It's no secret that AMD has had trouble competing with Nvidia in PC graphics in recent years. It is worth taking a look at the situation in more depth, especially with new console cycles on the horizon and a new array of matrix shrinkage: a contraction that is already being used for the next generation of 7nm Vega dedicated to artificial intelligence of AMD. also used for future Nvidia products (if that includes the next-generation GPU of the company is not something we know yet).
In Forbes, Jason Evangelho compiled some recent executive comments as well as a general description of AMD's GPU releases in recent years. David Wang, for example, commented that AMD had lost some momentum with players by pursuing artificial intelligence and machine learning markets. There are rumors of tension in AMD between focusing on courting explicitly semi-personalized customers like Sony and Microsoft, and focusing on the larger gaming market. Some of those tensions, and the need to get Ryzen out of the door, may have been responsible for Vega's less-than-stellar performance against Nvidia by diverting engineering resources from the GPU side of the equation.
I do not have any in-depth information to add to this, but it's true that AMD's GPD approach and gains were transformed beginning in 2013. It started with the launch of the Radeon HD 7790 – a GCN 1.1 GPU with a set of features similar to what would eventually debut on the Xbox One and PlayStation 4. Initially, AMD continued to pursue a competition strategy at the higher end, but Hawaii was the last AMD GPU to launch and unmistakably sweep the competition. After Hawaii, AMD's competitive positioning against Nvidia has generally been weaker.
Two aspects of AMD's global GPU strategy stand out: the continuous dependence and evolution of the GCN architecture, and the degree to which the new mid-range GPU versions have been aligned with semi-custom partner releases. Both AMD and Nvidia presented new architectures in 2012, but it seems that AMD has changed GCN less than the equivalent structural changes that Nvidia has made in its own architectures. Okay, AMD has not updated its chips as often as Nvidia did during the same time period, but that, in itself, is part of the problem.
Regarding the alignment update question, Evangelho states that both the RX 460 and Vega 56 were essentially commissioned by Apple, with the wider consumer market as a secondary feature. That's also true for the RX 400 family as a whole, which debuted on the desktop a few months before the PS4 Pro hit the market (the Xbox One X GPU, which came on the market a year later, is also derived from Polaris). Navi may reflect the work done for PlayStation 5. These GPUs work well for the medium and medium high range, but AMD has had trouble challenging Nvidia across the GPU stack.
Part of the problem, I suspect, is that AMD is a company with limited funds that has been extracted in a variety of directions. Taking into account that it can take 3 years to move a GPU from the first idea to the finished product; the products that went on the market in 2018 and 2019 were designed when AMD still operated in the red. HBM2 had known performance and ramp problems, which has undoubtedly impacted Vega's costs. The Artificial Intelligence and Machine Learning markets have been red-hot, and AMD is one of the few companies positioned to even take a hypothetical crack in that market (which is why the company decided to go ahead with a 7nm Vega GPU for that space). And the offers from Sony and Microsoft surely saved AMD as a going concern in 2013-2014. If the company had been left solely dependent on its x86 graphics and PC business, it might not have survived at all.
Now, think of the fact that the conventional PC graphics business, including professional and workstation products, AMD never gained much in the way of profit. Next, we have charted AMD's GPU operating revenues since 2008 (the first year the segment had revenues after the acquisition of ATI) until the first half of 2013, when AMD combined GPU revenues with console sales. .
More than five and a half years ago, AMD recorded $ 7,673B in GPU sales and operating income of $ 383 million, for a five percent operating margin . That is not a big thing. In fact, it is markedly worse than Nvidia's profit margin for consumers in GPUs same period, even though AMD's figures include its professional sales of GPU and Nvidia's. In short, manufacturing GPUs and selling them in the PC market was never something that made AMD gain many benefits.
The business segment EESC (Enterprise, Embedded and Semicustom) of AMD has been predominantly dominated by console sales from 2013 until mid-2017, when the ramp Epyc began incurring some additional costs that may have diluted the positive impact of the console's business margin. Even so, this chart covers the relevant time period, and AMD's absolute earnings and operating margins have been better than in the GPU business alone, even though the console margins are not really that good. On the other hand, artificial intelligence and machine learning margins are generally incredible, which sheds some light on why AMD wants to enter this space.
AMD and PC Gaming
I am convinced, based on conversations I & # 39; specifically, I have had several AMDers, that the company is absolutely concerned about games in general and PC games. But AMD's GPU divisions have also been building many parts throughout the market to meet the needs of multiple customers, and doing all of this without a ton of money for everyone. I also believe that it is true that the company has chosen to prioritize the needs of certain customers and align their GPU releases on the consumer side to fit those plans.
But here's the thing: I'm also not sure that AMD could have survived otherwise. And although the company has once again become profitable, it is hardly out of danger. AMD needs to navigate the 7nm transition, launch its next generation of Zen processors, continue to accelerate Epyc, launch new consumer GPUs and have $ 190M in debt maturity in 2019. None of this represents any kind of existential threat, but some Quarters in black does not mean that the company can afford to pretend that it has Apple's cash levels in the bank, either. Meanwhile, you will also need to invest in the AI / ML clients that you now court with Vega, those GPUs will not be worth it if AMD's tools and software can not compete with what Nvidia and Intel are taking advantage of in the same space. .
I think AMD can balance the demands of the semi-custom, consumer and AI / machine learning market. But I think it's going to need some sustained success in the longer term before we stop seeing some signs of tension as the company pivots from one market to another. In this, the situation is similar to Zen. While the core of the Ryzen CPU is excellent in general, there are some weak points where AMD could not improve the situation further by time and money. Certain database benchmarks that benefit from large L3 caches, or that create large amounts of cross-traffic, do not perform as well in Epyc precisely because they are not so favorable to your architecture. A specific variant of the Zen kernel server with a larger L3 cache on the die or faster Infinity Fabric links might have been useful for AMD in these cases, but that kind of flexibility was not possible given the circumstances. AMD uses the same Ryzen matrix from Ryzen 3 to Ryzen 7 for the same reason: personalized work would have saved some money by allowing more efficient utilization of the wafers, but the initial cost and time were obviously major concerns.  The good news is that if AMD continues to be successful with its current products, it can address these other problems more effectively in the long term. Whether you choose or not, that is a more complicated question. Answering it requires more information about where AMD is putting its R & D dollars today than what we have.
(Credit of the image above: AMD Zen Core / AMD)