Comments on: Data Analytics Can Be The Next HPC For IBM Power

By: James Reinhart

James Reinhart — Thu, 16 Jun 2022 02:29:51 +0000

I was at Motorola Semi when the 88K was first introduced and the MPC88100 Vector matrix extension processor was sat on by those who gave up on their own company and believed in the idiocy of anti-competative belief of best practices, only a fool would believe such nonsense. It was destroyed from within. Later, the PowerPC was introduced based on these same structure that was inexpensive but an outstanding performer which was ditched by Motorola by Management which moved to Intel for all sectors. IBM found the Giga-Processor Lite to encroach upon the profitability of their Power 4 CPUs and the Power PC was lost. Now with Power open, the architecture and instruction set is fabulous (Steve Jobs was so furious about loosing the 88100 processor that he melted down the Next computers and started OpenStep which I knew was going to happen as discussions became volatile between Motorola and Next. Ironically, the MPC7455 was going to be used by CISCO for their communication systems as well as Apple for their CPUs but our top Management destroyed the company for Intel’s sake and why, I have no idea as the forecast orders were very high for this particular processor.

It’s time to take a look at a broader based power open system for companies that use their own products rather than a group of fools making decisions that kill companies. BeBoxes were very fast and QNX showed just how powerful a well engineered Real Time OS can be when simulated on the Power PC.

It would be a revolution in staggering multi matrix computing and quantum computing to see this realized.

By: Matt

Matt — Sun, 08 May 2022 18:05:13 +0000

I can’t see how Hopper was in play for Frontier. Frontier was supposed to be up and running before Hopper was to be available. Frontier was planned to be operational in 2021. As Hopper is a 5 nm part it would have heen extremely unusual for Nvidia to push the supercomputer so far ahead of volume rollout of the GPU. And Nvidia certainly were never going to have commercial production of Hopper on 5 nm in mid 2021, which is what would have been necessary.

If the DoE knew Frontier wouldn’t be available for general usage until January 2023, as seems to be the current plan, they may have gone a different route. My understanding from servethehome.com is the main thing currently holding up Hopper is the lack of availability of any x86 PciE 5.0 CPU. So with the current Frontier timeline it would now only be delayed a further month or two, if at all, if it had been an IBM/Hopper system.

I don’t know if it’s accurate or not, but I believe I read somewhere that there have been issues with scaling out the Frontier topology efficiently, which Cray was also having problems with with Knights Landing.

By: Timothy Prickett Morgan

Timothy Prickett Morgan — Sun, 08 May 2022 15:01:35 +0000

In reply to Eric Olson. I could not agree more, and have argued this since I think Power6 or Power6+ on many different publications. IBM has made some headyway in Power on the cloud, but is focused mostly on IBM i and AIX customers. If IBM wanted to impress me, it would throw 500,000 Power servers, some small, some medium and some large, on the IBM Cloud and prove the merits of the platform and offer competitive pricing. It really hasn't done that, and as far as I know, Rackspace and Google haven't either.

By: Eric Olson

Eric Olson — Sun, 08 May 2022 14:40:04 +0000

While it’s possible to shrug off HPC as a niche area where it’s too expensive to compete, cloud servers and services are no longer niche and constitute an increasingly large part of the hardware market and software ecosystem.

The recent success of the Amazon Graviton and Oracle Altra-based instances suggest the hardware for cloud-based computing is quite fluid. Any deals to create large Power10-based cloud infrastructure would be significant. In fact, it’s arguable success in the cloud is necessary for whatever hybrid cloud means and could drive on-premise success.