Radeon HD 7000:  HD 7790 |  HD 7850 |  HD 7870 |  HD 7870 XT |  HD 7950 |  HD 7970 |  HD 7990 |  Radeon HD 8000:  HD 8750 |  HD 8770 |  HD 8850 |  HD 8870 |  HD 8950 |  HD 8970 |  HD 8990 
GeForce 600:  GTX 650 Ti Boost  |  GTX 660  |  GTX 660 Ti  |  GTX 670  |  GTX 680 |  GTX 690 |  Titan: GTX TITAN |  GeForce 700:  GTX 760  |  GTX 760 Ti  |  GTX 770  |  GTX 780 |  GTX 790 

April 25th, 2012

NVIDIA’s Kepler Efficiency with New Process Technology Explained

NVIDIAs Kepler Efficiency with New Process Technology Explained geforce 600
Interesting article was posted on NVIDIA Blog. A leader of the engineering team that helped TSMC in manufacturing first 28nm graphics chip explains how complicated was this development process.

Joe Greco — an author of this article — compares designing first NVIDIA 28nm chip to designing a new jet engine with a materials which are still in development. He explains how new technologies helped to increase performance per watt of this new chip.

TSMC offered NVIDIA new optimized process technology — 28nm high performance process — which used both the first-generation high-K HKMB process and second generation SiGe (Silicon Germanium). He also explains how big of an impact were those new technologies to performance:

Using TSMC’s 28nm HP (high performance) process enabled us to reduce active power by about 15 percent and leakage by about 50 percent compared to 40nm, resulting in an overall improvement in power efficiency of about 35 percent (see chart). Let me explain why this is so critical.

NVIDIAs Kepler Efficiency with New Process Technology Explained geforce 600 NVIDIAs Kepler Efficiency with New Process Technology Explained geforce 600

Take a moment for reading this detailed article.

Author: WhyCry  

  • BestJinjo

    Kepler is impressive but it still builds upon the foundation laid out by Fermi. In essence, it’s a re-tweaked/improve Fermi architecture. I can only imagine how impressive Maxwell will be in two years from now. Let’s just hope games catch up by then.

    Ya, the benefits of 28nm cannot be dismissed. The continuous node shrinks are critical for technological improvements to take place at the pace set out by Moore’s Law. Now if only NV grew some cajones and sold us a 500-550mm^2 GK110, then it would be even more impressive. 

  • Jerome

    Yes kepler is an improved version of Fermi, but seriously whats wrong with that?  More performance less power draw less heat.  You got to understand how companies like AMD and Nvidia approach a new architecture, theres just some many things.  AMD could make a 550 to 600mm^2 chip that eats 350 watts and it would be the one gpu to rule them all.  However will it be manufacturable?  How much would they pay for sku’s?  Its all about profit and not consumer needs!

  • BestJinjo

    I never said there is anything wrong with Kepler being built off Fermi. My point is normally we get 50-75% faster GPUs every major release. 8800GTX –> GTX280, GTX280/285 –> GTX480/580. GTX680 is not a true flagship product. So NV is shoving it for $500 when really the real Kepler is GK110 and it is MIA.  I think this round is just underwhelming in general. That’s my opinion and I followed GPUs for 10+ years. I am most disappointed by this generation than any other since FX5800Ultra dustbuster and 2900XT. GTX680 is just 30-35% faster than GTX580. That’s unacceptable to me. I’d rather buy a 250W GK110 for $600-650 with full dual precision capability than a $500 195W GK104. 

    Granted, I find no games worth upgrading for, so more likely than not I’ll skip this generation regardless. Just stating that the performance increase is not what we are used to, not at all. 

  • BestJinjo

    Also, to your point, Nvidia has a history of launching large die chips:

    November 8, 2006 – 8800GTX – 484 mm^2 on 90nm
    June 16, 2008 – GTX280 – 576 mm^2 on 65nm
    March 26, 2010 – GTX480 – 520 mm^2 on 40nmThere we have NV making massive chips on 3 different nodes. Each time the performance increase was at least 50-75% over previous generation.GTX680 on the other hand is just a 294 mm^2 die chip, 30% faster than GTX580. It’s only impressive because HD7970 is so lackluster. In the context of NV’s leaps, it’s underwhelming.