IBM unveils its Energy10 deployment plan and launches Energy11 | Zoom Fintech

November 23, 2020

Timothy prickett morgan

We have now followed the Energy10 processor event with great curiosity over the past few years and tried to determine exactly when – and how – Massive Blue will install its future processor in Energy Programs machines. At the Frequent Europe On-line vCEC 2020 event last week, Steve Sibley, vice president and supervisor of IBM’s Cognitive Programs division, spoke about IBM’s plan and set some tough dates.

Once we spoke to Sibley again in May, all he could tell us was that the impression of the coronavirus pandemic could trigger delays in the launch of Energy10. And this summer at the Sizzling Chips digital convention when the structure of the Energy10 chip was unveiled, Sibley mentioned that there would definitely be a delay, but that it was in the range of weeks, not months and that the Energy10 chip would certainly be shipped in turn. this point in 2021. Until we assume that this would happen around September 2021, if the story is any news, and towards the end of September for that matter, IBM could therefore just have a little nudge in it. during the third trimester and an excellent fourth trimester.

Sibley – and certainly anyone else in the Cognitive Programs division, which many people still call Energy Programs – made no mention of what machines would be released in the second half of 2021. Usually IBM starts to back and works his method all the way to the excessive end, but with Power7 it started in the middle of the road, then worked down, then up. We didn’t know what the sample might be for Energy10. Under usual circumstances, with Intel and AMD each bringing new X86 processors into service in early 2021 for entry-level machines – Intel’s “Ice Lake” Xeon SP chips and AMDEpyc’s “Milan” chips to be exact – you’ll assume IBM may want to get its SKUs for Energy10 servers in the industry.

However apparently not. At vCEC last week, Sibley said that “mainframe” using Energy10 processors won’t be released until the end of 2021, and that “scalable” machines, which he really named two of them, would be releasing one. day in the first half of 2022. From this, we assume that the Energy E1080 machine will come out first, and presumably in addition to the Energy E1050. And in fact, we have now heard through the vineyard that the division and distinction between the stand-alone four-socket node (the Energy E750, E850 and Energy E950) and the four-socket nodes which can be augmented with NUMA pipes which can be used to create the high-end NUMA machines (Energy E770 and Energy E780, Energy E870 and Energy E880, and Energy E980) will be phased out for good and these machines will likely be converged to the least one area that can do each job. It’s just like how IBM converged to a single Energy E980 instead of creating an Energy E970 server with half the options and completely different CPU benchmarks. IBM wants the best range of energy programs it can put in the industry, which is smart as long as this is a relatively small-scale company nonetheless.

Input methods based primarily on Power9 processors were the main ones to return to the market at this time, and did so in February 2018 with the “ZZ” variants of machines supporting the PowerVM hypervisor and, of so do IBM i, AIX, and Linux. Scale-out-only Linux machines, codenamed “Boston”, adopted in May 2018, after which the four-socket domain that does not support IBM i (which bothers us), then in August 2018 the midrange The four Zeppelin outlet Energy E950 and the high-end ‘Fleetwood’ Energy E980 complete the range.

The deployment of Energy10 is involved in reverse, it seems. Or rather, IBM is preserving the mid-range Energy E1050 node and the clusters built from it that include the high-end Energy E1080 where we intended and outperform the entry-level Energy10 servers. from where we expected them at the start of center 2021 to early 2022 with coverage in case the coronavirus pandemic results in a slower than expected business transfer until mid-2022. It will likely be a four-year hole for entry-level machines, but only a three-year hole for mid-range and high-end methods. 4 years is a very long time between generations of processors for single-socket and dual-socket servers, whatever the structure.

While Sibley did not specify these changes, they are quite apparent to us. We would expect that small and medium-sized businesses that rely on the input servers of energy programs probably do not want the raw efficiency of the Energy10, and and not using a substantial HPC and AI business for fostering the adoption of these servers, IBM is in a hurry to launch entry-level Energy10 machines for its prospects. If IBM had received one of the world’s multi-billion {dollar} exascale supercomputer awards over the next several years, we’ll probably be talking about a different story, but even when it does ‘is produced with the pre-exascale “Summit” and “Sierra” methods built by IBM for the US division of Vitality, the Energy AC922 system was a completely personal custom domain when it was launched in December 2017 and it did not. not actually marketed in the types of volumes IBM had hoped for. There are a selection of clusters based primarily on the Energy AC922 in the global market, and this “Witherspoon” machine is an excellently designed platform with all kind of capability that is not available on an X86 field. However, it is not cheap either.

Sibley wanted to reassure the COMMON Europe team, and as a result, the rest of us turned out to be a global opportunity, regarding the long term prospects of the Energy processor line and Energy Programs server line.

“We are already starting to design and review the key requirements for charting where Powernext is going to go, what chances will be called Power11 and which will come later in the decade,” Sibley said. “The same three to four year cycle we’ve seen with processors in general. “

Based mainly on the cycle so far, 4 years is more regular for massive jumps, as you can see from the roadmap above, although there has been a Power7 + shrinking course at 32. nanometers which is not proven to actually cycle 1- 3-3-Four and we presume from this collection, echoing a SAT exam, the next amount will likely be 4. So it’s 2025 for Energy11, and this may imply that the entry servers get it first and we go back to the midrange / high end sample adopted entry that we are used to. However who knows? The lots can change in 5 years in the IT sector. Regardless, if Energy11 comes out in 2025, the product line could very well be sustained with current processing capacity for around 5 years, hitting the 2030 promise that Massive Blue made to provide support to IBM i.

It’s about as well because it will be fine. You did not receive a higher guarantee from all other platforms.


IBM’s potential designs for Energy10 programs

Explore the structure of the Energy10 chip

Energy at tenth energy

Stoop energy programs are not as unhealthy as they seem

The way really opens up for alternative energy processors, but is it enough?

Powers of ten

IBM gives a glimpse of the future at POWERUp 2019

What does Open Sourcing Energy’s ISA mean for IBM i retailers

IBM’s plan for etching Energy10 and later chips

The road ahead for energy is paved with bandwidth

IBM puts the issue of future energy chips in the ground

What will IBM i do with an Energy10 processor?

Samsung joins the OpenPower Consortium social gathering

Tags: Tags: AIX, AMD, Cognitive programs, COMMON Europe, Epyc, HPC, IBM i, Linux, NUMA, POWER AC922, Energy E1080, Energy E770, Energy E780, Energy E870, Energy E880, Energy E950, Energy E980, Energy Programs, Energy10, Energy11, Power7 , Power9, PowerVM, Sierra, Summit, X86, Xeon SP

About Ricardo Schulte

Check Also

How to qualify for $10,000 fast

Our goal at Credible Operations, Inc., NMLS Number 1681276, hereafter referred to as “Credible”, is …

Leave a Reply

Your email address will not be published.