JPR Claims AMD Tiptoes Towards Workstation Exit

We've talked about AMD's server roadmap on multiple occasions this year while simultaneously covering the company's ramp of its FirePro graphics cards. With all the Bobcat, Llano, and Bulldozer chatter, the quiet passing of AMD's workstation offering has gone all but unnoticed. That, at least, is the argument from John Peddie Research, which chronicles the decline of AMD's efforts in this sphere even as it aggressively ramps its professional GPU products.

We wouldn't be surprised if AMD struck a different tone with regard to its workstation business, but the company's own website lists just two entries under "Workstation Products:" Professional Graphics and AMD FireStream Processors. Opteron is still occasionally mentioned as a workstation solution in various places, but only in passing.


Bulldozer is supposed to revitalize AMD's server performance—will it do the same for workstations?

We agree with JPR that AMD's lack of workstation CPU products is the result of the company carefully picking its battles. When AMD slashed Opteron prices early this year, the company focused its marketing efforts on proving how 12-core Magny-Cours products could compete with Nehalem thanks to a mixture of aggressive pricing and superior CPU scaling. Such a position makes sense when discussing 2S-4S systems; it's less effective when comparing single-core products.

We think AMD will have to wait for Bulldozer before it makes a play for workstations again—but it might be able to do so with a unique approach. The first generation of Bulldozer CPUs won't have an onboard ATI GPU, but our discussions with the company leave us thinking we could one day see this feature even at the higher end of the workstation market.

AMD has been talking about heterogenus computing for quite awhile now; one of the company's long-term goals is to allow workloads to execute seamlessly across both the CPU and the GPU. While high-end GPUs would still be necessary under heavy graphical workloads, a relatively small cluster of processors on-die could potentially boost performance significantly.