This week I was at the AMD Epyc processor launch in San Francisco. I’ve been at a lot of AMD events, and up until this one, the general message was that AMD was almost as good as Intel but not as expensive. This year it is very different; Intel has stumbled badly, and AMD is moving to take the leadership role in the data center, so its message isn’t that it is nearly as good but cheaper anymore; it is that it has better customer focus, better security and better performance. Intel’s slip really was around trust, and as Intel seemed to abandon the processor segment, OEMs and customers lost faith, and AMD is capitalizing on that slip.
Lisa Su, AMD’s CEO, did the keynote. She has been in the news of late, because she has been building her team up with well-connected top tier players ironically addressing the very problem that took Intel to its knees: an inadequate succession plan in the face of an unplanned CEO departure. While this has led to speculation that she may be leaving--certainly she’d be highly valued wherever she went--but she has indicated this is what she learned to do at IBM: ensure succession. It is this unwavering focus on doing what is right for the company and its customers that has allowed AMD to overshadow Intel and why the firm has been so successful of late, particularly in the cloud.
Go here to see eWEEK’s listing for Top Linux Server Vendors.
In this world where the abuse of power and misacting executives seems to define the space, it frankly is refreshing to see someone like Su just buckle down and do a great job.
AMD has always been relatively conservative, but Lisa stated that the company has broken 80 performance records and that this new processor is the highest-performing one in the segment. This is one thing Lisa’s IBM training helps validate; I went through that training myself and, at IBM, you aren’t allowed to make false claims. AMD isn’t making a false claim here.
The new Epyc 2 is 64 cores and 128 threads and with PCIe generation 4, it has 128 lanes on top its 7nm technology, which currently also appears to lead the market. Over the years the average performance for the data center chips, according to Su, has improved around 15% per year. The last generation of Epyc exceeded this when it launched, but just slightly. This new generation blows the curve out; instead of 15% year-over-year improvement, it is closer to 100%.
This isn’t just a theoretical performance increase. Using a running comparison between AMD’s fastest processor, the 7743, with Intel’s highest performer, 8780L, the AMD processor is showing 97% higher performance. For HPC testing, the performance drops a bit competitively to 88%, but it is still a nearly overwhelming win using industry benchmarks. Looking at diverse enterprise loads across a variety of platforms, AMD drops competitively again, but only to an 84% improvement over Intel’s current best. Demonstrations here at the event showcase between a 1.5X to 1.9X advantage, using a broad range of IT-focused applications.
Joining Su on stage was Mark Potter CTO from HPE. Third-party validation is critical to an event like this, and having a vendor like HPE back up AMD’s effort was important. HPE had, at launch, three systems immediately available that use this new processor, and they plan to have 12 systems within 12 months. These new HPE systems, according to Potter, have broken 37 world performance records. Loads include data analytics, enterprise virtualization, database virtualization, decision support, big data analytics, server-side Java, and SAP sales and distribution.
Potter attested to the fact that some of the most powerful supercomputers coming to market will use this processor, because it is higher performing. He contended that these new AMD-based systems are the most secure in the industry as well as the highest-performing. HPE Infosight loaded on these systems creates a self-managed, self-healing and self-optimized result. One interesting aspect of these servers from HPE is that they are wrapped with an app called Green Lake, which allows for the price of the servers to vary by use.
Jen Fraser, a senior director from Twitter, came on stage next and spoke to her company’s critical performance needs. One of their biggest problems is keeping energy use down, and the efficiency of this new second-generation Epyc processor is what attracted them to the product. Using this part, they were able to put 40% more cores into their data centers, using the same energy as what these parts replaced. Twitter plans to swap to Epyc 2 by year’s end.
AMD’s Mark Papermaster replaced Su on stage, and, as CTO, he was arguably one of the most influential people behind this change. He took some really big bets by pushing the company to 7nm technology at a time when Intel was struggling with 10nm. He rearchitected the cores to provide better performance and move further away from the Intel standard. In addition, he drove the infinity architecture, which improved performance, security and configurability.
Moving to 7nm was a massive investment by AMD. This resulted in faster, smaller lower-powered transistors. This change also gave AMD 2x density and the ability to provide the same performance as Intel--but using half the energy. This speaks to why Twitter’s senior engineer was so excited about this technology. Papermaster went through a long list of unique improvements to this part, fueling this big performance gain. AMD massively increased the memory density, I/O capacity and memory bandwidth, which all help anchor the impressive performance gains AMD is showcasing.
Papermaster bought up Robert Hormuth, Dell VP and CTO of Server Infrastructure and Systems, who spoke to Dell’s use of these new parts. Hormuth discussed how Dell PowerEdge portfolio was making use of this technology and agreed that it was allowing the company to push the envelope on performance and security. Dell’s use of Epyc 1 was very successful and allowed the company to bring to market well-priced servers that Dell customers were very excited about. This year Dell plans to have a new line of Epyc 2 servers that may be even more impressive.
Intel has had a number of dire security problems that it didn’t disclose in timely fashion, making their largest customers very nervous. AMD is going after this vulnerability aggressively and pointing to how they’ve uniquely hardened Epyc 2 so that customers that use it have few, if any, of the concerns they’ve had surrounding Intel parts. Part of this is jumping to more than 500 unique encryption keys tied to the platform.
Krish Prasad, SVP and GM for VMware’s cloud business unit, was brought on stage, and he was very impressed with this new version of Epyc. He supported the argument that Epyc 2 represents what is now the most scalable secure infrastructure. VMware has been collaborating with AMD for some time, and it is that collaboration which has resulted in this powerful third-party advocacy. He represented that through this partnership, together, the firms have created something that has massive performance and security advantages. It is interesting to note that there was a subtext in this part of the presentation that indicated that AMD and VMware have been working unusually closely. One of AMD’s lasting advantages is that it doesn’t dictate what its customers want; they actually ask the question, and I get the sense that VMware’s answers had a great deal to do with how this part was architected.
Norrod is one of AMD’s most experienced senior managers. He opened by reiterating that this new Epyc part is the most powerful in the market with an astounding 32 billion transistors. This is an optimized no-compromise single-socket system, and AMD’s single-socket solution often outperforms Intel’s dual-socket solution, which suggests AMD’s two-socket solution must be particularly impressive. Supporting this is the half-gigabyte of cache on the part. Norrod again pointed out that the highest-performance Epyc processors put out twice the performance as Intel’s current highest- performance Cascade Lake parts. (This explains why AMD had so many third-parties on stage, because this kind of an advantage from a firm that has historically lagged Intel at the top end wouldn’t be believable otherwise). One example was of a retailer who had 60 Intel Xeon 8280 servers, and the retailer switching to AMD dropped the number of servers by 27 (to 33), providing the same performance with far lower infrastructure, energy and support costs.
Validated performance advantages ranged from engineering solutions that have a 52% performance advantage to fluid dynamics, which showcase a 95% performance improvement.
Peter Ungaro, Cray’s CEO, came on stage and spoke to why he is excited about this new Epyc part. Cray is arguably the leader is supercomputers, and having its blessing for this part is particularly powerful. They introduced Epyc to their line around a year ago, and they have around $1 billion in sales tied to that platform. The coming Epyc 2 Frontier supercomputer, according to Ungaro, will outperform the current top 100 super computers collectively that are currently in production. He said that one of the F1 teams was able to showcase a 45% increase in performance with the last generation of Epyc.
The U.S. Air Force and Indiana University will both be early adapters of this new supercomputer technology. Indiana University indicated that it believes this technology will significantly improve the research-based advancements coming out and that its new Epyc 2-based supercomputer will be able to do in one second what prior technology took 28 days to accomplish. This will help the university continue to bring in the substantial amount of grant-based income ($750 million) that it currently receives.
Single Socket Advantage
This is fascinating, because it points to a problem with dominance. Norrod implied that Intel’s signal socket server has historically been crippled to favor two-socket configurations. AMD hasn’t done this, so its single-socket servers have massive performance advantages as a result. He represented that the AMD single-socket solution will outperform Intel’s entire line of dual-socket solutions. This indicates a 54% reduction in TCO over Intel for virtualized environments, he said.
Doug Fisher, COO of Lenovo’s Data Center Group, also validated these claims. Fisher pointed out that 90% of the data in the world is only 2 years old, showcasing how incredibly rapidly big data is growing. This is doing ugly things for operational expenses, and Lenovo is using Epyc 2 to provide customers with a path to significantly reduce the related rapidly growing costs. Fisher said that with Epyc 2, Lenovo has been able to set 16 new world records on its related servers. He indicated that currently Lenovo, with this technology, has the most efficient and most reliable server in the market. Fisher also claimed Lenovo has the most secure server in the market, overlaying AMD’s security technology with Think Shield, which is unique to Lenovo. One interesting customer is Pivot 3, which apparently has a video security solution tied to this platform that has to be seen to be believed (it is in the demo room and it is impressive). One of Lenovo’s other customers, Home Secure, added to this, stating that the Lenovo/AMD solution outperformed anything else in the market. Lenovo has also wrapped this platform with a pay-as-you-go pricing solution.
Azure and Cloud Computing
Cloud Computing was one of the major targets for this new Epyc part. Girish Bablani CVP from Microsoft’s Azure effort then came on stage and indicated that Epyc 2’s performance and workload potential was impressive with a 54x improvements over what they previously had deployed. He argued that the AMD solutions were the leading solutions on the Azure Cloud. Rescale, an aerospace company, in support of Aerion’s AS2 biofuel executive jet development has been using this technology heavily and they are apparently very excited about this massive performance increase. Microsoft is announcing HPV2, their next generation platform, based on Epyc 2. AMD squared is their new Virtual Windows offering also using this new Epyc part. Da_V3 and Ea_V3 are additional Epyc 2 based systems that will shortly be available on this part and likely represents one of the best places to test this technology on customer workloads.
Norrod discussed line pricing, and at every price level the AMD part nearly doubles the performance for similar or less cost. At the high end, AMD has twice the performance at around half the price per processor. Intel had a market lock at the high end, and this apparently has resulted in an excessively high price from that company. The confident Norrod said Epyc 2 “kicked Intel’s ass.” He argued that Epyc, with Epyc 2, is the new standard for the data center. This would be unbelievable if it weren’t for the massive third-party validation on stage.
Su, at the end, came on stage and brought Bart Sano, VP of Engineering from Google, on stage, who said he was feeling “epic” when he came on stage. He also is clearly excited about this chip. Google has some of the largest data centers around the world, and it is focused like a laser on TCO. They engineer their own systems and were so impressed with the AMD Epyc 2 part that they are moving aggressively to it. Google wants and needs the best of the best, and Sano clearly indicated that Epyc 2 is the best of the best. (As a side comment, IBM is typically my gold standard in terms of being able to get third-party advocates. I think AMD even topped them at this event.)
Google was an early adapter of this technology, and it has already deployed it. Their engineers particularly like the generation 4 PCIe, the high core count and the advanced memory fabric and capacity. Google plans to significantly expand its AMD Epyc commitment going forward. Google will be offering with the Epyc 2 the largest VMs it has ever offered.
Wrapping Up: Holy Crap
AMD came to play in San Francisco this week. I’ve never seen it go after Intel this aggressively and, to be frank, this would have failed had it not been for the massive third-party advocacy behind Epyc 2. I’ve been in this business since the mid-‘80s, and I’ve never seen this level of advocacy for a new processor ever before. And it was critical that AMD set this new bar; I guess this was an extra record they set, but AMD can legitimately argue that it is the new market leader, at least in terms of both raw and price performance, in the HPC in the server segment.
I think this also showcases how badly Intel is bleeding support after abandoning the IDF (Intel Developer Forum) conference. In any case, at least this week, AMD is the company to beat and its staff worked their asses off to get here. Sometimes just putting your head down, listening to customers and focusing is all you need to take a market.
Rob Enderle is a principal at Enderle Group. He is a nationally recognized analyst and a longtime contributor to QuinStreet publications and Pund-IT.