Nvidia said that its Arm-based Grace CPU Superchip processor in tests with real workloads shows a twofold increase in performance compared to x86-compatible CPUs in the same range of power consumption. With which processors Nvidia compared, it is not specified.

The company says these figures mean data centers can handle twice as much peak traffic for the same power consumption if they switch to Grace. Or they can cut their energy bills by almost half while maintaining productivity.

Nvidia itself says that Grace provides such results thanks to three main factors. Firstly, these are 72 Arm Neoverse V2 cores in one CPU (Grace CPU Superchip, we recall, contains two CPUs on one board), and secondly, in fact, the combination of two such CPUs via the NVLink-C2C interface with a bandwidth of 900 GB / s , third, the use of server-class LPDDR5X memory.

Asus introduced the external GeForce RTX 4090 in the form of the XG Mobile Graphics Dock at the beginning of the year. Now this product is on sale for a very impressive amount.

In Japan, they ask for $ 3,000 for the device, but Japanese prices are sometimes much higher than global ones. In this case, the device is also available in Canada for about $2,000.

True, we are talking here, of course, about the mobile RTX 4090 Laptop – it is she who underlies the module. In terms of performance, it roughly matches the desktop RTX 4070 Ti. The module itself consists of two parts: one contains the 3D card itself and a set of ports, and the other houses a 330W power supply. True, the TGP value for the adapter is limited to 150 watts.

Probably the biggest downside of the device is that it is only compatible with select Asus ROG Flow laptops.

Yesterday Nvidia unveiled the monstrous 188GB H100 NVL graphics accelerator for training language models, and now the company has added a card to its range that sits on the other side of the product line. The RTX 4000 Ada professional adapter is much smaller, much simpler, and clearly much cheaper, though we don’t know the price of the H100 NVL yet.

The RTX 4000 Ada, on the other hand, costs $1,250. And in terms of its parameters, this is a very unusual product. For example, it has 6144 CUDA cores (about as much is attributed to the GeForce RTX 4070 video card), 20 GB of memory, but only a 160-bit bus. At the same time, the consumption of the novelty is so small (70 W) that it does without additional power connectors!

Such a modest power consumption with good parameters is due to the GPU frequency, which reaches only 1.56 GHz. For comparison, the RTX 6000 Ada is 2.5 GHz.

The performance of the novelty (FP32) is 19.2 TFLOPS. Also noteworthy are four Mini DisplayPort connectors and a dual-slot design.