USB 4 and the compatibility to Thunderbolt 3
USB 4 was already announced by the USB Promoter Group in March 2019. Well over half a year later, the final specification was released on September 3rd. According to the press release of the USB-IF, the update complements the existing USB 3.2 and USB 2.0 and “delivers the next-generation USB architecture”.
Since USB 4 is based on Intel’s Thunderbolt protocol specification, on one hand the bandwidth is doubled up and 40 Gb/s are possible, and on the other hand multiple data and display protocols can be processed simultaneously. The USB Type C connector, which established as the external display port for host devices, also remains for USB 4.
So far, so good. Now the big disadvantage is: Manufacturers are not obligated to implement the Thunderbolt 3 functionality into their USB 4 specification. That means: USB 4 capable devices are not necessarily compatible with Thunderbolt 3. And since not every USB device can reach the promised 40 Gb/s, USB 4 devices may reduce their speed rate to fit the hardware. USB 4 will be available with 10, 20 and 40 Gb/s – which probably means that small and cheap devices will only be found with the lower rates.
More about the intelligent bandwidth of USB 4 can be found in this article.
The published specifications can already be downloaded on the website of the USB-IF. Until products with USB 4 will be released, it will probably take until the end of 2020.
Nervana NNP-T and NNP-I: Intel’s AI processors
The topic Artificial Intelligence is growing and growing, that’s no secret. To serve the demand for special chips in this sector Intel introduced their Nervana processors at the Hot Chips HC31 conference at the end of August. With these, the concern competes with Google’s Tensor processors, Nvidia’s NVDLA and Amazon’s AWL Inferentia chips.
The processor combination of the chips NNP-T and NNP-I is mostly made for Machine Learning. NNP stands for Nervana Neural Processor, but the chips are also known under their codenames “Spring Crest” and “Spring Hill”. In the use case they are coming into action complementarily: The NNP-T chip is used to train the AI model with Big Data, the NNP-I chip is responsible for the Inferencing, therefore to implement the training results into the AI workflow. The SoC are produced in Intel’s own 10-nm technique, NNP-T in the 16-nm technique.
With the introduction of Nervana, the Xeon CPUs which were responsible for AI tasks until now, are replaced. Although they still provide enough performance, they are less effective and efficient than Nervana.
AMD market share grows, Ryzen prices fall
Already in our last news we reported about the latest business numbers of AMD. In the second quarter of 2019, AMD as well as Nvidia noticed a lower turnover of graphics cards in desktop PCs, workstations and servers – Nvidia much more than AMD. That’s why AMD successfully increased their market share from 22.7 to 32.1 percent.
Looking at the whole GPU market, that includes integrated graphics cards, Intel still has the pole position with 66.9 percent market share. But AMD also scores here: In contrast to Intel and Nvidia, only AMD was able to increase the deliveries and thus, raise their market share to 17.2 percent.
One of the main reasons for AMD’s success are the thriving Ryzen APUs. The processors of the Ryzen 3000 series note drastic price cuts by now, only two months after sales launch. The models Ryzen 7 3800X and Ryzen 5 3600X, whose value-for-money ratio was rated as bad until now, have made the biggest price cuts. The price development of the Ryzen 7 3800X can be found here.