Sign in with
Sign up | Sign in

Nvida Updates Its GPU Roadmap at GTC 2013

By - Source: Nvidia | B 38 comments

At GTC 2013, Nvidia CEO Jen-Hsun Huang outlined a future where "mobile, desktop and supercomputer technologies intersect in powerful and surprising ways".

At this year's GPU Technology Conference, Nvidia's CEO outlined the company's roadmap for desktop and mobile platforms and described how "GPUs are already finding their way into applications that were undreamed of a decade ago".

He also noted that last year's Kepler GPU architecture has been a "runaway hit for gamers" and underpins a "new generation of hyper-efficient supercomputers" including the GeForce GTX Titan. The roadmap aims to "put the efficiency and speed of our massively-parallel GPU architectures into an ever-broader array of devices" ranging from mobile to supercomputer.

Starting with Nvidia's desktop GPUs, Maxwell will offer unified virtual memory, providing CPUs with access to the speedy memory built into GPUs and vice versa. Following on from Maxwell will be Volta that introduces the idea of 'stacked DRAM' that piles memory modules directly atop the GPU cores and gives the GPUs access to a bandwidth of 1 TB/s or enough to move a full Blu-ray disc of content in 0.02 seconds.

With regards to mobile processors, Logan will pair ARM based mobile processors with Kepler GPUs and provide mobile devices with access to technologies such as PhysX, CUDA 5 and OpenGL 4.3. The Parker architecture will combine 64-bit ARM CPU cores with the next generation Maxwell GPU to provide server-esque functionality including the ability to "gulp down big chunks of data".


We've been impressed by the content and ambitiousness of the GTC presentation and are looking forward to both Nvidia's upcoming architectures and perhaps getting our hands on Jen-Hsun Huang's jacket! For a play-by-play from the keynote, check out our liveblog!


Contact Us for News Tips, Corrections and Feedback

Discuss
Ask a Category Expert

Create a new thread in the News comments forum about this subject

Example: Notebook, Android, SSD hard drive

This thread is closed for comments
Top Comments
  • 13 Hide
    Memnarchon , March 20, 2013 10:15 AM
    "Nvida Updates Its GPU Roadmap at GTC 2013"
    I think its Nvidia. :p 
  • 11 Hide
    de5_Roy , March 20, 2013 10:08 AM
    nvidia is really saying,"we'll get there when tsmc gets there. until then you'll just have to wait and buy our overpriced cards(gtx 650ti, gtx 680, titan)."
    the jacket was the real highlight!
    nvidia doesn't have a capable or existent cpu to compete against hsa. build the cpu first, brag later.
Other Comments
  • 11 Hide
    de5_Roy , March 20, 2013 10:08 AM
    nvidia is really saying,"we'll get there when tsmc gets there. until then you'll just have to wait and buy our overpriced cards(gtx 650ti, gtx 680, titan)."
    the jacket was the real highlight!
    nvidia doesn't have a capable or existent cpu to compete against hsa. build the cpu first, brag later.
  • Display all 38 comments.
  • 4 Hide
    ibjeepr , March 20, 2013 10:11 AM
    So Maxwell in early 2014 by the looks of it. Glad I didn't wait.

    The mobile side of things likes interesting but not relevant to me.
  • 13 Hide
    Memnarchon , March 20, 2013 10:15 AM
    "Nvida Updates Its GPU Roadmap at GTC 2013"
    I think its Nvidia. :p 
  • 2 Hide
    A Bad Day , March 20, 2013 10:25 AM
    Quote:
    Starting with Nvidia's desktop GPUs, Maxwell will offer unified virtual memory, providing CPUs with access to the speedy memory built into GPUs and vice versa.


    HSA vs Nividia's version vs Intel's CPU brute force...

    Hm...
  • 3 Hide
    gm0n3y , March 20, 2013 10:35 AM
    Stacking the RAM directly on top of the GPU doesn't seem like a very good solution thermally. I'll be interested to see if the benefits outweigh the likely lower clock speed. On the plus side they should be much more efficient, running at lower voltages and clock rates.
  • 2 Hide
    memadmax , March 20, 2013 10:39 AM
    At some point, Nvidia is gonna have to enter the x86 market, it would be great if they released a x86 Tegra chip....
  • 1 Hide
    nukemaster , March 20, 2013 10:41 AM
    Not sure I would call the GTX 650 ti overpriced(I have one from MSI in my media center and its power consumption is VERY good for media playback[while being nearly silent] and it can do some med/high setting gaming.).

    When on sale its not bad. I mean for a small bit more, you get the 7850(this thing rapes the 650 ti).

    The card seems to fit right into its price range.
  • 5 Hide
    nukemaster , March 20, 2013 10:41 AM
    memadmaxAt some point, Nvidia is gonna have to enter the x86 market, it would be great if they released a x86 Tegra chip....

    I do not think Intel will let that happen.
  • 2 Hide
    nforce4max , March 20, 2013 10:58 AM
    Interesting that they are going with stacked dram and the lower latency will make for a huge boost performance wise but likely to be very expensive considering yield. I wouldn't be surprised if the failure rate goes up with possible thermal tolerances with such a high IC density. I wonder what material they will be using for the IC to IC bonding and thermal interfacing between each IC...
  • 5 Hide
    bentonsl_2010 , March 20, 2013 11:26 AM
    I The spirit of comic book fans here is how Kayla works in the naming convention. (For those of you wondering who Kayla is suppose to be)

    "NVIDIA also showcased an mITX-like board called Kayla, which features a Tegra 3 SoC and a low power Kepler GPU, presumably from Logan."

    For those of you who watch the Xmen origins movie for wolverine she was his wife in the movie

    "Kayla Silverfox was a mutant with the ability to persuade anyone she touched into doing what she wanted. Her sister is Emma Frost. After Logan left the Team X project and went back to Canada, Silverfox was tasked with keeping an eye on him. During their six years they resided together in a cabin high in the mountains, she manipulated him into a state of complacency with her power, calming him though the nightmares of the past. Kayla told him about a spirit who was tricked into being parted forever from his lover, the Moon. The spirit, whose name translated to 'Wolverine', was subsequently forced to look at the Moon forever and never be with her again. Kayla and Victor Creed faked her murder with hydrochlorothiazide for William Stryker so he could trick Logan into participating in his Weapon X project. When Logan was about to get his skeleton injected with adamantium, he asked for dogtags bearing the name "Wolverine", inspired by Silver Fox"

    So since this is an off product supporting Logan it makes perfect sense to use his former wife as the name.
  • 2 Hide
    blubbey , March 20, 2013 11:38 AM
    Remember that they've had delays before. Volta will probably be 2017 or so.
  • 2 Hide
    blazorthon , March 20, 2013 11:43 AM
    Volta just seems to take advantage of DDR4 or a GDDR based on DDR4. The claims made for it may not be outlandish.
  • 1 Hide
    CaedenV , March 20, 2013 12:09 PM
    memadmaxAt some point, Nvidia is gonna have to enter the x86 market, it would be great if they released a x86 Tegra chip....

    1) how could they? If I am not mistaken AMD, Intel, and IBM are the only companies with access to the tech. AMD and Intel will not let in another competitor, and IBM's x86 stuff is too dated to be of any use

    2) why would they? I can't think of an Nvidia product that really needs it. They have ARM for light weight stuff, and they have their GPU tech running more and more of the server space. These are 2 industries that are seeing tremendous growth right now, and Nvidia is still a relatively small company. They would have to make a lot of changes in order to add essentially a 3rd in-house company.
  • 0 Hide
    ashesofempires04 , March 20, 2013 12:19 PM
    Maxwell isn't going to offer the CPU access to any faster memory. The PCI-e 3.0 bus has a max theoretical transfer in the realm of just under 16 GB/sec (985 MB/sec per lane x16 lanes). Current DDR3 RAM has a transfer rate of roughly 14.9 GB/sec at 1866 mhz, and at 2133 it exceeds 17 GB/sec. Maxwell will only do two things: allow the CPU access to more ram, and cause problems because that RAM is of a different architecture than standard DDR3 RAM.

    Volta, on the other hand, will eliminate a bottleneck that has crippled poorly architected and low-end GPU's: memory bandwidth. Higher memory bandwidth in AMD's cards has given them better performance scaling as anti-aliasing gets more complicated, and as screen resolutions go up (and as monitors are added).

    Maxwell's bit about CPU access to the GPU's discrete RAM is probably more about CUDA/OpenCL software's increasing need to move code blocks back and forth more easily, and less about software using GPU RAM as more system ram. There's no point. Crossing the PCI-e bus cuts out any benefit you may derive except for sheer quantity available to write to.
  • -1 Hide
    nebun , March 20, 2013 12:21 PM
    Memnarchon"Nvida Updates Its GPU Roadmap at GTC 2013"I think its Nvidia.

    maybe nVidia
  • 1 Hide
    Memnarchon , March 20, 2013 12:42 PM
    nebunmaybe nVidia


    Just no.
    Both Forbes and Wikipedia has it as Nvidia.
    You are being confused with the company's logo.
  • -8 Hide
    madjimms , March 20, 2013 1:21 PM
    nukemasterI do not think Intel will let that happen.

    Intel? AMD owns x86..... Intel has to pay for a license.
  • 6 Hide
    s3anister , March 20, 2013 1:36 PM
    madjimmsIntel? AMD owns x86..... Intel has to pay for a license.

    This is seriously incorrect. Intel developed the first x86 processor with the 8086 in 1978 and as such x86 is Intel's original architecture. Now eventually AMD was given legal right to manufacture their own x86 chips to compete with Intel, however, what you're confusing is AMD's rights to the x86-64 architecture extension that they developed.
  • 0 Hide
    alchemy69 , March 20, 2013 2:00 PM
    MemnarchonJust no.Both Forbes and Wikipedia has it as Nvidia.You are being confused with the company's logo.



    I think nebun's comment was more about the missing 'i' than the capitalization.
  • 0 Hide
    ko888 , March 20, 2013 2:13 PM
    MemnarchonJust no.Both Forbes and Wikipedia has it as Nvidia.You are being confused with the company's logo.

    Both Forbes and Wikipedia are wrong.

    The registered trademark is NVIDIA®

    Look at their web site and their product literature.
Display more comments