The first part devoted to audio encoding will be supplemented today with the video encoding results. We used our integrated test packet based on the open technique of testing performance of video compression into DivX, XviD, WMV9, MPEG1 and MPEG2. I hope this review will be very useful to those who often deal with video compression and want to find an optimal configuration for his or her applications.

Testbed and software

Testbed

• CPUs:
• AMD Athlon 64 FX-51 (2200 MHz)
• AMD Athlon 64 3200+ (2000 MHz)
• AMD Athlon XP 3200+ (2200 MHz, Barton core)
• AMD Athlon XP 2500+ (1833 MHz, Barton core)
• AMD Athlon XP 2200+ (1800 MHz, Thorton core)
• AMD Duron 1400 MHz
• Intel Pentium 4 3.2 GHz 
• Intel Pentium 4 2.4C GHz (Hyper-Threading, 800 MHz FSB) 
• Intel Pentium 4 2.4 GHz (533 MHz FSB)
• Intel Pentium 4 2.0 GHz 
• Intel Celeron 2.4 GHz 
• Intel Celeron 2.2 GHz 
• Intel Celeron 2.0 GHz 
• Mainboards:
• Gigabyte K8NNXP-940 (BIOS F3)

on NVIDIA nForce3 Pro 150
• ASUS P4C800 Deluxe (BIOS 1014)

on Intel 875P
• ABIT KV8-MAX3 (BIOS 17)

on VIA K8T800
• Memory:
• 2x512 MB PC3200 DDR SDRAM DIMM TwinMOS
• 2x512 MB PC3200 Registered DDR SDRAM DIMM Corsair (äëÿ Athlon FX)
• Video card: Manli ATI Radeon 9800Pro 256 MB
• Hard drive: Western Digital WD360 (SATA), 10,000 rpm

Test results

Before we proceed to the test scores let me point out that the scores were obtained quite a long time ago, yet last year, with the older version of the test suite. That is why we can't include the scores of the Athlon 64 3400+ and Pentium 4 3.2 GHz EE as they were tested on the newer benchmark version, and the internal research shows that the absolute scores differ. However, it doesn't matter within this round because all the processors are tested on the same benchmark version.

Ripping

DivX 5.1.1 Pro

The NetBurst processors get an advantage here (Pentium 4 and Celeron). The bigger the cache, the better the scores (compare Celeron 2.0 GHz and Pentium 4 at the same clock speeds), but the higher the frequency, the less the effect from the large cache (compare Pentium 4 / Celeron 2.0 Ghz and Pentium 4 / Celeron 2.4 GHz). Anyway, the large L2 cache helps only within the same platform: it doesn't let jump over the threshold of codec's preferences regarding the CPU core: Athlon XP 2500+ can't beat Celeron 2.4 GHz, though the L2 cache of the latter is significantly smaller. But among the new and old AMD platforms (Athlon XP and Athlon 64), the new one wins exactly at the expense of the twice greater L2 cache. Duron is far behind because of the small cache and low clock speed. So, in DivX encryption Intel's processors of the certain clock speed, including Celeron, are faster than AMD's CPUs of the same rating. 

XviD 1.0 beta 2 (Koepi)

Although the MMX optimization was disabled for Pentium 4 and Celeron, it didn't have any decent effect - this version of XviD prefers AMD K7/K8. Like in the audio encoding tests, there's only the Pentium 4 3.2 GHz that managed to outscore the Duron 1400 MHz and take one of the upper lines in the table. The cache size helps only on the Intel platform, but even here the effect is inconsiderable. Well, the performance gain of the Pentium 4 relative to the Celeron of the same clock speeds is less than the price difference.

Windows Media Video 9 VCM

First of all, let's compare the platforms. The worst choice is the processors on the first-generation NetBurst core (without Hyper-Threading). It refers both to Pentium 4 and Celeron. The Athlon XP and Duron of the same rating look much better. The K7 core is one step above. The Pentium 4 2.4C GHz and Pentium 4 3.2 GHz managed to beat Athlon XP and Duron, i.e. Intel's CPUs with Hyper-Threading support. However, the first position is taken by Athlon 64 FX thanks to the combination of the architecture (K8 core) and the dual-channel memory controller. Even Pentium 4 3.2 GHz falls behind it. The cache size has the least effect again, both for Intel and Athlon XP / Athlon 64. Finally, we can suppose that the Athlon 64 3200+ clocked at the lower speed beats the Athlon XP 3200+ due to the SSE2 support.

MPEG1 (with Mainconcept MPEG Encoder)

As we found out it earlier, the Hyper-Threading makes even a worse effect on Intel's processors in case of MPEG1 encoding in this application. But users wouldn't restart the PC and turn off Hyper-Threading in BIOS Setup every time before encoding because of preferences of a single application, that is why we tested Intel's CPUs as is: those which supported virtual SMP were tested with Hyper-Threading enabled, those which did not were tested without it. That is exactly why the old Pentium 4 2.4 GHz beats the new Pentium 4 2.4C. The latter even loses to the Celeron 2.4 GHz because of the Hyper-Threading! It also indicates that the L2 cache size is not that influential in this subtest. The other processors of all architectures prove it. Thanks to Hyper-Threading Athlon 64 3200+ outdoes Pentium 4 3.2 GHz (the research shows that without the HT support the scores are almost identical). The Athlon 64 FX-51 is a champion here. But in general the situation is very complicated, and we can't say which architecture takes the lead. 

Movie director's tests

Mainconcept MPEG Encoder / MPEG2

For MPEG2 encoding the Hyper-Threading works correctly, and Intel's processors that support it get a good gain. Just compare Pentium 4 2.4C and its sibling without Hyper-Threading. By the way, if the bus bandwidth is not indicated, it means that the RAM speed has a weak effect on performance, and the comparison of the Pentium 4 2.4C and Pentium 4 2.4 comes to comparison of performance of the processors with HT and without it. Performance doesn't depend much on the cache size, though there's some difference between the processors with the large and small L2 caches. But the worst disadvantage of the Celeron is not the cache but the lack of the Hyper-Threading. The gap between Celeron 2.0 and Pentium 4 2.0 is not that great, it's the same for Celeron 2.4 GHz and Pentium 4 of the same clock speeds without Hyper-Threading. And the breakaway of Pentium 4 2.4C is caused not by the large cache... AMD's processors look flabby in this test - AMD's flagships Athlon XP 3200+ and Athlon 64 FX-51 go on a par with Pentium 4 2.4C and fall behind Pentium 4 3.2 GHz. But the worst thing for AMD is not the defeat of its top CPUs but the fact that Athlon XP 2500+ looks equal to Celeron 2.4 GHz! 

Canopus ProCoder / MPEG2

This time AMD gains revenge for its defeat in Mainconcept MPEG Encoder with encoding into the same format. Intel loses because one of its main trumps, HT, doesn't help it. There's no such a great effect like in Mainconcept MPEG Encoder for MPEG1 encoding, Canopus ProCoder hardly notices the second virtual CPU. The large L2 size helps a lot - Celeron 2.4 GHz beats Pentium 4 2.0 GHz. The scores of Barton and Thorton also indicate that the size is important here. The upper positions are taken by the processors with the largest cache - AMD Athlon 64 3200+ and Athlon 64 FX-51. Well, we can expect good scores from Pentium 4 Prescott as well...

Cinema Craft Encoder SP / MPEG2

Cinema Craft Encoder SP prefers the NetBurst architecture (Pentium 4 / Celeron) and is not indifferent to the L2 cache. At the same time, architecture has a greater effect than cache - Pentium 4 3.2 GHz takes the first place, Athlon 64 3200+ / FX-51 with 1MB L2 take the second and third positions. Celeron is far behind, the low-end 2GHz model is almost equal to Duron 1400 MHz (even NetBurst core doesn't help it). This test is very transparent - now when we know its preferences, it's not difficult to predict scores of other processors.

Conclusion

We can't single out the only preference for video encoding. However, there are some factors these benchmarks are influenced by. They are:

1. Architecture (Intel NetBurst, AMD K7, and sometimes even AMD K8);
2. L2 cache size;
3. Hyper-Threading support;
4. RAM speed / FSB throughput (memory bandwidth).

We excluded the clock speed because within the same architecture almost all tests react equally to its changing. Let's try to make a list of preferences...

1. DivX 5.1.1 prefers NetBurst core (Pentium 4 / Celeron), is not indifferent to a large cache, the memory bandwidth is not that influential.
2. XviD 1.0 beta 2 (not the latest version) fancies all AMD processors, sometimes reacts to memory bandwidth and L2 cache size.
3. Windows Media Video 9 VCM is almost indifferent to the cache size, improves performance with Hyper-Threading, and fancies AMD K8 core (either due to the SSE2 or due to the integrated memory controller; probably both factors are important).
4. Mainconcept MPEG Encoder 1.4 / MPEG1 slows down with Hyper-Threading enabled, doesn't depend much on the cache size and likes the K8 core much (probably for its K7+SSE2 because it's not sensitive to the memory bandwidth).
5. Mainconcept MPEG Encoder 1.4 / MPEG2 benefits a lot from Hyper-Threading, is not that sensitive to the NetBurst core (even without Hyper-Threading). Note how the preferences change regarding the format. 
6. Canopus ProCoder 1.25 is indifferent to Hyper-Threading but very sensitive to cache size; architecture and memory bandwidth are not that important.
7. Cinema Craft Encoder SP 2.67 prefers a large cache and the NetBurst architecture, with the latter being more vital. It doesn't benefit much from Hyper-Threading, as well as from memory bandwidth.

That's all. The rest depends on what applications you use most of all.

Write a comment below. No registration needed!