Intel’s New Xeon Scalable Processors Are Its Broadest Datacenter And Carrier Play Yet

By Patrick Moorhead - July 11, 2017
Intel is a giant in the in the server chip space commanding 99% market share, and in compute for the data plane of datacenter storage and networking, 50 to 75% share. This may lead some to believe the company’s datacenter growth prospects are dim. Not so fast. First off, cloud, HPC and AI are growing markets, while enterprise is flat to declining. Intel has already geared up to increase the datacenter market basket size with SSD storage, network and AI accelerators, and software stacks. Consider Intel has a very low share of the networking and storage control plane, platforms like base stations, SSD, and newer, specialized compute needs. To put a number on it, Intel estimates it only has 36% share of a $45B growing to $65B TAM.  Of course, these growth opportunities need to be measured against new competitive threats in the general-purpose compute space from Advanced Micro Devices and specialized AI accelerators from NVIDIA. Intel has announced some of these new growth products and initiatives, but not everything and all came together today with the launch event I attended in Brooklyn of their new datacenter Xeon Scalable processors, code-named “Purley”. Xeon goes broader but simultaneously gets more workload-targeted First and foremost, Intel’s launch today signified an expansion and broadening of Intel’s datacenter efforts and puts an exclamation point on how it’s targeting workloads for enterprises, carriers, HPC, cloud and AI workloads. The industry has been using the term “workload-based computing” for 25 years, but the big difference here is that this isn’t just talk, it’s reality, and it’s happening at the silicon level into a comprehensive platform. Intel isn’t just throwing a CPU at it and calling it a day- they are, by targeted workload, choosing the right general-purpose CPU, chipset, accelerators, SSD storage and specialty software stacks to attack the problem. These targeted workloads include many of the markets where Intel can still find growth and not compete with themselves, while defending their turf. One good example is in networking where by using Xeon, a chipset (PCH) with QuickAssist, and also the DPDK software stack, Intel says they can deliver up to 2.5X better performance than the prior generation. HPC is another one where, with Xeon AVX-512 as well as integrated Intel Omni-Path Architecture, Intel says they can provide up to a 2X FLOPs/clock improvement. Storage is another example where with Xeons with its Optane SSDs and Storage Performance Development Kit (SPDK), Intel says they can achieve up to 5X more IOPS and reduce latency by up to 70 percent compared to out-of-the-box NVMe SSDs. Intel isn't confused on AI either and I believe they know where the heat is going to be, what they have to do to capitalize, and now it's up to them to execute, particularly in training. Intel made a compelling case today with their new Xeons in heterogeneous machine learning inference workloads, but I need to investigate this further.
This isn’t a paper launch or a promise for the future, it’s here right now. Today’s announcement marks the general availability of the Xeon Scalable processors with more than 500,000 Intel Xeon Scalable processors already sold as part of Intel’s early ship program to big cloud players like Google Cloud Platform  and AWS and carriers like AT&T and three entries into the Top 500 Supercomputer list.
Big-time platform performance boost for targeted workloads
With Xeon Scalable Processor, Intel is claiming a performance increase of “1.65X on average” over the previous generation of Xeon processors with a combination of chip improvements, platform enhancements, accelerators and software stacks. It’s important to not get stuck on one chip here as Intel is bringing a full package to the table. Also, as Intel supports multiple environments at varying scale, you need a scalable architecture supporting one to eight sockets with varying levels of memory, IO and acceleration. One size doesn’t fit all in the Intel world, it’s more targeted.
The new Intel Xeon Scalable processor features a new core architecture, new memory controller and on-die interconnects. These new Xeons can support up to 28 cores and 56 threads per socket with support for memory frequencies up to a new 2666 MHz. The microarchitecture itself is being upgraded from Broadwell to Skylake, bringing a whole host of improved features like support for AVX-512, new cache management instructions, improved time stamp counter virtualization and a slew of new security features. Intel also highlighted the real-world benefits due to its long-term engagements with ISV partners on software optimizations.
When it comes to improving overall performance, Intel is bringing AVX-512 as well as a new Intel Mesh Architecture for reduced system latency and flexibility. I was expecting AVX-512, but I wasn’t expecting the new mesh architecture, a feature that is more about scalability and future-proofing than anything else. The company is also utilizing Intel QuickAssist Technology (QAT) interestingly enough in the chipset to accelerate hardware cryptography as well as data compression. Intel is also integrating their high-speed OmniPath Architecture fabric into the processor designed to improve the cost of deploying HPC clusters and improving density. Intel is also introducing a new Ultra Path Interconnect (UPI) that replaces QPI (QuickPath Interconnect) which increases the data rate from 9.6 GT/s (Gigatransfers per second) to 10.4 GT/s or about 10% and allows for up to 3 UPI in a 2 socket configuration. QPI and UPI are primarily used for inter-socket data transfers in multi-socketed systems, which is important when you look at how many multi-socketed systems are out there. Intel claims a data efficiency increase of 4% to 21% going from QPI to UPI and a reduction of idle power of up to 50%. This technology also enables a new level of scalability with support for up to 6 terabytes of memory in a 4-socket system and scale to support 2-socket through 8-socket systems and beyond. This is some serious scalability. When it comes to storage, Intel is introducing an integrated VMD (Volume Management Device) and VROC (Virtual RAID on CPU) inside of some of their Purley processors. This helps to accelerate the performance of RAID arrays using NVMe SSDs and to also reduce the complexity of the build out which includes reducing the amount of overall power consumption and the need for backup batteries. Intel claims an improvement of up to 15W per RAID volume with this new VMD and VROC storage solution built into their Purley processors. This allows Intel to build a 1 Petabyte storage system in a single 1U rack, which is far denser than what it used to take only a few years ago. I'm looking for independent, third party benchmarks and workload comparisons on Advanced Micro Devices EPYC, particularly single socket configurations. Security
Intel’s new Xeon Scalable Processors also come with improved security. Intel says the new Xeons deliver a 2X performance improvement in security (encryption, data protection, cryptographic hashing) performance compared to the last generation resulting in a performance impact of less than 1% for common workloads. These enhancements leverage AVX-512 and Gen 2 QuickAssist. Intel has also introduced their Key Protection Technology which is designed to guard against security key attacks, a common attack vector in enterprise data breaches. This adds onto Intel’s current security technologies like TXT-OTA, Boot Guard and PTT which will assist with business continuity.
New tiering, naming and solution packaging
With the Xeon Scalable processor brand announcement earlier in the year, Intel introduced a new product tiering system that utilizes a metallic Bronze, Silver, Gold and Platinum scale. Every product needs a tiering mechanism to segment the market and have logical upsells determined by some value metric or combination of a few. This is true for potato chips and datacenter chips.
Bronze is Intel’s ‘entry’ Xeon Scalable processor for ‘light tasks’ and is price sensitive for these ‘light workloads’. The Xeon Silver Scalable processor is considered their ‘Good’ processor for scalable performance at low power and ‘moderate tasks’ and counts among its features Intel Turbo Boost and Hyperthreading technologies. The Xeon Gold Scalable processors are up to 22 core counts and have support for both 2P and 4P socket support as well as 3 UPI uplinks. Finally, is Intel’s highest end Xeon Scalable processor, the Platinum tier deemed the ‘Best’. The Platinum Xeon Scalable processors feature up to 28 cores and up to 8 socket support including 2 and 4 socket as well as DDR4 2666 MHz support and the highest accelerator throughput.
While I’m not a huge fan of the “metals” analogy and prefer numerical families, out of the other side of my mouth, I do like the segmentation as it delineates between different levels of different kinds of performance characteristics. Metals will enable timeless introductions for years to come and by using metals, no one will be confused that this is a small launch. It’s a biggie. With Intel Select Solutions, the company is even building “near”-engineered solutions to the table with Select Solutions, an indication that it is moving up the food chain and enabling a much larger market-basket. I’m not surprised by this move as Intel has historically moved into the OEM and ODM territory when it made sense for them. Intel has historically done solutions testing and optimizations, and now with the work and bigger market basket opportunity for targeted workloads, why not package it all up and put huge marketing dollars behind it? Select Solutions is a solutions brand that is based on Intel’s Builders ecosystem collaborations designed to simplify and speed up the choice and deployment of datacenter and network infrastructure. I like to think of Select Solutions as the “Centrino” for the datacenter. Select Solutions is starting with solutions delivered on Canonical Ubuntu, Microsoft SQL 16 and VMWare vSAN 6.6. I am expecting many, many more. Companies like Ericsson, Hewlett Packard Enterprise, Huawei, Quanta, Lenovo, Sugon and Supermicro are already onboard to deliver these infrastructure Select Solutions starting today to drive enterprise refresh. Wrapping up There is a lot to absorb with Intel’s new Xeon Scalable Processors- in fact we’re only scratching the surface with all the improvements that Intel has made, particularly the many more workload specific improvements. The new Xeons represent growth and opportunity for Intel, and the company is doing it doing it by more finely targeting workloads and addressing it by offering a full platform of CPUs, chipsets, SSD storage, accelerators and software stacks, then tying a ribbon around it with the new Select Solutions. I see growth potential in spaces it hasn’t historically achieved high degrees of market share while defending their CPU turf at the same time. All in all, a very solid launch.
+ posts

Patrick founded the firm based on his real-world world technology experiences with the understanding of what he wasn’t getting from analysts and consultants. Ten years later, Patrick is ranked #1 among technology industry analysts in terms of “power” (ARInsights)  in “press citations” (Apollo Research). Moorhead is a contributor at Forbes and frequently appears on CNBC. He is a broad-based analyst covering a wide variety of topics including the cloud, enterprise SaaS, collaboration, client computing, and semiconductors. He has 30 years of experience including 15 years of executive experience at high tech companies (NCR, AT&T, Compaq, now HP, and AMD) leading strategy, product management, product marketing, and corporate marketing, including three industry board appointments.