A long time ago, but not in a galaxy far, far away, the world of CPUs was simple. For example, the x86 series was designed only for single-core personal computers. Ok, when the 80386 appeared, it was on a par with processors that were used in workstations just a couple of years before that, or similar to CPUs used in mainframes ten years before that. But nothing more: both classes also built up physique for that time. Some manufacturers even managed to make i486 multi-processor systems. But they did it with special chipsets, so these solutions were off the main course of development of x86 systems. It seemed that the situation wouldn't change ever.
However, Intel Pentium blew out simplicity of the computer world. The company did not stop at the usual rapidly growing segment of personal computers -- that was the time when SMP was introduced to x86 systems. Owing to relatively low prices of these processors, journalists often used the term cheap multiprocessing. Intel actually solved the main problems of interprocessor communication in its chipsets, so it was easy to manufacture dual-Pentium workstations and servers. All other components were absolutely standard, and there was no need in special processors either. So, dreams of symmetric multiprocessing came true fast and painless. And Pentium Pro processors contributed to this situation. The new architecture was designed for servers and multiprocessor workstations in the first place. These chips did not become popular, but they were not intended for that anyway -- the x86 chip for servers was born. There already appeared systems with four processors.
Rollout of Pentium II was accompanied by another small revolution. The fact is, system bus in this processors was the same as in Pentium Pro. But these products were designed for the mass market, so they were much cheaper than their predecessors. Besides, the first Celeron was launched over ten years ago to push Pentium and its clones out of the market. It was cut down to bare necessities (for example, its first models lacked L2 Cache), but it was based on the sterling core inherited from Pentium Pro. Moreover, practically all Intel chipsets for Pentium II supported SMP from the very beginning (even if dual-CPU only). There existed special "pared down" modifications in series 440, of course, but they were used for budget motherboards. And it was easy to buy a dual-slot motherboard based on 440LX or 440BX, they were quite affordable. Add two Celerons, for slightly more than $100 each, and... you got a dual-CPU computer for less than $1000. It certainly couldn't work performance miracles, but it still had two processors with all advantages of this approach. So the idea of the cheap SMP burst into blossom. First of all, only enthusiasts were interested in budget CPUs in multi-processor systems, but later on the entire industry came to their aid. Just think of the legendary motherboard Abit BP6 with two Sockets 370. When it appeared in stores, only Celerons could be installed into this socket.
Such ascent in popularity of budget SMP systems couldn't please the biggest CPU manufacturer. Just because it lowered its profits. So this idea was abandoned. In summer 1998, the first Xeon processors appeared -- Pentium II Xeon. They offered larger L2 Cache operating at the clock rate of the processor core (not at the half clock rate as in Pentium II) and a new socket. The company announced that "true" SMP should be based on Xeon, the other solutions are toys for enthusiasts. That was absolutely correct, if you needed more than two processors (Xeon, just like Pentium Pro and unlike popular products, worked well in quad-processor configurations), but the other users had a choice whether to trust Intel or save their budget. The company actually did not risk removing SMP support in popular processors. However, the trend was selected.
And the company abode by this policy. Unlike its predecessors, Pentium 4 did not support SMP. If you wanted to use two or more processors, Xeon was your only choice. These processors were absolutely incompatible in design, they used different sockets. However, they were based on the same architecture, used the same bus to communicate with the chipsets, but they were not interchangeable. That became a policy: Socket 478 for single-processor systems and Socket 604 for SMP. Then the former was replaced with LGA775, and the latter with LGA771. But the idea was the same. You couldn't buy a multi-processor motherboard and use it with a cheap desktop processor at least for the first time. You couldn't upgrade from a single-CPU system to dual-CPU by just changing a motherboard and adding another processor. And so on and so forth. Technically, Xeon sometimes outscored popular products (for example, Hyper Threading debuted in Xeon), sometimes it was outperformed (Pentium 4 already mastered the 800-MHz bus, while Xeon still used only 533MHz). But most importantly, these two lines stopped being physically compatible, remaining compatible on the logical level.
This approach had both pros and cons. Advantages for the manufacturer are apparent: it's easier to segment the market and flexibly control prices for maximum profits. Drawbacks are less noticeable, but they can hamper the industry. For example, different packages mean twice as many problems to package dice. By the way, it also creates extra problems to motherboard manufacturers, as they have to order and store different sockets. Besides, multi-socket motherboards for Xeon are designed for servers in the first place. And single-socket products for popular processors are intended for gaming and generic computers. But what if your objectives are somewhere in between? For example, you need a single-CPU system, but a motherboard with a server chipset with all its peculiarities, such as powerful hard drive or network controllers and proper slots (for example, PCI-X). Or on the contrary, you need almost a usual desktop computer, but it should have two processors. The industry learned to cater for such need of users. However, these solutions were always very expensive, because they were not manufactured on a big scale. Besides, they were not very easy to find in stores.
Why do we flash back to this situation (ancient history to many users)? Just because this article is devoted to Xeon processors, which are not often reviewed on our website. But! On the face of it, the new Xeon series that was rolled out today (two series to be more exact) differs much from the previous generations. We've mentioned sockets for a reason: they were always different for Xeon and popular desktop processors. But our new processors are designed for LGA1366 just like Core i7. Revolution? Actually, no, there is a simple explanation to it. Core i7 is not a processor for the masses either. Intel just had to keep its promise and present a processor based on the new architecture. To all appearances, it was not quite ready for the server market. Any defects could have had grave consequences for the manufacturer, so discretion is the better part of valour. "People's Nehalem" was also not ready for manufacturing, as it is not today. Moreover, rollout of Core i5 may be delayed for a year or so relative to the initial release date. It's too radically different from the old designs, having integrated the entire Northbridge of the chipset (including PCI Express controller). So engineers have to change the entire industry of motherboards and chipsets, to decide what to do with remaining components, to cut down production of Core 2 processors, etc. Lots to do. And what if the company must launch a processor, but it's not ready for any market? Intel has found an elegant way out of this situation. The company has chosen top Xeon products (they are ready), then rapidly designed a simplified chipset with QPI bus, based on the almost ready 5500 (perhaps, they didn't even have to simplify it much), and gave it a typical desktop name, X58 Express. Here we are, the new desktop series is ready. Why not? It does not sound like a desktop product, of course. For example, we cannot logically explain the Northbridge in X58, which is just a PCI-E controller connected to a processor with a very fast bus, apparently designed for other tasks. Prices for these motherboards also go beyond the general limits, and memory manufacturers were not ready for mass production of triple-channel memory kits. But it was still possible to position this solution as a top product for hard-driving users. Especially as the company is not new to this sort of thing. It has recently come up with the Skulltrail platform for games, which is actually a direct transfer of the server-workstation solution to the desktop market. But the way, Core i7 and Skulltrail have a lot in common: eight processor cores (even if four of them are virtual in the first case), support for SLI and CrossFire, etc. The latter actually worked only with one CPU model. As for the former, there appeared even three products for it (the i5 delays may expand the line to five processors). AMD also moved server solutions to the desktop segment many times. The first attempt was actually similar to Core i7: two Athlon FX models for Socket 940 (remarked Opteron per se). So, there is nothing new. This way or another, Core i7 was not the worst solution, just a bit strange (but Skulltrail and AMD 4x4 were much worse, to say nothing of strange).
And now we meet Xeon with the Nehalem in its pure form. There are many processors announced, but we already know some of them. For example, the Xeon 3500 series (previously known as Nehalem-WS) is designed for single-socket systems, that is practically in the same computers as Core i7. The only difference between these two series is support for ECC memory in Xeon and no such support in Core i7. Note that support does not mean necessity. You can still use usual memory modules, which will be faster by about 5%, but less reliable. In fact, many users combine Core i7 with 12GB of memory (two modules per channel) and do not report about any problems. The other aspects are the same. Three processors and the same TDP at equal frequencies. Xeon W3570 3.2GHz costs $999 and differs from Core i7 EE only in locked multiplier. Xeon W3540 (2.93GHz, $562) and W3520 (2.66GHz, $284) are counterparts of Core i7 940 and 920 correspondingly (they just have a full-speed QPI controller, which is of no use in single-processor systems). Nothing interesting.
Xeon 5500 (Nehalem-EP) is a more interesting series. Unlike Core i7 or Xeon 3500, all its processors support two QPI links, so they can be used in dual-processor configurations. However, such processors can be also used in a usual desktop motherboard. By the way, processors with one link can theoretically be installed into dual-socket motherboards. However, only the processor connected to the chipset will work in this case. Or it may fail to work. We don't have a proper motherboard to test it so far. On the other hand, this test will accomplish nothing but sate our curiosity. You cannot save on this configuration, and there will be no bonuses from extra investments here.
The question is what if you want a quad-processor configuration? In this case you will probably have to wait for Xeon 7500 to roll out. That is, the old top Opteron-based multiway servers can not be afraid of competition. There are still no Nehalems, and it's not always expedient to use several classic Xeons, because of the bottleneck, the shared bus between processors and the chipset. But Intel will eventually cover this market segment as well, it can't just give it away to the rival.
Let's return to our processors. There are a lot of them here, so it's more convenient to divide this series into groups. The top group includes only one device -- Xeon W5580 for $1600. Its clock rate is expectedly 3.2GHz, TDP 130W. Sounds familiar? Yes.
Other products are more interesting. For example, three processors -- X5570 (2.93GHz as in the i7 940 and W3540, $1386), X5560 (2.8GHz, no counterparts in other series, $1172) and X5550 (2.66GHz as in the i7 920 and W3520, $958) -- can boast of 95W TDP. So these products may also interest users of single-processor systems -- their power consumption is lower by one third. But this solution will be more expensive. The price of X5570 is higher by one third than that of the faster W3570 (we shan't even compare with W3540 that offers the same performance level).
Besides, if power consumption is more important than performance, there exists another series, its model names start with the letter "E". Its advantage is TDP of just 80W (over 1.5 times as low as in the Core i7 or Xeon 3500!), there are some drawbacks as well. To be more exact, there are three drawbacks: lower clock rates, slower QPI controller, and only 4MB L3 Caches instead of 8MB ones as in the other processors. In fact, the size of shared cache (especially if it's strictly inclusive, like in Nehalem) must have a strong effect on performance of quad-core processors, so the "E" series will be attractive only in saving terms, whether it's money or electricity. But they are really good at it. There are five quad-core processors in this series: E5540 (2.53GHz, $744), E5530 (2.4GHz, $530), E5520 (2.26GHz, $373), E5506 (2.13GHz, $266), and E5504 (2GHz, $224). The last one could have been the cheapest Nehalem but for another dual-core processor -- E5502 (1.86GHz, for just $188!) Frankly speaking, it's very tempting to find this processor and compare its performance with the ancient Core 2 Duo 6320, as these processors have identical characteristics, only their cores and memory controllers are different (by the way, we can still find the 6320 in stores right for about $180).
The "L" series is designed for systems, which power consumption is of primary importance. Such processors offer even lower characteristics, but they can provide TDP of just 60W, which is more than twice as low as in top processors or Core i7. There are two such products: L5518 and L5506. Both of them operate at 2.13GHz just like the E5506. What's the difference then? All 550x chips do not support Turbo Boost and Hyper Threading, and their QPI controller is designed for 4.8 gigatransactions per second (while top processors from the "X" and "W" series provide 6.4 gigatransactions, the E55x0 and L5518 -- 5.86 gigatransactions). And finally, the L5508 -- 2.0GHz and TDP of just 38W, similar to low-power Xeons based on the Core 2!
On the whole, we can establish a fact that this number of simultaneously launched processors renders Xeons on other cores obsolete. So, farewell to FB-DIMM (that failed to secure itself in the market) and shared FSB (at last). Nehalem's coming to Xeon changes the market much more radically than Core i7. The latter coexists with Core 2, and this situation won't change at least until the appearance of Core i5, which will replace both of them. As for the non-desktop market, the new processors are replacing existing ones completely. It will take some time, of course, but not very long. All server manufacturers have already announced their products based on Xeon 5500. Quite a warm welcome.
What does it give us, common users? Actually nothing. The pricing policy of this company will still make most users choose between Core 2 and AMD Phenom. Much fewer users will take a closer look at Core i7. And Xeon will interest only those people who need more than four cores, and virtual cores won't solve the problem. Using the 3500 series in popular single-CPU systems makes absolutely no sense. The 5500 series may give you only a tad higher energy efficiency and lower performance for the same money. At least we doubt that many users will prefer Xeon X5560 to Core i7 EE 965 just to conserve 35W. However, we've tested this processor using our standard method of testing. So that you could make your choice.
Write a comment below. No registration needed!