10,000 Processors Rome with 64 Cores Zen 2 in each are expected to provide computational capabilities of nearly 24,000 trillion operations per second - which also exposes us to one of the designed working frequencies of the innovative product
The positive momentum of AMD Continues, and after the announcement of the new Zen generation and collaboration with the giant cloud platform of Amazon - It's first customer turn for the processorsEPYC Which will be based on the formation of a truly formidable HPC system.
The customer in question is the powerful power center in Stuttgart, Germany, or HLRS, which exposed a program called Hawk based on 10,000 processors EPYC 2 leaders to create its most powerful flagship system - with 640,000 physical cores and over a million virtual logical capillaries alike, for a record 24TFLOPS performance that will make the work one of the most powerful in all of Europe.
The company that will build the supercomputer is HP (26 terabyte) for storage and 26,000 terabytes of dynamic memory as part of molecular dynamics calculations and fluid dynamics that require as high bandwidths as possible between processors and X-RAM Next to them - and this is something that the architectureZen 2 may well excel in it thanks to the new symmetrical arrangement.
The development and construction of the computer will cost about 38 million euros, starting next year - and ending with an actual operational operation only during the year 2020. We hope that by then we will have the opportunity to hear details of another variety of giant computers based on home hardware AMD, Although it should be noted that even the current situation is an improvement for the company. In the 500 list, the world's most powerful supercomputers do not have a single one based on the X-EPYC From the first generation.
In one of the slides we received a statement about a practical working frequency of 2.35GHz for the Zen 2 cores in the Hawk - a very modest frequency for Processors But it is important to note that server processors generally tend to cut at lower frequencies than their home counterparts, without a jump to high acceleration frequencies, to ensure uniform power consumption and to maintain optimal efficiency in massive arrays that are equally important And perhaps more) than the maximum performance.
About 2.4TFLOPS performance from a 64-core processor running at 2.35GHz is data that has refocused on our expected performance improvement as part of the transition from Zen to Zen 2 and between 14nm and 12nm to 7nm lithography - so a statement of AMD The power of 29 percent performance improvement under the same frequency (IPC - Instructions Per Clock) caused great excitement, in an era where we were used to getting a modest one-digit percentage improvement from year to year.
After the buzz was forced AMD Provide a clarification in which it states that a 29 percent IPC improvement between Zen and Zen 2 was observed in only one specific test that combines integer and floating point calculations, and no such general improvement is expected in the intergenerational transition - so we will continue to wonder what exactly the full potential of AMD Both in the world of servers and in the home processing market, and is there any chance that we will see a leap from the well-known 14 nanometer cores Intel.