Nvidia is probably the epitome of high graphics power. Users do not only enjoy this when playing demanding games. Crypto farmers also benefit from the hardware’s high VRAM. But did you know that the US manufacturer also develops processors? They don’t do things by halves, but put together matching components of entire supercomputers without further ado. Now Nvidia has presented its power CPU with the likeable name “Grace” in more detail.
A challenge to previous CPUs on the market
Although we know Nvidia primarily from the conception and production of graphics cards, the US-Americans can apparently also do otherwise. This has been known since last year. The company proudly announced for the first time that they were working on the so-called Grace CPU. However, all we got to see back then were just plain designs. Now the company has apparently gotten serious and presented its finished product.
And the result truly knocks the socks off industry experts and competitors alike. Nvidia has actually managed to put its specially combined single- and dual-chip design into practice. Grace consists of two chips in total, but they don’t have to work at the same time. So, depending on the system load, the processor relies on one or two chips. This should not only ensure an incredible performance that is far above the competition. On top of that, the choice between one or two chips allows for very good energy efficiency.
A total of 144 cores
But let’s go into detail. When it comes to the design, Nvidia surprisingly doesn’t rely on x86, but ARMv9, with each of the two chips offering 72 cores on their own. Consequently, 144 cores are available to Grace when both chips work. However, Nvidia doesn’t reveal more details about the architecture. The connection technology that is used for the interaction of the two chips is also exciting. The so-called NV-Link-C2C is supposed to enable speeds of up to 900 GB/s.
When it comes to memory, the US-Americans again rely on LPDDR5X technology, which is supposed to enable up to 1 TB/s. Incidentally, this is located on the package itself, which is supposed to be particularly beneficial for energy efficiency. With a total of 500 watts of required power, Grace is clearly below the values of AMD CPUs. The latter need a total of 560 watts in a dual-socket configuration. By the way, the placement of the memory in Grace is reminiscent of Apple. The US tech company takes a similar approach with its most powerful processor, the M1-Ultra.
Still available in 2022
Incidentally, Nvidia’s CPU was named after a famous computer scientist originating from the US. Among other things, the eponymous Grace Hopper made a great contribution to the development of the COBOL programming language. If you are interested in setting up your own supercomputer, you won’t have to wait very long. After all, Grace will probably be used in large server solutions and other supercomputers and not home computers for the time being. Nvidia wants to sell the first chips of its new processor in the first half of the year. Prices are still unknown, as are possible customers.