In yet more good news for AMD, the lack of a ’10-series’ chipset was just explained to SemiAccurate. If your browser does not support HTML 9.1ab </sarcasm> tags, the reason for it is not good news.
If you have not been following the chip and chipset world recently, AMD puts out a new chipset every year for their new CPUs. Whether the silicon is actually new or not, the model number of the chipset goes up by 100 or so, often modified by NMI (Nebulous Marketing Increments). The current high end chipset is the 990FX, and that replaced the prior 890FX, which in turn put the 790FX out to pasture.
We are long overdue for a 1090FX, or in marketing-speak, the Xr1090F1-eXXXtreme, but it has not broken cover. Why? Because chipsets are largely vestigial of late, being not much more than a boot rom, SATA breakout box, and low level PCIe chip, now with integrated USB3 hubs. This functionality is not exactly bleeding edge unless you are Intel, then USB3, like any open standard, seems to be a bit of a head scratcher.
Back to AMD, the current chipsets have SATA3, USB3, and most of the goodies other than PCIe3. Until that tech is ready in Austin, there is no point to a 10-series chipset. Guess what is delayed? No, not the new Brittney Spears album, but PCIe3 on AMD.
Sources tell SemiAccurate that you won’t see built-in PCIe3 on AMD products until late 2013. While we think that is a little pessimistic, mid-2013 does not seem out of line in light of current product schedules. Either way, this is quite frankly an unacceptable situation. If something doesn’t change here, AMD is out of the gaming market for desktop CPUs, PCIe3 is a ‘must have’ item.
More interesting is the server market. PCIe3 is not a big deal now for servers, bar a few bleeding edge HPC installs, but it does explain a very interesting roadmap slip. Remember how Vishera, the 20-core, two die AMD server CPU, lost four cores in February?
While that was probably due to memory changes, or lack thereof, on die PCIe also went AWOL at the same time. This precluded a socket change, and therefore a memory change.
So that means AMD is having some rather severe problems with desktop PCIe3. This is extremely curious in light of their lack of problems with GPU PCIe3, it quite simply works, and has been shipping in volume since January. At least some teams at AMD understand PCIe3, but that doesn’t seem to be helping the CPUs.
Financially speaking, how will this impact AMD? On the desktop and laptop fronts, not much. AMD is a low end player in both segments, and that is not where PCIe3 is used, much less needed. For the next year, only bleeding edge gaming rigs and high end workstations will have a real use for PCIe3, but even that is questionable. AMD’s marketshare in both categories is quite low, and the rest won’t care.
Marketing is another matter entirely, and that could affect sales. In the rather dubious world of ‘Big Box checklist’ sales, PCIe3 could matter, but SemiAccurate is tech focused, not retail, so we can’t comment about this with any authority.
Moving on to servers, PCIe3 is limited to a very small segment of the server market, very very small. Unfortunately for AMD, that segment, HPC, is one where AMD still has some decent marketshare. How much that market cares about PCIe3 for the next 12 months is an open question, but given the lack of non-GPU PCIe3 cards on the market, we don’t think it will be a dealbreaker in many cases.
In the end, this is just bad news for AMD, the lack of PCIe3 is pretty shocking. Given their overwhelming market leadership in PCIe3 cards, it goes from shocking to almost comical. If AMD can put PCIe3 in desktops and servers by the next product refresh, Kaveri and Vishera, then the impact will be limited. If it has to wait for the generation after, it will be a big big problem for the company.S|A
Latest posts by Charlie Demerjian (see all)
- Intel shows off 10nm 112Gbps SerDes - Mar 12, 2019
- Intel releases Compute Express Link spec - Mar 11, 2019
- Qualcomm rolls out a second gen 5G modem called X55 - Feb 19, 2019
- What is Intel’s Foveros tech and what isn’t it? - Feb 11, 2019
- Why SemiAccurate called 10nm wrong - Jan 25, 2019