Instinct MI300 next year, the El Capitan supercomputer at the turn of 2022/2023

The official CDNA computing architecture roadmap extends only to CDNA Generation 2, now released. How will it be? If some sources spoke of further developments, they built the next generation – CDNA 3 – by 2023. Such a development would mean that the release of CDNA 2 / Instinct MI200 a year after CDNA (1) / Instinct MI100 was an anomaly rather than a phenomenon that would define shortening the development cycle in this segment.

However, new reports published by various leakers confirm the increase in cadence. According to leaker Kepler, the Instinct MI300 is to be a product of 2022, and according to the relatively reliable leaker Bondrewd, the DOE (US Department of Energy) management has described the El Capitan supercomputer as a product coming 5 quarters after the current Frontier.

supercomputercomplete.FP64 powerconsumedCPU(GP)GPU
Summit20180,2 EFLOPS13 MWIBMNvidia
Sierra20180,125 EFLOPS11 MWIBMNvidia
Perlmutter20200,1 EFLOPS21,5 MWAMDNvidia
HPC Mega-Project?0,275 EFLOPS?AMDAMD
Fugaku20210,415 EFLOPS18 MWFujitsu
Frontier2021>1,5 EFLOPS27 MWAMDAMD
Oceanlite20211,3 EFLOPS35 MWSW26010
Tianhe-320211,3 EFLOPS?FeiTeng
Aurora2022?~2,4 EFLOPS60 MWIntelIntel
Captain2022/23>2 EFLOPS33 MWAMDAMD

This would confirm that the Instinct MI300 also cannot arrive later than five quarters after the Instinct MI200, so its release can be expected either late next year or rather in January at CES 2023. This would be roughly the same as the availability of the architecture. Hopper from Nvidia (it is worth mentioning here that the latest generation of AMD releases several quarters after actually delivering them to partners, while Nvidia usually in advance – to the formal release of the computational GPU architecture Hopper so it may happen sooner). Bondrewd further specified that El Capitan’s consumption should be around 33 MW, which is almost half of Aurora’s competitors. Compared to Frontier, there is a 20% increase in energy requirements. Peak performance is expected to reach over 2 EFLOPS, but how much is not yet known.

AMD Instinct MI250X (Aldebaran)

It is believed that Intel, which increased Aurora’s target to 2.4 EFLOPS, did so so as not to lag behind El Capitan. But no one knows what systems will eventually occupy the position, because there is still quite a lot of time and therefore room for certain changes to complete them. The cooling and power infrastructure for El Capitan should be significantly oversized, so that any interest in increasing the target output should not hit the limits of the background.

Forbes added that AMD and HPE had won orders for both supercomputers at a stage when both Instinct MI200 (CDNA 2) and Instinct MI300 (CDNA 3) were still on the drawing board. Their proposal thus reflected the specific requirements of the contracting authority, which AMD tried to meet as best as possible. These architectures, or products, can be understood as a certain boundary between classic development and semi-custom design at the customer’s request.


Source: Diit.cz by diit.cz.

*The article has been translated based on the content of Diit.cz by diit.cz. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!