Business, Hardware

AMD Moves Trinity to First Quarter 2012, Pushes GlobalFoundries

On Tuesday, September 13th Thomas Seifert, CFO and former interim CEO of AMD spoke at the Deutsche Bank Technology Conference in Las Vegas, Nevada. Among other interesting stuff, he announced, that the follow-up to the current A-series APU (Llano) codenamed Trinity will launch "at the beginning of next year."

This places the launch of Trinity into the first quarter of 2012 and the launch could happen at CES 2012, which takes place on January 10-13. Previously it was assumed that Trinity would follow Llano only in the second quarter of next year with the launch at Computex Taipei 2012, June 5-9. Trinity has already been demonstrated in action on the Fusion Developer Summit in June 2011. It combines two Bulldozer modules delivering four integer cores with two shared FPUs with a VLIW4 GPU based on the Cayman-architecture (Radeon HD 6900 Series). While CPU performance should be roughly flat compared to Llano, the GPU should provide a 50% increase in performance. Just like Llano, Trinity will be manufactured at the 32nm SOI node at GlobalFoundries in Dresden, Germany.

AMD will also have a chip ready for the upcoming Ultrabook category, heavily touted by Intel. The chip will have half the TDP of current mobile Llano SKUs, which pits it right against Intels 17W Ultrabook CPUs. This chip was first announced at AMD’s last earnings call and Mr. Seifert mentioned it will be ready next year when Trinity launches. It should be expected that it outperforms Intel’s CPUs in graphics performance, but might have a hard time in CPU-bound applications.

Regarding manufacturing at GlobalFoundries, Seifert gave the following statement: "We have been pretty open in that we see room for improvement on the GlobalFoundries side, I think that is very much true."
He continued: "Performance is not where it needs to be and we are driving them very hard to where we need them to be in order to continue to grow this partnership."

It is pretty obvious that currently AMD is not very happy with the performance of GlobalFoundries. From one side, it is obvious that GlobalFoundries is dealing with the aftermath of the manufacturing tools issues that former management at AMD (Hector J. Ruiz and Dirk Meyer) caused – resulting in lot of delays for the 32nm node, but there is also management situation, as GlobalFoundries is facing a brain drain, such as changing five leaders of Dresden Campus in a span of just 18 months and recently losing their Head of Communications to Intel. 

Said problems are what drove AMD to change the way they pay them, moving from wafer based pricing to good die pricing. Mr. Seifert explained, this is only natural as their relationship with GlobalFoundries moves to a standard foundry relationship. He said AMD has a similar agreement in place with TSMC on both the APU and GPU side and that deal worked out extremely well.

Earlier we revealed that while Bulldozer yields appear to be fine, there are problems with yields on the Llano side of things. While the CFO did not explicitly mention Llano, he said that the Bulldozer ramp went ok, while there are "problems with more complex products, products where we have significantly more room to improve." Since Llano is the only other product manufactured on 32nm at GlobalFoundries and the fact that a GPU integrated into a CPU on SOI technology can be called quite complex, it doesn’t take a rocket scientist to figure that out.

However, having a GPU on SOI yielded some impressive power savings, as we reported earlier – downvolting the Llano die resulted in 32% saving in power consumption.