by Ilya Gavrichenkov
04/13/2010 | 05:19 PM
Intel has long established itself as the maker of the fastest processors for desktop PCs. But while there can be arguments as to what CPU models are optimal today in the mainstream and low-end categories, there is no competition in the top price range. Intel’s Core i7 series has no alternatives in AMD’s product line-up. At least, this is true when I’m writing this and there are still a few weeks till the arrival of AMD’s six-core Phenom II processors codenamed Thuban. As a matter of fact, the already available 4-core Phenom II processors may even be more appealing as they are but a few dozen percent slower than the Core i7 series while being many times cheaper, but anyway. PC enthusiasts are ready to pay for the maximum performance they can get, therefore the Core i7 series enjoys high popularity.
Even without direct market competition, this consumer interest towards high-performance and expensive CPUs makes Intel keep on developing its top-end products by increasing their clock rates, introducing microarchitectural improvements, and endowing them with more and more cores. The hero of this review is the recently announced model of the Core i7 series which is the first desktop CPU with as many as six cores!
It must be noted that the arrival of the six-core Core i7 does not mark a beginning of a six-core revolution. Right now, Intel only offers one such CPU, the Core i7-980X Extreme Edition. It is a kind of a demo sample that is targeted as wealthy enthusiasts who are ready to shell out a thousand bucks for the processor alone! That’s how things will stand until this fall when one more, less expensive, six-core model is scheduled for release. And there will still be a long way till the mass arrival of CPUs with more than four cores to the market. I mean Intel’s CPUs, of course. AMD has its own vision and is going to start selling midrange six-core CPUs in near future, but we don’t yet have the opportunity to play with such products in our test labs.
The Core i7-980X is based on the new semiconductor die Gulftown that incorporates six processing cores and a 12-megabyte L3 cache. The implementation of all these units in a monolithic silicon die was made possible by 32nm tech process. The same manufacturing process is partially used for the Clarkdale series but the Core i7-980X is the first product to apply it all the way from top to bottom. Thus, it is the Core i7-980X that is going to best illustrate the evolution of the Nehalem microarchitecture. The recently announced Core i5 and Core i3 processors are a poor illustration because the distribution of CPU subunits in two semiconductor dies one of which is manufactured on 45nm tech process resulted in bottlenecks that had a negative effect on the consumer properties of the end products.
In other words, the Core i7-980X is the best Intel’s engineers can do at the moment by applying their progressive tech process to their most advanced microarchitecture. This makes the Gulftown interesting from a theoretical point of view. Practically, such CPUs will be limited to luxurious PC configurations and will not make it to the mass market this year. Intel has no plans to offer cheaper versions of the Gulftown in 2011, either, because the company is going to move on right to the next microarchitecture generation codenamed Sandy Bridge.
Although I have introduced the Core i7-980X as a revolutionary product, I have nothing sensational to tell about its microarchitecture. Intel’s engineers simply put together a six-core CPU out of their Nehalem building blocks: computing cores, L3 cache, memory controller and QPI bus controller. But now there are six cores and 12 megabytes of L3 cache memory. All of these parts fitted into a single die thanks to the new 32nm tech process. As a result, although a Gulftown die consists of 1170 million transistors, which is about 1.6 times the transistor count of the Bloomfield die, it has an area of 248 sq. mm compared to the Bloomfield’s 263 sq. mm.
Judging by the photo of a Gulftown die and the placement of subunits, the new processor is just the result of the older model’s transitioning to the new tech process with minor corrections.
The two additional cores put aside, this is indeed so. The computing cores and the memory controller of the Core i7-980X are absolutely the same as those of the Core i7-900 series processors that have been produced for over one year already. The difference is in the manufacturing technology. The single innovation is the introduction of seven AES-NI instructions that accelerate cryptographic algorithms but we already know this instruction set by Clarkdale CPUs.
Let’s compare the key specifications of the new model in comparison with the Core i7-975, the senior model of the Bloomfield generation the new six-core flagship has come to replace.
The memory and QPI controllers of the Gulftown have the same specs as those of the Bloomfield, meaning that these CPUs can be used on the same platforms. The Gulftown does not have a PCI Express controller. The graphics subsystem is supported by the well-known Intel X58 Express chipset.
Thus, the Core i7-980X comes in LGA1366 packaging and is perfectly compatible with LGA1366 mainboards. A BIOS update is the only thing necessary for such mainboards to identify the new CPU.
Notwithstanding its extra cores, the Core i7-980X has the same TDP as its four-core predecessors. The transition to the more progressive tech process is not accompanied with a reduction in CPU voltage as the CPU-Z screenshot shows.
However, Intel has equipped its 6-core processor with a new tower-like cooler that has four 6mm heat pipes and a 2-speed 100mm fan.
This is a step towards enthusiasts who don’t have to use a third-party cooler now. The boxed cooler is quite efficient.
Touting the Gulftown as the highest-performance CPU available today, Intel puts an emphasis on two of its features: the increased number of computing cores and the larger amount of cache. However, it is clear that there are not so many applications capable of loading six CPU cores with work. These are mostly 3D modeling and digital content processing/creation software. For more popular applications, the 12MB L3 cache is going to be the key performance-boosting factor. Thanks to it, the new CPU is going to be faster at old tasks which are not optimized for multithreaded execution. The L3 cache is common for all the cores and can be monopolized by one or several cores depending on load.
A simple increase in the amount of CPU cache is always accompanied with some negative effects. Intel’s engineers did not change the logical organization of the L3 cache, so it is still 16-way associative. Considering the increased size and the arbitration among more CPU cores, the cache latency has increased by 33%.
The second factor that may have a negative effect on performance is that the uncore part, including the L3 cache and the memory controller, has a reduced frequency in the Gulftown. Intel tried this with the Lynnfield series and achieved a considerable reduction in power consumption by reducing the frequency and voltage of the L3 cache. The same goes for the new CPU. Its memory subsystem performance has been sacrificed for the sake of the two extra computing cores. Otherwise, the six-core Core i7-980X would not fit within the 130-watt TDP specified for LGA1366 processors.
Summing up the cache memory specs of the senior Gulftown, Bloomfield and Lynnfield processors, we get an ambiguous picture.
Clearly, the Gulftown is inferior to its predecessor in cache and system memory performance. The difference can be measured with Everest Cache & Memory Benchmark (I used DDR3-1600 SDRAM with 9-9-9-24 timings).
Core i7-980X (Gulftown)
Core i7-975 (Bloomfield)
The difference in practical cache performance is obvious: the Bloomfield is 33% better than the Gulftown in the speed of reading from the L3 cache and 25% better in terms of latency. The new CPU is also inferior when it comes to working with system memory. The practical bandwidth and latency of the six-core CPU is 15-20% worse than those of its four-core predecessor that seemingly has the same 3-channel DDR3-SDRAM controller.
Thus, despite the extra computing cores and the larger cache, the Core i7-980X may be slower than the Core i7-975 in practical applications due to objective reasons. This explains the rather low model number of the new product, by the way. The new Gulftown is not better than the Bloomfield from every aspect. It is weaker in some important respects.
Turbo Boost and Hyper-Threading technologies were introduced in the very first Bloomfield processors. They have passed the test of time and proved their effectiveness. Hyper-Threading helps make the system faster under multithreaded load whereas Turbo Boost improves performance when only some of the CPU cores are loaded. Both technologies have been transferred to the new six-core Gulftown.
Hyper-Threading adds six more virtual cores to the six physical cores of the Core i7-980X, so the operating system sees as many as 12 cores as the result.
This screenshot raises the question of are there any applications capable of using up all these resources? Besides, the cores share a single memory bus, and the latter may prove to be a bottleneck, unable to deliver data quickly to all the cores. I carried out a small test to check this out. I measured the frame rate of a popular 3D shooter while running some resource-consuming applications in the background. To be specific, I benchmarked the system in Far Cry 2, running a few copies of the WinRAR-integrated benchmark (which itself supports multithreading). The computer’s memory worked as DDR3-1600. I also performed this test on platforms with top-end Bloomfield and Lynnfield processors for the sake of comparison.
The Gulftown copes better than its four-core cousins with the multithreaded load. Its performance lowers less than that of its opponents when the background load increases, so the 3-channel memory subsystem seems to be sufficient for multithreaded environments.
The Turbo Boost implementation in the Core i7-980X is somewhat disappointing. After Lynnfield processors for the LGA1156 platform acquired the ability to increase the clock rate by 667 MHz above the default one, I expected to see something like that in the Gulftown as well, but Intel decided differently. The new six-core CPU is as conservative as the Bloomfield series in terms of Turbo Boost. Its frequency can be increased by means of that technology from the default 3.33 GHz by 266 MHz only – to 3.6 GHz. The table shows the Turbo Boost frequencies of the senior models of Gulftown, Bloomfield and Lynnfield processors.
Thus, the highest frequency of all top-end CPUs with the Nehalem microarchitecture is the same 3.6 GHz. Officially, the Core i7-980X can maintain that frequency even for two CPU cores, but I could only observe it working at 3.6 GHz when executing a single thread. When a second core was under load, the clock rate lowered to 3.46 GHz.
However, it must be noted that the CPU overclocking based on Turbo Boost depends not only on the activity of the cores but also on the power consumption of the CPU. Perhaps, the inability of the Core i7-980X to work at 3.6 GHz under dual-threaded load is due to its power consumption going beyond the prescribed limits.
I am sure no one doubts that Core i7-980X is one of the fastest CPUs out there. Therefore, in our performance tests we are going to compare it against two of the fastest quad-core Intel CPUs from the Core i7 series and the top Phenom II X4 processor. As a result, we used the following hardware to build our testbeds:
SYSmark 2007 shows the performance of a computer when executing typical scenarios in real-life applications. This test highlights the downsides of the Gulftown I have talked about above. When the application is not optimized for multiple-core processor architectures, the Core i7-980X may fall behind its four-core predecessor Core i7-975. That’s what we see in the E-Learning and Productivity scenarios: the CPU with a faster memory controller and a more effective L3 cache is ahead of the CPU with more computing cores. The digital content creation and processing scenarios put the Gulftown on top, which is no surprise as such applications usually boast good optimization for multiple CPU cores. The resulting total scores of the new Core i7-980X and Core i7-975 are similar in SYSmark 2007.
Many modern games can make good use of dual-core CPUs. Some of them can effectively use four CPU cores, too. But they don’t seem to be able to fully load the six-core Gulftown, let alone with Hyper-Threading technology turned on. This explains the small difference between the Core i7-980X and the Core i7-975. The large 12MB L3 cache has a bigger effect on gaming applications. It is thanks to this factor that Intel’s new CPU can appeal to gamers.
3DMark Vantage can load any number of CPU cores. Therefore, the Core i7-980X delivers its very best here. It looks like from now on, new world records in this benchmarking suite are going to be set by computers with this CPU.
Adobe Photoshop is optimized for multiple-core CPUs but not all of its operations and filters can benefit from as many cores as possible. Therefore the six-core CPU does not enjoy an overwhelming advantage. Besides the two extra cores, the increased L3 cache contributes greatly to its result.
Video encoding is a task that can be easily paralleled to multiple CPU cores. The six-core Core i7-980X is over 40% faster than the four-core Core i7-975!
We’ve got the same picture at high-definition video editing in Premier Pro.
WinRAR can make good use of multiple CPU cores, too, but the performance benefits from a fourth core and more are really negligible. That’s why the Core i7-980X and Core i7-975 are similar in this test. The six-core processor’s 12MB L3 cache is not very efficient as the effect from its large capacity is negated by its high latency.
Excel 2007 can parallel its mathematics well enough. The test task is solved much faster on the newer CPU.
The software audio studio Sonar 8 Producer runs faster on the six-core CPU when mixing tracks. The Core i7-980X is about 5% better than the Core i7-975.
Final rendering is the type of load that eagerly reacts to any extra CPU core available. The 20% advantage of the Core i7-980X over its opponent is an expected result.
I will also check the CPUs out under single-threaded load by benchmarking them in the computing test MaxxPi and the chess program Fritz, in which I manually set the number of CPU cores at 1. These tests is interesting because senior CPU models in the Core i7 series feature Turbo Boost technology which can increase their clock rate to 3.6 GHz when only one core is in use.
As you can see, the Core i7-980X and Core i7-975 are roughly similar in these tests, the older CPU enjoying some advantage due to faster cache memory. The Core i7-870 comes closer to the leaders, being mostly limited by its lower memory bandwidth.
Formally, larger the number of cores in the new Core i7-980X processor didn’t lead to any changes in the calculated TDP. It is LGA1366 compatible in TDP due to more advanced production process used to manufacture semiconductor Gulftown dies as well as due to lower Uncore frequency and voltage. As a result, the calculated TDP of the Core i7-980X processor is 130 W, just like the TDP of the Core i7-975 CPU.
Nevertheless, in order to get a more detailed picture we also performed some power consumption tests. The graphs below show total system power consumption (without the monitor) measured “past” the power supply unit and representing overall power consumption of all system components. In this case we do not take into account the efficiency of the PSU itself. During our tests we used 64-bit LinX 0.6.3 utility to load the systems to the utmost extent. Moreover, to ensure that we estimate the power consumption in idle mode correctly we activated all power-saving technologies, such as C1E, AMD Cool'nQuiet 3.0 and Enhanced Intel SpeedStep.
In idle mode LGA1366 platform consumes more power than other tested platforms independent of the processor type used in it. it can be explained by the fact that Intel X58 Express chipset is pretty power-hungry. As for the actual CPU power consumption, in idle mode it doesn’t exceed a few watts.
Under heavy load things are much more interesting. The new six-core processor turns out even more energy-efficient than its quad-core fellow – Core i7-975. However, 32 nm production process doesn’t really work any specific wonders and Core i7-980X remains pretty power consuming: its numbers are higher than those of the top LGA1156 and Socket AM3 solutions. On the other hand, keeping in mind that Gulftown have 1.5 times higher computational potential, the energy-efficiency (the performance-to-consumption ratio) is also on a completely different level in this case.
The introduction of new manufacturing processor for CPU production usually leads to increase in the processor frequency potential. Core i7-980X is the first CPU manufactured solely with 32 nm process. Therefore, we are particularly interested in finding out how overclockable it is.
The only currently available Gulftown CPU model belongs to the Extreme Edition series. It means that Intel doesn’t lock its clock multiplier giving the user the opportunity to overclock this processor easily. We took advantage of this opportunity during out overclocking experiments. To dissipate the generated heat we utilized Thermalright Ultra-120 eXtreme air cooler.
First of all, we tried to find out what the maximum overclocking would be for our Core i7-980X without increasing its core voltage above the default 1.2 V. As we have just revealed in our recent article, this type of overclocking is the most effective in terms of energy-efficiency and doesn’t cause any dramatic increase in power consumption or heat dissipation.
Our practical tests showed that if we don’t tweak the CPU Vcore, the system loses stability at only 3.6 GHz frequency.
Unfortunately, this frequency is extremely close to the nominal that is why it will hardly make enthusiasts happy. Therefore, the second part of our experiments was performed with the CPU core voltage increased to 1.35 V. Especially, since we all know from Clarkdale’s example that processors manufactured with 32 nm process should respond very nicely to their core voltage increase.
Due to higher Vcore we could get our CPU to run stably at a much higher clock rate of 4.13 GHz.
Frankly speaking, this is not what we have expected to see during our Core i7-980X overclocking experiments. It turns out that although this processor was made using the most advanced production process, it doesn’t overclock any better than a one-year-old 45 nm processor. In other words, if you overclock without involving any special cooling techniques or equipment, then the frequency potential of the new Gulftown processor will be about the same as that of Bloomfield CPUs that can go as high as 4.0-4.2 GHz.
By the way, I would like to point out two things, whichw e noticed during Core i7-980X overclocking. First, Gulftown remains relatively cool even when its frequency increase is accompanied by the corresponding Vcore increase. 60 degrees under maximum load is very low temperature, especially compared with the temperatures of Core i7 Bloomfield processors overclocked by raising their core voltage. Second, you need to carefully select your voltage settings if you want to succeed in Gulftown overclocking, because by raising it too high you can damage its overclocking potential. For example, our processor sample worked at 4.13 GHz when its Vcore was increased by 0.15 V above the nominal, but as soon as we pushed it 0.2 V higher, the CPU failed to pass stability tests even at 4.0 GHz frequency.
Although Gulftown processor is not only the first six-core CPU, but also the first processor manufactured exclusively with 32 nm process, , we wouldn’t call it a new generation solution. In fact, Intel offered us exactly the same things that we have already seen in Bloomfield processors, with that only difference that they decided to introduce a new processor model in the Core i7 family by adding new cores rather than pushing the clock frequency up. And taking into account that Nehalem processors have internal block structure, this is not that innovative at all.
As a result, the new Core i7-980X theoretically has 1.5 times higher performance, which allows us to formally regard it as the fastest desktop processor these days. In reality, however, everything depends on the applications optimization. As our tests showed, there are not that many tasks that could benefit substantially from having six cores available to them and they deal mostly with digital content creation and processing. So, it appears that Core i7-980X is an excellent solution for a workstation, rather than a home computer system.
No wonder that at their Gulftown launch Intel decided to go with only one $999 model. In everyday work there is no real need for a six-core processor. Moreover, under special circumstances Gulftown CPU may be even slower than its quad-core counterparts because of increased L3 cache latency and slower memory controller. That is why the new Core i7-980X is obviously targeted for those wealthy enthusiasts who go for the new stuff primarily out of curiosity rather than reasonable interest. Even after the Core i7-980X launch, some pragmatic users will most likely be still extremely interested in the existing quad-core processors, which are fast enough for everyday work as well as contemporary 3D gaming. Besides, the new 32 nm process also doesn’t really bring in too many benefits. As our tests showed, Core i7-980X became only a little bit more energy-efficient than its quad-core LGA1366 predecessors, and its overclocking potential doesn’t even exceed that of the 45 nm CPUs.
Overall, we will have to wait at least until early 2011 in order to see some really innovative Intel CPUs, which could become interesting for a vast majority of users out there. At this time the microprocessor giant should release their dual- and quad-core solutions with refreshed sandy Bridge microarchitecture, which will be manufactured with 32 nm process. As for the new CPU we discussed today, all we can say is: nothing really special.