Gartner says that the edge will eat the cloud...we've been saying If software is eating the world - it's definitely going to eat the edge. Not sure I understand "edge eats cloud"
HP proved w/ 3PAR that it actually can do acquisitions - if it can repeat that w/ Nimble it will be another home run. And Aruba looks very promising as an acquisition
But no question HPE is financially much stronger - $5.8B in cash and an ability to do acquisitions such as Aruba, Nimble, Simplivity and other tuck ins - smaller co's that can scale
It's wasn't all pretty - some struggles with the team, the organization, mis-steps on the HP Public cloud, assessed a strategic merger w/ EMC...all leading up to a massive breakup of the company - which was the ultimate chosen path
It's been a 6-year journey for Meg Whitman - imo she took one for silicon valley when she took this job - definitely in better shape than when she inherited a mess from the prior regime
We are reaching a point where Moore's Law breaks down. We cannot increase the density of the compute chips any more. But we are constantly faced with more & more data. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
As we gain larger & larger models, we need more compute. Since we can't put more cores on the chip, the only way to increase the compute is to scale out. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
So that is a problem for those of us on the software side who want to speed up the computation of these larger models. That is where Cat can help us because she is working on new hardware. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
In our lab we think about new ways to do the computations & focus on the computations that really matter. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
We're leveraging novel devices. You've heard of #Memristor. Instead of using memristor for non-volatile memory for data-driven computing systems, we're using these devices for doing computation in the analog domain. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
So one of our first core computations we're going after is matrix multiplication. That is a fundamental mathematical building-block for a lot of machine learning, deep learning, signal processing, you name it. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
Of U remember Ur linear algebra in college, a dot product is exactly a matrix multiplication. It's a dot between the vector and the matrix. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
For me as a software person the end of Moore's Law is a bad thing because I can't increase compute speed. For Cat it is a good thing, because it forces her to think creatively about new devices. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
General computing is not always a good thing. Sometimes if you want to speed up a computation, you have to develop a device designed for that type of computation. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
In machine learning, those multi-vector computations are at the core. You spend 90% of the time doing exacdtly that computation. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
So if Cat can come up with a more effective way of doing those matrix computations, it would really help us. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
So some of our people took the dot product engine as the core and then thought about designing a system specifically for doing convolutional neural networks. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
We're seeing 10X to 100X speed improvements over GPUs & even a 15X improvement over a state-of-the-art, specialized, digital ASIC. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
So compared to even the beset e can do today we're seeing a potential for a huge amount of speed up & energy savings as well. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
There's a lot of interest in automotive industry, space, robotics, for more low power, very high performance, very efficient computation. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
When I was studying computing I saw 3 ways to improve performance: improve the technology, which is what Moore's Law measures; change the architecture, which is what RISC did and what GPUs, etc., do; and change the organization of components. @PLBurris#theCUBEhttp://bit.ly/2A
I presume that what you are talking about in part at least is changing that organization. And making it so the developer does not have to know everything about that organization. Is that getting at what you are talking about? @PLBurris#theCUBEhttp://bit.ly/2A5dhS...
Yes, you have it right. talking about some of the architectural challenges of today's processors, not only can't we increase power of a single device, but if we did the challenge would become how do you bring the data to the device fast enough. Natalia Vassilieva @HPE#theCUBE
What the dot computation engine does is computations in memory, inside. So U limit the # of data transfers between chips so U don't face the issue of feeding the chips. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
We are thinking about it holistically. In Labs e have software working with architects. It's not just a clock speed issue. It's thinking about what computations matter. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
One of the great things with the dot product engine & these new computation accelerators is with something like the memory-driven computing architecture we have an ecosystem that favors accelerators & these specialized hardware pieces. Cat Graves @HPE#theCUBE http://bit.l
In memory-driven computing, if all my data fits in the shared pool of memory, & I have different heterogeneous compute devices accessing thast data, its up to the system management software to allocate a specific tasks to the device that does that most efficiently. Cat Graves @
In memory-driven computing, if all my data fits in the shared pool of memory, & I have different heterogeneous compute devices accessing that data, it's up to the system management software to allocate each specific tasks to the device that does that most efficiently. Natalia V
Because the only want to increase speed is to scale out, I have to take the locality of the data into account in my software. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
If the data is in local memory it will take maybe 100 milliseconds to access it. If the data is in another socket, it will take longer. So to avoid having my software wait for the data, I need to schedule that very carefully. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dh
In memory-driven computing, where the assumption is not only that all the data is in a single pool but that it's also evenly accessible to every compute device, I don't have to care about thast any ore. You can't imagine what a relief that is. Natalia Vassilieva @HPE#theCUBE
When a new GPU comes out, they have to take months to redesign their algorithm to tune it to that specific hardware. That's with generations of the same product from the same company. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
I was always interested in a lot of things. I wanted to study & do everything. I got to a point in college where physics still fascinated me & I felt like I didn't know nearly enough. So I decided to pursue my Ph.D. in that. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
I really enjoyed research, but I wanted to do that on things that might have more of an impact in my lifetime. My Ph.D. work was in something that might be implemented in a couple of hundred years. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
There aren't many places in my field of hardware where you can do cutting edge research & see your work implemented. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
As a kid I always liked math puzzles. It became obvious that I liked solving the math much more than I liked writing about anything. In middle school I saw my first class in programming. I was right into that. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
My teacher suggested that I go to a specialized school. That led me to the Physics & Mathematics Lyceum, & then to the math department at the university. It was straightforward for me. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
I think the reason more women don't enter tech starts very early. Both my parents are scientists, so we always had books around the house. I was always encouraged to pursue that path and be curious. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
Various academic institutions have done studies that show that the issue is surmountable. Carnegie Mellon has a nice program for this where the percentage of women in their Computer Sci program went from 10% to 40% in 5 years. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
They did things like peer-to-peer mentoring, pairing freshmen with seniors, so you don't feel like you're the only one interested in what you're doing. Cat Graves @HPE#theCUBEhttp://bit.ly/2A5dhS...
For me it's not the percentage (of women in tech) that matters. It's don't get in the way of people who are interested in that area & give equal opportunities to everybody. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
I do think the industry provides equal opportunity. I came from St. Petersburg, Russia, & I believe that the ex-Soviet Union countries have a much better history in equal opportunity for women in technical fields. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
In the Soviet Union we didn't have men and women, we had comrades. After WW 2 women took all kinds of hard jobs. We got moms at work. All the moms of all my peers were working. My mom & dad were both engineers. Natalia Vassilieva @HPE#theCUBEhttp://bit.ly/2A5dhS...
#HPEDiscover
