Opinion: AMD's gamble now paying off

Bob O'Donnell

Posts: 81   +1
Staff member
Why it matters: For a company that just a few years ago some people had essentially written off as dead, AMD has certainly come a long way. Not only are they one of the top ten best performing stocks of 2019 so far (after enjoying similarly high rankings for all of 2018), the company recently announced major new partnerships with the who’s who of big tech companies.

Microsoft for the next generation Xbox One game console (Project Scarlett), Google for their Stadia cloud-based game streaming service, Sony for the PlayStation 5 game console, Apple for the new MacPro, and Samsung for GPU IP intended to power future generations of Galaxy smartphones and tablets.

On top of that, the company just launched its latest generation desktop CPUs, the Ryzen 3000 series at Computex two weeks ago, and yesterday at E3, debuted its newest Radeon GPU cards, codenamed Navi, that are based on a new GPU Architecture the company calls RDNA (short for Radeon DNA). The first commercially available products from the Navi effort are the 5700 line of desktop GPU cards, designed specifically for the gaming market. Also, in a nod to the importance of CPUs in gaming, the company added a new top-end addition to its 3rd generation Ryzen line: the 16-core, 32-thread capable Ryzen 9 3950X.

All told, it’s a broad and impressive range of offerings, and it’s tied together by a few critical decisions the company leaders made several years back. Specifically, AMD decided to aggressively go after leading-edge 7nm process architecture for both CPUs and GPUs and, importantly, chose to pursue a chiplet strategy. With chiplets, different components of a finished chip, made with different process technologies, could be tied together over a high-speed connection (AMD dubbed theirs Infinity Fabric) instead of trying to put everything together on one monolithic die. Together, these technology bets enabled the company to reach a point where it’s starting to do something that many thought was unthinkable: challenge Intel on CPU performance and challenge Nvidia on GPU performance. While final numbers and testing still need to be done before official winners are declared, it’s clear that AMD can now be considered in the elite tier of performance in the most important semiconductor markets, particularly for CPUs. In the GPU space, AMD chose not to compare its new 5700XT to Nvidia’s highest performance GeForce RTX 2080 and RTX 2080 Ti cards, but given the aggressive $449 pricing of the new AMD 5700XT, that certainly makes sense. (AMD is quick to point out that Apple claimed the new AMD Radeon Pro Vega II Duo powered multi-GPU cards in the Mac Pro are the fastest desktop GPUs in the world, but they’re really more of a workstation product.)

It’s great for both consumers and businesses to see a truly rejuvenated AMD, because it inevitably forces all of its competitors to get better, which in turn, leads to better products from all the major players

The momentum that AMD is currently enjoying is clearly due, in part, to those big technology bets, particularly around 7nm, as well as the fact that they are one of a few major semiconductor players with significant CPU and GPU assets. Again, many industry observers questioned that strategy for a long time, but now that the company is starting to leverage technologies from one side to the other and is really integrating its approach across what, admittedly, used to be two very distinct groups, the payoffs are starting to happen. In addition, the coordinated efforts are allowing them to do things like be the first company to integrate PCIe 4.0 across both CPUs and GPUs, as they've done with the latest Ryzen and Radeon products, as well as leveraging Infinity Fabric for both CPU-to-CPU connections (in the Ryzen line) and GPU-to-GPU connections (in the Pro Vega II inside the Mac Pro).

The company’s vision is now broader, however, as it’s started to reach into the server and datacenter market with its Epyc CPUs and Instinct GPUs, even launching what it claims will be the world’s fastest supercomputer in conjunction with the Oak Ridge National Laboratory. The overall Epyc and Instinct market share numbers are still small, and the cloud and datacenter markets are still generally very loyal to Intel and Nvidia, but the fact that AMD is back to being able to compete at all in the server market once again highlights the relevance of its core technology decisions. In addition, though it’s early, AMD’s newly announced partnership with Samsung could finally help the company make an impact on the mobile market—where they have been completely absent. With growing interest in cloud-based game streaming, we could even end up seeing AMD technology in the cloud talking to AMD technology in mobile devices, which is quite a stretch from where they've been.

In the end, it’s great for both consumers and businesses to see a truly rejuvenated AMD, because it inevitably forces all of its competitors to get better, which in turn, leads to better products from all the major players, as well as a more dynamic computing market. To be clear, AMD still needs to execute on the broad vision that it has laid out for itself—and unfortunately, execution issues have slowed the company in the past. However it’s encouraging to see some key strategies driving new opportunities, and it will be interesting to see what AMD is able to do with them as we head into the future.

Bob O’Donnell is the founder and chief analyst of TECHnalysis Research, LLC a technology consulting and market research firm. You can follow him on Twitter . This article was originally published on Tech.pinions.

Permalink to story.

 
Just to add to all this, I'm excited that AMD is doing better as well. I haven't been excited about AMD processors for about 15 years. I'm glad they're back to being competitive, if not better. Intel had too long of a run of really minor performance increases... and now I'm ready to replace my Ivy Bridge CPU with probably the Ryzen 3.
 
So far everything looks great except GPU's TDP and honestly I dont understand how a big company like AMD cannot make a GPU architect with low TDP
 
At the moment, AMD looks to be definitely ahead. But it may only be that way for a short time. Intel is apparently planning to include AVX-512 support on its consumer 10nm chips, and that will once again put them ahead of AMD by the same margin, about, as was the case with the previous Ryzen generation. So I find this good and exciting news, all right, but I'm not optimistic about AMD capturing a big lead over Intel; I think they'll still be needing to catch up instead.

Apparently I'm wrong. This article
https://www.digitaltrends.com/computing/intel-10nm-dead-leaked-roadmap/
says Intel won't have 10nm full desktop parts until late 2020 or even 2021. If that's right, AMD will keep leadership for a while.
 
Last edited:
At the moment, AMD looks to be definitely ahead. But it may only be that way for a short time. Intel is apparently planning to include AVX-512 support on its consumer 10nm chips

Even in the professional segment, AVX software ecosystem is thin; I wouldn't use it as an argument. Also, there's usually a lower-than-base clock frequency when running AVX-512 workloads.
 
At the moment, AMD looks to be definitely ahead. But it may only be that way for a short time. Intel is apparently planning to include AVX-512 support on its consumer 10nm chips, and that will once again put them ahead of AMD by the same margin, about, as was the case with the previous Ryzen generation. So I find this good and exciting news, all right, but I'm not optimistic about AMD capturing a big lead over Intel; I think they'll still be needing to catch up instead.

AVX 512 is only supported in a handful of professional applications and from those I see a 10 - 15% performance improvement over AVX 256. I'm willing to bet the much superior core count of the AMD CPUs will completely erase any potential lead those Intel processors might have even in those niche workloads. Mind you, that's not considering the downclock Intel processors will have to take to run those workloads, which is likely to be more severe on the more demanding 512 workloads. AMD processors do not underclock when using AVX.
 
Last edited:
One area where AMD needs to gain traction is in mobile computing: notebooks and tablets. It's happening, but very slowly. I'm only seeing it being marketed in low end consumer notebooks, and not in ultralights or convertibles (where Intel CPUs dominate).
 
One area where AMD needs to gain traction is in mobile computing: notebooks and tablets. It's happening, but very slowly. I'm only seeing it being marketed in low end consumer notebooks, and not in ultralights or convertibles (where Intel CPUs dominate).

GPD Win Max will be using Ryzen 3000, seriously looking forward to it, too bad it's only 12+nm though, guess mobile sector is of lesser importance for AMD for now... hope they gonna introduce super low TDP 7nm APU parts soon.
 
These announcements are no more than marketing until July 7th, but that won't stop hundreds of more articles and Youtube videos being made about how great these paper numbers provided by a manufacturer look.
As I see it, this article is an opinion piece.

As well, the market served by TechSpot is only part of AMD's market. They are gaining a foothold in the substantially more important and lucrative server server market. AMD's GPUs might not be gaming GPUs, but they are far faster in compute than the competition, and that fact, along with their superior multi-threaded CPU performance, is probably what landed the Supercomputer contract.

Not to mention the game console contract and the Samsung contract.

At this point, I don't think that those contracts are going to be affected by gaming benchmarks that will appear in July. About the only thing that might kill things is if those benchmarks were a total dud. I think we all know that is not going to happen.

So, whether they are good enough for you, or for me, for that matter, does not really matter at this point.

Perhaps the best aspect of AMD and Lisa Su's leadership is they have brought competition back to the CPU market. I think we will see that happen in the GPU market at some point.
 
Just to add to all this, I'm excited that AMD is doing better as well. I haven't been excited about AMD processors for about 15 years. I'm glad they're back to being competitive, if not better. Intel had too long of a run of really minor performance increases... and now I'm ready to replace my Ivy Bridge CPU with probably the Ryzen 3.
I am considering a similar upgrade from an Ivy Bridge 2011 socket proc.
 
So far everything looks great except GPU's TDP and honestly I dont understand how a big company like AMD cannot make a GPU architect with low TDP
AMD is smaller than Nvidia, and I'm fairly certain most of their R&D budget for the past few years has gone to CPUs, not GPUs. Also, since power efficiency is not a big concern for most PC gamers, AMD pushes their GPUs past the optimal frequencies even at stock, and with fairly high voltages as well(I've seen multiple people talk about how they've undervolted Vega56 and actually gained both power efficiency AND performance), which makes their cards perform better, but at a big cost to power efficiency.

Mind you, Nvidia's is still the more efficient architecture(probably by a significant margin, given that they're still on an older, less power efficient process), but AMD clocking their desktop cards aggressively makes the difference seem bigger than it really is.
 
So far everything looks great except GPU's TDP and honestly I dont understand how a big company like AMD cannot make a GPU architect with low TDP

The Navi’s will have half the power usage as the Vega at the same performance, as stated by Lisa Su herself. That’s impressive. Sure you will have higher draw at higher performance but it will still be a lot better than Vega. Also if you are really worried wait for Navi+ as that is the entirely new non GCN architecture. I doubt I’ll get Navi as I’d wait and see what Nvidia’s Ampere offers and Intel has cards coming. I think my GTX 1080 Ti will last me very nicely for many years to come. But if Navi really is faster than RTX 2070 for similar power draw and cheaper I may replace my older 1070.
 
How much ... OK I won't say it because it's against the rules but you all know what I would be saying.
 
So far everything looks great except GPU's TDP and honestly I dont understand how a big company like AMD cannot make a GPU architect with low TDP
One big reason: AMD could not build (at least until Navi) GPUs specifically for games. They had to do compute and gaming on the same GPU. Nvidia's R&D budget is much bigger and is mainly focused on GPUs, while AMD is making both CPUs and GPUs.
 
Just to add to all this, I'm excited that AMD is doing better as well. I haven't been excited about AMD processors for about 15 years. I'm glad they're back to being competitive, if not better. Intel had too long of a run of really minor performance increases... and now I'm ready to replace my Ivy Bridge CPU with probably the Ryzen 3.

Likewise, I will more than likely replace my 3770k with a Ryzen 3.
 
These announcements are no more than marketing until July 7th, but that won't stop hundreds of more articles and Youtube videos being made about how great these paper numbers provided by a manufacturer look.
Well, then since you are a fan of facts and benchmarks, you should keep your silence until they hit. Then you can blast whatever side you fell appropriate :)
Also, we do know some results at 5Ghz all core LN2, for example Cinebench R20, Geekbench and few others - you can extrapolate since they scale well with clocks. Only real gaming benchmarks are yet to be conducted (that we know of).
 
Sound engineering is not a gamble. Neither was hiring Jim Keller. The writing was on the wall since 2012.
Agreed. In my opinion, AMD have made better decisions than Intel over the past few years - fab scaling and associated costs are becoming increasingly more challenging for the current and near future economic climate. A multichip module, utilising different fab companies and process nodes, was the obvious choice - AMD just needed to make their bus design work. Intel have stuck to their monolithic, "just make it longer" design for far too long and heavily gambled on having it all on their 10 nm node, with good yields across the range. The challenge for AMD is to maintain the momentum.
 
While I might agree regarding Ryzen, I dissent regarding Navi. It seems to me that all the improvements announced are based merely in the 7nm process instead on improvements on the architecture. And they consume significantly more than Nvidia cards for the same performance.

May be they should send the threadripper engineers to the radeon division and prioritize.
 
While I might agree regarding Ryzen, I dissent regarding Navi. It seems to me that all the improvements announced are based merely in the 7nm process instead on improvements on the architecture. And they consume significantly more than Nvidia cards for the same performance.

May be they should send the threadripper engineers to the radeon division and prioritize.
Navi definitely made big arhitectural changes. It has a very forward thinking design that can net devs huge performance gains if they make AMD specific optimisations while also keeping complete compatibility with the code written for GCN. This is why next gen consoles are going to have great backwards compatibility and still be able to get big gains in perf.
This video covers some of them:
 
Back