facebook rss twitter

Nvidia reveals details about its next-gen Pascal GPU

by Mark Tyson on 18 March 2015, 09:43

Tags: NVIDIA (NASDAQ:NVDA)

Quick Link: HEXUS.net/qacp2l

Add to My Vault: x

In addition to the unveiling of the Nvidia GeForce GTX Titan X last night, at the GPU Technology Conference keynote in San Jose, Nvidia CEO Jen-Hsun Huang sketched out some details concerning the company's next generation of GPUs. An updated graphics processor roadmap was revealed with Pascal GPUs arriving in 2016 and the next-next generation Volta GPUs set to arrive in 2018.

Lets first look at what Nvidia claims will be the qualities and strengths offered by Pascal. The new and improved features are summed up nicely in the Nvidia live blog from the GDC Keynote (time: 10.31 onwards), as follows:

"Pascal has three great features. One is mixed precision, which is provides at 3X the level of Maxwell. Another is 3D memory, which provides more bandwidth and capacity simultaneously. A third is NVLink, which is an ability for multiple GPUs to be connected at very high speeds. It also has 2.7X more capacity than Maxwell."

Pascal's new architecture including 3D memory will enable frame buffers up to 32GB in size with approximately 3x the bandwidth available. NVLink is said to offer data transfers between 5x and 12x faster than today's PCI-Express standard and will enable 8-way SLI configurations.

Weighing in all the factors and technology mentioned above the Nvidia CEO summed up that "Pascal will be 10X faster than Maxwell". However it is noted that "this is just a very rough, high level estimate." It applies to at least one use case highlighted - 'Deep Learning' tasks. That sounds interesting but it feels like Nvidia really had to scrape around for use cases to come up with that headline figure which makes one wonder how much of an uplift it will represent to PC gamers, for example.

Other highlights of the Nvidia presentation include the launch of the DIGITS Devbox. This is an energy-efficient, quiet, cool-running PC system put together by Nvidia for data scientists and researchers. It packs four TITAN X GPUs and is set to retail at $15,000. A great tool for speeding up deep learning research, we are told.

Another important development for Nvidia's business is the Nvidia DRIVE PX. Last night we learned of the powerful self-driving car computer's pricing and availability. The system, based around dual Tegra X1 processors, facilitates real time processing and analysis of the data streaming in from sensors and cameras mounted all over the car. It will be available to automakers from May this year, with a development kit costing $10,000.



HEXUS Forums :: 7 Comments

Login with Forum Account

Don't have an account? Register today!
Wonderful, even faster graphics that 99% of game devs won't be able to utilize due to cost of development.
abaxas
Wonderful, even faster graphics that 99% of game devs won't be able to utilize due to cost of development.
I don't think the intention of the video linked in the article is to talk about Pascal's gaming prowess, I think it has more to with the academic nature of nVidia's research programs and partnerships, for data centres that use GPGPU arrays and other areas where the workloads can be run across large numbers of parallel processors instead of confining it to a smaller limited number of CPU cores.
will enable 8-way sli configurations

1.21 gigawatts!!!!!
I may be getting a new PC in 2016 I reckon :P
2016 looks like a good year to upgrade my 680s

sick of SLI - too many driver issues