Upgrade Trends: AMD has announced new high-frequency processors: EPYC 7FX2
AMD EPYC processors have considerably influenced the server market: even their first-generation offered 32 cores and good performance. There is no need to talk about the second: 64 cores in each processor socket and improved architecture. EPYC chips are now found everywhere — from ordinary servers to supercomputers and clouds.
The company has introduced new EPYC versions of the 7FX2 series with even higher clock speeds.
AMD EPYC processors with higher clock speeds had been seen in the list of ASUS server systems. As you know, even in our era of optimizing software for multithreading, there remain areas in which the game is still ruled by the clock rate. This is characteristic both for tasks that are poorly parallelized in principle, and where minimal latency is essential at any price.
In the EPYC lineup, there have never been similar models. In the chip segment proffering a small number of cores, 3.35 GHz was the maximum figure. The flagship EPYC 7742 also boasted a mere 3.4 GHz, and even then not for all cores and in its every-day mode. However, AMD has concluded this omission with today’s announcement introducing three new processors — the EPYC 7F32, 7F52 and 7F72.
The F symbol in the name clearly indicates frequency, the clock frequency that AMD has emphasized in this case. The company is not chasing the number of cores in this series: new processors have 8, 16 and 24 cores, respectively. These are all the same as the Rome with Zen 2 architecture.
However, the frequency formulas here are much more serious than before: the junior 7F32 model has a base frequency of 3.7 GHz, and at the highest possible mode 3.9 GHz. The 16-core 7F52 starts at 3.5 GHz, but it is also able to increase its core frequency to 3 9 GHz. The senior model can accelerate from 3.2 to 3.7 GHz — and this is significantly faster than the old 24-core 7352 and 7402 with their base frequencies of 2.3 and 2.8 GHz and their maximum of 3.35 GHz.
Interestingly, we have managed to keep the TPD within acceptable limits: the company says the values are 180, 225 and 240 watts respectively. The L3 cache is 128 MB in the eight-core chip and 256 MB in the 16-core. Though the 24-core 7F72, due to heat dissipation restrictions, is limited to 192 MB, although this is a significant step forward — previously the 24-core EPYC Rome had only a 128 MB L3.
The combination of high clock speeds and a serious cache size makes the new AMD processors unique and perfectly suited to a wide variety of tasks. The company itself speaks of their focus on hyper-converged infrastructure, commercial tasks of the HPC class and DBMS. AMD has dubbed these new products leaders in terms of price to performance ratio, and in this they seem to be right.
Major manufacturers of server and HPC equipment have already announced their solutions based on the new EPYC 7FX2 processors. HPE announced new Nutanix HCI-certified servers, IBM is talking about new 48-core cloud-based solutions, and Dell is discussing new PowerEdge R6525 servers based on the EPYC 7F72.
Supermicro also has something new: this is the first high-density (20 single-processor blades) SuperBlade computing complex equipped with AMD processors in the company’s history. It supports all AMD processors with TDPs up to 280 watts and uses the InfiniBand 200G HDR interconnect.
Unlike Intel, AMD can rightfully say that it adheres to its schedule and does not experience problems with new architectures and processors. With the new announcement, the series of Rome server chips can be considered basically complete. But lying ahead are the new red heights that they have yet to conquer, the Zen 3 Milan and Zen 4 Genoa.
