Recomiendo mirar los gráficos de un artículo aparecido ayer en Digit-life que muestra las diferencias entre los procesadores de dos y cuatro núcleos. El análisis realizado por Digit-life se hizo basado en un equipo base ASUS Maximus Extreme, una tarjeta gráfica NVIDIA 8800GTX y 4GB de memoria RAM. Con dicho equipo han pasado diversos tests, tanto sintéticos como de tareas reales. Y los resultados dejan claro lo que significa aumentar el número de núcleos de una CPU.
Los procesadores comparados son el E6600 y Q6600 (ambos de 65 nm y 2,4 Ghz) y el E7200 y Q9300 (ambos de 45 nm y frecuencias respectivas 2,53 Ghz y 2,5 Ghz). Los resultados de las pruebas s se han comparado teniendo como referenci los del Q6600. Los resultados hablan por sí solos y com oera de esperar no es un aumento lineal de rendimiento. Según cuán optimizada esté la aplicación así se verá beneficiada de tener más hilos, pero por lo visto en el caso más favorable para los procesadores de cuatro núcleos el rendimiento es algo superior a un 30% de sus equivalentes en velocidad de doble núcleo.
El mundo de procesadores multinúcleo avanza muy rápido pero ahora le toca al software dar el salto. Un dato a tener en cuenta es la diferencia de consumo de los procesadores de cuatro núcleos, que en el caso de 65nm es un 42% superior. Y en los de 45nm duplica y pasa de largo el consumo de su equivalente doble núcleo.
Dual-Core Versus Quad-Core
Intel Core 2 Duo/Quad Processors of New and Old Generations
This time we’ll try our new test procedure in an interesting but “academic” research. This article won’t feature many new processors. All the four contenders belong to the Intel Core 2 family, and the main problems addressed in this article have nothing to do with the question which model is the fastest.
It’s very interesting to compare quad-core processors with dual-core ones, as the cheapest Intel Core 2 Quad (Q6600) already costs about $200. This price border separates CPUs for well-off users from popular processors on the Russian market. So common users may be tempted to buy a real quad-core processor, even if it does not operate at a high clock rate. We only have to find out whether it’s an objectively expedient decision. Of course, a quad-core processor will always be more expensive than a dual-core processor on the same core operating at the same frequency, it will consume more power and get hotter. Besides, a quad-core processor needs a more efficient and thus either a more expensive or noisier cooler, and probably a more powerful power supply unit… So, if the above-mentioned issues are not compensated by a serious advantage in performance, this whole idea will make no sense. That’s what we’ll find out in this article.
Our contenders include only two relatively new processors – Intel Core 2 Duo E7200 on Wolfdale core and Intel Core 2 Quad Q9300 on Yorkfield core. If you are interested in details, you can read our article about Core 2 Extreme QX9650, which describes differences between Yorkfield and older quad-core Kentsfield. What concerns differences between Wolfdale and Conroe, they fall within the description in this article for the only exception – they are dual-core processors.
Hardware and Software
|Intel Core 2 Duo E6600||ASUS Maximus Extreme||Corsair CM3X1024-1800C7DIN||GeForce 8800 GTX|
|Intel Core 2 Duo E7200||ASUS Maximus Extreme||Corsair CM3X1024-1800C7DIN||GeForce 8800 GTX|
|Intel Core 2 Quad Q6600||ASUS Maximus Extreme||Corsair CM3X1024-1800C7DIN||GeForce 8800 GTX|
|Intel Core 2 Quad Q9300||ASUS Maximus Extreme||Corsair CM3X1024-1800C7DIN||GeForce 8800 GTX|
- Memory: 4 GB (4 x 1 GB modules)
- HDD: Samsung HD401LJ (SATA-2)
- Cooler: Thermaltake TMG i1
- Power supply unit: Cooler Master RS-A00-EMBA
|Processor||Core 2 Duo E6600||Core 2 Duo E7200||Core 2 Quad Q6600||Core 2 Quad Q9300|
|Process technology, nm||65||45||65||45|
|Core clock, GHz||2.4||2.53||2.4||2.5|
|# of cores||2||2||4||4|
|L1 cache, I/D, KB*||32/32||32/32||32/32||32/32|
|L2 cache, KB**||4096||3072||8192||6144|
|FSB clock***, MHz||266 (1066)||266 (1066)||266 (1066)||333 (1333)|
|Heat dissipation****||65 W||65 W||95 W||95 W|
* – in multi-core processors – per single core
** – “X x Y” means “X KB per each of Y cores”
*** In AMD processors this is memory controller bus clock rate;
**** Measured differently for Intel and AMD processors; impossible to compare directly.
|64-bit application||Multi-threaded application*|
|Microsoft Windows XP Professional SP2||+||+|
|Microsoft Windows Vista Ultimate SP1||+||+|
|Autodesk 3ds max 9 SP2||+||+|
|V-Ray 1.5 SP1||+||+|
|Autodesk Maya 2008 Ultimate||+||+|
|NewTek Lightwave 3D 9.2||+||+|
|SolidWorks 2007 SP0.0||+||+|
|PTC Pro/ENGINEER Wildfire 3.0 M120||+||–|
|UGS NX5 126.96.36.199||+||+|
|Wolfram Research Mathematica 6||+||+|
|MapleSoft Maple 11||–||+|
|MathWorks MATLAB 2007||+||+|
|Adobe Photoshop CS3 10.0||–||+|
|Microsoft Visual Studio 2008||+||+|
|Apache HTTP Server 2.2.8||–||+|
|MySQL Community Server 5.0.51a||–||+|
|ACDSee 10 Photo Manager||–||+|
|xat.com Image Optimizer 5.10||–||–|
|Musepack MPC Encoder 1.16||–||–|
|Nero Digital Audio Encoder 188.8.131.52||–||+|
|Ogg Encoder 2.83 (Lancer)||–||+|
|Canopus ProCoder 3.0||–||+|
|DivX Codec 6.8.2||–||+|
|XviD Codec 1.1.3 Final||–||–|
|x264 Codec rev 807||–||+|
|Call of Duty 4: Modern Warfare (Patch 1.5)||–||+|
|Call of Juarez (Patch 184.108.40.206) + DX10 Enhancements Pack||–||–|
|Crysis (Patch 1.2)||+||+|
|S.T.A.L.K.E.R. (Patch 1.006)||–||+|
|Unreal Tournament 3 (Patch 1.2)||–||+|
|Company of Heroes (Patch 1.71)||–||+|
|World in Conflict (Patch 1.007)||–||+|
* – not just generating several threads by the process, but two or more simultaneously active threads in the process of running tests.
Necessary preface to the charts
Our test procedure features two peculiarities of data representation: firstly, all data types are reduced to one – integer relative score (performance of a given processor relative to Intel Core 2 Quad Q6600, its performance taken for 100 points), and secondly, detailed results are published in a Microsoft Excel spreadsheet, while the article contains only summary charts for benchmark classes. Nevertheless, we’ll sometimes draw your attention to detailed results, if they are worthy of your attention.
Professional group of tests
3D Modeling and Rendering
Quad-core processors seem to perform very well in this group (their advantage over dual-core processors is impressive). But if we have a look at the detailed results, you will see where it comes from in the average score. Owing to rendering speed tests, of course. It’s not a secret that rendering can be easily distributed between many processors, and performance gains are close to ideal (100% per each new core). What concerns tests of interactive operations in 3D Modeling applications, the situation with quad-core processors is not that peachy anymore, and performance gains are limited to 10%.
It’s an excellent example of absolute uselessness of quad-core processors for such software. Cannot be better: E6600 scores as many points as Q6600, E7200 is on a par with Q9300. The choice is crystal clear: “unassuming” dual-core processors, preferably on the new Wolfdale core.
On one hand, we cannot say that there is no difference between dual-core and quad-core processors here. On the other hand, it’s not that big. We monitored CPU load during compilation of this project (about 30 minutes) and noticed that it rarely used all four cores simultaneously, while periods with only one loaded core were quite lengthy. However, it’s quite a typical project from the point of view of modern C++ style – so its compilation peculiarities are also typical.
Professional Photo Processing
It’s the second test, where quad-core processors “broke loose”. On the average, four cores are 30% as fast as two cores! You may object that the ideal advantage must reach 100%… But let’s be realists: even 30% is a very good result these days, it’s quite rare.
Scientific software cannot boast of optimizations for quad-core processors: even if we have a look at detailed results, we’ll see that the maximum performance gain from additional two cores is about 10%.
As we use two benchmarks, one of which (PHP Calculator) can use several threads, while the other (PHPSpeed) rarely uses even two cores – the average score is something in between. Indeed, quad-core processors enjoy almost 100% performance gains in PHP Calculator. And PHPSpeed needs mostly performance of a single core.
Total Professional Score
A difference between dual-core and quad-core processors in each of our competing pairs amounts to ten points. On one hand, it proves that our result is certainly not random or caused by a measurement error. On the other hand, 10-11% of performance gain as a result of twice as many cores is not impressive… But let’s not forget that the total score is nothing but “the average temperature of patients in a hospital.” If we analyze charts with detailed results, we can clearly see two groups on the common mediocre background – these are programs for 3D Modeling and professional photo processing (Adobe Photoshop). Quad-core processors demonstrate impressive results here. The other programs from our list of professional software still do not need four cores…
It’s the second group of tests, where we suspected that the new core would be hardly better than the old core, because archivers, just like compilers, need a large cache. However, our doubts were unfounded: Wolfdale/Yorkfield performs well even with the smaller L2 Cache. Its representatives even take the lead on the chart. But we shouldn’t forget that they operate at a higher clock rate. I’d call it a draw. It seems that only WinRAR can benefit a little from four cores (see detailed results).
Funny results: a dual-core processor on the new core almost catches up with a quad-core processor on the old core in media encoding! It’s an illustrative situation for this class of software. The table with detailed results shows two codecs that can use four cores to this or that degree: DivX (its results are not impressive) and x264 (excellent results – four cores are almost twice as fast as two cores).
If we have a look at the table with detailed results, we can easily single out champions of truly multi-core optimizations (not limited to dual cores). They are two games: Unreal Tournament 3 (a quad-core processor outperforms a similar dual-core processor by 26-40%) and World in Conflict (quad-core processors are faster by 12%). The others demonstrate much lower results. We can mention Crysis (5% performance gain for quad-core processors) and Call of Duty 4 (performance difference between quad- and dual-core processors appears only with the old core). Those seven points of advantage on the total score chart in E6600/Q6600 and E7200/Q9300 pairs are earned practically solely by Unreal Tournament 3. The other games cannot load quad-core processors.
Non-Professional Photo Processing
Quad-core processors take the lead. If you have a look at the table with detailed results, you will see why: excellent multi-threaded optimization of Paint.NET helps quad-core processors outperform dual-core models by almost twofold. But you should keep in mind that additional two cores are almost of no use in all other applications.
Total Non-Professional Score
Pairs of dual-dual and dual-quad processors demonstrate practically identical pictures, which illustrate the advantage of new cores over the old ones: Q6600 lost six points to Q9300, E6600 lost six points to E7200. If we compare an old dual-core processor with an old quad-core processor and a new dual-core processor with a new quad-core processor, results will be identical: quad-core processors get nine points more. Is this result high or low? It depends on whether you are ready to pay at least 1.5 times more for these nine points… We can only add that the advantage of quad-core processors over dual-core ones in the professional score is one point higher than in the domestic score. It’s not surprising: heavy tasks favor “heavy” processors.
Estimated Power Consumption*
* – we actually measure power consumption of the on-board VRM, so our readings may be higher, because VRM does not have the efficiency factor of 100%.
There is nothing surprising about results of E7200 and Q9300 – 45-nm processors naturally consume less power than 65-nm products. However, one especially effective comparison is really striking: the higher-clocked quad-core Q9300 consumes less power in idle mode than the lower-clocked dual-core E6600.
The situation repeats: the fully-loaded quad-core processor on the new Yorkfield core still consumes less power than the dual-core Conroe.
The new core is apparently much faster than the old one. The main “sufferers” here are old quad-core processors (Kentsfield): very few applications can use four cores; besides, dual-core processors have been overhauled so that a dual-core Wolfdale performs on a par with an equally-clocked quad-core Kentsfield (3% of performance difference for more than a 1.5-fold difference in their prices.) However, two groups of users don’t have to worry about their investments: those who use 3D Modeling applications and Adobe Photoshop. Quad-core processors are almost always faster than dual-core products here.
The final chart shows no radical changes: Core 2 Duo processors (especially Wolfdale) are still the most reasonable choice in this family. Quad-core processors make sense only for certain applications – you must be sure that your application can effectively use all four cores. Even attractive prices for the cheapest models will hardly be a ponderable argument for quad-core processors, if you are not sure that performance in your application of choice will come up to your expectations.
If you didn’t test your tasks for performance gains on quad-core processors versus dual-core products, or you didn’t obtain definitely positive results – you will most likely fail to get significant performance gains from using a quad-core processor instead of a dual-core model these days. That’s the main tendency now. There are only a few exceptions. Advantages of quad-core processors are open to those users, who know for sure what they want from a CPU and how their applications of choice can use it.
In fact, dual-core processors came through the same period in the beginning of their brilliant career not long ago. The history is repeating…