Rory Read and his new management team pretty much cleaned house after taking over at AMD. We've seen a striking number of big names and familiar faces leave the company in the past year and a half. More recently, though, the firm has been hiring new leadership to help complete its hoped-for corporate turnaround. The latest addition to AMD's revamped leadership bench is something of a homecoming: Raja Koduri is back.
Koduri worked at ATI prior to the merger with AMD, and he then rose to become the CTO of the Graphics Product Group at the combined company. Koduri has made a number of noteworthy contributions to real-time graphics over the years. While at S3, he invented the texture compression method that became the DXTC, or DirectX texture compression, standard. At ATI, he was involved in the skunkworks project that devised the original CrossFire scheme—using an external FPGA chip and a pass-through video cable—as a response to Nvidia's SLI. Later, at AMD, Khoduri helped create the custom filter antialiasing scheme that offered improved image quality via shader-based custom kernels—important because the R600 chip's ROPs didn't quite work out as planned. (Also notable because Nvidia's much-ballyooed TXAA works very much like CFAA.) After leaving AMD, Khoduri most recently worked at Apple, where he influenced that firm's graphics roadmap, a noteworthy role given the sheer amount of die space that company has dedicated to GPUs in its SoCs.
Koduri left Apple a little while back and didn't jump to AMD immediately, as I understand it. He now joins AMD as Vice President, Visual Computing, reporting directly to CTO and SVP Mark Papermaster. AMD's graphics hardware, multimedia, and software development efforts will be united under the umbrella of the Visual Computing group, which Koduri will head.
Late last week, I had the chance to talk with Raja about his new role, and he spoke briefly about his reasons for returning to AMD at a time of considerable challenges. One of those reasons is the chance to push the envelope in graphics. He told me he believes there's another "10 years of work or more left on the table" in GPUs, that "we know the pixel we want to get in real time."
Koduri also said he was motivated by the chance to move hardware and software together. He cited the typical lag between new features being implemented in chips and properly supported in software as an irritation, and said he looks forward to the chance to "shorten the cycle."
One thing Koduri emphasized in our talk was his—and by extension AMD's—enthusiasm for and commitment to high-end graphics solutions for gaming. AMD is very much focused on new business avenues, including the development of low-power SoCs for mobile devices and licensing opportunities like it's found with Sony and the Playstation 4, but one gets the sense that high-end Radeons will continue to feed both of those other businesses going forward. On the subject of game consoles, Raja said AMD sees "several such opportunities in other segments in the coming years," which fits with some hints we've heard from Papermaster and others. AMD must have other business lined up, beyond the Wii U, PS4, and the next Xbox. We don't yet know what it might be, though.
I asked a few questions about potentially important issues Koduri will confront in formulating AMD's future roadmaps—and we're probably talking three to four years before his influence is felt, given how long it takes to build a chip. He was cautiously willing to venture a few answers.
On the subject of whether to pursue separate graphics architectures for mobile SoCs and more traditional PCs, he said he's willing to consider using an older architecture for mobile devices if necessary, for the sake of efficiency, but wary of how such a split would increase the firm's software development workload. He said, for now, he'd "try hard" to keep the graphics architecture consistent from "top to bottom."
On the question of whether to develop mid-sized GPUs and avoid larger chips, like AMD did in the past several generations (though it fudged a little with Tahiti and its 384-bit memory interface), Koduri expressed flexibility. He said that if, in a given generation, a big GPU has the potential to solve a performance problem with a high-profile application (he cited Battlefield 4 as an example of a key app) then he might choose to "go for it." If not, then the tradeoffs may not be favorable.
Again, none of this says much about what AMD will do in the near term, but it is a quick look into the mind of one of the firm's top decision makers.
At any rate, an accomplished engineer and a veteran of ATI and Apple is taking the helm of AMD's graphics business at this critical point in its history. Koduri joins a list names—including CTO Mark Papermaster, parallel computing guru John Gustafson and CPU architect Jim Keller—stepping into key roles at the re-tooled AMD. We're curious to see what they build.