AMD processors (Athlon 64, Athlon 64 X2, Athlon 64 FX) have been the best choice for gaming computers of late. The situation was doomed to change, of course. Intel would have hardly tolerated this state of affairs for long. Engineers from this company designed a new architecture and processors, which should take up the lead in performance, including gaming. The processor core, codenamed Conroe, should demonstrate outstanding performance and so-called energy efficiency in usual desktop applications.
It became known in the beginning of the year that Intel planned on giving up the Pentium brand. Renunciation of this trademark meant upgrading desktop and mobile processors to a new architecture. At the same time, the mobile line of Intel processors on Yonah core got the names Core Solo and Core Duo. Future desktop processors of the next generation on Conroe core got the name Core 2 Duo.
Core 2 means that it's the second generation of Intel Core technologies, though mobile processors Core Solo and Core Duo differ much from the second modification. Intel Core microarchitecture consists of three cores: desktop Conroe, mobile Merom, and server Woodcrest. The new Intel Core 2 architecture resembles the architecture of Pentium III, Pentium M, Core Duo and Core Solo. It's a heir to dual core Yonah. You can read the details about Conroe architecture in the following article on our web site: Conroe: Grandson of Pentium III Processor, Nephew of NetBurst Architecture?
The new generation of Intel microprocessors will replace the Netburst generation, which had served the company for six years. "Core 2" trademark will also push Pentium (it has been used since 1993) out of the desktop segment. The company hasn't yet officially announced the names of models. But according to some sources, the desktop series will be marked E4000 and E6000, they will differ in FSB clock - 800 MHz and 1066 MHz correspondingly. But only the latter models will appear first. Prices for processors of the new line will start from $183 for Core 2 Duo E6300, operating at 1.87 GHz and possessing 2MB L2 Cache. The top Core 2 Duo - E6700 (2.67 GHz and 4MB L2 Cache) will cost $530. There are another two models in between: E6400 and E6600. They differ in clocks and L2 Cache size. The costs are $224 and $346, correspondingly. They all have a 1066 MHz FSB and 65W TDP.
Besides, Core 2 Extreme will also appear this year. It will be the highest performance desktop processor from Intel. It's based on the Conroe XE core and is intended to replace Pentium 4 Extreme Edition and Pentium Extreme Edition. CPU model - X6800. It will run at the highest clock in the series - 2.93 GHz, will have 4MB shared L2 Cache, and 80W TDP. FSB clock for the extreme model will most likely be higher, for example, 1333 MHz. The estimated price is $999. The extreme series will be expanded in future, most likely at the end of 2006 - Core 2 Extreme X6900 3.2 GHz.
Processors will be manufactured on the company's plants by the 65nm process technology. According to Intel's data, Conroe will demonstrate 40% as high performance at 40% as low power consumption compared to Pentium D. While Pentium 4 and Pentium D processor were notable for high clocks, even lower-clocked Core 2 processors will offer better performance than Netburst-based processors. Their power consumption and heat dissipation are noticeably lower compared to Pentium. All non-extreme processors up to E6700 offer 65W TDP, which is nearly twice as low as in powerful dual core processors of the current generation. And it's much lower than in the fastest AMD processors (Athlon 64 FX-62).
The approximate release date of the new desktop Intel Core 2 Duo processors aka Conroe is July 23, 2006. New processors should be launched on that day and prices for processors of the previous generation should go down. Conroe will live in Intel processors for several years. Successors to the first processors based on this architecture will have more cores, larger L2 cache, and they will be manufactured by new process technologies. We are already curious how good these processors of the new architecture will be in 3D games, the most resource consuming applications for desktop computers.
We have been provided with a test system, based on an engineering sample of Intel Core 2 Duo (Conroe) 2.13 GHz with 4MB L2 Cache. Its technical characteristics are somewhere between E6400 and E6600. Our processor has the clock as in E6400, but its L2 Cache is twice as large - 4MB, like in E6600. We have run standard game tests, included into 3Digest, to analyze performance of this system, but only with two highest-performance video cards from the key chipmakers: NVIDIA GeForce 7900 GTX and ATI RADEON X1900 XTX. Our set of tests is rather wide and allows to evaluate performance of the new processor core in 3D games. For comparison sake, we have published standard results obtained on our testbed with Athlon 64 4000+. Though the comparison is not well posed (Intel processor has two cores operating at 2.13 GHz, while the AMD product has a single core operating at 2.4 GHz), there are traditionally no gains from the second core in games so far and the high clock frequency of the Athlon 64 processor will make things harder for the new processor from Intel.
Athlon 64 - based system
Testbeds run under Windows XP Professional Service Pack 2, DirectX 9.0c API. We used the IIYAMA Vision Master Pro 514 (22") monitor.
We have come up with two fastest video cards for our performance tests:
The list of software that we used as benchmarks
Results of Performance Tests
We have tested the games in all resolutions, usual for 3Digest. But in case of CPU tests, the most interesting ones are always low resolutions with disabled anisotropic filtering and antialiasing. Such modes show a clear-cut difference in performance and CPU dependence. But we'll still be interesting in high resolutions with antialiasing and anisotropic filtering, as we play games in such resolutions, or at least we try to. Heavy modes will help us evaluate the issues of global compatibility of the new system, its memory performance, etc.
Let's start the analysis with the results of games that demonstrate evident dependence on CPU performance. Green bars on the diagrams stand for a system with a video card from NVIDIA, red bars - ATI. Light bars show the results of the engineering sample of Intel Core 2 Duo, dark ones - AMD Athlon 64 4000+.
Far Cry is notable for vast open spaces. Game performance on such levels evidently depends on CPU, especially in low resolutions. That's exactly what we see on the diagram. Modes without anisotropic filtering and antialiasing are limited by the performance of Athlon 64 4000+ on the level of 89-90 FPS. But the Conroe system lifted this limit to 124-126 FPS. That is the new processor from Intel is approximately 40% as fast in such conditions, its clock frequency being lower at that! Quite a big difference. But let's see what happens in heavy modes. While the Intel system enjoys some advantage up to 1280x1024 with enabled anisotropic filtering and antialiasing, it wears thin with each resolution increase. It's even outperformed by 5-7% in 1600x1200 and 2048x1536. Perhaps, it may happen because the Conroe testbed is still not fine-tuned - for example, high memory timings and relatively low memory frequency.
Let's have a look at another application, evidently limited by CPU performance - Half-Life 2.
Even when Half-Life 2 just appeared, its performance in low video modes used to be evidently limited by CPU. Game levels are split into small pieces. The load on a video card is not very large, especially for such monsters as GeForce 7900 GTX and RADEON X1900 XTX. That's why the diagram with test results really shows that the performance is limited by CPU. You can clearly see the 93-95 FPS limit on the Athlon 64 testbed in all resolutions. The video cards are fully loaded only in 2048x1536 with antialiasing and anisotropic filtering. A change in the testbed to Intel Core 2 Duo again makes a great performance difference - CPU limit's lifted to 125-135 FPS, depending on a video card. We again got the 35-40% performance difference in favour of the Intel processor. Similar test results are demonstrated by Intel and AMD only in the highest resolution, where CPU dependence disappears.
And now we are going to analyze games, where CPU dependence is clearly demonstrated only in low resolutions. Such games are in the majority now, but their 1024x768 and 1280x1024 mode can still tell us something about comparative CPU performance.
Call of Duty 2 has rather complex pixel and vertex shaders, it uses several texture layers and complex special effects, so it mostly depends on a video card. But we can still see the 7-24% advantage (it depends on a mode) of Intel's processor in low resolutions without antialiasing and anisotropic filtering and in 1024x768 with enabled AA and AF. Performance in the heaviest resolutions is determined by video cards alone, there is no difference between AMD and Intel systems.
The graphics engine of DOOM III uses a CPU, for example in rendering shadow volumes. So the game demonstrates some dependence on CPU performance in low resolutions without antialiasing. That's what we see on the diagram - the engineering sample of Core 2 Duo turns out faster by 7-20% than Athlon 64 4000+ in all modes without anisotropic filtering and antialiasing. There is almost no difference between the systems in heavy modes in case of the card from ATI. And the difference reaches 17% on the system with NVIDIA GeForce 7900 GTX. The Conroe processor has again shown itself to advantage, performance of this system in DOOM III is higher in all video modes.
The situation in The Chronicles of Riddick: Escape From Butcher Bay is almost a complete copy of what happens in DOOM III, except for tests with enabled anisotropic filtering and antialiasing. While the Conroe system scores 17% more FPS than Athlon 64 in usual modes, the difference in heavy modes is almost always negative to Conroe, though it's still within the error of measurement - up to 3%. Perhaps, performance in such modes is limited by another component, for example, memory bandwidth or latencies.
Like its predecessors, the last game in this series (Splinter Cell) mostly depends on the fillrate of a video card. It does not depend much on CPU performance. We managed to obtain different test results only in the lowest resolution (1024x768) and only on a testbed with a video card from NVIDIA - the difference is 10%. Both systems offer identical performance in all other cases. The difference is below 5%. But it should be noted that the performance difference varies from minus too plus.
Even before we look at the results in F.E.A.R., we can assume that there will be no difference between testbeds with different processors. This game is too heavy a load for a video card. It depends on its fillrate in the first place. This assumption is confirmed by practice. The positive difference for the new processor from Intel (greater than the measurement error) can be seen only in the lowest screen resolution (1024x768) and only for RADEON X1900 XTX. The same combo demonstrates the maximum negative result for Conroe in the opposite conditions - 2048x1536 with enabled anisotropic filtering and antialiasing. The question is whether we should take the difference in a couple of average FPS into account. So let's just say that this test demonstrates no evident dependence on CPU performance.
In conclusion, let's analyze the results of famous synthetic 3D tests from Futuremark. We shall publish only the total scores.
Conclusion on the previous generation of the test: 3DMark 05 performance depends little on CPU in most resolutions. ATI video card demonstrated more or less significant difference only in the lowest resolution. However, it's even below 5%. No difference is demonstrated in all other cases, give or take 2-3% both ways. It's quite a good result, considering that this Conroe is an engineering sample, the memory clock is not very high, and its timings are far from the best.
Well, let's have a look at the new test from this company. But I personally have no doubts that we won't find any dependence on CPU performance there either.
In my opinion, 3DMark 06 results are very strange. No, they are quite clear for the ATI video card and they correlate well with the results in 3DMark 05. A difference between the systems on Intel and AMD processors does not exceed 2-3%. Athlon 64 is more often victorious here. This insignificant difference can be explained by the same factors: the Conroe testbed is not fine-tuned, frequency and timings of its memory. But I don't understand what happened to GeForce 7900 GTX. It's either the problem of NVIDIA drivers or something went wrong during our tests. I haven't got a better explanation to these figures. The 15-30% difference in favour of Athlon 64 is evidently anomalous, judging by the results of absolutely all previous tests and the results of the same system with the ATI card in the same 3DMark 06. Therefore this benchmark results on GeForce 7900 GTX are most likely incorrect, but we are bound to provide them because this popular 3D benchmark must be a part of the article. Let's consider this a preliminary review of a sample that, along with the infrastructure, is to be refined.
The figures we obtained in our video lab show that Conroe processors will be the fastest CPUs in the desktop segment in a couple of weeks, when this series is launched on the market. The new processor from Intel is an evident winner in the majority of game tests, indicative of CPU performance, in low resolutions and without anisotropic filtering and antialiasing, especially considering its lower clock and possible defects of our engineering sample. You can just have a look at the average FPS rate in such CPU-intensive games as Half-Life 2 and Far Cry. Even in some modes with a relatively heavy load on a video card, the performance gain compared to a rather fast processor from AMD is several percents. To say nothing of low resolutions and modes without antialiasing and anisotropic filtering, where the performance difference reaches 43% in favour of Conroe!
In fact, there is no need in special conclusions, the test results are illustrative. The future processor from Intel completely outperforms AMD Athlon 64, when test performance depends on a CPU. Unfortunately, we cannot say the same about high-quality modes. That may be the effect of defects in the Conroe testbed. That's excusable, as the system is based on an engineering sample of a processor. And the BIOS settings were far from optimized for maximum performance. The situation must change in future, when the tests are run on production-line samples, identical to those in stores. Perhaps we'll manage to get back to this topic and test such a system before the processors are launched on the market.
As for now, we can draw a conclusion from the excellent results of the engineering sample that Intel processors are ready to offer top gaming performance again. It should happen in a couple of weeks already. If early engineering samples demonstrate such good results with memory operating at 667 MHz and nonoptimal timings, the 2.93 GHz model and the future 3.2 GHz extreme processors are sure to offer the best performance in 3D games. I'll remind you that we have tested only a medium-clocked processor, 2.13 GHz. We can only wait for the forthcoming launch of Core 2 Duo on the market. AMD should hurry up with its next modification of the K8 architecture, because the launch of Core 2 threatens its temporary leadership on the market of high-performance gaming systems.
Alexei Berillo aka SomeBody Else (firstname.lastname@example.org)
June 14, 2006
Write a comment below. No registration needed!