In recent weeks, we have repeatedly said that the recession will accelerate the transformation of technology, but it is not the root of the transformation. Now, we can see the competition of server CPU and GPU between AMD and Intel, as well as the hope of AMD and Intel to change NVIDIA's dominant position in the data center GPU market.

To be blunt, amd has never had a better chance than now to try to compete with Intel in the CPU market and try to weaken the attack it will bring to the GPU market. But recessions, both in the long and short term, make implementation more difficult because the process can be volatile.

Surprisingly, amd was not able to take more shipments and share from Intel overall, especially when the company reviewed its server spending plan in the first quarter and predicted 2020 and 2021. This allows Intel to keep in touch with customers to ensure that the data center division is again at a new high, as Intel did in the first quarter of this year, and again in the second half of 2019.

As far as we know, the large-scale service providers have made strict adjustments to the Xeon processor design, and the cloud service providers are worried that users may encounter similar problems when they bring their applications to the cloud they sell for operation. We don't think that's the case, but we're not sure. It's hard to believe that server buyers still doubt AMD's commitment to server CPUs and server GPUs. Amd has been involved for a long time and has 10 billion supercomputing capabilities to prove this.

Since the launch of "Naples" epyc 7001 processor in June 2017 and "Rome" epyc 7002 processor in August 2019, amd has been developing its own data center business for the past three years, but Intel's long-term success is unmatched.

After "sledgehammer" Opterons in 2003 and 2004, amd has been exerting pressure on Intel in an unprecedented way. Rome epycs can fight alongside the "cascade Lake" Intel Xeon sp. In terms of core and total throughput, amd more than doubled the number of cores, increased the number of memory controllers by another third, and increased the operation speed of PCI express I / O channel by 60%. At the same time, amd has a price advantage.

That's all we can think of. Intel's 14 nanometer technology for making cascade lake and cascade lake-r chips must be very mature, with high output, price and cost advantages. This also means that the 14nm output used for "Broadwell" Xeon E5 / E7 and "skylake" Xeon SP chips a few years ago is not as good as many of us think, because today's sharp price cut still can maintain the profit margin.

We don't know how much the price reduction will be, but if Intel can make up for the huge performance and price / performance difference of AMD epycs at the list price, the price reduction will be very steep. If so, we put forward some strange ideas in the article. If this does happen, we can assure you that this is not a strategy planned by Intel, but an inevitable one, because it has been striving for mass production of 10 nanometer manufacturing technology for many years. It runs counter to everything Intel has believed in since its inception.

Obviously, Intel has gained an absolute advantage in a very bad situation. To be honest, it's amazing.

In this case, it is not clear what price range AMD epyc can reduce, because the company has reduced costs by turning to small chips, and proposed radical pricing to obtain market share today. It uses TSMC's advanced 7Nm technology for Roman core microchips, but still uses the very mature 14nm technology from global foundries to manufacture the I / O and memory hub microchips of Roman packaging center.

Small chip package also increases the cost, but the net cost of processor slot is lower than that of single chip design. If this kind of chip package has a certain scale, it will become easier and cheaper with the passage of time, so that AMD has more computing power in this field and can obtain higher profit margin.

In the quarter to March, AMD's revenue increased by 40.4% to $1.79 billion, and its net profit increased more than 10 times to $162 million, accounting for 9.1% of its revenue, which is very healthy. As of the end of the third quarter, the company ended the quarter with $1.39 billion in cash and investment.

Sales in the computing and graphics division were $1.44 billion, up 73%, and revenues jumped 16.4 times to $262 million. Obviously, the client CPU and GPU services perform well. The sale of game consoles has always been a concern and must start to increase as soon as possible to meet the demand of the holiday shopping season.

Depending on the magnitude of the recession, new consoles may be a big selling point for household consumption, but it's hard to say. In the short term, the game console chip business in the embedded and semi customized segments of the enterprise dropped 21.1% to $348 million in the first quarter of 2020, and the segment released an operating loss of $26 million, compared with $68 million in the same period last year.

Amd earned $60 million in one-time licensing revenue in the same period last year, so the profit at that time was lower than expected, which also means that the operating loss of core business was not as bad as it seemed.

In a conference call with Wall Street analysts, amd CEO Lisa Su talked about data for the quarter, which was driven by data center sales (i.e. epyc CPU and radeon insight GPU sales). As a result, we will set this ratio at 18.5%, compared with about 15% in the previous quarters.

According to the simplified data, past trends and forecasts, we believe that epyc CPU brought us $252 million in sales in the first quarter, an increase of 2.2 times over the same period last year, and a growth of 9% over the previous quarter, which is inconceivable. We believe that the data center GPU sales (including the radeon and radeon insight cards manufactured by the computing and graphics departments) reached US $78 million in the first quarter, a little higher than the same period last year, a little lower than the fourth quarter of 2019.

Su said in a conference call that since the fourth quarter of 2019, server shipments have increased by two figures month on month, more than three times higher than the same period last year, which means amd has the upper hand in price / performance to capture share from Intel.

"We get growth momentum from cloud, enterprise, and HPC customers."

Su explained: "we saw the strength of cloud providers, introduced new instances and accelerated the current deployment. Microsoft azure, Google and IBM have all announced new products supported by the second generation of epyc processor, especially Google has released multiple general VMS, Microsoft has launched all amd virtual desktop products, including the radeon insight GPU. Several cloud providers have accelerated their infrastructure deployment to meet the growing demand of more and more users working and learning from home. For example, one of our large cloud customers can deploy 10000 second-generation epyc servers in less than 10 days to support the surge in demand for collaborative services. "

The huge sales impact on the large-scale producers and cloud builders is the decline in the average sales price. And we believe that even for corporate customers, there should be some price reduction. The good news is that the SMB (server message block, a protocol name) business of AMD's epyc server chip is not large, and the SMB business of the radeon insight GPU accelerator is not, so those companies most adversely affected by the coronavirus pandemic will not purchase these data center products in any case.

Another interesting anecdote, Su added, is that the majority of the growth in epyc CPU sales for large-scale producers and cloud computing builders is internal rather than public cloud building. Su said HPC workload sales were particularly strong across the enterprise.

Many of us are impatient with adopting the epyc CPU immediately and quickly, because it shows that the competition is extremely fierce. But from 2003 to 2008, it wasn't Opteron versus Xeon. Although it doesn't look like that, Intel isn't as vulnerable as it was then. Su is advising people to expect stable and positive earnings, rather than a substantial increase in earnings.

"From my point of view, this is very similar to the growth we see in the PC business." Su explained: "with the development of the server, the operation of the server has been stable. Every quarter, there are more qualified platforms, and they all increase. So, related to our expectations, it actually went well. Because it's related to cloud acceleration, I think we're satisfied with that. At present, we are not ready for growth figures. I think we've got a very radical growth assumption. "

Amd may be patient, but that doesn't mean the rest of us will. AMD is still expected to launch the "Milan" epyc 7003 processor with zen3 kernel by the end of 2020, and will launch a new rDNA GPU at the same time (it is expected to have the version of radeon insight).

We will soon see what the third round of fighting brings, but for at least the other two quarters, we are still in the second round of fighting. It's a very complicated battle, complicated, epic battle.