Jonathan Seckler, Dell Technologies & Ariel Pisetzky, Taboola | Dell Technologies World 2023

Jonathan Seckler, Dell Technologies & Ariel Pisetzky, Taboola | Dell Technologies World 2023

Show Video

welcome to Dell Technologies world it's the Premier Technology event of the Year join John Furrier Dave vellante and Lisa Martin as they talk to the Trailblazers and trendsetters of future technology Dell Technologies World 2023 and the cube the leader in live an emerging high-tech coverage evening everyone and welcome back to thecube's coverage of day one dell Technologies World 2023 coming to you live from Las Vegas Mandalay Bay Lisa Martin Dave vellante Dave we've had a great first night so far we're just scratching the surface yeah well and having some great conversations so far a lot of news today I'm excited about this next segment because you had me at data so data and AI one of our alumni is back with us Jonathan seckler joins us senior director of product marketing at Dell Technologies and Arielle piezetsky vpit and cyber at taboola guys welcome thank you thank you yeah Aria let's go ahead and start with you give the audience an overview of tabool you're a Dell customer but tell us what you do your vision mission how are you using AI give us all that scoop okay sure thing so taboola is a Content recommendation platform so we provide approximately 4 billion web pages a day we see about 1.5 billion unique users a month and we do all of that powered on Dell servers so so we have about 12 000 Dell servers and for each and every customer for each and every user that comes to our website or comes to a publisher website we provide personalized content to do that personalized content we do it with AI so inferencing at the edge and really deep learning and machine learning at the back end to make sure that we provide the best user experience for any user out there what's what's for the name what is that signify so taboola comes from the Latin phrase tabula rasa which means Blank Slate so we we don't know who the user is we're not like any of the other big ad tech companies you don't log into our service you don't tell us anything about your affiliation we just infer everything in real time and and so did you have something to add no no no so so why didn't you just do this in the cloud why did you why are you doing this on-prem so the cloud while very trendy is also another thing very expensive to operate heavy compute loads to operate machine learning to operate AI in the cloud is just so super expensive that our TCO our total cost of operations on-prem is that much cheaper using Dell servers using our own it systems and just making it work that much faster and cheaper for the business and for the End customer so Jonathan it's not like you just oh you know that last week invented this class of servers that could serve us a customer like Ariel right right you've been at this for a while what's the what's your sort of AI server Journey been like it's obviously been accelerated with all the hype right just in terms of the awareness but from a from a product standpoint a reality standpoint take us through that yeah exactly so so we've been working with taboola for three or four or five years now at least right and so you know we've been providing the infrastructure for artificial intelligence uh you know during all of that time and you know it started out as uh you know starting out as just being a great platform uh for hosting the data for for processing the the models and and for you know delivering the inferencing but you know we've then evolved that into a set of validated designs for artificial intelligence that we do for um we do we have natural language processing we have uh uh you know General artificial intelligence uh and uh we do uh some converged I O type machine learning Ops Solutions and things like that and uh you know it's been uh really a journey to get customers is over that hump of going from a pilot or a proof of concept that like I said at a in the public Cloud to actually like putting something in production and and I want to like appreciate that almost everyone who does put something in production is going to end up like taboola in building their own single tenant infrastructure because of the cost of operations right I mean artificial intelligence is still very much a high performance Computing solution right yes did it start out like as a reference architecture then it evolved into a validated design exactly the reference architecture and then we we had design guidance and and uh testing and validation uh tools and things like that well it's interesting because we we're in conjunction with Dell tech world we did a bunch of pre-recorded interviews for ISC which is in Hamburg and it was very interesting to speak to some of the customers and who said look I'm going to do this stuff on-prem because first of all it's too expensive at some point I don't know what the crossover point is some number of thousands of cores it gets way way expensive when you go beyond experimentation and they're worried about IP leakage and they said they want to show it off all right they got this cool data center or even a super computer that they're like hey check this out you know why should the cloud guys have all the fun yeah I mean I mean you would know more than I but artificial intelligence is still very much you know unique IP no matter where you you know the stuff that you're doing no one else is doing they can't do what tabula does right so it's you know it's a it requires some investment yeah so Ariel talk to us about why Dell obviously we always thought customers have a choice you mentioned working together for quite a few years now what what is it about Dell it's Innovation it's technologies that really LED you to make that decision this is the right one to power our business so the short answer it's easy it works it's reliable and then we can go into a much longer answer of our transition from other vendors into really being a Dell shop through and through because we wanted to have that One-Stop shop for support for the management of the servers themselves for the security making sure that we're secure from the chipset to the firmware to the software that we place on the servers the ability to get the architecture help from the sales engineering team and the validated Solutions all of that helps us squeeze more performance out of every core that we bought so as you said from thousands of cores we're running over half a million cores and at that point you really really want to get as many Performance Cycles as you can out of at each core so even a one percent bump is really a lot when you're talking about that that many servers so Dell was a Natural Choice easy to work with secure reliable good support I don't think you can ask for more well energy has to be a huge concern obviously so how are you to get you now the good news is okay you've got your you guys are smart enough to know how to manage your own infrastructure and keep your costs down Etc but now you you of course inherit all these other data center issues like power and cooling how are you addressing that that's that's a great question and really thank you for that so part part of efficiency is of course not only squeezing as much as you can in terms of compute but also really bringing as much power to the I.T side in the hosting environment so not wasting that cost and Power on Cooling and sorry on Cooling and other things that you might have in the data center so with Dell we are able to control fan speeds we are able to run the servers a bit hotter on the cold side than with other solutions that we have we are able to really see a full view of the server within our Management systems and optimize the airflow optimize the power pull optimize the power usage for the different levels of the day where we have more compute coming in or less compute power coming in so the load balance software load balancer would bring in and distribute all of that into all the different servers but when the traffic goes a bit lower in the day we're able to cycle back with the Dell technology and also cycle back on the power you mentioned you can run hotter on the cold aisle or side you said right that means you could you could you're saying you can let the temperature rise just a bit yes just a little bit I mean even but even a couple of degrees is going to make a big difference right why is that Jonathan well I mean computers at the end of the day I like to think of it it is it's like you're putting electricity in you get mathematics out yes and at the end of the end and unfortunately with the mathematics you get heat right right and one thing that Dell is really good at is getting as much math out as you can but but dealing with all of the Heat and so you know we have you know for example in our latest poweredge uh servers we've we've altered the airflow in the servers to improve the airflow through the server so that we can run at higher temperatures right and when you're talking about Ai and you're talking about millions of cores and thousands of of nodes you know you've got to be able to know to get that like you said that that air uh from the cold aisle to the hot aisle as efficiently as possible to take that heat with you so you got the a plus and thermodynamics so so what is the AI stack look like can you describe that and how is that different than sort of the everyday stack to run whatever sap or or or you know just general purpose applications well I mean you should tell a talk about I'd be happy to share yes yes so when you're talking about sap as an example that's a monolithic application or usually an application that is very heavily bound to a single server let's call it when you're talking about AI you're talking about grids talking about super computers you're talking about hyper Computing it depends on the use case so we have thousands of computers or thousands of servers I should say in the different ranks all connected to each other through a non-blocking network so they can interact and you don't actually have only one CPU solving something or one system with a four-way CPU you have thousands of CPUs and tens of thousands of cores solving problems as they come in or trying to infer different uh I'd say conclusions for the users that are coming into to the system you said a half a million cores did I hear that correct yes okay that's incredible and so what's the networking like inside there what's what is it is it ethernet talking to yes so we we are strong Believers in Ethernet so we're using 25 gig networking and really that is our easiest go-to solution where we have each server hooked up to a 25 gig port and then then we have the top of rack hook up at 100 and 400 gigs is 100 Gig in your your future or is it is it is the cost Delta too much right now and because they're coming down compressing absolutely so 25 gig is our default go to today and I would guess for the next approximately two years that's where the this port is kind of going it's it's here to stay but in terms of new servers coming in for at least the next two years we're going to be on 25 gig and then we're probably going to see the 100 Gig to the server itself as the CPUs themselves become bigger hotter and able to also absorb more processors there will be a crossover I presume it's like today you wouldn't do a 10 because you got to go to 25 it's the same price as good for free so Jonathan can you comment obviously what Ariel has described here is a huge deployment tremendous power going on can you comment on the learnings that Dell has gotten from this customer relationship and as AI is evolving how are they helping to evolve Dell in that respect that's a great question so with tabula we've like I said we've been working with with tabula uh on their deployments uh in in Israel but also all across the world and it has really um spurred the need to really understand you know from a customer standpoint or a support standpoint being able to deliver that Global support no matter where tabula is you know not just in terms of of breaking and fix but but that AI expertise that that uh configuration expertise Etc and you know we're taking that and then we're building that back into the portfolio even as we talked today so uh you know the the sexy cool AI technology out there now is this thing called generative AI like chat GPT and I I will say that while that is really an exciting being technology to see demonstrated the application is going to be more like what taboola is doing where you build it a unique model and a unique set of uh of tools to help your Enterprise solve those problems you know underneath the cover so that's where the real value is going to be in the future I saw I think it was a Twitter maybe it was LinkedIn somebody posted have you figured out your AI strategy yet well before you figure out your AI strategy you better figure out your data strategy right do you buy that and and what is your data strategy yeah does it care about data at all when you think about about AI obviously you need to feed it it's like a monster that just demands more and more so the data must be pushed into the algorithms so or on the training side or eventually when you want to get answers from them so for us on the on the training side we do nlu which is natural language understanding where we are NLP natural language processing where we would like to understand what an article is about so we will ingest a whole lot of Articles a whole lot of publish like publisher web pages online bring them into our systems understand what they are and categorize them so we can provide the best content recommendation for that moment so we'll do that on gpus but just think of the sheer amount of data when we serve four billion web pages a day how many web pages we need to understand ingest and have the computer kind of understand the language and what they are about not to speak of at least 30 different languages that we support today and this is all with our internal AI so when you first saw chat GPT were you were you like yeah I got this what was your reaction take us inside sort of uh uh it's a it's a it's a term of endearment an alpha geek's brain when you you first saw that so Chachi PT is a revolution because really if you think of what my kids your kids will be talking about in in 20 30 years ago oh you had computers wasn't that like a typewriter you had to actually type on them because chat GPD is changing the way we will interact like emails will suddenly my English will suddenly be that much more polished my my spelling would be that much better just because of these uh Solutions out there and who knows what's coming 10 years down the road so it's totally a revolution and it revolutionizes the ability of it teams to work better so if I'm taking it just back into our Tech world for a second for anyone that is coding anyone that is writing scripts the productivity boost there is astounding so we for us we run those 12 000 servers with about 15 sres cyber liability engineers and now I Envision 20 and 30 000 servers with the same amount of people I don't need to grow anymore because chat GPT is like another friend sitting here helping me whispering in my ear how to code better how to write better and how to be better I think that's the key to all of artificial intelligence you know we've been looking at chat GPT as this cool sexy thing but what you don't realize is that it's it's a it really is a public demonstration of what can be done but the real value is going to be boosting and productivity of employees in all kinds of ways both predicted right you know like I.T coding those are those are those are those are going to be like low-hanging fruit right uh but then you know who knows what else is coming right Michael Dell was interviewed uh a month or two ago and he said something around you know when you when you when the cognitive power goes to zero think of the things that you can do and and that that really resonated with me that this idea that you know in every job that you're in no matter what you do you spend a lot of that time just trying to find the information you need to make a decision or you're trying to um you know make sure you've got all your ducks in a row before you pull the trigger or something like that and and Technology like generative AI inside the Enterprise is going to do all that kind of grunt work for you and it's going to make people I think it's going to make them more productive it's going to allow them to be more creative and so you're you know it's going to create all kinds of unpredictable consequences some of them might not be wonderful and I know everyone's worried about that but I think that at the end of the day it is a it's a game changer for the for the for the for the economy and presumably like other mind-blowing Innovations the graphical user interface when you first saw that you're like wow the web browser oh my gosh and then you look back and you're like wow that was horrible I can't imagine what this is going to bring exactly what it's going to bring what's your what is your outlook so I think it will really improve us it will help us be more productive find things faster the ability to to really ingest a whole lot more information make it more user friendly just think of help pages today sometimes when you're looking for for a technical solution and you find this this help page and it's hard to read but if you search it on Bard on chat GPT and suddenly you get something that is user friendly that is for us as users how will it change publishing is a really interesting question how will it how will it change search because suddenly instead of going into just a normal search bar normal for the last 15 years where you get all these answers and links to the publishing World suddenly you're getting the answer and who's going to pay for that because that has scanned content that someone actually worked hard to create so who's going to pay for that what's going to happen there there's a whole lot of questions to be asked you know even search when you think about it you had the way you prompted search in the early days you had to really think about this plus that and now it's all ads so Ariel take us out with obviously phenomenal use case that tabula has with Dell with AI what's next well what's next I would guess that a whole lot more compute in terms of the ability to see more compute units within a single rack I'm I'm looking at if today we're looking at a 15 and 20 kilowatt racks we're going to see much hotter racks with much better cooling so we will be more environmentally friendly the whole IT industry needs to be more environmentally friendly and Dell is really leaning that way and helping us as tabula see the the future and really being carbon neutral so a better so it will be a better place for the planet and we will be able to be more user friendly and provide more I'd say intuitive Services much easier awesome guys what a great use case thank you so much for joining Dave and me on the program today talking about tabula and Dell what you're doing what you're enabling your customers to do and the The Horizon seems Limitless talking to the two of you we appreciate your your enlightening enlightening comments thank you so much thank you thank you for my pleasure for our guests and for Dave vellante and this is Lisa Martin signing off from day one of our coverage of Dell technology tomorrow right we get Chuck Whitten in the Keynotes I think right tomorrow morning yes we do and then Michael's coming on Chuck's coming on we got a big big big day we do and while the wall tomorrow and the next day we hope you have a great night we'll see you tomorrow foreign foreign foreign

2023-05-28 04:37

Show Video

Other news