The move is not an official split into a separate company just yet, but it was likely the first steps in doing so. Before the final move AMD would want to gauge the ability of the group to work and also see if there is serious investor interest in the GPU business as a stand-alone company (there is). Now the RTG team is working very hard to get the word out about their GPUs and the technology they plan to drop into them.
So far we have seen the first generation of GPU leveraging 2.5D HBM (High Bandwidth Memory). These cards have fairly solid performance according to most reviews, but are not really the game changer that AMD had hoped for. This has not stopped AMD and they are back at work on the next generation GPUs that they hope to launch before the end of Q2 2016. Most of these will feature GDDR5 which can increase the amount of memory available while keeping the performance relatively high.
HBM will still be in the mix though and we expect it to come out on AMD’s high-end cards in the form of HBM 2.0. The new generation of HBM will allow for greater memory densities and also higher bandwidth per chip. This means that AMD will have to ensure that their cards can handle that amount and speed of memory. In the Fury X there were a few areas where the memory was underutilized due to limitations of the GPU architecture. Still as a first generation product goes it was pretty spot on. The problem ended up being that NVIDIA had a card that matched it almost perfectly for performance and did not use the more costly HBM.
As GDDR5 has gone through some impressive changes since its introduction in 2007 we expect this memory to continue to feature in all but the top-end cards. After all with densities of up to 512MB per chip available now and 1GB chips rumored to be out soon AMD could drop in a large amount of memory onto their mid-range and even low end cards without push prices too high. This will let them reserve HBM 2.0 for the top-end and optimize those cards for its use. They will still have competition though as NVIDIA is planning to use HBM 2.0 on their Pascal GPUs which should also launch in 2016.
The RTG team has a lot of work to do to recapture the market. They not only need to have better concepts and ideas, but also need to execute them more efficiently and faster than NVIDIA. Now that they are no longer hindered by the CPU side of the house they just might be able to pull this off. When ATi was still around they were able to go toe to toe with NVIDIA for a long time. It was not until NVIDIA bought 3dfx and their SLI core technology that NVIDIA really pulled away. ATI had no competing technology and the first couple of generations of Crossfire were certainly not up to the task.
Still the RTG team has learned a lot over the past few years and should have the talent to pull this off and we have a feeling that they have been directed to do so. The information coming out from them seems to indicate that they also working hard to get more information about direction and the technology they plan to implement. This is always a good marketing move, but does not always work. If you have a history of good and timely execution on ideas and concepts then dropping a hint from time to time will yield results. If your history is more checkered… well people might not want to wait to see if you got things right this time. Again this also plays into the move from being AMD to now being the Radeon Technologies Group. It helps to breathe new life into the brand and also affords the possibility that the market will give a second chance to prove the Radeon group can execute.
Things could get really interesting in 2016 for the GPU market.