As promised, in the continuation of the first part, we are going to review the performance of low-end processors for Intel LGA775 platform. So, a lengthy introduction is out of place here, the main objective of all the articles in this series being the same. We only want to note that we decided to diversify the article aspects, so the second part will pay more attention to the scalability of Celeron D — an obvious favourite in the Intel low-end sector and actually the only low-end CPU for LGA775.

Testbed configurations

• Processors
  • Intel Celeron D 335 (Socket 478, information from the previous article)
  • Intel Pentium 4 2.8E GHz (Socket 478, information from the previous article)
  • Intel Celeron 325J (LGA775, 2.53 GHz, 256 KB L2, 533 MHz FSB)
  • Intel Celeron 330J (LGA775, 2.66 GHz, 256 KB L2, 533 MHz FSB)
  • Intel Celeron 335J (LGA775, 2.80 GHz, 256 KB L2, 533 MHz FSB)
  • Intel Celeron 340J (LGA775, 2.93 GHz, 256 KB L2, 533 MHz FSB)
  • Intel Celeron 345J (LGA775, 3.06 GHz, 256 KB L2, 533 MHz FSB)
  • Pentium 4 520 (LGA775, 2.80 GHz, 1 MB L2, 800 MHz FSB, Hyper- Threading)
• Motherboards
  • ASUS P5GDC-V Deluxe (i915G chipset)
• Memory
  • 2x512 MB PC3200 (DDR400) DDR SDRAM DIMM 2.5-2-2-5 (Corsair TwinX)
  • 2x512 MB PC2-5400 (DDR2-533) DDR-2 SDRAM 3-3-3-8 (Corsair XMS2)
• Video card: ATI Radeon X800 256 MB
• HDD: Samsung SP1614C (SATA), 7200 rpm, 8 MB Cache
• Windows XP Professional SP2, DirectX 9.0c
• ATI CATALYST 5.4 (Display Driver 6.14.10.6525)

This time we have also taken a motherboard on the mainstream chipset (i915G). The model choice is conditioned by the fact that ASUS P5GDC-V allows to install both DDR and DDR-2, so that we can test performance of the same processor on the same motherboard but with different memory (we've done it with Pentium 4 520). The P5GDC-V turned out the only motherboard of this kind available in our lab at the time we carried out our tests, so we actually had no choice :). However, we still tested Celeron D with the regular DDR400 memory, because it's more preferable for a low-end system due to its democratic prices.

You may have also noticed that we have also replaced a video card that was used in previous tests. It was inevitable, because there are no sterling AGP video cards for LGA775. That's why we decided not to create artificial parity and just take video cards, which are approximately equal relative to top solutions for this platform. Of course, it inevitably means that we cannot compare directly the results of Socket 478 and LGA775 in those applications that actively use a video card. We are going to remind you about it in comments to the corresponding tests, we even highlighted Socket 478 results with a different color on some diagrams. Why only on some of them? You are in for a little surprise...

Test results

SPECapc for 3ds max 6

At first let's clarify the main point: what scalability do you expect from Celeron D? It's clear that in the ideal case it's linear, that is performance must grow in proportion to the clock. It means that Celeron D 345J must be faster than Celeron D 325J by 3.06/2.53 ~= 21%. That's the ideal case. Remember this figure.

What do we see in the rendering test from SPECapc for 3ds max? Celeron D 325J got 0.85 point, while Celeron D 345J — 1.01. It's easy to calculate that the real performance gain is about 19%. It's a very good result, just 2% less than the ideal case. However, the good scalability of Celeron D does not change the alignment of forces in rendering, because the result of Pentium 4 is initially too high.

Celeron D scalability in the interactivity test is much lower (just 13%); and its absolute results don't strike us, so everything is clear here. But the comparison between Pentium 4 2.8E GHz with its LGA775 counterpart, Pentium 4 520, is suggestive... to all appearances it has to do with the video card. Indeed, interactions mean immediate rendering of the results among other things, which directly implies a 3D accelerator. The above said seems the most logical explanation of the Pentium 4 2.8E defeat. However, Pentium 4 for Socket 478 with a weak card is still better than Celeron D for LGA775 with a powerful video card.

The final score is a compilation of the two preceding tests, it demonstrates nothing new.

SPECapc for Maya 6

Everything goes as it should in graphics performance and mysterious I/O tests: Celeron D gradually scales up, but Pentium 4 is always in the lead. The situation in the CPU test (according to the SPEC interpretation) is quite different: Celeron D catches up with Pentium 4 starting from 2.93GHz, and the top model even outperforms them. Thus, the CPU test from SPECapc for Maya 6 almost doesn't depend (within the same architecture) on cache size. And to all appearances, the Maya operations used to calculate this score do not have the SMP optimization. It's very strange from the one hand. But on the other hand, SPEC is too respectable an organization to make a slipshod. We should probably think about introducing the Maya rendering speed test, even if it's a separate test.

The total score takes into account all the previous tests with the following weight coefficients: graphics — 70%, CPU test — 20%, I/O — 10%. As the developers clearly gave preference to the graphics test, the result is quite expectable. Celeron D scalability is just 9%, which is approximately twice as low as the ideal case. Of course, this figure evokes no optimism, but it's better than nothing. I worry about the other issue: according to Celeron D 340J vs. Celeron D 345J results, the performance will grow increasingly less with the further clock increase.

Lightwave 8.2, rendering

No miracle happened: Celeron is much worse at rendering than the sterling Pentium 4. As we already know from the previous article, Lightwave favors a large cache size as well as Hyper-Threading support. Celeron lacks both features, so it expectedly brings up the rear. But its scalability is just excellent — 17%. That is it's theoretically possible to beat Pentium 4 thanks to the clock, but it will require a higher clock than the top Celeron can offer in this test. We also didn't notice the tendency to slow down rendering speed with the increasing Celeron clock, so we can hope that even higher-clocked models will not be limited by the bus.

SPECapc for SolidWorks 2003

You cannot say that Celeron looks like a total outsider — the performance difference between Celeron and Pentium 4 is not that large. But it's still large enough not to recommend computers based on this processor for such a serious CAD. However, the diagrams clearly demonstrate another tendency — each next clock increment gives Celeron a lower performance gain. To put it simply, it will probably never manage to catch up even with Pentium 4 2.8 GHz with this bus and cache size, irregardless of its clock.

Adobe Photoshop CS (8)

Celeron demonstrates excellent scalability (18%), steady performance growth, no signs of insufficient bus throughput (the steps are even, there is no tendency to decrease the gap between higher-clocked models). Pentium 4 has a convincing advantage in absolute speed. That is we have a situation similar to that in Lightwave: a higher-clocked Celeron would probably outperform Pentium 4, but not this time.

Here comes the first surprise: the color model conversion demonstrates complete disregard of the large cache and Hyper-Threading support, so that two top Celeron models outperform their "elder brothers". But don't take it to heart — it's only one operation among many...

We can see two bursts hard to explain here: Celeron 330J, which heavily outperformed the 325 model; and Celeron 345J, which actually managed to catch up with Pentium 4. Frankly speaking, considering the steady performance of the other models, we cannot think of any logical explanations. We can only take this result for a fact...

Nothing seditious: Celeron scalability is good, but it's not enough to outperform Pentium 4. Just remember the previous article: you could clearly see there that the rotate operation prefers a fast bus and Hyper-Threading.

The diagram shows an even "ladder" and... Pentium 4 processors blend very well as one of the steps! One can say that the top Celeron has almost caught up with them. Mind that Unsharp Mask filter is a widespread Adobe Photoshop operation.

The situation is absolutely the same: Celeron manages to make up for the lack of large cache, fast bus, and Hyper-Threading support due to the higher clock.

It's all very well, but almost no difference between Celeron D 340J and 345J is disturbing. It looks like the performance is limited by the bus. In that case, no matter how much you raise the clock above this level, it will have almost no effect on the performance.

Celeron D demonstrates excellent scalability in total score (17% and a little more) and Celeron D 345J with the highest clock almost manages to catch up with Pentium 4 520 operating at 2.8GHz. But let's not jump at conclusions: it hardly makes sense to buy high-clock Celeron D instead of Pentium 4 for Photoshop. It's more like a reason to be pleased, if your computer is already based on Celeron D: your computer is not that bad for this bitmap graphics editor. At least, within the simplest and most frequent actions.

Adobe Acrobat 6.0

Two Adobe products demonstrate almost an identical picture. But in this case our conclusions will be slightly different: if Celeron D copes so well with this test, probably it makes sense for prepress specialists to assemble an inexpensive computer based on this processor, which will be solely used for distillation? It's an attractive opportunity to use a low-end solution to deliver DTP computers from this dull work...

All-purpose data compression (archiving)

7-zip archiver runs to Hyper-Threading, large cache size, and fast memory — we already know that. That's why Celeron was doomed to defeat. WinRAR has the most democratic requirements, you can also see it on the diagrams. We'd like to digress a little from low-end CPUs and to draw your attention to another funny fact: it's probably for the first time that we can see a significant breakaway of a DDR2 system based on Pentium 4 from the similar system based on DDR. Strange as it may seem, this breakaway is demonstrated by "more democratic" WinRAR. But on the whole, the total score shows that Celeron is not a very good processor for archiving. Besides, its scalability in these operations is rather weak: just 12%.

Multimedia lossy compression (MP3/MPEG2-4)

Pentium 4 shoots forward in encoding MP3 with average quality (note that it's a very fast operation — pay attention to its duration, our 150MB WAV file takes about a minute). Both top Celerons greatly outperform it in encoding with the highest quality! The irony is that Pentium 4 turns out fast where it's needed the least — in already fast operations. But it loses all its advantages in slow, high-quality encoding and the difference between this processor and the equally clocked Celeron is almost not noticeable.

The twin brothers DivX and XviD are totally indifferent to Pentium 4 capacities: they are impressed neither by a fast bus, nor by Hyper-Threading or cache size. To all appearances, the only preference is a clock. Naturally, this results in a top Celeron easily outperforming Pentium 4 operating at 2.8 GHz. It's a mini-shock, isn't it? Celeron turns out the most optimal among all Intel processors by its price/performance ratio in MPEG4 encoding by two most popular codecs!

Windows Media Video prefers multiprocessing, even if virtual (remember the previous article), so Celeron is out of the question here because it does not support Hyper-Threading.

Canopus ProCoder also loves Hyper-Threading, but it's more loyal to the clock, so the top Celeron model manages to bring down the score considerably. Celeron demonstrates excellent scalability in this test, you can see it with the naked eye.

On the whole, we can say that Celeron is a very good choice in terms of the performance/price ratio. If you are interested only in media data encoding performance, of course. Even Windows Media Video and Canopus ProCoder couldn't spoil much the total score of these processors. Excellent scalability (19%!), no signs of the bus throughput limitation even in the highest-clocked model.

CPU RightMark 2004B

CPU RightMark does not look relevant in this review, because it's theoretical rather than practical. So we decided to limit ourselves to publishing total scores. There are no surprises: as CPU RM is not very critical to the memory system, Celeron has good scalability. But the lack of Hyper-Threading does not allow these processors to catch up with Pentium 4 (CPU RM has a multi-threaded rendering module).

3D games and graphics visualization
in professional packages

I'll remind you one more time that the diagram results of Socket 478 processors were obtained on a different video card, so it would not be correct to compare them directly to the results of LGA775 processors.

DOOM 3

As in many other cases, Celeron offers great scalability (19%), but it's not of much help, because Pentium 4 is actually much faster. Differences between Pentium 4 processors of the same architecture and clock for two different platforms are not that much noticeable, but don't forget that we currently analyze the lowest graphics quality mode.

LGA775 (Radeon X800) starts to significantly outperform Socket 478 (with Radeon 9800 Pro) with an increase of resolution and graphics effect quality. But the situation with the relative performance within the platforms remains the same.

And here is the answer why it does not change: the performance drop at a considerably increased load on a video card is negligible among all LGA775 processors. It actually means that the processor cannot load a video card with work even in the simplest mode, Radeon X800 is too powerful solution for all these CPUs. Even for Pentium 4 520.

Far Cry

It's not that Far Cry prefers Celeron. The point is that performance in this game depends on a video card, so it doesn't need a top processor to run fast. On the whole, Far Cry runs well on Celeron (you can easily see that by the absolute fps values on the diagrams) — the performance does not drop below 70 fps even in the high resolution mode with high quality even on the slowest model. Economical users will also be glad to hear that LGA775 started to outperform Socket 478 with a less powerful video card only in the heaviest mode.

What concerns the "satisfaction level" of the game with the CPU power, it's not very high. That is our conclusion based on the analysis of DOOM 3 results is confirmed: all LGA775 CPUs are not powerful enough for Radeon X800, they cannot load this accelerator with work completely.

Painkiller

We are quite at a loss what this game needs! Pentium 4 outperforms Celeron only a little. A more powerful video card installed on LGA775 platform almost did not improve the performance either. We can only assume that Painkiller "does not like" the Intel platform in general, irregardless of details. It will be interesting to see how this game behaves with AMD processors...

The same jumpy, chaotic graph, as in the first article of the series. To all appearances, the built-in Painkiller benchmark is not very precise, that is its results vary from session to session within a rather wide range.

Unreal Tournament 2004

You can also see that Unreal Tournament is limited by the processor from practically the same results for Pentium 4 on Socket 478 and LGA775, despite different video cards with significant performance difference.

Total scores in games

There is nothing much to comment on: you can clearly see on the summary graphs that Celeron is not a gaming processor, even if it significantly surpasses Pentium 4 by the clock. The scalability breaks no records either: from 15% in low resolution up to 13% in high resolution. You can also see that the LGA775 system is in fact much less balanced: it could have done with a less powerful video card. Well, one cannot know everything, we found it out only after the majority of tests had been already completed. On the other hand, one cannot say that the results are useless: on the contrary, we experimentally confirmed the thesis that a powerful video card needs a powerful processor, economy is out of place here.

SPEC viewperf

The situation in SPEC viewperf is similar to that in games. Pay attention that the Socket 478 platform with a less powerful video card was considerably outperformed by LGA775 only once.

Conclusion

Well, we are again left without sensations: Celeron is no good for complex professional packages anyway (though Photoshop was a pleasant surprise), the same concerns games, at least if you expect high performance. However, there are still a couple of applications, where these processors are very good - MP3 and DivX/XviD encoding. In other respects, it's a regular low-end, nothing special. In most cases its scalability is not bad, sometimes it's even very good, which is good news. However, not that many Gigahertz are left for the top Celeron to reach the limit that stopped even Pentium 4. The situation with Hyper-Threading is also ambiguous: on the one hand, there are tasks where Celeron lacks this feature badly; on the other hand, this technology may have a bad effect on this processor with its small cache.

There are two results worth mentioning, which are not directly related to the object of the review: (1) DDR2-533 with good timings slightly outperforms DDR400; (2) a powerful video card in middle and low end systems is hardly justified, because it provides no real performance gain in most cases.

Write a comment below. No registration needed!