By advantage of its relentless pursuit of ever quicker, ever extra highly effective GPUs, Jen-Hsun Huang has claimed that Nvidia, over the previous 20 years, has pushed the “value of computing down by a million occasions”.
When you have a look at the rising prices of contemporary graphics playing cards in contrast with their forebears, that is possibly onerous to fathom. It certain appears to be like like the price of a GPU has simply been steadily rising to most of us once we have a look at the objects of our silicon wishes. But while you have a look at simply what the graphics chips of immediately are able to, the extent of uncooked computational energy on the disposal of even a lowly RTX 4060 would have appeared borderline legendary 20 years again.
A GeForce 6800 Ultra from 2005 delivered a whopping 6.4 GFLOPS, whereas the underside of the Ada Lovelace technology comes with 15,100 GFLOPS of processing grunt. That’s a complete world of distinction from a $499 card of 20 years in the past versus a $299 GPU of immediately.
And that is not even a card wherever close to the highest of the stack, nor near what you will get from Nvidia’s strongest enterprise GPUs.
Jen-Hsun, at immediately’s morning-after-keynote Q&A session, in contrast what Nvidia has accomplished in growing extra highly effective graphics silicon, pushing down the relative value of GPU computational energy, to the affect of Moore’s Law.
“The cause Moore’s Law was so essential within the historical past of the chip is that it drove down computing prices,” Huang remarks. “In the course of the final 20 years we have pushed the marginal value of computing down by a million occasions.
“So a lot that machine studying turned logical: ‘simply have the pc go determine it out.'”
Basically, there’s a lot computational energy accessible for such comparatively little money that you just would possibly as effectively simply begin throwing it at AI to resolve all our issues. Or, you realize, draw us an image of a gold fish while you completely, positively simply have to have a freshly generated image of a fish.
There’s no getting away from it, the graphics card, and its constituent part, the GPU, have change into crucial items of silicon in our fashionable time. There’s additionally no getting away from the truth that Nvidia is liable for among the most essential silicon of our time, nonetheless you’re feeling in regards to the rise and rise of synthetic intelligence and its potential affect on the world and humanity.
Does Jen-Hsun’s maths add up? I do not know, he did not present his workings, however what’s true is that for the reason that delivery of the GPU as we all know it, the cost-to-performance ratio has solely been going in a single path.
