NVIDIA releases fresh performance data for its latest 144-core server processor Grace Superchip on Arm architecture. During the presentation at the end of March, NVIDIA compared the new product with AMD EPYC chips, and now shared a comparison with Intel Xeon, which is of somewhat more interest.
The thing is that in March, Grace Superchip was compared with AMD’s previous generation server chips – EPYC Rome – which will be four years old by the time NVIDIA’s solution is released in 2023. According to NVIDIA, its processors are 1.5 times faster than a pair of 64-core EPYC 7742s. From this we can conclude that the Grace Superchip will match the performance of a pair of current flagships of the EPYC Milan series. And even this is not so important: by 2023, AMD will release EPYC Genoa – even more powerful chips on a new architecture and platform.
Therefore, a comparison with the current Intel Xeon Ice Lake, namely with a pair of 36-core and 72-thread Xeon Platinum 8360Y models, looks a little more interesting. Although, once more, in 2023, Intel will have promising Xeon Sapphire Rapids in its arsenal. Nevertheless, solutions with the same number of threads are also compared here – 144 each in both the Intel and NVIDIA systems.
NVIDIA claims that its Grace Superchip is twice as powerful as an Intel system, and, importantly, it is 2.3 times more energy efficient than its competitor. Note that NVIDIA processors benefit from a much higher memory bandwidth, which is extremely important for this test. Paired with Grace Superchip, LPDDR5X memory with a total bandwidth of more than 1 TB / s worked – the system on Intel was equipped with only DDR4 memory.
If you notice an error, select it with the mouse and press CTRL + ENTER.