foreign [Music] I'm a senior director in the AMD Cloud business unit mainly responsible for driving go to market strategy and execution for our Cloud business today marks exactly my third anniversary at AMD and every single moment of those three years I've been really excited in bringing compute Solutions in partnership with Google Cloud to our customer base all of you guys I'll spend the next 30 to 35 minutes to walk you through all the GCE and gke solutions powered by AMD epic CPUs in Google Cloud that would leave us roughly about five to ten minutes in the end for Q a and I recommend if I suggest if you can hold on to your questions till the end come to the mic and ask your questions that way the broader audience will be able to listen to your question and we'll have a more meaningful discussion with that what are we waiting for let's get started before I actually dive into the AMD epic driven GCE and gke Solutions in Google Cloud let me give you a little bit of overview of AMD as a company AMD is not only just a server CPU provider anymore we have established ourselves in the last couple of years as true holistic data center Solutions company starting to the left if you see far left it's the AMD epic line of server CPUs targeted towards Enterprise and Commercial HPC workloads both on-prem and in the cloud with generative AI large language models inference and training being the most talked about the most sought after workloads we're very excited with the AMD Instinct line of gpus they are really the Mi 250 in production right now and Mi 300X coming out in Q4 we are again very excited to really put the solution in front of our end customers which is again targeting AIML large language models last year we acquired two companies xilinx which is the leading provider of programmable fpga Solutions and adaptive socs we acquired xilinx around the first quarter of last year and what that did for us is it increa it increased or expanded our Damned from just server data center to like markets such as Telco markets such as embedded fintech so on and so forth we also added pin Sandos which is a programmable P4 based smartnic dpu company and what that did for us is it provided us with a companion smart neck as companion chip which actually offloads some of the infrastructure workloads such as crypto storage compression decompression networking encryption decryption it uploaded the CPU and brought it into the smartnet dpu which is P4 programmable again and that leaves more cores for end user application so this basically completed our holistic data center solution kind of strategy and now we are in execution mode all right so in this session we're going to mainly Focus out of the four uh Solutions I showed in the previous slide we're going to mainly focus on the AMD epic line of server CPUs there are lots of benefits multiple benefits from the AMD epic architecture starting with uh TCO if you look at AMD epic all the way from Naples to Rome to Milan and now to Genoa it provides the highest x86 core density with the industry leading performance per core if you multiply the core density per socket with performance per core what that gives you is an extremely compelling PCO which is not only beneficial to a hyperscaler Partners like Google Cloud but a lot of that is passed on to the end customers in terms of either performance per core or in terms of dollar per VM and that if if you are optimizing for cost that gives you an excellent uh cost efficient infrastructure if you're optimizing for performance same thing a lot of Enterprises now are also optimizing I I the sustainability aspect of it their social responsibility aspect of it the Enterprises are not only optimizing or reducing their scope 1 and scope 2 emissions they want to really optimize their scope three emissions for workloads they are migrated to Cloud AMD epic CPUs with the latest tsmc 5 nanometer technology and the chiplet architecture is the most energy efficient x86 CPU that you'll find out there so that's again a big Advantage from the AMD epic architecture next is the ecosystem lots of Enterprises have been running their x86 applications for 10 years 20 years 30 years you really did not want to go touch those applications that have been bringing business to you that have been interfacing to your customer base and add risk and effort by putting that to a different architecture so AMD epic again based on the x86 architecture and available in all GEOS within Google Cloud gives you that ease of migration for your workloads that you were hosting on-prem but now you can easily do that in the cloud last but not the least confidential compute verticals and markets such as Healthcare FSI public sector one of the key hesitant hesitation in moving the workloads to cloud is will my workload be secure will my application be secure will my application my data in the cloud be secure so confidential compute in Google cloud is exclusively powered by AMD epic VMS both confidential VMS and confidential containers Within gke all right so now let's really go dive into I'm going to skip this one really dive into the different families uh compute families we have powered by AMD epic in Google Cloud one size does not fit all anymore when we started our journey and our partnership with Google Cloud we only had the general purpose VMS powered by second gen AMD epic code named Rome and since then we have gone from general purpose we have added to the portfolio of general purpose n2d VMS computer optimize c2d VMS Cloud native optimized Tau vmc 2D VMS and confidential VMS so starting with n2d VMS to your left these are really targeted towards your Enterprise lift and shift workloads such as web and app servers small to medium sized databases line of business applications like media encoding transcoding it really provides you a very easy button lift and shift from on-prem to Cloud for mainstream Enterprise workloads but as I said one size does not fit all when you look at the cloud HPC Market which basically right now uh according to Hyperion research is about six to seven billion dollars and it is gonna go at a cargo of about 17 percent between now and 2026 to about 11 to 12 billion dollars your commercial HPC applications Automotive Aerospace r d designs computational fluid dynamics drug Discovery Oil and Gas and energy exploration these kind of workloads really require compute intensive infrastructure with very Ultra fast High performing cores the highest memory bandwidth per core highest memory capacity per core and also very tight very fast networking low latency networking for your tightly coupled workloads that's where we kind of launched c2d VMS uh in 2021 in partnership with Google Cloud now right at that time we also realized that there is a need to really provide a highest price performance Cloud native uh architecture or VMS which are targeted towards scale out workloads and AMD and Google Cloud combined and co-engineered this new family of VMS called Tau VMS t2d which was based on third generation AMD epic Milan and these provide up to 42 percent better price performance and 56 percent better performance compared to any other VMS available on Google cloud or elsewhere and I kind of touched a little bit about confidential compute both entity and c2d families have confidential VM capability and feature support and also confidential gkes supported on these two VMS okay let's kind of double click into each of these VM families starting with n2d I just mentioned that n2d is really targeted towards your mainstream Enterprise workloads uh web serving think about app serving Java applications think about databases both SQL and nosql think about uh just line of business applications like media encoding transcoding uh gaming and so on and so forth so these general purpose VMS entity we started off with second gen epic with Rome in 2020 and they were refreshed with 3rd gen epic Milan in 2021 these provide these entity VMS general purpose VMS or the highest VM density or highest VM size you will find in Google cloud with 224 vcpus as the highest VM size there are also 13 percent lower cost compared to comparable n-series instances available on Google Cloud so again think about highest performing VM per core multiplied by lower 13 percent lower cost per VM really gives you that TCO advantage that I was talking about earlier and if you see the charts to the right what we did is we benchmarked entity Milan which is the third generation epic with n2d ROM and it provides about 30 percent Better Price performance for workloads such as nginx databases MySQL so on and so forth so now let's talk about couple of partners and customers who have really leveraged and benefited from migrating to entity VMS sap which is the leading Enterprise business software provider think Erp CRM so on and so forth they chose entity VMS for their as a default for their rise with sap program the main reason why they chosen to do VMS was the price performance benefit and the global availability of these VMS the other two customers I want to highlight that really benefited and adapted the entity VMS is Salesforce we all know about Salesforce here Salesforce for similar reasons to sap adapted entity VMS to host its industry leading Enterprise CRM SAS solution in Google Cloud Vimeo is a leading content and uh video provider it's also a SAS provider hosting on Google Cloud they saw 20 lower transcoding time or faster transcoding video transcoding using n2d VMS compared to the previous infrastructure they were using they also at the same time save 20 cost while hosting on entity VMS on Google Cloud all right so we talked a little bit about general purpose entity VMS let's switch over to compute optimized HPC focused c2d VMS these VMS again are targeted towards more technical workloads which are very compute intensive computational fluid dynamics for automotive and Aerospace energy exploration for oil and gas weather forecasting by by national agencies those are the workloads that are very compute intensive Network intensive because they are tightly coupled and also memory bandwidth intensive so with c2dv amps it's available it's also the highest VM size as far as compute intensive workloads or VMS are concerned in Google Cloud it provides up to 112 vcpus at the largest size and you can rent out the entire node to really host in a sole tenancy mode just like entity VMS compared to other n-series V and c2d VMS are also offered by Google Cloud at 13 lower cost compared to comparable C Series VMS it also provides 30 Better Price performance compared to general purpose entity VMS as you could see from the chart to the right but more interestingly I want to kind of quote two customers who benefited from migrating to c2d VMS air shaper is an online aerodynamic simulation platform company and they saw 50 reduction in their overall simulation costs and 30 percent faster simulation times when they migrated to c2d VMS and clutch which is a marketing and data analytics company providing the features and data analytics features at real-time analytics to end Enterprises saw similar benefits of adopting C to D on this slide let me sum up all the features that are available on n2d and c2d starting with the top left I mentioned earlier and we're going to talk a little bit about confidential VMS both entity and c2d offer confidential compute capability and comp confidential VMS and confidential container capability powered by AMD secure encrypted virtualization sap technology we're going to talk about it in just a little bit they both also support um 100 up to 100 gigabit per second networking very fast networking which provides enough networking bandwidth per core really required for HPC workloads in c2d and large size Enterprise workloads VM size and n2d both also offer local ssds n2d offers up to nine terabytes of local ssds and c2d offers up to three terabytes of local ssds for our for your really IO iops intensive workloads and finally both these are available in sole tenancy configuration so if you belong to an Enterprise which is concerned about security from neighboring VMS or multi in a multi-tenancy mode or let's say if your Enterprise which you're mainly running on-prem and you're doing lift and shift to the cloud and bringing your own license these VMS in a soul tenancy mode allows you more flexibility it allows you more direct lift and shift to Google Cloud um compared to multi-tenancy options all right so so far we talked about general purpose and to dvms and compute optimized c2d VMS let's kind of double click a little bit about these completely new family that Google and AMD co-engineered which is the Tau VMS t2dvms these are available from vcpu sizes from 2 all the way to 60 vcpus and and again as I mentioned they were co-engineered really to serve one purpose a lot of cloud native scale out workloads needed a reason to really go migrate to the cloud because obviously the cloud architecture provides that scale out capability one of the key um blockers if you if I if you will were like does it provide me the right price performance so AMD and Google co-engineered this from a performance site to provide the industry best performance so it provides up to 42 sorry 56 up to 56 percent better performance compared to any other VMS in gcp or elsewhere but to top it off and another icing on the cake Google Cloud really priced this very aggressively to give you up to 42 percent Better Price performance uh compared to any other VMS available elsewhere and you can see some of the charts to the right which shows the price performance comparisons two of the leading social media and collaboration companies Twitter now X is hosting on t2d VMS because of the price performance value prop we provide in the cloud native scale out capability because these these workloads these applications are very elastic in nature depending on how much news how much uh users are using it so they needed that kind of flexibility at a good price Performance Point which was the main reason that brought Twitter onto t2d same with Snapchat snap really adapted t2d technology or t2dvns exactly for the same reasons all right so let's recap we talked about general purpose and 2D VMS we talked about compute optimized HPC c2d VMS we talked about Cloud native scale out Tau VMS one of the features that is exclusively available only on AMD VMS in Google cloud is confidential compute what is confidential compute historically if you look at data and the journey of data from on-prem to Edge to Cloud data is moving it's in transit data is in rest at rest within the disks and nvmes but there's also data in use being processed in the CPUs for decades there have been Technologies to really encrypt and secure both data at rest and data in motion data in motion is secured by ipsack kind of at the network data at rest in the distance secured by all the AES encryption Technologies the hole we had over here was how do you secure the data in use or being processed the data and use are being processed if you think about multiple multi-tenancy VMS running on the same host you need to make sure that your data in your VM is protected and secured from Rogue in malicious VMS neighboring VMS and it's also protected from the Rogue and malicious hypervisor and also the zero trust model says that it needs to be protected by your cloud service provider and the Silicon provider like Google cloud and AMD itself it's a completely zero trust model that's where amd's secure encrypted virtualization SUV technology comes to play and Google was the first hyperscaler to adopt this SUV technology and launch confidential VMS and confidential compute containers based on this technology we can literally spend one hour going deep in depth but don't worry I'm not going to do that let me narrowed out of what SUV is what SUV is is it provides within every AMD epic chip there is a hardware there's a secure processor which provides a hardware route of trust that encrypts the data not only in the main memory but also the data in the CPU registers that is being processed it manages the hardware encryption Keys such that only the hardware knows what those keys are and your data center operator does not have access to those keys what that does is provides a zero trust model uh to secure your Federated data or your regulated data I think the key points over here without going into the details is you guys probably are not interested in the sausage making behind this but the key takeaways from this is two things one you can enable Sev and confidential VMS and containers by a push button within the Google Cloud UI there is no need to touch your application code that you have been running stably for 10 years 20 years 30 years uh to make it secure just click a button in your workload all the hardware encryption root of Russia is enabled and all taken care of by Hardware it provides confidentiality which means nobody can go and steal your data it provides and it provides secure Integrity which means nobody can go and corrupt your data and it also provides attestation at a hardware route of trust level the last and final thing that you should remember is we spend a lot of time benchmarking what is the performance impact if you turn on This SUV confidential compute feature and it is almost minimal to zero impact on your workload performance so you are getting a lot of automated security without having to give up too much on your performance or at all so there are a couple of things I want to mention here uh just what's available available today with AMD SUV technology that provides your confidentiality um confidentiality not Integrity yet is in general availability on entity and c2d VMS both confidential VMS and containers and what is in preview which was announced is SCV SNP which actually provides the memory Integrity protection what that means is nobody else a malicious hypervisor or malicious VM or a data center operator not only can they not steal your data they can also not corrupt your data so that was just announced it is in public preview and will go in GA very soon and uh SUV feature is available on n2d and c2d whereas the Sev SNP feature in public preview is enabled on n2d right now all right more importantly I want to talk about couple of customers who belong to a very regulated Federated industry AstraZeneca that we all know about is a Healthcare biosciences Company who really develops uh vaccines for many different types of diseases they store a lot of confidential client data and it is of utmost importance for them that that information is not leaked out that information is not corrupted that information is not stolen they really Leverage The confidential VMS for n2d and c2d to do that and icon is another blockchain FSI fintech company which really leveraged the confidential VMS powered by AMD CV technology on Google Cloud as well all right so so far just a quick recap talked about history talked about AMD as a company as a solution provider talked about General per person to dvms talked about compute optimize c2dvms talked about Cloud native T to dvms typed a little bit about confidential VMS this brings us to the exciting news for this week Google Cloud just announced yesterday that next gen general purpose VMS c3d VMS powered by amd's latest 4th gen epic CPUs code name Genoa so this was announced yesterday we are very excited about it before I get into c3d VMS and what it provides uh to you as a customer base let me talk a little bit about uh Genoa first Genoa is Industries highest core density highest performance per core x86 CPU it packs 96 up to 96 cores of highest performing x86 course within a single socket and that's the part that has also gone into the c3d VMS it also provides up to 9 2.9 terabytes of memory per node on top of it it provides 12 ddr5 memory channels attached to the CPU provides the highest memory bandwidth per socket and highest memory bandwidth per core this is very critical for HPC workloads and even some in-memory databases and then there are 160 PCI 5 Lanes which really drives the highest iops available on this platform for your i o intensive workloads so I talked about c3d a little bit uh this is in preview this will go into preview very soon but uh it's available uh in various shapes uh all the way up to 360 vcpus with 2.9 terabytes of um of ram as I mentioned of ddr5 and it's 4800 Mega transfers per second is the bandwidth or the speed uh and again targeted towards your mainstream general purpose workloads so the first thing we do when we get this is uh and sorry before I get there we also upped the total networking bandwidth on this compared to n2d it now offers 200 gigabit per second networking bandwidth per node compared to 100 gigabit per second that was there in n2d so if you add up highest number of course highest performance per core 200 gigabit per second networking fastest i o available iops available that translates into real world performance uplifts and we benchmarked this with the Google Cloud engineering team and we are seeing 45 better performance on c3d VMS on an average across all the critical workloads like web serving engine X in-memory caching databases like readers relational databases like MySQL and Java workloads as benchmarked by spec jbb so 45 average higher performance which is a huge step function jump generationally compared to the Milan based into dvms all right so this is my last slide it's no good if we have great technology but not available in your GEOS and regions so we work constantly and continuously with Google Cloud to really grow the footprint quarter of over quarter so with the exception of c3d which is just in preview right now all of our other VMS are available in multiple GEOS across the world wherever your Enterprises or your data center operations are and with c3d now announced in preview and going in GA soon you will see a lot more availability for c3d to a data center near you thank you
2023-12-24