AMD's Processor Shift: The Future Really is Fusion
Retiring the Workhorses
There is an inevitable shift coming. Honestly, this has been quite obvious for some time, but it has just taken AMD a bit longer to get here than many have expected. Some years back we saw AMD release their new motto, “The Future is Fusion”. While many thought it somewhat interesting and trite, it actually foreshadowed the massive shift from monolithic CPU cores to their APUs. Right now AMD’s APUs are doing “ok” in desktops and are gaining traction in mobile applications. What most people do not realize is that AMD will be going all APU all the time in the very near future.
We can look over the past few years and see that AMD has been headed in this direction for some time, but they simply have not had all the materials in place to make this dramatic shift. To get a better understanding of where AMD is heading, how they plan to address multiple markets, and what kind of pressures they are under, we have to look at the two major non-APU markets that AMD is currently hanging onto by a thread. In some ways, timing has been against AMD, not to mention available process technologies.
The Desktop and the End of AM3+
Currently AMD has two levels of desktop products; the FM2 based APUs and the AM3+ based FX CPUs (as well as the older Athlon II and Phenom II products that are still available). The foundation of the AM3+ socket infrastructure is a mature one, but it still very fast and effective for what it does. For a while AMD had some of the most feature packed chipsets available for their processors, but often these were overshadowed by the lower performing CPUs as compared to Intel’s latest products of the time. The release of the 890FX with the SB850 southbridge was a watershed event as they arguably had the most advanced chipset out at the time. The 40 PCI-E 2.0 lanes supported by the 890FX were complemented by the first SATA 6G based southbridge available. Not only that, but it was the first to natively support six SATA 6G devices.
The Asus Crosshair IV was an excellent example of the then cutting-edge 890FX/SB850 combo. Too bad the base chipset has not been improved upon in years... and likely never will be.
AMD had a chipset that hit all the major checkmarks for OEMs. They complemented the 890FX with cut down versions of the chipset to hit different price points. They still had very competitive integrated graphics with the 880F and 890G. The Phenom II CPUs were doing “ok” against the latest Intel offerings of the time, but they were certainly not at the performance level of the i7 series of the time. The promise of Bulldozer was initially good, and AMD was expecting to compete with Intel’s fastest. Things went sideways at this time. Early Bulldozer results were very disappointing. The fastest chips off the line were barely hitting Phenom II X6 1090T performance. A few delays later and we finally got to see the much-ballyhooed FX processors, and were duly unimpressed.
To ready the market for Bulldozer support, AMD released the 990FX chipset family. The 990FX was simply a rebrand of the 890FX. They rebranded this to insure the change in VRM specifications from the AM3 to AM3+ socket infrastructure, so as not to confuse customers. During this time the decision was made to cancel the upcoming 1090FX northbridge and the SB1050 southbridge. This was the first real tip-off that AM3+ would not be a supported platform after a certain point.
This was later confirmed when roadmaps were released and showed that the Vishera lineup would be supported for the AM3+ platform throughout 2013 and 2014. The Steamroller architecture, which will be introduced with the Kaveri parts, will not be represented on AM3+. Steamroller does appear to be a big step up in both IPC and multi-processing as compared to Zambezi and Piledriver architectures. Sadly, for those users hoping for one last upgrade on the AM3+ platform, it looks like prospects of a large Steamroller based product on that platform are slim.
The Server Market
The original Opteron was another great success for AMD, and it catapulted them into the server market in a big way. These processors featured HyperTransport connectivity, integrated memory controllers, native 64 bit processing, and performance that outpaced every other Xeon on the market. Unfortunately, the salad days did not last terribly long for AMD. Intel came back with Core 2 based products and soon regularly outpaced AMD and their latest.
Over the years AMD’s marketshare in the server space has continued to erode. They have been able to stave off extinction here by focusing on high core count products at very reasonable prices. AMD keeps the TDPs low and the core count high, and have been able to maintain a presence in this lucrative market. The cracks are showing though. AMD has yet to really update the server side chipset offerings since around 2008. The southbridge that is implemented in nearly every board is based on the old (and somewhat buggy) SB750. This means no SATA-6G capabilities, not to mention the slower I/O speeds even with modern SSDs.
Things are slightly different here than with the desktop. AMD announced the Warsaw based processors for the server market. Very little information has been released about the specifics of these products, but the one thing they were quick to point out is that they will be faster and more power efficient than the current Opteron 6300 series of products. These claims point to these chips being based on the new Steamroller architecture and being manufactured on 28 nm rather than the larger/older 32 nm PD-SOI process current AMD CPUs and APUs are manufactured on.
The Supermicro H8SGL is still considered a modern motherboard supporting G34 based Opterons. Unfortunately, it is still several years old and relies on a chipset design that was released in 2008/2009.
Here is where things get interesting. If AMD was going to go ahead and design Steamroller parts for C32 and G34, why wouldn’t they port these numbers over to AM3+? Also, why have they not updated their server chipsets if they are going to keep pushing these sockets? There will be no PCI-E 3.0 northbridges, and I do not believe that AMD is putting in the extra effort on their latest A85X and A88X I/O hubs so that they can be certified to work on server platforms. So why even go there?
The answer to that one seems to be that Warsaw is a stopgap measure to keep AMD’s foot in the door in the server market. AMD has put in the time to create the Warsaw design, but they will not be focusing on optimizing the design for high clock speeds. Manufacturing and design appear to be aimed at getting the new product into the upper 2 GHz range while still supporting 4, 8, 12, and 16 thread models. They look to be aiming at keeping these faster than previous Interlagos based chips and not going overboard on the TDP. While designing any modern CPU is a time and manpower intensive operation, this particular scenario looks to be one where AMD is abandoning the high end and putting out a product that is better than previous, and good enough for what that aim is. Warsaw is designed to make AMD stay relevant in the server market and keep their current partners pleased with the offerings. It really does look to be a stopgap measure as the rest of the Opteron ecosystem has remained essentially untouched (and un-updated) for years.