The 6 Phenoms, power consump., overclocking

Six Phenoms in tests

For this test, we were able to get our hands on the following processors:
- Phenom X4 9850 Black Edition (x2) (2.5 GHz, 1.3v)
- Phenom X4 9750 (2.4 GHz, 1.3v)
- Phenom X4 9550 (2.2 GHz, 1.2v)
- Phenom X3 8750 (2.4 GHz, 1.2v)
- Phenom X3 8450 (2.1 GHz, 1.2v)

Besides their frequencies, note that the default voltages of these processors also vary.

Power consumption

We measure the power consumption of the processor’s power supply stage with the help of an ammetric clip on the ATX12V connection line which is exclusive to the CPU. This allows us to better isolate CPU power consumption instead using a method with overall values. The only thing we should keep in mind is that the CPU power supply stage has an efficiency of between 80 and 90%.


Given the difference in fabrication processes, comparison with Intel 45nm processors isn’t too flattering whether in idle or load. And in fact, the latter is particularly disappointing for the Phenom despite the presence of Cool’n’Quiet.

Note that there is a 27.5 watt difference in favor of the X3 8750 compared to an X4 9750 when functioning at the same frequency.

Overclocking

We should distinguish between two types of overclocking with the Phenom: that via the multiplier and only authorized on « Black Edition » versions; and overclocking via the initial frequency of the hypertransport bus. The first is of course easier because we “simply” have to play with this parameter as well as the processor’s voltage.
Strictly speaking, the base frequency of the hypertransport bus doesn't just influence the processor’s frequency. A coefficient is also applied to this frequency in order to obtain the frequency of the memory controller integrated to the processor as well as that of the hypertransport bus. In both cases, this multiplier is x9. Also, increasing the base frequency of the HT bus means overclocking these two frequencies in addition to that of the core. Therefore, potential for poor functioning is triple.

Luckily, it is possible to lower these coefficients on AM2+ motherboards, for example, in order to not surpass a ceiling of 2150 MHz for the memory controller. According to our tests, this seems to be the limit after which crashes are more likely.

We started with trying to overclock the two Phenom X4 9850 BEs. In both cases we went from 2.5 GHz to a stable 2.9 GHz (2x30 min of Prime95) without increasing the voltage from 1.3v. On one of the two processors, 3 GHz proved to be stable at 1.35v. Otherwise, it was impossible to go above this value and obtain stable results even by increasing the voltage.


The same result was reached with the Phenom X4 9750 by overclocking the bus this time. By changing the bus to 242 MHz, 2.9 GHz was stable without increasing voltage. For 3 GHz, we had to increase to 1.35v from 1.3v and lower the memory controller coefficient to x8.

The Phenom X4 9550 initially functions with 1.2v and 2.2 GHz. Without touching anything, we were able to attain 2.6 GHz, while 2.7 and 2.8 GHz were reached with 1.25v and a controller coefficient of x8. Finally, to stabilize the CPU at 2.9 GHz we had to increase to 1.35v and a coefficient of x7.

The Phenom X3 8750 is the only processor in our test which surpassed 3.1 GHz. From its starting 2.4 GHz / 1.2v, we were able to attain 2.8 GHz / 1.2v and 2.9 GHz / 1.25v. At 3 GHz we had to lower the memory controller coefficient to x8. Finally for 3.1 GHz, a voltage of 1.35v was applied.

The least expensive processor of the Phenom line, the X3 8450, also reached 3 GHz. At 1.2v, it holds at 2.3 GHz and then at 2.5 GHz with 1.25v. Beyond this, we had to use 1.3v and a coefficient of x8 which enabled 2.8 GHz. Otherwise, at 1.35v with a coefficient of x7 a stable 3 GHz was obtained.


So in the end, without being exceptional, overclocking the Phenom « B3 » is much better than the former revision, the range being 2.9 to 3.1 GHz.

The test

We now move onto the comparison of these processors with other dual and quad cores in our test suite. The following configuration was used:
- GeForce 8800 GTX / ForceWare 169.01
- 2 x 1024 MB DDR2-800 4-4-4
- 2 x Raptor 74 GB
- Windows XP SP2 French
- Socket 775 : ASUSTeK P5E Deluxe
- Socket AM2+ : ASUSTEK M3A32-MVP Deluxe

3ds Max 9 and Maya 8

3ds Max 9 and Maya 8

In this test, we used two test scenes from Maya and 3dsmax developed by Yann Dupont of 3DVF (whom we thank for their use) and which employ the MentalRay rendering engine. This isn’t a random choice as this engine is now a standard part of both programs and is the most commonly used in production.
- the scene with 3dsmax is very heavy in terms of polygons and the number of objects. The objective was to test processor capacity and manipulate a heavy flow of data.
- Maya's scene is much lighter, but uses MentalRay's advanced lighting algorithms and employs the processors’ raw power in terms of mathematical calculations.

We could have thought that in this type of application, which can fully benefit from multiple cores, the Phenom X3 would do well compared to the Core2 Duo. Of course, at equivalent prices (the X3 6850 vs. the E8300, for example) they are ahead but not by much. The frequency as well as higher efficiency at equal frequencies helps the Core2 to absorb the blow. AMD’s most expensive quad core does not manage to challenge Intel’s least expensive quad core; however, we should add that the two models have similar prices.

TMPGEnc 4.0 XPress and DiVX 6.8

TMPGEnc 4.0 XPress

Version 4 of this video encoder integrates SSE4 optimizations in its latest patch. The test consists of the encoding of a 10 minute 16 second DV file in MPEG-2 format in 720x576 with an average bitrate of 4500 kbits /s and in 2 paths. The video preview display is activated during this test.

This time results are even more to Intel’s advantage because at equivalent prices, their dual core solution holds its own against AMD’s tri-core.

VirtualDub & DiVX 6.8

We now use the version 1.7.8 of VirtualDub and the version 6.8 of DiVX, which has SSE4 optimizations. We encode the same video source as with TMPGEnc in Fast recompress mode and with the DiVX 6.7 codec in one path with an average bitrate of 1500 Kbits /s, the highest quality encoding. Also, Experimental SSE4 full search is activated in SSE4 or SSE2 mode. The video preview mode is activated during this test.


SSE4 optimizations integrated to DiVX 6.8 enable Intel processors to hold a large lead.

WinRAR 3.7 and Nuendo 3

WinRAR 3.7

Since its version 3.6, WinRAR was given multithread optimizations. We compress in RAR at the highest level a total of 588 MB of files comprised of 493 Word & Excel files (69 MB), 22 e-mail Eudora files (251 MB) and a single wav audio format file (268 MB).


While WinRAR benefits from multiple cores, the gain isn’t significant beyond 2 cores. For this reason, the Phenom X3 and X4 are penalized and Intel solutions widen their lead. In fact, none of the Phenoms finish at the same level as a “good ole” Athlon 64 X2 6400+!

Nuendo 3

Something new in our test suite is Nuendo in version 3. This is a solution devoted to production and audio post-production. The test consists of exporting a relatively heavy project to an audio file (thanks to DraCuLaX for the file).


Unlike WinRAR, Nuendo strongly benefits from multithread optimizations and this enables AMD tri and quad core solutions to do significantly better. The only thing is that once again AMD’s tri-core has only a slight advantage compared to the Intel dual core.

Crysis and World In Conflict

Crysis and World In Conflict

Games do not really benefit from more than two cores and Crysis is no exception. In this way, the Athlon 64 6400+ is AMD’s fastest solution while the Phenom has some difficulty in rivaling Intel.

It’s the same in World In Conflict with even bigger leads between the different processors.

Conclusion

Conclusion

Despite AMD’s efforts, the Phenom is still in for some hard times. Of course, the lower prices of this line due to, amongst other things, the arrival of tri-core versions make access to this family of CPUs easier; however, rivaling the prices of the Core 2 is something else.

And certainly in terms of raw performances the Phenom X3 sometimes manages to surpass competing Core 2 Duos in applications that strongly benefit from multithreading. But the lead is never that great (less than 10%), while with applications that do not benefit from more than two cores, the advantage is more obvious for Core 2 Duos.


AMD even has trouble with the price positioning of its CPU as the Phenom X3 8750 and X4 9550 are equivalent in this area. The first is slightly faster than the second in a few applications (Crysis, World In Conflict) while the X4 is largely faster in others (3ds, Maya, TMPGEnc, DiVX, Nuendo). So the interest of the X3 8750 therefore seems quite limited.

Given these performances and combined with the advantages of the Core 2 in terms of overclocking and power consumption, the situation seems clear. The Phenoms are good processors but the Core 2s are excellent! For this reason, besides the upgrade to AM2 based machines, the AMD tri and quad-core offers aren’t really too interesting – other than just prolonging the fight of David versus Goliath.

With lower performances at equal frequencies, lower frequency increases and all the while costing more to produce, the Phenom is a veritable headache for AMD. The Phenom X3 is the epitome of this and does not enable giving a second wind to K10 architecture to rival Intel’s Core 2.

Will this change? The transition to 45nm and the evolutions that go with it are one possibility but this will not happen until the end of the year. By then Intel will have launched the successor to the Core 2, the Nehalem. The potential well-being of AMD in terms of CPUs could then be the Bulldozer, its next architecture, which unfortunately shouldn’t be out before 2009/2010.