At the end of last year, she presented the company Amazon (or its AWS, Amazon Web Services) new cloud processor Graviton4. It is built on the ARM architecture and was supposed to offer significantly better performance compared to its predecessors, which was confirmed. A test of 64 processor instances appeared on the Phoronix server, and the new product was compared to previous generations of Graviton processors (Graviton2 and Graviton3) as well as competing processors Intel Xeon 8448C (48 cores) and AMD EPYC 9R14 (96 cores).
click for more tests at Phoronix.com
In HPC deployments, the Graviton4 excels and wins almost every test in this group (and often by quite a margin). On the contrary, the processor from AWS did not cope with cryptographic tasks, hashing and the like at all. It showed mostly very low numbers compared to AMD and Intel. It surprised with a very good performance in 7-Zip compression, it also performed very well in source code compilation, where it was usually in first or second place with EPYC. In terms of database deployment, it has significantly improved compared to previous generations of Graviton and is a worthy competitor to AMD and Intel (it often won and if it was, for example, third, the difference was not great). In addition to cryptographic tasks, Blender is also not good for him.
As a result, it is 31% more powerful than the original Graviton3, which is a very nice intergenerational increase. It then beat instances with Xeons by less than 5%. It should be noted here that the Xeons had only 48 cores, while the EPYC and Graviton4 have 96. Intel does not offer Xeons with a significantly larger number of cores, so far it ended up with 64 cores, and only the new generations will reach much higher. But the new product from AWS is far from the EPYC 9R14, which still has 25% more performance than the Graviton4.
Source: www.svethardware.cz
