It's finally happened. The LGA1156 platform has been finally rolled out. It won't be an exaggeration to say that it's the most welcome event in the second half of the year, as we've all been waiting for the new processors for too long. Until now things have been shaping up by classical canons: Intel wouldn't start to sell the "true desktop" Nehalems, and users didn't want to buy the outdated Core 2 processors, pay too much for LGA1366 (its cost was mostly determined by features that were not necessary to home users), and switch to AMD. Now aspirations and possibilities have finally matched. However, Intel cannot offer new products to thrifty users so far -- the manufacturer's price of the cheapest new processor (Core i5 750) is 194 dollars, so retail prices will exceed $200 for some time (preliminary orders for this processor were taken for 205-207 USD), which is above the psychological barrier. Thus, economic shoppers will have to do with processors from the previous generation for now. It's not much of a problem though -- this category of users is not inclined to jump at new products, they make their choice for purely practical reasons. And if you are ready to pay two or three hundred dollars for a processor, you will have to make a weighted choice. The top Core i5 only insignificantly exceeds the barrier, while the top Core i7 870 has the same price as another top processor (Core i7 950), and the processor 860 costs just like Core i7 920 (which is still available). The last two models will compete with the old quad-core processors Q9550 and Q9650 (still available in stores) for some time, which makes the choice even harder. Add a couple solutions from AMD (Phenom II X4 955 and 965), and you will agree that competition in the price range of $200-$350 is getting very intense. So, the platform cost (easy to calculate) and performance (to be measured here) become the main choice criteria.
Architecture and naming
There is no point to delve into theory these days -- the new devices are architecturally the same Nehalems. There is only one major difference from its predecessors and Xeons for LGA1366 -- support for two memory channels instead of three. Hence the new socket: it needs fewer pins. It simplifies PCB layout (and makes it cheaper), allows to augment memory at smaller steps -- that is it reduces the platform costs. Especially as it comes almost "free of charge" -- we found out that usual desktop tasks did not need the third memory channel. To make up for the theoretically decreased memory bandwidth, the company officially allowed to use DDR3 1333 with the new processors (this memory is also supported by Xeon X55x0, but not by Core i7 for LGA1366).
Another factor that has a positive effect on the total platform costs is abandonment of the three-chip design (CPU - Northbridge - Southbridge) for the two-chip design. The complex large QPI bus controller is replaced with PCI Express and DMI controllers in LGA1156 processors. The former supports 16 lanes, corresponding to the PCIe 2.0 standard. They can be used either for a single full-speed graphics slot, or for two 8x slots for multi-GPU configurations. DMI interface (which is actually an electrically modified version of PCIe 4x) still has the throughput of 2 GB/s, which hasn't changed since i915 series chipsets, where it debuted. So all it formally takes to create a ready system is to solder one of the existing Southbridges (all of them, from ICH6 to ICH10R, are designed for DMI) to the corresponding pins of the processor. But the manufacturer claims that it takes chipsets of the fifth series, like iP55.
Interestingly, QPI support hasn't gone anywhere, that is we cannot say that one complex controller has been replaced with two simple ones. Only now this bus does not go outside, connecting PCIe controllers and main processor units inside the die. Its throughput is still 4.8 GT/s as before. So what's all this fuss about? It's just that high-speed "global" and "local" bus implementations differ in complexity. Link length used to be measured in centimeters, and now it's millimeters, which allows lower signal levels, etc. Besides, they saved a little on modifications in the core -- they would have to design an interface between a processor and PCIe controller anyway, so why not use the simplified QPI link instead?
What else has been changed besides peripheral and memory controllers (where one of the three controllers was removed)? In our today's processors -- nothing. They are the same quad-core processors (Nehalem) equipped with three cache levels (8MB L3 Cache). All three support Turbo-Boost. Two more expensive models also support Hyper-Threading, so they are no different from Core i7 processors for LGA1366. Much attention is paid to HT support now (or the lack of it) -- in fact, that's exactly what divides the three announced processors into two groups.
Up to recently the Internet has been full of rumors how LGA1156 processors will be called. Most of them agreed that LGA1366 processors will preserve the name Core i7, while models for the new platform will be called Core i5 and i3 depending on the number of cores: the first family will include quad-core processors, and the second -- dual-core products. Intel has been ignoring all these speculations for a long time, but it announced not long ago that the matter would be more complicated. Indeed, 800-Series processors as well as representatives of Series 900 (LGA1366) we already know are called Core i7. What must a processor do to qualify for this class? To execute eight computing threads. And we do mean threads, not cores -- short-range plans of the company do not have true eight-core processors (we've heard only of six-core models so far). And eight threads can be provided by four cores and Hyper-Threading support.
What is a Core i5 processor? Apparently, it's a device that can execute four instruction streams. That's where several options are possible: it's either a quad-core processor without Hyper-Threading (Core i5 750), or a dual-core processor supporting this technology. And a Core i3 processor will be able to execute... no, not two threads, but four as well. However, these processors will always operate at a starting frequency -- they will be able only to drop it when necessary, that is they won't support Turbo-Boost. If we remove all the new features from Nehalem (that is TB and HT) and reduce the number of cores to two, we'll get... another Pentium. No, seriously -- the company decided to prolong the life of its favorite brand once again.
Simple? Logical? We don't think so. Except for Pentium really looks good -- this trademark has been used for entry-level dual-core processors, better than Celerons, but still not sterling CPUs. Everything's fine with Core i7 as well -- they are High-End models. The problem is with Core i5. It includes 45nm quad-core processors with 256K L2 Cache per core and 8M Shared L3 Cache as well as 32nm dual-core processors with 1M L2 Cache per core (there are some rumors about 2M per core, but it's very strange considering the inclusive architecture of cache memory in Nehalem) and 4M L3. The latter products are actually very similar to Core i3. It goes without saying that they will have different indices and all, but we are already missing the old naming system -- at least it showed how many cores a processor had.
However, we'll face all these problems next year. As for now, the only problem that awaits an inexperienced user is buying an LGA1156 motherboard for Core i7 900 (or vice versa). In other respects, all the three processors are sterling quad-core CPUs with the same architecture. Core i5 just has no support for HT, but it's the lowest desktop Nehalem in the other parameters as well, so it's easy to sort it out. Turbo-Boost works as +5-4-2-2 in all three models (not as +2-1-1-1, like in the Core i7 900). DDR3 1333 memory is officially supported -- so it will be very interesting to see how these differences affect performance. Besides, Intel surreptitiously applied another steak to the new processors -- the default UnCore frequency is 2.4 GHz. It falls in between 2.13 GHz for Series 900 and 2.66 GHz for Xeon X5500. Its increase will apparently affect performance, especially in tasks sensitive to memory speed -- it means that L3 Cache and memory controller will work a tad faster now than in the cognominal family for the other platform (in particular, we may actually see performance gains from DDR3 1333, almost like in Xeon, instead of no changes in extreme models from Series 900). However, it concerns only the new Core i7 processors -- Core i5 750 has the same UnCore frequency as the old Core i7 models -- 2.13 GHz (lest it should get close to top models, when it works with four or fewer threads). So, modern processors spawn more and more technical parameters, which gives manufacturers freedom of movement -- they had much less of it, when the only way to expand the product line was to produce processors with different clock rates.
Not only we'd like to compare new and old Core i7 processors, we'd also like to examine the performance of the top Core i5 CPU (which will remain the top for a long time). Besides, we should take a look at other competitors.
Write a comment below. No registration needed!