Intel’s Pentium M 1.4GHz processor

As far as I’m concerned, the real gem of Intel’s Centrino brand is the new Pentium M processor, which integrates all sorts of neat power saving features to preserve notebook battery life. The Pentium M is loaded with cache and other goodies that let it execute a much higher number of instructions per clock (IPC) than mobile versions of Intel’s Pentium 4, too.

I was lucky enough to get my hands on a Pentium M 1.4GHz-powered notebook for a few days, and I ran it through a gauntlet of performance benchmarks against my own Mobile Pentium 4 1.8GHz laptop. What’s the Pentium M all about? Can it perform enough work per clock cycle to make up for a 400MHz clock speed disadvantage? Read on to see.

Meet Banias
Code-named Banias, the Pentium M actually owes a lot more to Intel’s previous generation of Pentium III processors than it does the new Pentium 4 line. Here are the key ingredients of Intel’s new mobile chip.

• A monster L2 cache – The Pentium M’s 1MB cache is massive. Intel’s current high-end mobile and desktop CPUs have only 512KB of L2, which makes the Pentium M look very different right off the bat. The Pentium M’s L2 cache is 8-way associative, just like the Pentium 4’s, but to conserve power, elements of the Pentium M’s L2 cache are only activated when needed. Having to power up parts of the L2 cache before use will add latency that can degrade system performance, but it should reduce the Pentium M’s power requirements, which is key for a mobile chip.
• Don’t skimp on the L1 – The Pentium M’s huge L2 cache is complemented by 64KB of L1 cache that’s split evenly between instruction and data caches. Intel has yet to divulge the exact size of the Pentium 4’s L1 instruction cache, but the Pentium 4 M’s 8KB L1 data cache is a quarter the size the Pentium M’s.

  One more difference between the Pentium M and Pentium 4’s L1 cache is that the former is a write-back cache, while the latter is a write-through cache. With a write-through cache, data is written to L1 and main memory simultaneously; write-back caching only writes L1 data to main memory when absolutely necessary. In theory, a write-back cache should be faster than a write-through cache because the write-back cache does fewer slow memory writes.

• A 400MHz system bus – Like early Pentium 4 desktop processors and Intel’s current mobile Pentium 4 offerings, the Pentium M uses a quad-pumped 100MHz front-side bus, or effectively a 400MHz bus. The Pentium M’s 400MHz front-side bus is actually identical to that of the Pentium 4.
• 0.13-micron tech – Intel’s next Pentium M revision will apparently be spun using 0.09-micron manufacturing technology, but the current Banias core is built using 0.13-micron tech. With 77 million transistors, nearly half of which are used for L2 cache, the Pentium M has 23 million more transistors than the Pentium 4.
• SSE2 support – The Pentium M supports MMX, SSE, and SSE2 instructions, though it won’t support the 13 new instructions that will reportedly be incorporated into Intel’s upcoming Prescott-core Pentium 4 desktop chip.
• Somewhere between 10 and 20 – Taking a cue from NVIDIA, Intel isn’t revealing exactly how deep the Pentium M’s main pipeline is. They have said that it’s deeper than the Pentium III’s 10-stage pipeline, but shallower than the Pentium 4’s 20-stage pipeline. A deeper pipeline does help enable higher clock speeds, but it also increases the penalty for branch mispredictions. The Pentium M aims to avoid branch mispredictions as much as possible using a swanky new branch prediction unit.
• Micro-ops fusion – With an eye towards overall efficiency, the Pentium M sends micro-ops down the pipeline to execution units in bundles rather than individually. This fusion of micro-ops gives the pipeline a little extra idle time, which saves power. However, the bundling process increases latency, since micro-ops aren’t always executed immediately.
• Low power consumption – Many of the Pentium M’s features aim to improve the chip’s performance, but some are there solely to conserve power. Given that the Pentium M is primarily a mobile chip that can also be used in dense, clustered server environments, customers may first look at power consumption before they examine the chip’s performance. The Pentium M 1.4GHz that we’ll be looking at today requires between 0.96 and 1.48V, although ultra-low-power versions of the chip clocked between 600 and 900MHz need only between 0.84 and 1.0V.

  To help conserve power, all Pentium M chips use third-generation SpeedStep technology to raise and lower their clock speeds and core voltages. SpeedStep lets the Pentium M deliver more performance when it’s needed and consume less power when it’s not. The chip is also able to shut down internal components such as unused segments of L2 cache to draw even less power. As if that weren’t enough, the Pentium M actually boots in a sleep state and only activates internal components as they’re needed; the Mobile Pentium 4 boots at full power and disables functional units as they become idle.

The Pentium M is currently available with clock speeds between 600MHz and 1.7GHz, which offers users a wide range of performance and power consumption options to choose from. Today, we’ll be looking at the 1.4GHz version of the chip, which seems to be in the sweet spot for notebook pricing right now.

A word on the chipset
Intel designed the Pentium M to work with the 855 core-logic chipset, which has a 400MHz front-side bus, support for DDR333 memory and AGP 4X, and a few extra power saving features of its own. The 855 chipset can send power-down instructions to the Pentium M and is also capable of shutting down elements of the system bus to further conserve power. A series of ultra-deep sleep states rounds out the 855 chipset’s power-saving features.

The 855 chipset is available with and without an integrated graphics core as the 855PM and 855GM, respectively. Both north bridge chips interface with a mobile version of Intel’s ICH4 south bridge, the ICH4M. The desktop and mobile versions of the ICH4 are identical in terms of features, but the ICH4M supports deeper power saving states.

For notebooks, Intel is bundling the Pentium M with its 855 chipset and Pro/Wireless 2100 802.11b Wi-Fi networking technology under the Centrino name. Centrino is little more than a branding ploy that ties the Pentium M to Intel’s own chipset and Wi-Fi card, but that’s not necessarily a bad thing. At least when buying a Centrino-branded notebook, consumers know they’re getting a Pentium M with all the power-saving features of the 855 chipset and 802.11b Wi-Fi connectivity.

Our testing methods
All tests were run three times, and their results were averaged, using the following test systems.

Because the Pentium M is a mobile chip, it’s hard to cobble together perfectly comparable systems for testing. Luckily, my Mobile Pentium 4-equipped Inspiron 8200 is similar enough to the Latitude D600 system to warrant a comparison, but there are a few key differences to note. First, notice that the Mobile Pentium 4 1.8GHz system is only using DDR266 memory while the Pentium M is using DDR333. Also note that, despite using the same graphics chip, the two notebooks differ in their graphics memory sizes and driver versions. With mobile graphics applications, you’re at the mercy of notebook vendors for drivers. I used the most recent driver versions from Dell’s web site.

Since I only had access to the Pentium M-equipped Latitude D600 notebook for a few days, I was only able to quickly run it through our standard suite of processor benchmarks. I would have loved to do some real-world battery life testing with the Pentium M, but there just wasn’t time.

We used the following versions of our test applications:

• Cachemem 2.65MMX
• SiSoft Sandra Standard 2003 (2003.3.9.44)
• Compiled binary of C Linpack port from Ace’s Hardware
• ZD Media Business Winstone 2002 1.0.1
• ZD Media Content Creation Winstone 2002 1.0.1
• Cinebench 2003
• Sphinx 3.3
• LAME 3.92
• Xmpeg 4.5 with DivX Video 5.02
• FutureMark 3DMark03
• Unreal Tournament 2003 demo benchmark
• Comanche 4 demo benchmark
• Quake III Arena v1.31
• Serious Sam SE v1.07
• SPECviewperf 7.1

The test systems’ Windows desktop was set at 1024×768 in 32-bit color at a 60Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests. Most of the 3D gaming tests used the high detail image quality settings, with the exception that the resolution was set to 1024×768 in 32-bit color.

All the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

Memory performance

The Pentium M 1.4GHz’s use of DDR333 memory gives it a clear edge in Sandra’s memory bandwidth tests. With notebook motherboard BIOSes, there really isn’t any room for memory tweaking, so scores are a little lower than we’re used to seeing with DDR266- and DDR333-equipped desktop systems.

Our Linpack results clearly illustrate the Pentium M 1.4GHz’s large L1 and L2 caches. The Pentium M is very fast until it hits matrix sizes near 900KB, where performance drops off as we move from the chip’s massive 1MB L2 cache to main memory.

In Cachemem’s memory bandwidth tests, the Pentium M 1.4GHz is way out ahead in the read bandwidth test, but slightly trails the Mobile Pentium 4 1.8GHz when it comes to write bandwidth.

With a 2048KB block size and 512-byte stride, the Pentium M 1.4GHz has slightly higher memory latency than the Mobile Pentium 4 1.8GHz. Let’s widen the scope a little and look at the full set of Cachemem results. I’ve color coded the levels of the memory hierarchy, from L1 cache to main memory, to make them easier to see.

Despite its power saving features, the Pentium M 1.4GHz’s latencies are actually quite reasonable when compared with the Mobile Pentium 4 1.8GHz.

Business and Content Creation Winstone

The Pentium M 1.4GHz embarrasses the Mobile Pentium 4 1.8GHz in the Business and Content Creation Winstone benchmarks, despite the Pentium M’s 400MHz clock speed advantage.

Media encoding

The Pentium M 1.4GHz is faster with MP3 and DivX encoding, too, though the Mobile Pentium 4 1.8GHz is quite a bit closer this time around.

Gaming

In 3DMark03’s CPU tests, the Pentium M 1.4GHz is nicely out in front of our Mobile Pentium 4 1.8GHz, but things change when we switch over to real-world games.

In our real-world gaming tests, the Pentium M 1.4GHz system gets smacked around by the Mobile Pentium 4 1.8GHz system, but the performance difference likely has much more to do with each notebook’s graphics subsystem than with actual processor performance. After all, the Mobile Pentium 4 1.8GHz-powered Inspiron 8200 is running with an extra 32MB of graphics memory and a more recent driver revision than the Pentium M 1.4Hz-equipped Latitude D600.

Cinebench 2003

In Cinebench’s processor-intensive rendering and shading tests, the Pentium M 1.4GHz is back on top. However, its performance advantage isn’t as dramatic as it was in the Winstone tests. The Pentium M is being held back by our test laptop’s graphics subsystem in the OpenGL hardware shading test.

POV-Ray

In POV-Ray’s ray tracing tests, the Pentium M 1.4GHz is faster for both scenes. However, the Mobile Pentium 4 1.8GHz does narrow the gap when rendering the complex “glasschess” scene.

SPECviewperf

In SPECviewperf’s suite of workstation graphics tests, the Pentium M 1.4GHz actually fares rather well, despite its graphics memory disadvantage. With the Pentium M and Mobile Pentium 4 each winning three tests, I’m going to have to call this one a draw. However, the Mobile Pentium 4 1.8GHz probably owes its success to its graphics subsystem advantage more than anything else.

ScienceMark

ScienceMark’s tests show the clock speed-deprived Pentium M 1.4GHz out ahead again overall—and in all three individual calculations, too.

Speech recognition

The Pentium M 1.4GHz is just able to run Sphinx’s speech recognition engine in real time, while the Mobile Pentium 4 1.8GHz trails by more than 20%. Of course, the holy grail for Sphinx is to run at least 20% faster than real time to allow speech-based applications to run in the background; the Pentium M 1.4GHz isn’t there yet.

Conclusions
Despite only scraping the surface of the Pentium M’s power saving features, I’m already enamored with the chip. It routinely beats a Mobile Pentium 4 that’s running 400MHz faster. That fact is a testament to the Pentium M’s intelligent design and high IPC. Given that some of the chip’s power saving features actually increase latency, it’s impressive to see overall performance so competitive. We can probably thank the chip’s comparatively large L1 and L2 caches for some of that impressive performance.

Unfortunately, I wasn’t able to test the Pentium M 1.4GHz against anything other than my Mobile Pentium 4 1.8GHz-based laptop. Pentium M processors are currently available with speeds as high as 1.7GHz, but the Mobile Pentium 4 tops out at 3.06GHz on a 533MHz bus with Intel’s new 852 chipset. Even though we saw the Pentium M 1.4GHz take it to the Mobile Pentium 4 1.8GHz, it’s unlikely that a Pentium M clocked at 1.7GHz will beat a Mobile P4 at 3.06GHz on a 533MHz front-side bus. Battery life will be a different story, though.

Still, the Mobile Pentium 4’s performance lead in the notebook space may be short-lived. Intel is supposed to roll out a new “Dothan” Pentium M core later this year that will reportedly be built using 0.09-micron process technology, have higher clock speeds, and feature a massive 2MB L2 cache. As always, there’s something better just over the horizon.

Today, the Pentium M 1.4GHz delivers compelling performance with conservative power consumption that makes it ideal for just about every notebook short of a high-end desktop replacement aimed at maximizing performance. Remember, for some notebook users, power consumption may be a more important factor than benchmark scores, which makes the Pentium M’s long list of smart power saving features that much more important. Heck, I think the Pentium M would even work in small form factor desktop systems, and it definitely has potential in high-density server clusters.