AMD's Role in High Performance Computing

Over the last years, a huge hype has been built around AMDs desktop Ryzen CPUs, and more recently their mobile chips proved to deliver outstanding performance at low energy usage as well. It is remarkable what the company has achieved in just a couple of years, going from an architecture worse than what Intel had in 2010 to being a market leader. With Intel’s recent fabrication problems, many are concerned that AMD might become the new Intel, charging ridiculous prices for their products, while others remain skeptical that the company will be able to sustain their lead over the blue giant in the long term. However, I think it is unlikely that either of those options will become reality in the near future, but that AMD will rather establish itself as a true market leader in the industry.

To understand why this might happen, it is important to realize where the company stands right now. Their Ryzen line-up of chips is a lot more energy efficient as well as optimized for heavily multi-threaded workloads than anything Intel has to offer but is also manufactured on a much more modern process node. The new chiplet design allows them to keep cost low, but because yields for Intel’s 14Nm are so high now, it is still cheaper for Intel to make a comparable chip.

On the graphics side of things, AMD has now reached the same efficiency level as NVIDIA with their RDNA-architecture, which is a huge step after multiple minor refreshes of GCN. But, just like with their CPUs, AMD has a significant process node advantage over NVIDIA, so many Analysts rightfully claim that if both products were manufactured on the same node, AMDs products would be inferior. The Radeon Cards also miss out on some truly unique features like DLSS and hardware-accelerated Raytracing, which will play a significant role in future games.

So, if AMD can barely achieve parody with their competitors on a much more advanced node, how could one think they are the future of computing? First of all, all the products I have mentioned so far were released around mid 2019. This year, AMD won’t use a more advanced process node, but they will instead improve their architecture to overtake their competitors. While it is likely that NVIDIA will still be able to match AMDs upcoming line-up in performance with their Ampere cards, they will have to use much larger dies resulting in much higher power consumption. NVIDIA simply won’t be willing to jump on the most cutting-edge process node like AMD, in order to keep their profit margins high and their investors happy. So, while the dedicated GPUs of the two companies might perform the same, AMD might offer the more power-efficient product, giving OEMs more flexibility in how they want to design their cards. Raytracing will also finally be added, so despite DLSS there should be feature parody between the two companies, and NVIDIA might have to seriously re-consider their future plans when AMD brings out their 5Nm GPUs one generation later.

With Ryzen, AMD is in an even better position right now. Even if Intel was able to bring out Icelake-based desktop CPUs this year (which they certainly won’t), Icelake would only have an IPC about 20% higher than their current chips, while having to be clocked much lower. In total, I would expect around 15% more single core performance than with coffee lake, so Ice-Lake could at best match AMDs Ryzen 4000 line-up coming this year, given that those chips don’t clock much higher than Ryzen 3000. More importantly, though, 10Nm chips will cost Intel much more to manufacture than their current ones do. I simply cannot imagine Intel to significantly reduce their profit on CPUs, considering their current ones are so cheap that they could easily undercut Ryzen in price. Also, by the time those 10Nm chips finally come out, AMDs Ryzen 5000 products manufactured on 5Nm will be right around the corner, leaving Intel completely in the dust. So, both in stand-alone gaming CPUs and GPUs, AMD seems to be well positioned for the future.

The most significant advantage AMD has over their competitors are not their CPUs or GPUs, though, it’s the scalability of their entire line-up. RDNA2 will most likely be used in Samsung’s own Exynos chips from next year on, and this finally gives developers the opportunity to optimize their games for all mobile, console and desktop devices at the same time. Sooner or later, they will make great use of the strength of AMD’s architectures, knowing that these tweaks will be beneficial for all platforms at once. Optimization is often much more powerful than any new chip, and the best example is the Xbox One. The console from 2013 is still able to run all modern AAA-titles on a chip not much more powerful than the ones used in our smartphones, while using extremely-low-bandwidth memory, all thanks to optimization. Of course, both NVIDIA and Intel will utilize their enormous cash to convince developers otherwise, but AMDs approach will eventually win.

In other words, it is very likely that the company will be able to power the majority of the gaming market on all mobile, PC and console, and their architectures and ecosystem will therefore become the standard in the near future. But what about other applications, you may ask. Gaming and high-performance computing in servers and workstations is only one part of the entire semiconductor industry, right?

Well, AMD has developed great products for the normal consumer as well. For a person who only surfes the web and does some occasional office work, the processing power of a device is not all that important. Instead, these customers will care about factors like long battery life, a slim design and no disturbing fan noise. Since AMDs newest processors are a lot more efficient than anything their competitors have to offer, while integrating both a moderately powerful CPU and GPU onto a single die, many OEMs will choose them over Intel, while potentially saving costs. As a result, sooner or later Ryzen-based chips will also be used in many ultra-books, too.

Finally, there is another group of customers, who are not interested in gaming but who require a lot of computational power for tasks like photo- or video editing or animation work. And a large majority of them rely on one company to provide them with the tools they need for getting their work done: Apple. In the recent months, Apple has proven that it truly cares about its creative audience, and that it will continue to develop high-performance (high cost) products for them. Apple can sustain their position, because a lot of the professional software simply is not available on other platforms. Currently, Apple is using AMD graphics, but sticks to Intel for their CPUs, and many sources point out that they might move towards their own, custom Arm-based chips once this situation changes. Apple has invested heavily in their custom silicon, which raises the question how AMD might make their way into future Macs.

The problem is, that even Apples’ strongest Arm cores won’t be able to match x86 performance in the near future, even though the company often claims they do. Arm was never an architecture designed to rival Intel or AMD in performance, but to beat them in power efficiency. While Apple may be able to equip some of their lower-power tiers with their own chips, it will still take some time until they are ready to give the creators an Arm-based machine. If they want to remain competitive in this space, they will have to use Ryzen for their high-end macs, which also has the appeal of more powerful integrated graphics to them.

What many don’t know, however, is that AMD themselves have an Arm-license, allowing them to use the architecture in their own product. If Apple truly wanted to differentiate themselves from their Windows competitors, they would have to come up with a truly heterogenous architecture, which consists of many small, power-efficient Arm-cores and a few x86-cores working together. While Intel tries something similar with their Lakefield-series of chips, of what we know so far, Lakefield is far from impressive considering the radical changes in architecture. Meanwhile, AMD does not yet have the recourses to work on such a huge project, and Apple will not get an x86-license in the near future. So, while AMD might try to develop a heterogeneous chip someday, the only way for this to happen in the near future is AMD and Apple working closely together.

While it is more than unclear if Apple and AMD have such plans, they could very well consider it if Ryzen turns out to play nicely with MacOS, which they will have to try for their high-end products in order to remain competitive. Either way, AMD processors and graphics will be implemented in Macs, and it will have to be seen how long they will stay.

Overall, it is amazing to see how broad AMDs future plans are designed, covering almost every field in the industry. This becomes especially clear when looking at the situation of the company just five years ago, when many believed it to run insolvent soon. Hopefully, their competitors do not make the same mistakes as they did in the past and remain somewhat competitive in order to keep prices low. As for now, AMD will have to capture back a considerable slice of the market before they will risk raising prices in a way Intel and NVidia once did.