However, on the servers and workstations front, Intel is expected to unleash a whole spread of speed grades across the low-power, mainstream and high-end ranges. Simply put, the dual Xeons can be priced much higher per chip, and secondly, these little babies will still have to hold AMD's Valencia and Magny-Cours processors at bay as well, just in case.
Talking about these two, expect AMD to quite possibly use the Ides of March period to announce its updated six-core and double six-core processors. The latter is what Magny-Cours is after all, two six core Valencia chips on a large MCM and even larger socket. Even with stepping improvements, Magny-Cours is still not expected to go much above 2.2GHz in production, even though overclocked units could reach close to 3.5GHz in lab tests. The constraints imposed by keeping the power and thermals right across 12 hungry cores and 4 memory channels are clear.
The single socket six-core chip - including its desktop version, Thuban, finally appearing - should come dangerously close to 3GHz this time. It would really be lovely if AMD could fairly quickly offer 3GHz six-core Phenom II X6 and Opteron units, as those would bring back a semblance of competition, albeit a limited one, to the high-end CPU arena. That's something that has been sorely missed for the past three years, frankly. I'd expect selected Phenom II X6's to even come close to 4GHz using good air cooling when overclocked, but let's see how it goes.
Talking about the high end, AMD's Magny-Cours does provide a very easy 4-socket glueless solution with up to 48 cores. So will the Intel Nehalem-EX or Xeon 7500 family, Chipzilla's top end chippery for ultra-reliable enterprise servers and, why not, large memory EDA or scientific desktop 3D supercomputers. However, the Xeon 7500 will also do it glueless for 8 sockets as well, delivering one up on the competitor, very likely before the end of next month.
Yes, there is a clock speed penalty for squeezing the 8 full cores and 24MB cache onto a single die, compared to 12 cores and 12 MB of total cache on two dies for AMD, but Intel's clock speed should, based on my expectations, still at the very least match that of Magny-Cours. On the other hand, Nehalem-EX will provide triple the amount of cache per core, as well as double the memory bandwidth per core. It is four serial links or a total of eight DDR3 memory channels per socket on the Intel entry, versus four DDR3 channels per socket on the AMD design. This may help quite a bit in memory-intensive applications.
In summary, announcements of six-core chips and those with multiples thereof will abound in the coming few weeks, with the sole exception of the "lucky" eight-core server Nehalem-EX in the latter part of this month. The performance gap between Intel and AMD will still remain at the usual 10 per cent to 15 per cent per core at the same clock and then add the clock speed difference, with the resulting price disparity, but at the very least there'll be more to choose from and compare, both on the market target and core count. So, look forward to a CPU-exciting month of March.
There will be plenty of new introductions, with associated system platform refreshes, new reviews and comparisons. We'll cover each family and market segment more in the individual post-launch stories, of course.
theinquirer.net (c) 2010 Incisive Media
Issue: 137 | June, 2012