Qualcomm, Who Use ARM Architecture To Develop Server Chips Is Preparing To Abandon It.

- May 25, 2018-

Bloomberg News quoted sources as saying that the world’s largest maker of mobile phone chips, Qualcomm, is preparing to abandon the development of data center server chips.

Informed sources also said that Qualcomm is still considering two options to close the business or find new buyers for it.

Qualcomm as the dominance of the global mobile phone chip is difficult to shake. When it comes to the server chip field, Qualcomm hopes to use ARM architecture to develop server chips to break the dominance of Intel in this lucrative market, but today Qualcomm’s rumors of exiting the server chip market indicate its dream. Is bursting.

Of course, Qualcomm is just one of the failures of many servers to challenge the Intel X86 server dominance. Further, Qualcomm's exit is also a defeat of ARM. For Huawei, which has just entered the server chip field in 2016, it will eventually lose out.

Qualcomm is not surprised to exit the server chip market

In 2017, Qualcomm’s Centriq 2400 server chip was launched. This chip is based on Samsung's 10nm process, and its power consumption and cost are superior to those of the Intel Platinum Xeon 8180 processor.

When the chip was released in November last year, Microsoft also took the stage to express interest in this chip, but after this Qualcomm has been silent on progress.

The news of the withdrawal from the server chip market, a high-pass spokesman declined to comment.

However, Qualcomm's CEO Steve Mollenkopf said in a financial teleconference last month that Qualcomm’s focus was on compressing non-core product spending.

In response to Broadcom’s hostile takeover, Qualcomm’s management had promised investors in January that it would squeeze annual expenses by US$1 billion to increase profitability.

In fact, before Qualcomm officially demonstrated its first server chip in this field in 2015, several giants have already challenged Intel's results with ARM server chips.

NVIDIA had explored the use of its 64-bit Tegra processor to enter the server market. It only provided graphics chips that were compatible with other companies' ARM server chips, but trying to compete in the server chip market was almost a disaster.

Samsung had also vigorously developed ARM server chips, and also dug executives from the AMD server division to take charge of related projects, but this project was eventually abandoned.

In addition, AMD had high hopes for ARM server chips when its server market share continued to decline in 2014. At that time, AMD expects its share in the ARM server chip market will reach 25%.

However, the actual situation is that Intel's x86 server chips still occupy about 99% of the market share in this area. AMD's share in the server chip market also fell from 25% at the peak time to less than 1% in 2017. Obviously, ARM server chips have not improved AMD's profitability and market share.

The AMD chip that AMD tried to return to the server market in 2017 is also the x86 architecture. ARM products have been abandoned.

Qualcomm's withdrawal is also a defeat of ARM

Samsung and AMD have failed to leverage the ARM server to shake the status of Intel x86. Before further discussing the reasons for the failure, it is also necessary to understand why giants, including Qualcomm, want to enter the server chip market.

The reason is simple. The server can handle data in the enterprise network and it is also the backbone of the Internet. Although its shipment volume is much smaller than that of smartphones and PCs, it is attractive because the server chip charges high costs for high performance and the profit rate is considerable. .

However, with Intel x86 servers occupying the upper hand and having the absolute right to speak, they want to challenge Intel in this market, only to find a new architecture.

At this time, Qualcomm, Samsung, AMD and other cooperation with ARM is a win-win choice. On the one hand, it can help the ARM architecture that has achieved great success in the mobile processor market to advance into the highest-end computing market. On the other hand, Qualcomm, NVIDIA and others can also borrow. This is a share of the server chip market and even challenges Intel.

From a performance point of view, the "mars" performance of the 2015 National Defense University's 64-core ARM server chip is very strong. According to the SPEC2006 simulator performance, the multi-core performance can be compared with the Intel E5 (high-end workstation and server processor series), but the single-core performance Weak; the Qualcomm Centriq 2400 mentioned above consumes more power and costs than the Intel Platinum Xeon 8180 processor.

Therefore, performance is not a major obstacle to the advancement of the ARM architecture into the highest-end computing market, and the software ecosystem is more critical. Just as Intel x86 entered the mobile phone chip market and was constrained by the fact that the software ecosystem led to more than half the effort, ARM encountered the same difficulties when it entered the PC and server chip market.

ARM certainly realized this problem. In order to quickly promote the 64-bit instruction set and establish a corresponding software ecosystem, ARM changed its past caution on the authorization of the ARM 32-bit instruction set, and authorized 64-bit instruction sets to accelerate the ecology of many powerful manufacturers. As a result, a large number of units or companies that do not have ARM instruction set authorizations in the era of ARM 32-bit instruction sets, such as Huawei and National University of Science and Technology, have also obtained authorization through the promotion of 64-bit instruction set by ARM.

However, the construction of the ecosystem is not an overnight event, and Intel has already dominated the server chip market. It is easy to imagine how difficult it is to construct its own ecological environment.

Qualcomm today will be Huawei's tomorrow?

In addition to ecological deficiencies, Intel's already-launched low-power server chips have completely blocked the possibility of ARM eroding the market through low-power server chips. Therefore, competition between ARM server chips and Intel can only rely on differentiation or policy protection.

We also saw that Qualcomm had found the Guizhou provincial government to establish a joint venture company Guizhou Huaxintong Semiconductor Technology Co., Ltd. to rely on the protection and investment of Chinese local governments to obtain ARM server chips. However, Qualcomm is an American company after all. It is not too Great advantage.

Huawei, as a local company in China, acquired the ARM 64-bit instruction set. At the launch of the Kirin 950 in 2016, Huawei announced that its proprietary chips for servers are in full swing.

Unlike other vendors who have entered the field of server chips, Hass server chips can serve the "cloud" of Huawei's "cloud, pipe, and end" strategy instead of simply selling chips.

Therefore, if Huawei's ARM server chip is successfully developed in the short term, and it is not inferior to the performance of the "Mars" of the National University of Science and Technology, it can not only be produced and sold, but it can also use its good relationship with the government and operators to obtain a living space.

However, the problem of the ARM server software ecosystem is difficult to solve, and more commercial companies still use Intel x86 chip servers as the mainstream.

Therefore, Huawei has paid a high cost for the development of ARM servers. Under the premise that ARM servers have little room for survival, is Qualcomm's today the future of Huawei?