
bandwidth on this card is a huge 900GB/s. The Tesla V100 has
over 100 tFLOPS of deep learning performance, 16GB of HBM2
running at 1.75GHz, 640 Tensor cores, 5120 CUDA cores, 21.1
billion transistors, a 4096-bit memory bus width, and 80 enabled
streaming multiprocessors. However, it has a 300W TDP, and costs
around $18000 each. GV100's price may be justified, however,
because of the 3 billion dollars spent by NVidia in research and
development for this GPU. The V100 can connect via PCIe or
NVLink 2, which allows 6 links per GPU, for up to 25GB/s of
bandwidth.
On the consumer side, the GPUs, just like the V100, will be
created by TSMC. They are expected to have GDDR6 memory, with
768GB/S memory bandwidth, as well as 10nm, 12nm, or 16nm
processes. The GV104 is expected to run the 2080 and 2070 cards
while the GV102 is expected to run the 2080ti and the next Titan
card. Speaking of Titan, a summer intern at NVidia leaked a
photo of a Volta Titan.
[attachimg=1]
Not much information can be taken from this, but it confirms
that they are currently working on creating another Titan, but
hopefully not a Titan X. Bad new for consumers however, as
prices are expected to be higher than Pascal.
*****************************************************
Page 1 of 1