In recent months, with the hot sale of a flagship mobile phone in the Green Factory, the Mariana X independent NPU chip equipped with it has once again been paid attention to by consumers, yes, "again"! The chip, which had been released by PPT as early as last year, was finally installed on the flagship mobile phone and mass-produced.
From Liu Zuohu's tireless speech and unsparing praise at the FindX5 conference, it is not difficult to see that OPPO has high hopes for Mariana X. After all, this is a chip with a very high concentration of self-research.
After Huawei was included in the entity list sanctions by the United States, the domestic self-developed high-end chips were no longer brilliant and fell into a trough. Where is the self-developed chip? Does our self-developed chip work? For a time, it became the focus of discussion among many people, and there were not a few people who held various views.
OPPO released a high-profile release at this point in time and soon mass-produced the Mariana X chip, and this wave of hot spots is really level. Naturally, consumers are also paying more attention to the performance of the Mariana X chip. Anyway, Liu Zuohu's cattle on the release have been blown out, but is it really so powerful?
First, let's look at the parameters of the Mariana X chip. It has the terrifying 18TOPS AI computing power, using TSMC's advanced 6nm process, memory bandwidth of up to 8.5GB per second, can process 20Bit HDR data.
Even Apple's proud A15 chip built-in NPU unit AI computing power is only 15TOPS, from the data alone, OPPO seems to have surpassed Apple. It does seem like that, after all, the parameters are there.
For OPPO's attitude of insisting on self-developed chips, I am positive, after all, the gold content and difficulty of self-developed chips are understood.
OPPO with a chip team of more than 2,000 people to commercialize this NPU chip on a large scale, you know, a chip from project establishment to design, tape-out, trial production, mass production to the final large-scale commercialization is an extremely difficult process, but OPPO did it, which would have been a good thing.
But then I found that the painting style was not quite right, and after a period of time, many self-media on the Internet praised OPPO's Mariana X, and even could not pick out any shortcomings. So much so that when I wanted to know more about this NPU chip with an objective and neutral mindset, I found that there was very little really valuable content.
On major video websites and information platforms, the major UP owners and self-media are almost all praising how powerful the Mariana X chip is, how high the self-research concentration is and so on. In short, in such media reports, we can only see the good side of the Mariana X chip, and there is almost no mention of its current problems. Of course, I understand OPPO's pride in the Mariana X chip, and I do not rule out that there is an interest in individual media and OPPO.
But is Mariana X really that perfect? I don't think so.
Having top-of-the-line performance is just the threshold for a good chip, and just doing that isn't enough to make Mariana X one of the top chips. Equally important for chip development and tuning optimization. Honestly, Mariana X did a poor job in this regard.
Li Jie, president of oppo Find product line, said: "For the Mariana X, we have just touched the door, and currently only play about 30% of the performance of this NPU. The FindX5 equipped with this chip will also be continuously optimized and will continue to use this NPU for the next two or three years.
At present, the Mariana X chip can only be regarded as a semi-finished chip, and its related algorithms and underlying optimization are not perfect enough, after all, the actual performance of Mariana X is far from meeting the user's expectations, and Liu Zuohu has bragged about many of the bulls at the FindX5 conference. Some bloggers on social media have also pointed out that OPPO is not making full use of the chip.
As such, Mariana X seems to be becoming a marketing gimmick for OPPO, which most people don't want to see. After all, consumers have paid for the real money of the Mariana X chip, and what they want is tangible results. If OPPO spends too much energy on publicity and marketing, it will seem a bit upside down.
In fact, all self-developed chips inevitably have various problems in the early stage, and even the industry giant Apple has to face such problems. Due to the change in chip architecture, the initial compatibility of the M1 chip to the application is very poor, and inefficiency, stuttering and even flashback occur from time to time. M1 has also been criticized for this.
Apple is still like this, let alone OPPO?
Even compared to the peak of the self-developed chip: Huawei Kirin 9000, Mariana X is too far behind. In terms of the number of transistors alone, Huawei's Kirin 9000 is 15.3 billion, and Mariana X is only 3.6 billion, which is less than a fraction of kirin 9000. In terms of process, the Kirin 9000 has adopted the TSMC 5nm process as early as two years ago, and the Mariana X is only 6nm, and its adaptation is even more incomparable.
In contrast, Mariana X really has little capital to brag about.
Apple and Huawei have decades of hard work in the field of self-developed chips, starting from the Apple A4 chip in 2007 and taking 15 years to have today's Apple A15. Huawei HiSilicon's first chip, K3V1, was released as early as 2009 and has been more than 10 years now. Its process technology has gradually evolved from the original backward 110nm to today's 5nm. This is all the result of down-to-earth investment and research and development.
Marketing has always been OPPO's best play, as early as 2019, OPPO's marketing expenses reached nearly 2 billion yuan, about more than 300 million US dollars, and various popular star endorsements are wave after wave. But it is clear that for chips, research and development is the focus, excessive publicity will only make the chip a marketing gimmick, become a roadblock on the road of self-developed chips.
We certainly can't deny what the Mariana X chip has achieved at this stage, but OPPO's recent marketing seems to be a bit off track, which is a more dangerous signal.
The effect of marketing can be seen in a short period of time, but chip research and development is a long-term and arduous project, and there are almost no shortcuts to take, and there is no so-called cornering overtaking. If domestic manufacturers want to really make achievements in the field of chips, they can only guard against arrogance and impetuosity, and adhere to investment and research and development.
In contrast, Huawei, according to reliable information, the research and development cost of the Kirin 980 chip is 300 million US dollars, and the total cost is more than 500 million US dollars.
At the recent Huawei earnings conference, Huawei executives announced the research and development expenses spent by Huawei in 2021, and in the case of a decline in revenue business, last year's research and development expenses reached 142.7 billion yuan, a record high. In the period from 2012 to 2021, Huawei's cumulative investment in research and development expenses reached 845.6 billion yuan. It is second only to Google in terms of R&D expenses.
From these astronomical figures, it is not difficult to see how much Huawei attaches importance to chip research and development. However, Huawei has not become complacent and over-marketed because of this. Even after being sanctioned, he still tried his best to insist on self-developed chips, and this down-to-earth approach is indeed worth learning from other friends.
In fact, the lack of continuous research and development investment in self-developed chips and even the problem of excessive marketing have been particularly obvious in recent years, not only OPPO, xiaomi, vivo seem to have similar problems. Xiaomi has not had the following since the launch of the surging S1, and although it has launched two chips, C1 and P1, the latter two belong to different types of chips, which always give people the feeling of a shot for a shot, and more seems to be serving the marketing.
Our self-developed chips have never lacked heat and marketing gimmicks, but we really don't want self-developed chips to become marketing tools for manufacturers. What we want to see is a radical transformation brought about by continuous deep research and development.
Tan Zhaofei