Video encoding
Nehalem architecture agrees with modern video codecs (and old as well) perfectly. However, we cannot dispute usefulness of HT here either. As we can see, competing products have no chance to succeed even without this technology. Even significantly higher clock rate won't save them.
Among other interesting results we can mention the 870 almost catching up with the 950 for the second time. But "almost" does not count. On the whole, we can conclude that the 950 is preferable in this pair, if we consider only performance or CPU price. The only thing that stands in its way is a higher price of the platform.
3D games
Having looked through our test results, we want to call Core i5 750 the best processor for games. No, it's not the fastest processor, but faster models are much more expensive. So much that the difference in price can recover upgrade to multi-GPU, which will certainly affect total results. However, we won't be surprised, if future Core i5 processors will be even better for gaming computers (they'll have just two cores, but they will support HT and operate at higher frequencies). That's in future, and Core i5 750 is available right now.
Conclusions
So, the new platform has been launched. Is it successful? Yes, more than somewhat. If Core i7 for LGA1366 hadn't spoiled the triumph, we could have compared this situation with the transition from NetBurst to Core 2: when new Mid-End processors were faster than top solutions from the old product line (both from the same company and from its competitor.) But the fact that we already examined the Nehalem architecture turns this situation from revolution to evolution. The new microarchitecture was available only to those users, who were ready to pay at least $500 for a motherboard and a processor. The rest could only dream about it. But now it has become much more affordable to consumers, its price dropped by $150-$200. Capacity for price cuts on P55-based motherboards is even higher than that for the P45. Besides, we may expect cheaper chipsets in future -- as soon as cheaper processors come out.
It's not the easiest time for AMD -- Phenom II has finally met a worthy competitor, unlike older platforms from Intel. In fact, the company can only play with its prices and rely on non-CPU advantages of its platform, until it significantly improves the architecture -- Intel has nothing to oppose to AMD 785G and 790GX after all. And only time will tell whether the integrated graphics core in dual-core processors for LGA1156 will be competitive on this level. And these methods will only help to compete with Core i5. What concerns lower modifications of Core i7, the company will most likely have to resort to another method: to give users more cores than Intel. Only this time it will be more difficult to pull off -- triple-core processors were a good alternative to dual-core ones. But now Intel offers virtual multiprocessing, so we cannot predict the outcome of a competition between desktop six-core processors and quad-core models supporting HT. By the way, in several months Phenom II X3 will have to compete not with Core 2 Duo, but with Core i3/i5 (two physical cores that can execute four threads) and discontinued Core 2 Quad processors. AMD has its response ready (Athlon II X3 and X4, very cheap because of the lack of L3 Cache). But it's hard to say now how adequate it will be. So, the company is practically unarmed until the transition to 32nm process technology (which will allow to raise frequencies and make six-core processors affordable) and Bulldozer's rollout (by the way, we've heard the rumor that it will also support additional computing threads -- it will be a new feature for AMD). As all these events will happen by 2011, year 2010 will pass under the sign of Intel. It will be similar to 2007 (we can only hope that 2011 won't repeat 2009 for AMD).
But let's return to Intel. As we can see, the entire family (except for Core i7 920) lines up in a perfect ladder of performance, and the 920 apparently falls out of the 750-860-870-950-975 line. Is it an accident? No, it's quite predictable. This year in May the company apparently came to a conclusion about the new processors for LGA1156. And the desktop models for LGA1366 were updated in May as well. If Intel had wanted to discontinue it, it could have just ruled this update from its plans. As we can see, the 920 is significantly worse than the 860, and the 870 is practically similar to the 940 (just take a look at the table with detailed results). With the same prices in each pair, non-extreme models for LGA1366 instantly lose their significance, so they can be discarded. If Intel has wished long and happy life for the 1366, it should have replaced the 940 with the 950, and launched the 930 instead of the 920. In that case the lowest processor for this platform would have had parity in performance, and the highest model would have looked well. However, this turn of events is apparently just as undesirable for the company -- why spoil the sales of the new Core i7 processors? But the company does not wish to part with LGA1366 for desktop computers either. Hence this compromise decision -- we don't need the 920 anymore, so it will be "killed" indirectly, but we need at least one non-extreme model, so it's made quite competitive.
OK, we've sorted out positioning of the 950: it must be a tad faster than the equally-priced processor for LGA1156, which is compensated to a certain degree by the platform cost. And what about the rumors of replacing the 950 with the 960? We should analyze relative positioning of the 860 and 870 here. As we can see, these processors differ in performance only insignificantly. We couldn't expect the opposite -- the clock rate of the 870 is higher only by 133 MHz. However, their prices differ almost twofold: $284 and $562. Isn't it too much? Yes, it is. The company has always given more for the same price difference on the LGA1366 platform -- 266 MHz prior to May and 400 MHz afterward. Why so? We can only guess here. The most logical explanation is this: the price of the top model for LGA1156 is $562, but Core i7 870 won't necessarily be that top model for long. To be more exact, it will be for now, as it serves the company's end -- it does not interfere with the sales of the i7 950, and it shifts the demand to the i7 860, which turns out to be a better choice at the initial stage: its clock rate is lower, so the yield of effective dice is higher. Especially as performance of the 870 is actually a tad higher (even if just a little, and not always) than can be expected by extrapolating results of the 860 in frequency. It proves that the 870 uses specially selected dice with lower power consumption. They perform better in boost mode, as the 3-GHz line is crossed, still staying below 95 W even at a tad higher frequencies. It's even more true for Xeon X5500, the company can afford it. Fortunately, the cheapest processor from this family costs about a thousand dollars, while defected chips can be used for cheaper processors like Xeon E5500 or Core i7). And later on, it will be quite reasonable to roll out the 880 processor to take the 562-dollar place, lowering the 870 to $360, for example. However, Core i7 880 will catch up with Core i7 950 already. But that's not a problem -- the latter can be replaced with the i7 960. So this scenario is quite possible. Especially if AMD manages to launch Phenom II X4 975 by the end of the year.
On the whole, two desktop sockets (LGA1366 and LGA1156) may coexist peacefully, if they are positioned reasonably. The more so, as AMD proved it possible with Socket 754 and 940: one is for the mass market, the other is for extreme models (compatible with server processors). The situation with LGA1156 and LGA775 is a tad worse -- AMD already demonstrated that nothing good comes out of this mess (754 and 939). Intel has its own experience -- Socket 478 and 370. At that time the company had to fight the outdated solution rather aggressively, so that it wouldn't interfere with the new product. And the company didn't like the results so much that it never allowed such situation to repeat. Perhaps the late transition of Intel to the integrated memory controller was dictated by the desire to use the same implementation for all desktop processors. But now the company failed to avoid this situation, so users are forced to choose between three sockets. The most popular (up to $200) and high-speed processors are incompatible with each other. So the company should master the 32nm process technology as soon as possible for Core i5/i3 processors with two cores, increase the number of 45nm quad-core models (Core i5 700 and Core i7), and gradually move the LGA775 remains to the budget segment. But these are steps to be taken in future. As for now, we have a ready platform LGA1156 and three processors for it. According to our tests, the top model is not very attractive with its current price. But the medium and the lower products are designed for common users (for relatively well-off users so far).
Write a comment below. No registration needed!
