A couple of seemingly unrelated things happened in the past week. First, AMD's new chief of the Radeon Technologies Group, David Wang, commented at Computex that the company lost their momentum with gamers chasing AI. Then an article at The Information was published in which AMD executive Forrest Norrod took some shots at former RTG chief Raja Koduri -- who landed at competitor Intel -- accusing Koduri of "wastage on the GPU roadmap" during his tenure.
Viewed one way, this is simply AMD "righting the ship" under new management. Viewed another way, it's setting up Raja Koduri -- someone who can't legally defend himself at this point -- as the fall guy for AMD's weak graphics roadmap and loss of a competitive edge against graphics giant Nvidia.
The timing for these comments is curious given Koduri's recent departure to Intel, which was followed by former Ryzen architect Jim Keller landing at Intel. Weeks later AMD's former global marketing director Chris Hook joined them. While not confirmed officially, I have it on good authority that Intel will enter the dedicated gaming GPU market (again) by 2020, which has been widely reported. Curious timing indeed.
UPDATE: Intel just confirmed it.
Since then I've spoken to sources in the industry (speaking under conditions of anonymity to protect their careers) about AMD's position in the consumer graphics space and its motivations behind designing graphics architectures and products.
Let me introduce these findings with questions: Is AMD really a graphics company? Is AMD really a PC company? Or are they instead rooted and driven by semi-custom designs, such as the ones seen in the Xbox One X, Sony PlayStation 4 Pro and Apple MacBook Pro? Have these in the back of your mind as you continue reading.
Polaris, Vega And Navi: Built For Customers, Adapted For Consumers
The Radeon RX 460 graphics card for gamers launched in June 2016, but that part was allegedly built from the ground up for one specific company: Apple. While the 2016 MacBook Pro with its Radeon Pro 460 graphics chip (and lesser variants) was announced a few months later, my sources tell me that part was designed with a low-power envelope exclusively for the MacBook Pro, and that the reason it exists at all is because of Apple.
AMD marketed the desktop offspring as primarily targeted at eSports and entry-level gamers.
Speaking of low-power envelopes, that's also what AMD built under Apple's direction for the newer iMac Pro, which features a Vega Pro 56 GPU and was unveiled in June 2017. AMD later launched RX Vega for gamers in August of 2017. Anyone who studies the performance, noise and heat outputs of desktop RX Vega knows that once you crank up the clocks things start to fall apart. That's because Vega wasn't built for gamers, or even for the AI and machine learning industries.
It was built for Apple because that's where the money is.
Don't even get me started about AMD's best integrated graphics solution in history, which rests under the banner of Intel's fantastic Kaby Lake-G.
*Continued on Page 2*
Polaris and Xbox One X
At E3 in June 2016, Microsoft announced an upgrade to its Xbox One, then called Project Scorpio and later the formally-named Xbox One X. It featured a semi-custom solution built largely on AMD's Polaris architecture. The Polaris-based Radeon RX 400 series launched that same month.
But Polaris was built for consoles because that's where the money is. Low margins but consistent high-volume orders.
While we'll never know exact figures, there's no question that sales of its semi-custom chips to massive platform holders like Sony and Microsoft far outstrip sales of its desktop RX GPUs. It's not a stretch that sales of Vega chips in Apple's Mac Pro will exceed sales of Radeon RX Vega 64 and RX Vega 56. The same argument can be made for Apple's MacBook Pro lineup with its series of Radeon Pro 400 and 500 GPUs.
Navi Was Built For Sony
Some new rumors popped up today about AMD's Navi powering the PlayStation 5. That makes sense, and according to my sources are factual. Another rumor popped up claiming that Navi won't be competitive against Nvidia's high-end consumer graphics cards and will instead be positioned as another mid-range competitor to replace AMD's Radeon 500 series.
While I can't corroborate that particular rumor, it too makes sense once you zoom out and view the entire picture. My sources tell me that not only was Navi expressly designed for Sony, but a staggering 2/3 of the engineering team under RTG's Raja Koduri was dedicated to it against his wishes, as he wanted to focus on gamers. The unspoken implication being that AMD CEO Dr. Lisa Su, coming from a semi-custom background at Freescale Semiconductor, made and enforced that decision.
Allegedly many cycles and resources were spent troubleshooting Navi and getting things perfect for Sony. Meanwhile the Vega desktop line suffered, and Navi for desktop lost traction. Beyond that, one source says that even Ryzen further depleted Koduri's engineering resources, though I can't substantiate that claim.
David vs Goliath...vs Goliath
That's a lot to unpack and it only raises further questions. The evidence paints a picture of AMD designing graphics products and architectures with its semi-customs clients first, and later adapting them for the PC gaming market. A practice that's been in play for the last several years at least.
AMD has multi-generational semi-custom product wins with Apple, Sony and Microsoft and those are worthy of celebration and continued resource allocation. There's nothing inherently wrong with this approach other than the obfuscation AMD seems to hide behind.
AMD is eager to steer its marketing around concepts like Radeon being the #1 gaming platform -- once consoles are taken into account. That's the only way it can win against Nvidia, aside from its excellent and widely-adopted FreeSync brand. It is eager to promote architectures like Polaris and Navi as being rooted in gaming.
And they are, just not PC gaming.
The truth may lie somewhere in between. These are products built for semi-custom clients, and then adapted for industries like datacenter, professional graphics and gaming. Typically we see professional and workstation products enter the pipeline first because they require less optimization on the driver front. This also gives engineers more time to raise clock speeds and extract the most amount of gaming performance possible.
But I think the most critical question is how long the Radeon brand as a desktop gaming product can survive. AMD has done an admirable job scraping back some market share from Nvidia with its Polaris and Vega GPUs, specifically if we look at Q4 2017. As reported by Jon Peddie Research, Radeon improved its market share by 6.5% quarter-over-quarter, but AMD's Dr. Lisa Su attributes 1/3 of that quarter's $140 million growth in the Computing and Graphics segment to cryptocurrency mining sales.
Those sales have peaked, AMD has no new GPUs to announce and they're staring down a future competing in the consumer graphics space not just against Nvidia (who is certainly primed to pull the trigger on their next-gen cards but has no competitive reason to) but also Intel. This could soon be a David vs Goliath vs Goliath tag-team battle against two companies with war chests and R&D budgets that dwarf their own.
Perhaps AMD is reticent to admit that, at least on the graphics side of the house, they are very much a semi-custom company first. But if the Radeon brand is to survive and thrive, it may want to figure out its identity crisis. It can be a competitor in desktop graphics, or it can be a successful leader in the semi-custom space. I don't believe it can be both.
I've reached out to AMD for any further comments or clarification and will update this article should a timely response be issued. If substantial, I will write a separate post.
