THE YEAR about to end was a more or less uneventful one from the processor competitive standpoint - after all, Intel just continued its performance leadership across ever more price points. It started the year with the top desktop, server and workstation Core i7 and Xeon LGA1366 parts at the uni- and dual-processor high-ends, and expanded into the second half of 2009 launching the mainstream LGA1156 socket Core i5 and i7 - yes, naming confusion included - quad core chippery at prices as low as $200 apiece.
As the year ends, we have the $1,000-plus parts, the Core i7 975XE, Xeon W3580 and Xeon W5590 for uni- and dual-CPU machines, all 3.33GHz three-channel memory LGA1366 parts, firmly ruling the performance realm in the PC space. In the mid-to-high range, the 2.93GHz Core i7 870 in LGA1156 has no competitor in the $500 space, while the $200-plus realm is guarded by the low-end 2.66GHz Core i5 750, holding out well against the newest stepping of the AMD Phenom II 965BE at somewhat lower power consumption. And then, the just introduced Core i3 low-end dual-core processors mop up the sub-$100 price class as well, at a very low power budget too.
In this situation, as a short term measure, AMD might want to consider further slight downward price adjustments for its otherwise pretty good Athlon II 620 and 630 quad-core CPUs that - officially at least - have no L3 cache, but do manage decent performance and some overclocking up to roughly 3.4GHz just on 'autopilot' with good boards, and more if you're willing to tinker around. At around $100, these CPUs can still be quite competitive against Intel's, say, dual-core 3.46GHz Core i5 670, offering equal or better performance - but still at over half more power budget needed.
Put aside the little performance comparing nitty-gritties, the key problem is that, in the 2009 PC arena, Intel completely pushed AMD away from all the high margin, above $200 each, processor market segments. To keep its PC CPU business more profitable, and enhance the 'brand performance image', AMD needs competitive high-end entries above the Phenom II 965BE. Yes, the extreme high-end quad-core eight socket servers are, until the Nehalem-EX arrives, still an AMD stronghold, but their quantities aren't large, and there are only two months or so of that 'advantage' left.
And this is all before 2010 starts.
Intel is expected to unleash a barrage of 32nm "Westmere" class chips across many market segments in the first quarter of the coming year. You'll already be seeing the entry level Core i3 and i5 parts, including some with an integrated - supposedly very decent this time - GPU, in the Socket LGA1156, in early January. A month later, it will be followed by the - not Core i9, mind you - but Core i7 980X six-core Gulftown processor with 12MB cache and the same 3.33GHz clock and TDP as the current cream-of-the crop, the Core i7 975XE. The processor, which we covered and saw in action over the past few months, overclocks at least as well as the current 975XE, but with 50 per cent more cores and L3 cache. Its equivalent on the Xeon DP front will do the same for the dual socket workstations, servers and extreme desktops.
Interestingly, if you look at the 2010 Intel roadmaps circulating on the websites, the high-end will stay pretty monotonous, with the Core i7 980X taking the top, and the current Core i7 960 LGA1366 and Core i7 870 LGA1156, both 45nm quad core parts, sharing the step below the top. In fact, there will be no mainstream 32nm quad-core Westmere shrink at all, as that will only appear in the high-end six-core and low-end dual-core parts. And that won't even happen until 2010 year-end, when Sandy Bridge is supposed to start surfacing. Now you see what lack of competition can do.
Why do we say so? Well, now it is clear that Bulldozer, the next major AMD processor core, will surely not surface in any way, shape or form in 2010. It will have to face Sandy Bridge from Intel's Israeli labs in early 2011 at best, it seems. Therefore, all that faces the new Gulftowns is the current AMD core in the Phenom II parts, at most maybe in a Phenom II X6 'desktop Istanbul' but at a reduced clock to manage the heat.
And, there is no indication that AMD is doing what it should have done in the first place - tweak the current core to extract another 10 per cent at least extra performance per clock, coupled with increasing the L3 cache size and lowering the L3 cache latency, as well as improving the DDR3 memory controller a bit to support the currently quite common DDR3-2000 RAM. Too much to ask? No. Look at the six-core Istanbul example, where the same old 6MB cache as on the four-core Shanghai has to feed all the cores. Contention is bound to happen. AMD does have those eDRAM dense cell capabilities, why not use them and make the L3 cache eDRAM based? The current latency is bad already, and can't get much worse, but doubling the capacity to 12MB with similar transistor count, and tuning up the rest to reduce the controller latency, would help a lot.
Right now, across two dozen top benchmarks, the AMD cores lose out to the Intel ones by between 10 per cent and 25 per cent clock for clock, and that is without multithreading, Turbo Boost and such stuff turned on. Even if Intel did 'optimise' for some benchmark, it couldn't do that for like twenty or thirty very different tests as some netizens claim. Simply put, its cores are faster. If AMD did as suggested above and, while we wait for the Bulldozer to appear more than a year from now, tuned up its current core and uncore to get a total of 20-plus per cent extra oomph clock for clock, it would be quite a different outlook.
In absence of that, on the desktop, AMD probably can still work out another stepping upgrade in the current 45nm process to propel the Phenom II X4 to 3.6GHz, and possibly a six-core Phenom II X6 to 3GHz. Still, this would only keep things alive for it in the $200 price segment, nothing above. And, Intel can continue to charge a grand a piece for the Core i7 980X throughout 2010, and half that for chips like the Core i7 960 or Core i7 870. No competition, what to do?
In summary, the desktop PC CPU 2010 futures aren't bleak, but aren't that exciting either. Intel will execute its 32nm migration at both the high and low ends as planned, on time, but the mainstream will continue to run off the current 45nm fab lines, and create steady profit as there is no need for new entries until early 2011 Sandy Bridge chips with integrated quad-core CPUs and GPUs on one die. No competitive threat means no need to act fast. If AMD did something to be more competitive, I am sure Intel would have to accelerate its rollout, but then....
So, if you've got money to burn, and want the best, look forward to the Core i7 980X or whatever Intel decides to call it at the launch. Or, even better, get two of them under the Xeon 5600 moniker and run 12 cores with six DDR3-1600++ memory channels. At the $200 level, you'll still have a two vendor choice, though. Anything more? It's really AMD's call this time - we need it up and running to keep up the speed of PC processor development. µ
It's time for our regular two-step through the Google news
Bug bounty offer: accepted