Slashdot Mirror
Intel and AMD May Both Delay Next-Generation CPUs
MojoKid writes "AMD and Intel are both preparing to launch new CPU architectures between now and the end of the year, but rumors have surfaced that suggest the two companies may delay their product introductions, albeit for different reasons. Various unnamed PC manufacturers have apparently reported that Intel may push back the introduction of its Ivy Bridge processor from the end of 2011 to late Q1/early Q2 2012. Meanwhile, on the other side of the CPU pasture, there are rumors that AMD's Bulldozer might slip once again. Apparently AMD hasn't officially confirmed that it shipped its upcoming server-class Bulldozer products for revenue during August. This is possible, but seems somewhat unlikely. The CPU's anticipated launch date is close enough that the company should already know if it can launch the product."
There might be good reasons on both sides, but the tinfoil hatter in me believes this might have more to do with fact that both companies might want to see a little more profit out of the R&D that went into the current generation of products before obsoleting them. The performance of the current generation is high enough that it is getting harder to introduce a new generation at a price point that could both recover R&D and provide reasonable value for the customer.
The current situation is Intel is slaughtering AMD. AMD hasn't had an architecture update in a long, long time and it is hurting them. Clock for clock their current architecture is a bit behind the Core 2 series, which is now two full generations out of date. Their 6 core CPU does not keep up with Intel's 4 core i7-900 series CPU, even on apps that can actually use all 6 cores (which are rare). Then you take the i5/7-2000 series (Sandy Bridge) which are a good bit faster per clock than the old ones and there is just no comparison.
On top of that, Intel is a node ahead in terms of fabrication. All Sandy Bridge chips, and many older ones, are on 32nm. AMD is 45nm at best currently. Not only does that equal more performance but it equals lower heat for the performance, particularly for laptops. Then of course Intel is talking about Ivy Bridge, which is 22nm, another node ahead. Their 22nm plant is working and they've demonstrated test silicon so it will happen fairly soon.
The situation is not good for AMD. All they've got is the low end and that is getting squeezed hard by Intel too. They need a more efficient CPU and they need it badly. Delaying is not something they want to do, Bulldozer has been fraught with delays as it is. They've been talking about it for a long time, like since 2009, and delivered nothing.
They have every reason to want to get Bulldozer out as soon as possible and preferably before Ivy Bridge. Each generation that Intel releases that they don't have a response for just puts Intel that much farther ahead.
Now that said, Intel may well have decided to hold Ivy Bridge if AMD can't deliver Bulldozer because they don't need to. Sandy Bridge CPUs are just amazing performers, they don't need anything better on the market right now. However I can't imagine AMD colluding with Intel on this. They are not in a good situation.
The most naive question to ask if is this sort of delay is relevant to Moore's law and similar patterns. There are a variety of different forms of Moore's law. We've seem an apparent slowdown in the increase in clockspeed http://www.tomshardware.com/reviews/mother-cpu-charts-2005,1175.html. The original version of Moore's Law was about the number of transistors on a single integrated circuit and that's slowed down also. A lot of these metrics have slowed down.
But this isn't an example of that phenomenon. This appears to be due more to the usual economic hiccups and the lack of desire to release new chips during an economic downturn (although TFA does note that this is a change in strategy for Intel's normal approach to recessions.) This is not by itself a useful data point, so this is not further need to panic.
On a related note there's been a lot of improvement in the last few years simply by making algorithms more efficient. As was discussed on Slashdot last December http://science.slashdot.org/story/10/12/24/2327246/Progress-In-Algorithms-Beats-Moores-Law by a variety of benchmarks linear programming has become 40 million times more efficient in the last fifteen years and that only a factor 1000 or so is due to the better machines, with a factor of about 40,000 attributable to better algorithms. So even if Moore's law is toast, the rate of effective progress is still very high. Overall, I'm not worried.
I was surprised that the original date when it was planned to be released back than would have made it way ahead of the curve predicted by Gordon Moore. I was saying that it should be delayed some time so it actually is more accurate to the prediction and it has been delayed, however I will never know whether it had been done for that reason.
So... If I'm reading you correctly... the reason that they should have delayed the hardware wasn't because of something like it being too expensive to produce for expected market, too difficult to produce in sufficient yields, or any other technical or business reasons that might exist, but because the number of transistors involved didn't match up to a prediction made 30-some-odd years ago?
You realize that prediction has only "come true" when you average the graph over a very long period, and there are significant statistical outliers (that represent significantly successful chips in their day) along that plot?
Wait, wait... you're trolling right? I admit, you got me!