News & Analysis
Comment
docdivakar
@DrDSP, the GPU usage in servers is also driven by high performance computing, ...
BobsUrUncle
There's no imagination evident in this market space. Die shrinks, and more cores ...
Nvidia hails Kepler, Maxwell in GPU roadmap
Peter Clarke
9/22/2010 6:36 AM EDT
LONDON – Nvidia Corp. has unveiled a road map of graphics processors according to online reports from the GPU Technical Conference staged by the company in San Jose, Calif.
The company has just begun shipping a processor known as Fermi and it is due to be followed by Kepler manufactured on 28-nm process technology in the second-half of 2011 and then by Maxwell two years after that. Kepler is expected to provide three to four times the performance of Fermi.
"We expect to go to production later next year, the design is progressing very rapidly. There are hundreds of engineers working on it," CEO Jen-Hsun Huang said of Kepler, according to a Reuters report.
Nvidia (Santa Clara, Calif.) with an established its position as a maker of graphics processors and graphics cards but faces a change in the market as PC processor makers Intel Corp. and Advanced Micro Devices Inc. start integrating GPUs within the CPU.
Nvidia is moving in the same direction by integrating its own GPUs with processing cores licensed from ARM Holdings plc (Cambridge, England) under the Tegra brand.
"The Tegra business is probably a year behind my goals. However, the Tegra 2 uptake, as I hope you will see later in the year, is really quite phenomenal," Reuters quoted Huang saying.
Related links and articles:
News articles:
MathWorks supports Nvidia's graphics chips
Nvidia takes license from Rambus in wake of ITC ruling
|
|
|
|
|
garydpdx
9/22/2010 10:24 AM EDT
I understand that the Tegra is a Nvidia product but I'm wondering if their GPU's have been integrated onto SoC's from other makers? Nvidia is facing a challenge in being independent, with AMD producing and integrating CPU's and GPU's (from their Canadian ATI acquisition) coming out as Fusion as well as Intel putting their homegrown GPU's onto coming chips.
Sign in to Reply
Rick Merritt
9/22/2010 11:11 AM EDT
I have not heard of Nvidia licensing its GPU cores for use on anyone else's processors, but 2011 will be a big year for heterogeneous processors from Intel, AMD and others--so this will be an important space to watch.
Sign in to Reply
qwerty5
9/22/2010 12:29 PM EDT
NVidia should just acquire Imagination Technologies. It is the most widely used GFX core on cell-phones.
Sign in to Reply
Dr DSP
9/24/2010 1:40 PM EDT
The GPU market seems to be driven by the high performance gaming segment. This segment will only go for an integrated solution if it offers the best performance. Can an integrated solution even get close to an add-on device?
Sign in to Reply
BobsUrUncle
9/24/2010 1:43 PM EDT
There's no imagination evident in this market space. Die shrinks, and more cores is the order of the day.
These companies need to work on making PCs easier to use and program. Improve the man-machine interface. Use 3D to create immersive environments.
Make it so easy my mother use it. Then we'll see some real growth.
Sign in to Reply
docdivakar
10/4/2010 2:23 AM EDT
@DrDSP, the GPU usage in servers is also driven by high performance computing, as I commented elsewhere in EE Times. About two weeks ago at Ansys User Conference in Santa Clara, I saw a a quad processor server with 4 GPU's (NVidia's Tesla) that showed some screaming performance for high end simulations.
Dell, IBM and even lesser known names (Mercury?, Appro?, etc) are all beginning to use Nvidia's Tesla M2050 in multicore multiprocessor servers mother boards. How ever, problem with GPU's is the heat dissipation that adds to the cooling costs of the server chassis.
There may be a business case for integrating GPU functions in CPU's but for higher bandwidth computing, products like Tesla, Tegra are going to have a longer shelf life in servers. This trend is already well supported by the shift to cloud computing in many enterprises.
Dr. MP Divakar
Sign in to Reply