Starting with Sandy Bridge, each subsequent generation Intel processors accelerates worse than the previous one, but in the new generations through microarchitectural improvements up slightly performance. Let’s evaluate in practice, which of the Core i5 I choose to overclock the system
Almost every day, as reports from the front, we read with sadness the news that the desktop market continues to be deprived of his loyal supporters. The loss is not just an army of users. One by one drop of the number of adherents of the classical desktop and hardware manufacturers. But the most insulting happens when some of the companies that have made a name for himself and earn huge capital is on the desktop market, found traitors and saboteurs, verbally declaring unwavering loyalty to the old ideals, but in reality – not just watching, but actively go forth “on the side “(mobile devices, of course). A blatant example of such a treacherous infidelity, which has not eclipsed in the memory of some new terrible betrayal, recently showed us by Intel.
Yes, we are talking about Haswell. About the processor, which was originally presented as a new cycle of high-performance micro-architecture design, but in fact was directly and profoundly adapted for use in low-power portable computing systems. The same Haswell, who eventually got a users desktop, wags dubbed Hasfail not just happen. Core desktop processors fourth generation, based on a new microprocessor design, steel for Intel-product with all the ensuing consequences. Our review of the Core i7-4770K has exposed major flaws: the lack of clear progress in computational performance and overclocking potential deterioration. The conclusion from all of this was done then is simple: upgrade the existing system and the transition to the new LGA1150 platform makes no sense.
However, since the announcement of Haswell’s been a few weeks, and former anger subsided a little. In head began to creep into the thought of me not too hasty in branding the new processor design shame? Perhaps Haswell desktop-can still be interesting, because these processors are still present some improvements. In other words, there is a need in a fresh look.
But to repeat the second time we have made tests, of course, will not. Today we look at Haswell different angle. Namely – to try to understand which of Intel’s processors should be available to the enthusiast, is located for this purpose budget of about $ 200-250. That is, try to answer the question of which of the available in stores overclocking Core i5 has the greatest practical value to date. Since the days of Sandy Bridge in each new generation of desktop CPU we have seen small steps toward improving productivity, on the one hand, but a systematic rollback in overclocking – on the other. Therefore, choosing a modern platform, advanced users today actually face the trilemma: Sandy Bridge, Ivy Bridge and Haswell. In this article, we have decided to directly compare all three available options: Core i5-2550K, Core i5-3570K and Core i5-4670K.
We are all accustomed to the fact that the newer processor, the better it is. And until recently, it did work. Improved manufacturing processes. This results in a potential increase in the frequency and an increase in the processing of semiconductor crystals. The increased transistor budget is consumed either on the microarchitectural innovations, or to increase the number of cores or an increase in the cache memory.
However, with the advent of Sandy Bridge generation of processors usual pace of progress was slow. Even despite the fact that for the production of Sandy Bridge is used by 32-nm process technology, and for a new Ivy Bridge and Haswell – 22-nm technology, all of these three generations of desktop multi-core processors have a similar structure, working at very close clock speeds and have the same Amount of cache memory. In fact, all affect the performance differences were now recessed into the bowels of microarchitecture.
In principle, the fact that the formal specification of processors for desktop systems since 2011 has stopped the growth of the baseline, there is nothing to worry about. As we know from previous experience, the microarchitectural improvements capable of much. Especially that Ivy Bridge, and Haswell – it’s not simple “tick” in Intel’s terminology. Even on Ivy Bridge, the output of which was associated with the change of process technology, Intel said as a measure “tick +”, stressing that it is not a simple transfer to the new Sandy Bridge technology rails, and a comprehensive revision of the old design. Haswell is generally refers to a cycle of “just so”, that is, is a new version of the microarchitecture without any reservations. Therefore, it was possible to improve the performance expected from the development of existing Intel processors, though it is not accompanied by a change in the numbers in the list of formal characteristics.
However, no rapid productivity growth of desktop processors are not really there. The reason is that the main developers of Intel’s efforts are not directed towards the improvement of computing power – it is more than enough to leave the competition far behind – and on improving the parameters that are critical for the mobile market. At the same time wanting to outdo and hybrid processors AMD, and mobile processors with the architecture of ARM, Intel systematically optimizes heat dissipation and power consumption, and is engaged in pulling up your own graphics core. For the same desktop processors, these parameters are insignificant, therefore, from the point of view of the users of desktop computers, the development of Sandy Bridge → Ivy Bridge → Haswell looks like a manifestation of technological infantilism.
Let’s try to remember what happened to the processor cores since 2011, when markets were first Sandy Bridge c truly innovative micro-architecture with a completely redesigned circuit extraordinary performance teams. The original design of Sandy Bridge has a strong basis for all subsequent generations microarchitecture. That’s when there are key and pressing still elements such as ring bus, cache decoded instructions “zero level” fundamentally new block branch prediction scheme performance 256-bit vector instructions, and more. After Sandy Bridge Intel’s engineers were limited to a few changes and additions without affecting laid in the foundation of the microarchitecture.
In published a year later Ivy Bridge processor family touched the progress of processing cores in a very small degree. As the front of the pipeline, designed to process four instructions per clock cycle, and the whole scheme of order execution of commands kept fully intact. However, the performance of Ivy Bridge still has a bit higher than its predecessors. It was achieved by three small steps. First, there was a long overdue for dynamic resource allocation between the internal data structures flows while before all queues and buffers based on Hyper-Threading divided into two streams rigidly half. Secondly, the assembly is optimized performance and a real fission rate whereby these operations has doubled. And third, the task of processing data transfer operations between registers has been removed from the actuators and the corresponding commands are translated into a simple dereference registers.
With the advent of computing performance Haswell again grew slightly. And while talking about a qualitative leap there is no reason, a set of innovations seems not erundovskim. This processor design engineers dug deep in the middle of the assembly line, so that in Haswell has increased the number of execution ports (by the way, for the first time since 2006). Instead, they became six eight, so in theory throughput pipeline Haswell was a third more. However, a number of steps have been taken to ensure that these ports to provide all the work, that is to improve the capabilities of the processor for parallel execution of instructions. To this end, the algorithms have been optimized branch prediction and increased the volume of internal buffers: first of all – the windows of the extraordinary performance of the teams. At the same time, Intel engineers have expanded instruction set by adding a subset of instructions AVX2. The main property of this set – FMA-team, combining at once a couple of operations on floating point numbers. Thanks to them, the theoretical performance Haswell at operations on floating-point single-and double-precision has doubled. Not to be overlooked and left subsystem work with the data. Expanding domestic processor parallelism, as well as the emergence of new instructions, tossing and turning large amounts of data, require the developer to speed up the cache. Therefore the capacity of the L1-and L2-cache in Haswell, compared with previous generations of processor designs has been doubled.
However, enthusiasts when new generations of processors want to see is not so much an extensive list of changes made as the increased bars on the charts with the performance in applications. Therefore, our theoretical calculations and we will add the results of practical tests. And for the best illustrative first of all we will resort to a synthetic benchmark, which allows to see the change of the isolation of the various aspects of the performance of the overall picture. For this purpose the great popular benchmark tool SiSoftware Sandra 2013, using which we compared among three quad-core processors (Sandy Bridge, Ivy Bridge and Haswell), clock frequency which has been brought to a uniform and constant value of 3.6 GHz. Please note figures are shown in the graphs Haswell twice. Once – when testing algorithms are not used new sets of instructions embedded in the processor design, and the second time – with activated instructions AVX2.
The usual arithmetic test reveals that Haswell was a noticeable increase in performance integer operations. The increase in speed is clearly associated with the appearance of the micro port, specially allocated for additional integer arithmetic logic unit. As for the speed of standard floating-point operations, it is the release of new generations of processors does not change. This is understandable, because the rate is now done on the introduction into practice of new instruction sets with higher bit depth.
In evaluating the performance of multimedia on the first place there is the speed of the vector instructions. So here Haswell advantage is particularly so when using a set of AVX2. If the new regulations to exclude from consideration, we see only a 7 percent increase in performance compared to Ivy Bridge. Which, in turn, rapidly Sandy Bridge only 1-2 percent.
Similarly it is with the speed of the cryptographic algorithms. The commissioning of the new generation microarchitecture boosts productivity by only a few percent. Significant increase in speed can be obtained only if use Haswell and new commands. But we should not delude ourselves: Benefitting from AVX2 in real life requires a rewriting of code, and it is known – the process is not fast.
Not too optimistic and look what happened to the latency of cache memory.
Sandy Bridge Ivy Bridge Haswell
L1D-Cache 4 4 4
L2-cache 12 12 12
L3-cache 18 19 21
Cache in the third level in Haswell really works with long delays than the previous generation of processors, as Uncore-part of this processor has received relatively asynchronous clocking cores.
However, increasing the delay is compensated by a two-fold increase in bandwidth occurring not only in theory but also in practice.
Bandwidth, GB / s
Sandy Bridge Ivy Bridge Haswell
L1D-Cache 510.68 507.64 980.79
L2-Cache 377.37 381.63 596.7
L3-cache 188.5 193.38 206.12
But in general, against the Haswell microarchitecture Sandy Bridge did not seem noticeable progress. The principal advantage is observed only for making use of the instruction set AVX2, and watch it while you can only synthetic benchmarks, since the actual software must still go through the long haul optimization and adaptation. If the new regulations do not take into consideration, the average level of excellence Haswell on Sandy Bridge is about 10 percent. And the gap is old men Sandy Bridge should be fully able to overcome due to overclocking. Especially when you consider the fact that the old processor frequency potential is higher than that of their modern followers.
If you go to the store and see what kind of overclocking Core i5 processor family is available, then the choice will come down to three options relating to different generations: Core i5-2550K, Core i5-3570K and Core i5-4670K. For clarity comparable to their characteristics:
Core i5-2550K Core i5-3570K Core i5-4670K
Microarchitecture Sandy Bridge Ivy Bridge Haswell
Cores / Threads 4/4 4/4 4/4
Hyper-Threading Technology No No No
Clock frequency of 3.4 GHz, 3.4 GHz, 3.4 GHz
The maximum frequency of 3.8 GHz in turbo mode 3.8 GHz, 3.8 GHz
TDP 95 W 77 W 84 W
Manufacturing Technology 32 nm 22 nm 22 nm
No HD Graphics 4000 4600
GPU frequency – 1150 MHz 1200 MHz
L3-cache 6 MB 6 MB 6 MB
Support for DDR3 1333 1333/1600 1333/1600
Extensions instruction set AVX AVX AVX 2,0
LGA1155 LGA1155 LGA1150 Package
Three Core i5 different generations look at this table, almost like twins. However, a more detailed introduction to each of these three processors reveals some interesting nuances.
Core i5-2550K. This is – one of the latest models of Sandy Bridge. It was released a year after the announcement of the main and discontinued only recently and therefore still widely available in retail. But if you’re serious about building a system with a processor Core i5-2550K, it is our duty to recall some of the highlights.
First, despite the fact that the formal characteristics of the operating frequencies of all the older Core i5 models are designated the same way: from 3.4 to 3.8 GHz, in fact, Core i5-2550K in a regular mode operates at a slightly lower frequency than processors later versions microarchitecture. The fact that the Turbo Boost technology in Sandy Bridge is not as aggressive as in the Ivy Bridge and Haswell, and at full load exceeds the rated frequency is 100, not 200 MHz.
Secondly, processors Sandy Bridge – and the Core i5-2550K among them – have somewhat less flexible memory controller, rather than Ivy Bridge and Haswell. Overclocker memory with frequencies up to DDR3-2400 support it, but here is a step change in this frequency is 266 MHz. That is, the choice of modes of memory with Core i5-2550K is somewhat limited.
And thirdly, Core i5-2550K – the only one of Intel’s overclocking processors, devoid of the GPU. In actual fact the core of the semiconductor chip is, but it severely disabled during the assembly of the processor. This, by the way – one of the reasons why the Core i5-2550K overclocks well.
However, the main basis of attractiveness Core i5-2550K as an object for overclocking is that Sandy Bridge – this is the last of the families of desktop CPU Intel’s mid-range, where the thermal interface between the semiconductor chip and the processor cover is used for special solder flux-free soldering, and not plastic material with questionable thermal conductivity. Followed later translation of semiconductor manufacturing at the 22 nm technology and the accompanying this step reduces heat generation Intel crystals considered a sufficient argument to simplify the procedures for the CPU assembly by eliminating soldering. However, overclockers from severely affected, since the thermal interface between the processor die and its lid has suddenly become a significant obstacle to the transfer of heat flow and the organization of a good cooling.
Core i5-3570K. A typical support design Ivy Bridge – the first generation of Intel’s processors released on 22-nm process technology. The use of more sophisticated than before, the process allowed Intel to significantly reduce the CPU heat dissipation and power consumption. The systems constructed on the base of Core i5-3570K, obviously more economical than similar configurations on Sandy Bridge. However, this advantage Intel convert to higher clock rates did not. Operating frequencies of the older third-generation Core i5, Core i5-3570K, the frequencies Core i5-2550K is almost identical.
What’s worse, despite a lower operating voltage and heat generation in the nominal mode, dispersed generation Ivy Bridge processors are much less likely than their predecessors.