Nvidia Bends Moore’s Law

The Pascal architecture from Nvidia will be based on multi-layer chips and stacked memory for a huge increase in bandwidth to/from the GPU.

Combined with their CPU roadmap, this technology will allow personal computing or HPC modules to triple in performance every year for the next few years. Are we there yet?

About Robert Pogson

I am a retired teacher in Canada. I taught in the subject areas where I have worked for almost forty years: maths, physics, chemistry and computers. I love hunting, fishing, picking berries and mushrooms, too.
This entry was posted in technology and tagged , , . Bookmark the permalink.

5 Responses to Nvidia Bends Moore’s Law

  1. oiaohm says:

    Why is stacked memory into dia the last optimisation. Speed of transmission can not be beatten. Why keep the memory out of the dia for so long. Simple allows more models on shelf.

    JD there is a physical limit. We are coming up on the physical limit at the moment.

  2. JD says:

    Here is an alternative perspective published a year ago:

    Nvidia GPU roadmaps aren’t an improvement with Volta

  3. JD says:

    We have been hearing this is the last optimization, etc for the last 20 years. Then magically, they always seem to come up with something else which increases performance again and so everyone has to buy new chips again.

    Companies like NVidia and Intel plan their products 5 to 10 years ahead. I’m pretty sure they have more optimization tricks up their sleeve, only to be revealed when they sense sales are slowing down.

    The sole purpose of industry—including management and labor—is to serve the consumer, and this can only be done successfully by giving the consumer a better and better product at lower and lower cost. If industry fails in this, it may have good reason to expect retaliation in the form of totalitarian government in place of our present democracy.—/. F. Lincoln, President of the Lincoln
    Electric Company.

  4. ram says:

    While I’m still in the process of benchmarking them, I can certainly say the new NVidia SLI CUDA cards truly provide seriously impressive performance under Linux (Debian Squeeze in my case). We are talking at least 40 GFLOPS for under A$400. That is a staggering “bang for the buck”. Power consumption is not too bad either.

  5. oiaohm says:

    Really this is about one of the last major performance optimisation left. CPU GPU and Ram in one dia with the shortest possible wires is really the last optimisation.

Leave a Reply