iXBT Labs - Computer Hardware in Detail






Processors for common people
Part 1: Intel Socket 478

Tests may have different objectives. Some tests are carried out to give readers an idea of the performance of the brand new processor, launched by one of the manufacturers. In this case the main objective is the fast responsiveness. Some contradictions in the testing method, a strange choice of software, partially obscure results everything is pardonable because the fact itself is more important: "we got something". That can be already used to take bearings (even if with some reservations).

Sometimes, on the contrary, a single program is used to test thoroughly several processors (may be even one...), which performance had been reviewed many times: with details, analysis of how its various options influence the performance, tests of the performance in separate operations, etc. That's actually testing the program itself rather than a processor: how it works with various CPUs, which one is the best for it, what non-obvious issues and rocks there are...

There are tests targeting complexity rather than innovations and details. Out of the "let's try to close this question" series. It would be great to do it for good, but ideals are unattainable, so at least within some wide range of tests. We'd like to hope that this series of articles "Processors for common people" will be a good example of such tests. We'll try to give a complex answer to a frequently asked question: "Why are low-end processors called low-end?" Indeed, can it be pure marketing? Can be Celeron and Sempron processors more than enough to make us happy, while Pentium and Athlon CPUs are just a way to beat us out of more money?

But it would hardly be reasonable to lump all low-end CPUs together: firstly, such an article would have taken much time to prepare (and readers got used to finding new materials in their favourite sections from time to time); secondly, a lot of test objects deaden the perception of fine points, which finally results in determining another "king of the hill" such issues as why it's a king and how it became one are omitted. That's why it has been decided to write a series of articles instead. What concerns the most logical way to subdivide processors, it was prompted by manufacturers themselves: it's... platforms! Socket A and Socket 478, Socket 754, Socket 939, LGA775... It's quite logical to examine them one by one and only then, in the final part, to compare the best representatives between each other. And at the same time (exactly in the final part) with top processors as well to see what can you aspire to and whether it's worth it.

In this first article we are going to examine mainstream and low-end processors for Intel Socket 478. It's a good start: one can say that it's a well known elderly platform, which lost its exclusive air and has long become habitual. It's now associated with the middle- and low-end, because all high-end models have gone to LGA775. And the choice of processors is again prompted by manufacturers: the fact is that Intel contrived to release five (!) processors for Socket 478... with the same 2800 MHz clock! They are the "new" and the "old" Celeron CPUs, Pentium 4 (Northwood core), and two Pentium 4 processors (Prescott core). None of these CPUs can be considered high-end now these are the workhorses, bought by people who can count their money. Of course, some of them have more money than the others... but the choice is large as well five options! So let's see who's the best in this zoo...

Testbed configurations

  • Processors
    • Intel Celeron 2.8 GHz (128 KB L2, 400 MHz FSB)
    • Intel Celeron D 335 (2.8 GHz, 256 KB L2, 533 MHz FSB)
    • Intel Pentium 4 2.8C GHz (Northwood, 800 MHz FSB, Hyper-Threading)
    • Intel Pentium 4 2.8A GHz (Prescott, 533 MHz FSB)
    • Intel Pentium 4 2.8E GHz (Prescott, 800 MHz FSB, Hyper-Threading)

  • Motherboards

  • Memory
    • 2x512 MB PC3200 (DDR400) DDR SDRAM DIMM 2.5-2-2-5 (Corsair TwinX)

  • Video card: ATI Radeon 9800 Pro 256 MB (Manli)
  • HDD: Samsung SP1614C (SATA), 7200 rpm, 8 MB Cache
  • Windows XP Professional SP2, DirectX 9.0c
  • ATI CATALYST 5.4 (Display Driver

You can easily notice that we decided to take a mediocre motherboard, not on the top i875P chipset; we have also set timings strictly "By SPD", no manual configuration, no fine-tuning. It seems to us that it conforms well to the main objective of testing: to analyze the performance of low-end solutions. The same concerns a video card as well: it is surely fast enough, but not a top model, even considering AGP solutions.

Test results

In this article we decided to try another way of presenting the results: absolutely all tests and subtests are included into the article in the form of diagrams (i.e. we don't omit anything), but the titles of summary diagrams are highlighted. So, if you don't want to analyze intermediate results, you can skip "unnecessary" diagrams and focus on the main ones.

SPECapc for 3ds max 6

As we have found out many times, the SPEC tests for old program versions work well with newer software as well and output adequate results. That's why we thought it fit (moreover potentially more interesting) to use SPECapc for 3ds max 6 with new Version 7 of this package.

Good old Northwood remains the champion in rendering with the built-in 3ds max engine, despite the latest version of this software. Prescott with Hyper-Threading support is outperformed; the lag is not catastrophic, but it's still enough to pay attention to the core when you consider various Pentium 4 options. And the "junior" Prescott with a 533 MHz bus without Hyper-Threading looks like a nothing. So, if you are pressed for money, you'd better save much and buy a new Celeron D. Old Celeron... well, no comments...

It's quite a different matter in case of interactivity operations: to all appearances, these operations are not optimized for multi-threaded operation, so both Prescott processors are on a par. Northwood is outperformed significantly. Celeron D also lags behind the nearest contender by a significant percent figure. We can assume that the cache size becomes critical in interactive 3ds max operations.

Pentium 4 on Prescott core is overall victorious, however Northwood is still not bad. What concerns both Celerons, it was obvious that they were not intended for such "serious" software. Anyway, Celeron D looks much more attractive than the old one.

SPECapc for Maya 6

We played the fox again and used the newer version of the package in the SPEC test. However, it had absolutely no effect on its performance. In this case our decision to use Maya 6.5 instead of 6.0 was also dictated by the information we got that many functions became significantly faster in Version 6.5.

Strange as it may seem, Graphics and I/O subtests react violently to CPU changes, though they theoretically have nothing to do with a processor. Graphics favours processors on Prescott core, I/O operations all Pentium 4 processors (compared to Celeron). By the way, we are going to return to the good results, demonstrated by Prescott, when a 3D accelerator is critical...

That's even funny: the difference between the CPUs is the least noticeable... in the processor subtest! That is, neither Hyper-Threading support nor a large cache have a strong effect on the performance actually. For the only exception the cache size in the old Celeron seems too small.

The total score is easy to predict from the above said: both Prescott processors are in the lead with a small breakaway, they are followed by Northwood, Celeron D is significantly slower than all mentioned processors, but it's still much faster than the old Celeron. The summary is simple: there is no significant difference between various Pentium 4 processors; both Celeron processors are traditionally unsuitable for serious 3D graphics packages.

Note that there is no rendering subtest in SPECapc for Maya 6, so the issue of CPU rendering performance is "off-camera" in this case.

Lightwave 8.2, rendering

There seem to be no interactive tests for Lightwave (at least we don't know anything about them), so we traditionally test performance in this program by rendering a test scene. The Hyper-Threading effect is prominent: both processors supporting this technology are in the lead. In the competition between Northwood and Prescott, the former is the winner. That's not surprising, we can only state that the situation has not been improved even in the latest patch.

Celeron D surely lags behind all Pentium 4 processors, but the gap from the only Pentium 4 processor, which does not support Hyper-Threading, is not that large. So we can draw a conclusion that Lightwave depends on L2 Cache size, but not much.

We don't want to comment on the old Celeron (that's becoming a tradition) because there is nothing much to comment on. There is probably no reason to apply the performance word to this CPU. It does not mean that it's useless: there are a lot of tasks, where the CPU performance does not play a crucial role.

SPECapc for SolidWorks 2003

Remember that SPECapc for SolidWorks 2003 scores points instead of real performance units, but the better score is the lowest, not the highest. That's not logical, but it's the tradition here... It's easy to notice that CPU changes have an effect on absolutely all subtests, but this effect is nearly the same. There are no significant performance differences between various Pentium 4 processors.

What's much more interesting Celeron D is only slightly outperformed less than by 20%! It actually means that Celeron D surely cannot be recommended for this package, but if you have no other choice, the work process won't be that inconvenient. Funny. Such a serious CAD and so low requirements to CPU...

Adobe Photoshop CS (8)

This time we are going to publish all the eight diagrams again: for each subtest and a summary. Despite the same clock, the contenders are five processors, which are rather different from each other. It will be interesting to see whether all tested operations in Adobe Photoshop have the same preferences.

Cache size (Celeron D demonstrates excellent results for its class) is not that critical to the blur variations (our script uses Gaussian, Motion, and Radial), the new Prescott core is a tad more preferable than the old Northwood. But the cache size being not critical does not concern the old Celeron: it's probably too small in this processor.

All contenders manage identically well the color model transformations (RGB --> CMYK --> LAB --> RGB). The old Celeron is still worse, but only a little, which is even unexpected...

Both Pentium 4 processors on Prescott core cope well with lighting effects, the old Pentium 4 on Northwood core is a tad worse, and Celeron D is still worse. Then again, the latter demonstrates very good results for a low-end processor.

The core plays even a more important role in rotations than the L2 cache size Celeron D outperforms Pentium 4 Northwood! However, the most impressive result is demonstrated by the combination of Prescott core, a fast bus, and Hyper-Threading support (Pentium 4 2.8E).

Unsharp mask demonstrates a tight group of leaders including all the three Pentium 4 processors plus Celeron D and the only outsider — the old Celeron.

To all appearances, resizing is well optimized for multi-threaded operations, because both CPUs supporting Hyper-Threading are victorious. Pentium 4 2.8A (Prescott without HT and with a slow bus) is only insignificantly better than Celeron D, so we cannot call it a good choice in this case: it's more expensive.

Free Transform (as well as Scale, Rotate, Skew, Distort, and Perspective) evidently prefer the new Prescott core, it's actually the second situation when Celeron D (it's also based on the Prescott-like core) outperforms Pentium 4 Northwood.

The nominal champion is Pentium 4 2.8E (Prescott, Hyper-Threading), but the true startler of the test is certainly Celeron D. It's only 12% slower than the most advanced Pentium 4 modification with the same clock, but it's a low-end processor! It's an excellent result. However, don't forget that our Photoshop tests include only the most common and typical operations. The situation may be quite different in complex filters. On the other hand, most users apply complex filters much more rarely...

Adobe Acrobat 6.0

PS into PDF conversion is a frequent and rather boring task, which is eminent for those people who are responsible for preprint operations. But this test shows no special preferences: even the old Celeron copes with this task almost as well as the fastest Pentium 4.

All-purpose data compression (archiving)

Archiving ranks are predictable: 7-Zip, which can use multiprocessing, prefers those Pentium 4 processors that support Hyper-Threading. More traditional and conservative RAR cannot take advantage of the new technologies so far. As a result, the overall winner is Pentium 4 with Hyper-Threading support (irregardless of the core), Celeron D lags slightly behind them, and the old Celeron expectedly brings up the rear. Let's note once again that Celeron D is closer to the Pentium 4 group in performance than to its predecessor.

Multimedia lossy compression (MP3/MPEG2-4)

The recently found LAME version, which is called LAME MT by the author (it's easy to guess that MT stands for Multi Threading), can use the second CPU and consequently it can potentially work faster with processors supporting Hyper-Threading. Considering this fact and that it's no slower than the traditional program on regular CPUs, we decided to use it in our tests.

In VBR encoding (160-320 Q=2 V=0) both processors with Hyper-Threading support take up the running. Considering the key feature of LAME MT, it's not difficult to guess the reason.

But in the best quality encoding (CBR 320, Q=0) all CPUs demonstrate nearly the same results. Northwood is still a tad faster; and Celeron on the old core suddenly outperforms Celeron D. However, we could see all these tendencies on the previous diagram as well, they were not that prominent though.

DivX demonstrates a staggering devil-may-care attitude to everything: core, cache size nothing can accelerate this codec significantly. Certainly, there may be other options, but in our opinion it indicates a weak optimization of the codec in the first place.

However, XviD demonstrates a similar behaviour, singling out (in the worst sense of the word) only the old Celeron.

MPEG4 codec Windows Media Video is obviously attracted to multiprocessing, even if it's virtual; but it's as indifferent to the cache size as the previous encoders. You can see it for yourselves by comparing the junior Prescott with 533 MHz FSB and Celeron D the only difference between them is L2 cache size.

Canopus ProCoder 2 prefers the Prescott core, but only in combination with Hyper-Threading. But even here Celeron D demonstrates more than worthy results.

The summary diagram clearly demonstrates two tendencies: firstly, "virtually multiprocessing" systems are victorious on an average across all the tests, because some codecs support multi-threaded operations. In other respects, architectural differences, even L2 Cache size differences, have little effect on the performance level. The same clock and the results are approximately the same. Let's make a reservation — we cannot check whether this is true for all processors so far, because in this case the tests don't include competitors from AMD.

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. No surprises, everything is as usual: Prescott is better than Northwood, processors with Hyper-Threading support are better than those without it. The junior Prescott is almost on a par with Celeron D in performance, which demonstrates the low requirements of this test package to cache... however, we already knew that well before.

3D games and graphics visualization
in professional packages

This time we decided to pay more attention to game tests, so each one is represented by four diagrams: performance in 640x480x32 with low graphics quality and minimum graphics effects, 800x600x32 with average graphics quality, and 1024x768x32 with high graphics quality (we didn't use Very High settings, though, and other extreme options). Besides, the fourth diagram shows fps percentage drop for each processor when switched from the lowest graphics mode into the highest mode. Theoretically, this parameter allows us to evaluate how much the game performance depends on a video card: if the percentage is high, then the processor manages more or less well and the system bottleneck is a 3D accelerator, if the percentage is low, then the performance depends more on the processor.


Quite an unexpected result: Prescott, considered much less appropriate for gaming, turns out better than a Northwood with the same clock. Could they fine-tune the video card drivers? Did the latest game patches contain some changes? Unfortunately, we cannot expect to get this information from the manufacturers...

On the whole, the situation completely repeats the previous one in 800x600x32 with the average quality settings.

With the further increase of the resolution, Celeron D is getting closer to the three Pentium 4 processors, that is the speed of a 3D accelerator is gradually becoming the stumbling block.

There is a reason that the column for the old Celeron is absent on this diagram: the system with this processor demonstrated the same performance (by the way, it was very low) both in 640x480 and in 1024x768 modes. The performance drop on all Pentium 4 processors with the increased resolution and effect complexity can be characterized as "significant, but not fatal". So we can draw a conclusion: a video card influences the performance as well as the processor. Celeron D is obviously not enough for this game: fps values are almost the same in the low mode as well as in the relatively high mode. It's not serious — just 16% of the performance drop with the quality settings switched from the lowest to the high mode and the resolution increased by two steps...

Far Cry

On the whole, the picture is similar to DOOM: the old Celeron is drastically outperformed, Pentium 4 processors come in a tight group, Celeron D catches up with this group with the increasing resolution.

The dependence on a video card breaks all records. More than a hundred percents even in the worst case, i.e. the performance in the High Quality mode is more than twice as low as in the Low mode!


This game does not like the Northwood core so that it even prefers Celeron D to one of sterling Pentium 4! The bus frequency and Hyper-Threading support have no effect on the performance, you can easily see that if you compare Pentium 4 2.8A and Pentium 4 2.8E.

The situation does not change with the increased resolution and graphics complexity... because it is practically the same! It seems that the video card is almost not loaded and the engine is totally up to the processor. Maybe that's because the game uses the prolicensed Havok engine (realistic physics)?

It's a motley picture, which probably shouldn't be taken into consideration (except for the total low percentage). Even the difference of one unit becomes significant with these low fps values, resulting in unexpected changes in columns. However, the general conclusion is clear: Painkiller depends much on a processor.

Unreal Tournament 2004

The situation resembles Painkiller, but the latter diagram is steadier. Another CPU-critical game, you can see it well by the fact that none of Celeron processors starts to catch up with the Pentium 4 group even in the most complex graphics mode.

Total scores in games

You can easily see two tendencies: firstly, Prescott-based processors demonstrate generally higher performance in games, it also concerns the junior model with a 533-MHz bus and without Hyper-Threading support. Secondly, the higher the resolution as well as the complexity and quantity of effects are, the lower the difference between Pentium 4 and Celeron D is. Thirdly, you can still see well that Celeron D (at least with the clock equal to that in Pentium 4) cannot be considered a serious competitor in games. Yes, it's not as bad as the old Celeron, but it's still not good enough for gaming.

SPEC viewperf

Remember, as our contenders differ much by their key technical parameters (L2 Cache size varies from 128KB to 1024KB, bus frequencies range from 400 QPB to 800 QPB MHz!), we decided to publish all diagrams, including intermediate ones. If this is too tiresome, you can always skip them and look for the summary diagrams (their titles are highlighted with a different color).

There are no deviations from the general tendency in most subtests, despite our expectations. The only exception is the maya subtest, where Northwood moves forward a little. On the whole, everything is simple: Prescott is either better or at least not worse than Northwood, Celeron D is behind them, but the gap is not that fatal; what concerns the old Celeron, it's somewhere far behind... you can see with your own eyes where it is... Strange as it may seem (on the other hand, why is that strange?), the situation in games and 3D visualization in professional packages is approximately the same.


On the whole, we can say that no miracles happened. The overall winner is still Prescott with Hyper-Threading support, but its breakaway is not so large to pay much attention. The second (it's very relative, mind it...) place is taken by Northwood on the 800MHz bus, which also supports Hyper-Threading. The slowest Pentium 4 processor, which took part in the tests, is Prescott without Hyper-Threading on a 533 MHz bus. Remember that it applies only to the total score. All Prescott processors are good in games, for example. Actually, all the three Pentium 4 processors come in a tight group, the difference between them is not large, so you'd better take into account the clock and the price rather than the core.

But the old and new Celerons are different as heaven and earth. Actually, performance of the new Celeron D is more like Pentium 4 than the old Celeron. It's great. On the other hand, pessimists may say that the reason to be happy is questionable: compared to such a slow processor as old Celeron with 128KB L2 Cache, any normal low-end processor would look like a champion. Well, perhaps pessimists are correct in parts.

Anyway, we could hardly take old Celeron processors for Socket 478 seriously (they were scornfully called socket stubs); but unlike those processors, Celeron D is just a regular low-end CPU, which performance is proportional to its senior brothers.

We are going to publish a small table with approximate prices for the reviewed processors to facilitate your choice. Note that all the prices are rounded (in compliance with regular rounding-off rules) accurate to $10. It seems quite reasonable to us, because they are approximate anyway, they cannot be 100% precise.

Price (for 04.20.2005)
Pentium 4 2.8E GHz BOX
Pentium 4 2.8A GHz BOX
Pentium 4 2.8C GHz BOX
Celeron D 335 (2.8 GHz) BOX
Celeron 2.8 GHz BOX
In the next part we shall review low-end processors for Intel LGA775 compared to Pentium 4 processors for the same platform and the best representatives of Socket 478 (you understand already that it will be Celeron D 335 and Pentium 4 2.8E GHz). Main heroes of the third part will be AMD Duron and AMD Sempron for Socket A and Socket 754.

Stanislav Garmatiuk (nawhi@ixbt.com)
April 20, 2005.

Write a comment below. No registration needed!

Article navigation:

blog comments powered by Disqus

  Most Popular Reviews More    RSS  

AMD Phenom II X4 955, Phenom II X4 960T, Phenom II X6 1075T, and Intel Pentium G2120, Core i3-3220, Core i5-3330 Processors

Comparing old, cheap solutions from AMD with new, budget offerings from Intel.
February 1, 2013 · Processor Roundups

Inno3D GeForce GTX 670 iChill, Inno3D GeForce GTX 660 Ti Graphics Cards

A couple of mid-range adapters with original cooling systems.
January 30, 2013 · Video cards: NVIDIA GPUs

Creative Sound Blaster X-Fi Surround 5.1

An external X-Fi solution in tests.
September 9, 2008 · Sound Cards

AMD FX-8350 Processor

The first worthwhile Piledriver CPU.
September 11, 2012 · Processors: AMD

Consumed Power, Energy Consumption: Ivy Bridge vs. Sandy Bridge

Trying out the new method.
September 18, 2012 · Processors: Intel
  Latest Reviews More    RSS  

i3DSpeed, September 2013

Retested all graphics cards with the new drivers.
Oct 18, 2013 · 3Digests

i3DSpeed, August 2013

Added new benchmarks: BioShock Infinite and Metro: Last Light.
Sep 06, 2013 · 3Digests

i3DSpeed, July 2013

Added the test results of NVIDIA GeForce GTX 760 and AMD Radeon HD 7730.
Aug 05, 2013 · 3Digests

Gainward GeForce GTX 650 Ti BOOST 2GB Golden Sample Graphics Card

An excellent hybrid of GeForce GTX 650 Ti and GeForce GTX 660.
Jun 24, 2013 · Video cards: NVIDIA GPUs

i3DSpeed, May 2013

Added the test results of NVIDIA GeForce GTX 770/780.
Jun 03, 2013 · 3Digests
  Latest News More    RSS  

Platform  ·  Video  ·  Multimedia  ·  Mobile  ·  Other  ||  About us & Privacy policy  ·  Twitter  ·  Facebook

Copyright © Byrds Research & Publishing, Ltd., 1997–2011. All rights reserved.