The road to any new microprocessor design is by no means simple. Planning for a major GPU like NVIDIA's Kepler starts four years prior to the chip's debut. In a world that's increasingly more focused on fast production and consumption of everything, it's insane to think of any project taking such a long period of time.

Chip planning involves figuring out what you want to do, what features you want, what the architecture should look like at a high level, etc...  After several rounds of back and forth in the planning stage, actual architecture work begins. This phase can take a good 1 - 1.5 years depending on the complexity of the design. Add another year for layout and validation work, then a 6 - 9 month race from tape out to products on shelves. The teams that spend years on these designs are made up of hard working, very smart people. They all tend to believe in what they're doing and they all show up trying to do the best job possible. 

Unfortunately, picking a target that's 4 years out and trying to hit it better than your competition is extremely difficult. You can put in an amazing amount of work, push through late nights, struggle with issues, be proud of what you've done and still fall short. We've seen this happen to companies on both sides of the fence, whether we're talking CPUs or GPUs, you win some and you lose some

 

Today NVIDIA unveiled Kepler, a more efficient 28nm derivative of its Fermi architecture. The GeForce GTX 680 is the first productized Kepler for the desktop and if you read our review, it did very well. As our own Ryan Smith wrote in his conclusion to the GeForce GTX 680 review:

"But in the meantime, in the here and now, this is by far the easiest recommendation we’ve been able to make for an NVIDIA flagship video card. NVIDIA’s drive for efficiency has paid off handsomely, and as a result they have once again captured the performance crown."

We've all heard stories about what happens inside a company when a chip doesn't do well. Today we have an example of what happens after years of work really pay off. A trusted source within NVIDIA forwarded us a copy of Jen-Hsun's (NVIDIA's CEO) email to all employees, congratulating them on Kepler's launch. With NVIDIA in (presumably) good spirits today, I'm sure they won't mind if we share it here.

If you ever wondered what it's like to be on the receiving end of a happy Jen-Hsun email, here's your chance:

-----Original Message-----
From: Jensen H Huang 
Sent: Thursday, March 22, 2012 9:48 AM
To: Employees
Subject: Kepler Rising
 
Today, the first Kepler - GTX 680 - is on shelves around the world!
 
Three years in the making.  The endeavor of a thousand of the world's best engineers.  One vision - build a revolutionary GPU and make a giant leap in efficient-performance.
 
Achieving efficient-performance, great performance while consuming the least possible energy, required us to change our entire design approach.  Close collaboration between architecture-design-VLSI-software-devtech-systems, intense scrutiny on where energy is spent, and inventions at every level were necessary.  The results are fantastic as you will see in the reviews. 
 
Kepler also cultivated a passion for craftsmanship - nothing wasted, everything put together with care - with a goal of creating an exquisite product that works wonderfully.  Let's continue to raise the bar and establish extraordinary craftsmanship as a hallmark of our company.
 
Today is just the beginning of Kepler.  Because of its super energy-efficient architecture, we will extend GPUs into datacenters, to super thin notebooks, to superphones.  Not to mention bring joy and delight to millions of gamers around the world.
 
I want to thank all that gave your heart and soul to create Kepler.  You've created something wonderful.
 
Congratulations everyone!
 
Jensen

 

POST A COMMENT

31 Comments

View All Comments

  • kn00tcn - Thursday, March 22, 2012 - link

    that's not the right way to think, engine is irrelevant

    i can dump all the art & effects i want into a dx9 game & it will be demanding & worthy of a benchmark

    & UE3 has had dx11 for a year with those exact features that magically make something worthy of a benchmark
    Reply
  • Targon - Friday, March 23, 2012 - link

    A big part of the argument has to be how well the game engine can put things on the screen. Resolution is only a small part of the picture for what drives performance. Many games today may include support for DX11, but they are not DESIGNED around DX11. CPU performance can even play a part in how many frames per second are pumped out.

    The engine is a part of it, but it will take a while before we see an end to DX9 being the target, and that in part is because you still have far too many people running Windows XP.
    Reply
  • Sivar - Friday, March 23, 2012 - link

    I bought a GeForce FX 5900 for $590 when it was released. It was my first "top of the line" card and my last until the 580.
    If ATI's 2900XT was a loser, then nVidia's FX5900 was its retarded, sociopathic cousin.
    Reply
  • piroroadkill - Friday, March 23, 2012 - link

    Ouch. FX series cards.. not exactly NVIDIA's brightest moment.
    The 680 is a fine beast though, so celebrations are in order at NVIDIA tonight!
    Reply
  • SlyNine - Saturday, March 24, 2012 - link

    At least you dodged the 5800 ultra. But yes both the 5900 and 5800 as well as the 2900XT were bastard cards that should have never been.

    But the 7970s and 680GTX cards are not the 9700 pros or 8800GTs of their days either
    Reply
  • xytc - Friday, March 23, 2012 - link

    Kepler launch day was like one of those good old days of PC history when you knew that something good to the industry was just happened.
    Great product NVIDIA and congratulations to you all for creating an outstanding product.
    Reply
  • Urbanos - Friday, March 23, 2012 - link

    this is great news and gets me really excited to see the breakthroughs that will be possible in new pcie3 tesla cards for HPC. Reply
  • saurabhr8here - Friday, March 23, 2012 - link

    Thanks Anand for the introduction highlighting how much work a chip-design team puts in to get a product to market. Benchmarking final products and with people writing off products in the comments sections because they did not get a 'feature' they wanted, I think its important for everyone to realize how complex and time-consuming this whole process is.

    However, I would also like to point out that not just the NVIDIA work cited here but nearly all big CPU/GPU design work takes about 3-4 years to get to market. And with technology nodes beyond 28nm, this has become an even longer affair. Not considering Intel, which does its own manufacturing, R&D, early engagement with foundries, test-chip tape outs etc., are a part of the product development cycle and all of this takes a full year before the actual product design work starts.
    Reply
  • Booster - Friday, March 23, 2012 - link

    Are there any PC gamers left out there? Really? They need to catch up. There are no good computer games anymore. Gaming on a PC is ancient history.

    So what's the point in a product like Kepler/whatever?

    It's very much like the Android story. 4 core CPUs, powerful GPUs are advertised - but what are you going to do with all this hardware if there are no games, no apps that could take advantage of all this 'power'?

    Heck, 99% of PC users don't need anything other that Intel's integrated HD Graphics. Kepler-Shmepler.
    Reply
  • SlyNine - Saturday, March 24, 2012 - link

    Other then the games looking much better on PC... I'm sorry but everything you said is just wrong.

    You're only remotly close in that, there are to many ports and no PC exclusives.
    Reply

Log in

Don't have an account? Sign up now