The US Department of Energy is looking to suppliers that will assistance develop supercomputers up to 10 times faster than the justrecently inaugurated Frontier exascale system to come on stream inbetween 2025 and 2030, and even more effective systems than that for the 2030s.
These information were revealed in a demand for details (RFI) released by the DoE for computing hardware and softwareapplication suppliers, system integrators and others to “assist the DoE nationwide labs (labs) to strategy, style, commission, and acquire the next generation of supercomputing systems in the 2025 to 2030 time frame.”
Vendors have till the end of July to respond.
For this RFI, the DoE states it is interested in the release of one or more supercomputers that can fix clinical issues 5 to 10 times faster than present cuttingedge systems “or fix more complex issues, such as those with more physics or requirements for greater fidelity.”
The present state of the art is possibly best represented by the Frontier exascale system setup in the Oak Ridge National Laboratory, which was stated functional at the end of May and clocked at 1.102 Linpack exaFLOPS of calculate power, however which is anticipated to hit a peak theoretical efficiency in excess of 2 exaFLOPS in future.
In line with this, the DoE states that its rough pricequote – based upon patterns covering the past 20 years – consistsof standard HPC systems at the 10-20 exaFLOPS level or beyond in the 2025+ time frame, and 100+ exaFLOPS and beyond in the 2030+ time frame, which it anticipates to be provided “through hardware and softwareapplication velocity systems.”
Any such supercomputer will be anticipated to run within a power envelope of 20-60MW, according to the DoE. For contrast, the Frontier system currently takesin about 20MW, and can obviously hit a peak of over 30MW.
Systems must likewise be “sufficiently durable” to hardware and softwareapplication failures to decrease requirements for user intervention.
Interestingly, the DoE states that it is lookingfor to relocation away from “monolithic acquisitions” towards a design that would permit more fast upgrade cycles of released systems to makeitpossiblefor quicker development on hardware and softwareapplication.
One possible technique that may be followed is an increased reuse of existing facilities so that the upgrades are modular, with an acquisition procedure that enables constant injection of technological advances to centers, maybe every 12 to 24 months rather than on a 4 or five-year cycle.
- US weathercondition forecasters triple supercomputing zest with newest makers
- Lenovo, Barcelona Supercomputing Center indication joint researchstudy offer
- Businesses brace for quantum computing disturbance by end of years
- Germany to host Europe’s initially exascale supercomputer
This sounds rather comparable to the technique that hasactually been embraced for Europe’s initially openly stated exascale computersystem, the Jupiter system being built in Germany by the European High Performance Computing Joint Undertaking (EuroHPC JU).
Jupiter will be based on a “dynamic, modular supercomputing architecture,” according to the Forschungszentrum Jülich where it will be based, makingup a universal cluster module paired with a GPU accelerator module and storage modules, however it is prepared to be broadened in future with extra modules that might consistof a quantum processing system or a neuromorphic processing module.
The actions to this RFI will aid the DoE and the nationwide laboratories to upgrade their long-lasting sophisticated computing roadmaps, as well as notify the requirements for one or more DoE system acquisition RFPs (request for proposition) for systems to be provided in the 2025–2030 time frame.
The level of info askedfor from suppliers by the DoE is rather detailed, covering not simply the kind of processors, memory, storage, and adjoin alternatives the suppliers predict utilizing within the 2025 to 2030 time frame, however likewise what production procedures they anticipate the chips to be made with, whether the processors will be some type of APU/XPU system-on-chip mix, the expectations for the bandwidth and power of interconnects, the possible node setup, and so on.
Perhaps the DoE hasactually been stimulated on by fears that the UnitedStates might fall behind China in the supercomputing arms race, as reported earlier this year.
While the UnitedStates has 3 exascale systems in the pipeline, China intends to have up to 10 functional systems by 2025, and reports in the Financial Times declared that UnitedStates professionals feared that China might beat them to essential science and innovation advancements by fielding a bigger number of exascale makers. ®