Putting It in Perspective: The Atom Takes on a Single-Core Pentium 4

When the Atom first appeared I immediately did my best to characterize its performance. Intel always referred to it as having the performance of a 90nm Pentium M but never really got as specific as I would like. In my subsequent testing I found that at 1.6GHz, Intel’s Atom performed like a 1.2GHz 90nm Pentium M (Dothan). If you had a Pentium M notebook back then, the comparison makes sense to you. If you didn’t, however, then you’d need another reference point.

I realized I had no idea how Atom compared to an old Pentium 4. To remedy the situation I dug up an old Dell based on a single-core, Northwood based (130nm) Pentium 4 running at 2.66GHz. These chips used a 533MHz FSB and had a 512KB on-die L2 cache. The system had an ATI Radeon X800 XT and 2GB of DDR memory. This Pentium 4 did not have Hyper-Threading, this was before HT made its official desktop debut.

For my first test I turned to Adobe Photoshop CS4:


Wow. This is huge. Intel’s dual-core Atom 330 is 25% faster than the 2.66GHz Pentium 4. The advantage is all in the extra core though. If you look at the Atom 230 the Pentium 4 beats it by almost 20%.

Next I wanted to look at video encoding performance, one of the Pentium 4’s strongpoints. If you remember back then, even the Athlon 64 had difficulty competing with the P4’s encoding prowess. We’ll start with Windows Media Encoder:

Windows Media Encoder 9 x64 - Advanced Profile Transcode

Again, the Atom 330 is faster than the Pentium 4 and again it’s due to the extra core. The Atom 230 is noticeably slower.

x264 encode performance is even more embarrassing for the Pentium 4; the Atom 330 is 62% faster in the first pass and 94% faster in the 2nd pass of the encode. The P4 system was 15% faster than the Atom 230 in the first pass and around 5% faster in the second pass.

x264 HD Encode Benchmark - 720p MPEG-2 to x264 Transcode

x264 HD Encode Benchmark - 720p MPEG-2 to x264 Transcode

Next up was Cinebench. This is an interesting benchmark to look at because it offers both a single-threaded and multi-threaded benchmark. The Pentium 4 without Hyper-Threading can only execute a single thread, so the multi-threaded test can’t run. I’ve combined the meaningful results in a single chart:


Cinebench R10

If we look at single-threaded performance, the Pentium 4 destroys the Atom - offering more than twice the performance of the svelte Atom core. Enable the multi-threaded render and the Atom 330 pulls ahead by 33%. This is the strength of the Atom; it’s clearly not faster than the Pentium 4 when working on a single thread, but exploit its ability to work on four threads simultaneously and it can actually be faster than a much larger, much more power hungry Pentium 4. This is a very important point to remember because despite losing to the Atom 330 in nearly all of the benchmarks, the Pentium 4 system it felt much faster than the Atom system in normal usage.

Simple things like opening up Control Panel or switching between windows felt faster on the old Pentium 4 system, which makes sense given that those tasks only spawn a single thread. In Cinebench we saw that the single-threaded performance of the old Pentium 4 was over twice as fast as the Atom - you tend to notice performance differences of that magnitude.

The final application test I ran was our WinRAR compression benchmark:

WinRAR 3.8 Compression - 300MB Archive

Once more, the Pentium 4 gets beat by the Atom 330 but turns the tables on the Atom 230. Even file compression is multi-threaded these days, and that is the Atom’s saving grace.

I wanted to know how well the GeForce 9300 could stand up against the old Dell’s Radeon X800 XT so I ran my WoW test on both systems. This is less of a CPU test and more of a CPU + GPU test obviously, but the results are interesting:

World of Warcraft - 800 x 600 - Good Quality

The Pentium 4 + X800 XT system is clearly faster, but not by as much as I expected. The advantage amounts to 22.5%, which is very noticeable, but close enough that if you had an even older system (or one with a lesser video card) you might not notice the performance difference between it and the Zotac Ion.

The last benchmark I ran comparing the Atom to a Pentium 4 was Futuremark’s Peacekeeper - a benchmark for measuring web browser performance. The benchmark is unaffected by internet connection and simply measures, once loaded, how fast your PC can work with various forms of commonly used javascript. I ran all of the Peacekeeper tests using Google Chrome:

The Pentium 4 absolutely demolished the Atom here. Once again it’s almost twice as fast as the Atom 330. In a multi-threaded environment the Atom 330 can pull ahead, but in most typical day-to-day tasks the Pentium 4 is going to be a much faster solution.

Power efficiency is obviously where the Atom based Zotac Ion wins out. AMD and Intel both viewed the move to multiple cores as a performance and power efficiency win, and the results below help show that. Granted Atom is built on a 45nm process and we’re looking at a 130nm desktop Pentium 4 system, which is where much of the difference really comes from. But the results do show you that if you don’t need top bin performance you can build a very capable PC these days with significantly better power consumption characteristics.

Idle Power Consumption

Load Power Consumption (x264 HD 1st Pass)

NVIDIA is forever arguing with me about how Atom and its Ion platform is a significant upgrade to PCs that are a few years old. My Pentium 4 results show that unfortunately, that’s not exactly the case. The Pentium 4 system I compared Zotac’s Ion to could never function as a modern-day HTPC and it won’t play Blu-ray discs. In those senses, Zotac’s Ion would be a tremendous upgrade. Even in multi-threaded CPU bound scenarios such as video encoding or 3D rendering, a dual-core Atom 330 is appreciably faster than the old Pentium 4 based Dell system. However, in normal web browsing and simply interacting with Windows, the Pentium 4 is significantly faster. The benchmarks show a nearly 2x performance advantage in single-threaded applications and I’d say they are accurately reflected in the usage experience.

Zotac’s Ion in My Theater Zotac's Ion vs. Intel's D945GCLF2 in Application Performance
Comments Locked

93 Comments

View All Comments

  • KidneyBean - Tuesday, May 12, 2009 - link

    Isn't it stealing to use Media Player Classic if you use patented decoders without paying for them?

    If so, I would hope you don't publish such information in articles in the future.
  • Penti - Friday, May 29, 2009 - link

    Nope infringement is never steeling. It's pretty okay as long as it isn't used commercially. Usually your allowed to use any patents for personal use though. So it's fine. You won't go to jail. Of course he won't publish stuff that discourage the use of homebrew codecs and software that are needed for playing warez and illegally ripping your movies. It would be silly as that's what people use computers for, especially htpcs. It would really be a limited worthless device without it.
  • sprockkets - Tuesday, May 12, 2009 - link

    With MPEG4, you only pay what you publish with it, not decode.

    DXA for nvidia cards is clearly documented on how to use it.

    Dolby Digital is another story, but seeing as how you can send the bitstream to your receiver to decode it legally, what's the big deal?
  • Jeffk464 - Tuesday, May 12, 2009 - link

    I don't understand what the major need is to have a low power desktop. If it plugs into the wall its not really an issue like it is with notebooks. Its seems like the main focus for a Home Theater PC should be in keeping things quiet. Why not use a notebook processor and focus on making a super quiet heat sink and fan. I like the idea of compact pc's but HP has managed to squeeze a full on systems with a high end cpu onto the back of a lcd display in a nice little package.
  • cosmotic - Tuesday, May 12, 2009 - link

    How are Flash (normal animation and (HD)video) and Java graphics performance?

    For Java, you can install the JDK and double click the jar:
    C:\Program Files\Java\jdk1.6.0_13\demo\jfc\Java2D\Java2Demo.jar

    The composite and transform demos would be particularly helpful.
  • Anand Lal Shimpi - Tuesday, May 12, 2009 - link

    I've been meaning to talk about FLV performance, I'll try and include that in a short followup soon.

    Take care,
    Anand
  • strikeback03 - Tuesday, May 12, 2009 - link

    The version of the Intel LF2 board you have here is already being pulled from the channel. It's replacement is the LF2D, which loses the S-Video port. Not a huge loss, as most people who tried to use it could not get what they wanted out of it (does not support different content to both the VGA and S-Video ports).

    Was Vista the OS for all the tests? I am using an LF2 board in my carputer with 1GB RAM, running a somewhat stripped-down version of XP on a 7200RPM 2.5" drive. Running on the car touchscreen (7", 800x480) or in PIP on my desktop monitor, and it feels pretty snappy in general usage. If I give it the full 1920x1200 though it slows down a lot, especially in my front-end software. Also, the image quality over VGA is not as good as some laptops I have hooked up to that monitor over VGA.

    For carputer use I would like to see a better onboard audio codec and more ports. The 3-port thing doesn't give much flexibility on inputs and outputs.

    Is the increased power consumption of the Intel boards due to using a standard ATX PSU? Did you consider getting one of the Pico-PSUs (like this: http://www.mini-box.com/picoPSU-150-XT-102-power-k...">http://www.mini-box.com/picoPSU-150-XT-102-power-k... ) to try on a more comparable basis? Also, is there any way to test the draw on each rail? The Intel boards are not really compatible with the lower-powered Pico-PSUs due to rather excessive draw on the 5V rail.
  • Anand Lal Shimpi - Tuesday, May 12, 2009 - link

    Vista was the OS used for all of the tests and I'll second what you're saying, I've found XP far snappier than Vista with these Atom based systems. I just stuck with Vista to keep the results somewhat comparable to other CPU results.

    I did use a standard ATX power supply on the Intel boards but that's why I also tossed in the Eee Box results to show you what an Atom system with a smaller PSU can do.

    Take care,
    Anand
  • Jeffk464 - Tuesday, May 12, 2009 - link

    I don't understand what the major need is to have a low power desktop. If it plugs into the wall its not really an issue like it is with notebooks. Its seems like the main focus for a Home Theater PC should be in keeping things quiet. Why not use a notebook processor and focus on making a super quiet heat sink and fan. I like the idea of compact pc's but HP has managed to squeeze a full on systems with a high end cpu onto the back of a lcd display in a nice little package.
  • sprockkets - Tuesday, May 12, 2009 - link

    Look at the prices on logic supply for a laptop socket itx board and how much the processor is and you will see how much cheaper this setup is.

    Besides, the vast majority of Anandtech readers prefer to build ourselves, not buy OEM crap.

Log in

Don't have an account? Sign up now