....
Indeed, one of the highlights of yesterday evenings conference call with analysts is that Sus chief finanical officer, Devinder Kumar, raised the companys outlook for the year.
AMD now expects to have mid-20% revenue growth this year, up from an expectation offered back in January for double-digit percentage growth, said Kumar.
In addition, gross profit margin is now expected to be greater than 37%, said Kumar, up from the prior forecast for greater than 36%."
Storming Nvidias Fiefdom
I asked Su if she really thinks she can catch up with CUDA, the software tools that Nvidia has used to achieve and maintain dominance in machine learning and artificial intelligence.
"I think we are offering something different than what CUDA offers, Su replied.
There are lots of parts needed to make machine learning successful. We believe that we can build an alternative that will work with AMD but also with other accelerators [chips]. That area will really grow over the next five years. We do not believe that battle is over. With all the growth in that market, there is definitely room for an alternative. You have numerous [software] frameworks [for ML], you have MXnet, and Caffe, and Tensor Flow, and many others. CUDA is a proprietary technology. I would say that it used to be the case that with everything, you needed to use proprietary software, until Linux came along. Now, in machine learning, and deep learning, and AI, there is a large and growing feeling that people want alternatives.
Custom AI Chips?
I asked Su about the rise of custom chips by some of the AI leaders, be it Alphabet (GOOGL) unit Googles TPU chip, or rumors that Facebook (FB) wants to pursue its own custom parts. Is AMDs biggest competition Nvidia, or is it something else, like the TPU?
The goal is to come up with the right solution, and probably, thats some combination of CPU and GPU and also some custom solutions. I dont view custom as a bad thing. If a customer wants custom, we are happy with that. Putting aside who makes the chips, we have always believed there is merit in custom silicon. Once things reach a high volume of shipments, something like an ASIC can make sense. The thing to keep in mind is that machine learning and AI are is still very early in their development, they are still evolving.
....