Advanced Computing in the Age of AI | Friday, May 20, 2022

Intel Maps Out The Year Ahead For Xeon Server Chips 

Intel has a busy couple of months ahead refreshing its Xeon server processors, and companies that are getting ready to launch new software projects this year will be in position to take advantage of better hardware.

That may sound backwards, but software drives hardware, not the other way around, and by the middle of the year, server makers will have a variety of "Ivy Bridge" family Xeon systems out the door to run both old and new code.

It is difficult for a hardware guy to admit that software drives hardware purchases, but this is clearly the case for most enterprises. Shannon Poulin wears two hats at Intel – one is as vice president in charge of the enterprise division in the Data Center and Connected Systems Group and the other is as general manager of marketing for the whole group – and admits this freely.

"At most of the enterprises that we talk to, I would love for the decision that they are making to be driven by the underlying hardware," says Poulin with a laugh. "But that is generally not what is driving their purchases right now. Decisions are based on whatever new services-driven software they need to deploy, and then they have to get some hardware to support that. They are not going to wait nine months for something that might be 30 percent better. They buy the hardware they need today, and they will add newer hardware to this as it becomes available. With virtualized pools, there is no reason to wait. If the pool needs to be 220 machines now and in the future I can take it down to 180, that works."

(There are exceptions to this rule, of course. Sometimes an increase in raw performance is all that matters, as is the case in some supercomputing simulation or high frequency trading applications, just to name two.)

That the Xeon motors from Intel have come to dominate the – dare we go retro and say it? – data processing in the datacenter is without question. So is the ever-increasing appetite for processing capacity. But there are other trends at work in the datacenter, too, with different kinds of processors, such as low-powered Atoms or massively parallel Xeon Phis, entering the picture. The prospects for ARM server chips remain good, despite the collapse of Calxeda last month, with other vendors, including AMD, Marvell, AppliedMicro, Cavium, and others, still working on 64-bit chips. IBM, Oracle, and Fujitsu are still selling systems based on their RISC processors. Still, for most workloads in enterprise datacenters, for now at least, the Xeon processor is the motor of choice.

Intel wants to keep the Xeon front and center, and the workhorse "Ivy Bridge-EP" Xeon E5-2600 v2 chips launched last September will drive the majority of Intel's sales in the first half of 2014. But there are other parts of the market that need a Xeon refresh, and Poulin walked EnterpriseTech through the timeline of this year's announcements.

"We have four good pops across the product line, depending on whether we are ticking or tocking," he says.

As we previously reported, the high-end "Ivy Bridge-EX" Xeon E7 v2 chips have been sighted here and there, with vendors promising to support them in products this year. These chips are aimed at machines with four and eight sockets, but we can expect to also see variants supporting two or sixteen sockets for specific workloads. (Supercomputing clusters for the former and large NUMA shared memory machines for the latter.)

Poulin confirmed to EnterpriseTech that these Xeon E7 v2 processors are, as has been rumored, set to be launched in the first quarter. Intel already started shipping them to its server customers in the final quarter of 2014 so they could qualify them in new products. Intel did not ship a "Sandy Bridge-EX" variant of the Xeon E7 line, so machines using the current "Westmere-EX" Xeon E7 v1 chips have to be completely refreshed with the "Brickland" platform. (Brickland is the name Intel gives for the combination of the Xeon E7 v2 processor and its related C600J chipset.) Intel has separately tipped its hand and let it be known that the top-end Xeon E7 v2 part will have fifteen cores, and it has previously said that an eight-socket system will be able to support up to 12 TB of main memory. That is 50 percent more cores and memory than a Westmere-EX machine that is nearly three years long in the tooth.

"We are focusing a lot of our effort on the Q1 launch of the Ivy Bridge-EX," says Poulin. "It will have very compelling performance, and it probably could not come at a better time given the competitive Power and Sparc products. We think we will be very well positioned once all of the data comes to light. We think it is going to be a great transition vehicle. We have seeded hundreds of the systems to software vendors and end users so they can tell us how it is going. It is Nehalem-like in its transformation versus the previous generation. We haven't had that in the EX space for a few years."

When Intel launched the "Nehalem" family of Xeons starting in early 2009, with their new QuickPath Interconnect and more sophisticated on-chip cache memory, that was when Advanced Micro Devices started to lose ground in the datacenter that it had fought so hard to win with its Opteron chips.

That's not the end of the line for the Xeons this year, of course. An Ivy Bridge Xeon E5-4600 v2 that plugs into the existing four-socket "Romley" platform will come out in the first quarter, a little bit after the Xeon E7 and its Brickland platform debuts. This will be a drop-in upgrade for existing Sandy Bridge systems, so it should not be a big hassle for server makers.

All of the Xeon E5 and E7 chips using the Ivy Bridge architecture are etched using 22 nanometer processes.

The Xeon E3 chips, aimed at single-socket servers and workstations, debuted with the new "Haswell" microarchitecture last year, along with Intel's Core processors for desktops and laptops. This new microarchitecture sports some modest performance improvements, slightly higher clock speeds, and much lower power usage at idle compared to Ivy Bridge chips. Rather than take its foot off the gas at the low end of the server line, Intel is gearing up for a Xeon E3 bump, based on the "Broadwell" design, which is coming in the middle of the year. The Broadwell chips will use Intel's 14 nanometer manufacturing processes, and the shrink from 22 nanometers used with the Haswell chips will allow Intel to push up performance and lower the power draw. Broadwell will come in a Xeon E3 socket as well as in a system-on-chip variant that is soldered right onto motherboards that is aimed specifically at densely packed microservers.

Interestingly, as EnterpriseTech has pointed out in a detailed overview of Intel's massive chip designing cluster, the company is moving from two-socket to single-socket Xeon servers for this system so it can benefit from the smaller footprint, higher clock speeds, and more affordable software licensing terms that its EDA software vendors have for single-socket machines.

Towards the end of 2014, the cycle that started with the Haswell Xeon E3s back in March 2013 will start anew, with the "Haswell-EP" Xeon E5-2600 v3 and its and "Grantley" platform coming out. The Grantley platform is expected to have a new chipset called "Wellsburg," but Intel has not confirmed this and has not said if these Haswell Xeon E5 chips would require a new socket, but it probably will. A good guess for the timing of the Haswell Xeon E5 chips is the annual Intel Developer Forum in San Francisco in September, which sets server makers up to be able to peddle machines using these chips during the fourth quarter IT budget push.

In general, Intel is optimistic about Xeon processor shipment and revenue growth in 2014. Poulin says that most of the server workloads that could have been virtualized have been and that this year will therefore be driven by new workload deployments. The cloud business is expected to see more than 30 percent shipment growth for Intel in 2013 and north of 25 percent growth in 2014, according to the company's latest projections, and Poulin adds that about three-quarters of that growth comes from companies like Pinterest, Instagram, Netflix – new businesses with new business models that represent net new growth for Intel's server chips.

"If we look at the enterprise segment, which is a little more than half of our market in terms of shipments, in 2013 it contracted a few points and in 2014 we are projecting it to grow a few points," says Poulin. "We tend to track gross domestic product numbers pretty closely, and the numbers were fairly negative in the first half of 2013 and turned a little positive in the second half. So we think that is going to carry through."

2 Responses to Intel Maps Out The Year Ahead For Xeon Server Chips

  1. Matti Patari says:

    Intel E5-4600/2600/2400 processors are used in numerous 4P and 2P x86 servers. To compare these server of different vendors is a pain. The information is scattered in data sheets in network myriads.
    Visual Configuration comparison map does a great difference. It show all 30 servers of Dell, IBM, HP, Fujitsu and Acer. The picture logical format is easy to understand. This single picture allows you compare these servers. The diagram does contain links to more detailed documents.

    A wall paper size hard copy is recommended for better readings ! Very dense stuff.

  2. So true. I’m always able to figure it out, but it usually involves a combination of Wikipedia and Intel’s website. It really shouldn’t be this difficult.

Add a Comment