Intel’s Pentium 4 3.4GHz processors

Amid the spectacle of wardrobe malfunctions and abnormally large caches, some noteworthy developments got a little lost. For one, the new Prescott processor was a shockingly major replumbing of the familiar Pentium 4, with an uber-deep 31-stage pipeline and a host of internal tweaks. These changes made Prescott usually a little slower, clock for clock, than the previous Pentium 4. What’s more, Intel was launching a pair of products, the Pentium 4 3.4GHz and 3.4’E’ GHz processors, that didn’t appear to exist yet. We were testing old and new Pentium 4 cores, Northwood and Prescott, at 3.2GHz, but not beyond. Only the P4 Extreme Edition was available for review at 3.4GHz.

The fact that samples weren’t available to the press was a bad omen. Over the ensuing days and weeks, it became clear Prescott supplies were tight at any speed. However, about three weeks ago, Intel made good on its promise to follow up with samples of its Pentium 4 3.4GHz chips when they became available. Today, at last, we can show you how they perform.

Reintroducing Prescott
If you aren’t familiar with the new Pentium 4 Prescott processor core, you really should go read our initial review of it. The changes from the older Northwood core are too complex to summarize easily. That said, I’ll make a feeble attempt. Prescott includes larger caches, a much deeper main pipeline, SSE3 instructions, improved data prefetching, revised Hyper-Threading, and enough microarchitectural tweaks to kill a horse. Most importantly, Prescott is manufactured using Intel’s new 90nm fabrication process, making it a smaller chip than the 130nm Northwood, despite having a much higher transistor count.

Test notes
Generally, the deeper we get into the technical stuff, the happier I am. However, I’m none too excited about making these articles difficult to read. So, as you peruse the following pages and glance over test results for no less than eighteen different CPUs, please keep a few things in mind.

First, Intel distinguishes the new Prescott Pentium 4 from the regular ol’ Northood P4 by tacking an “E” on to the end of the name, after the clock speed. So when you’re reading our graphs and see a Pentium 4 “E” processor, think Prescott. The Pentium 4 “C” you’ll see here is simply a Northwood chip. It gets the “C” tag because it has an 800MHz bus, while older versions of that chip did not. (All the Pentium 4s we’ve tested here have an 800MHz bus.) Forgive the confusion, but we’ve tried to stick to Intel’s naming scheme where possible, and confusion may be the natural outcome of that effort.

Next, you may recall from our original Prescott review that we actually published a complete set of benchmark results for the Northwood 3.4GHz processor way back then. We accomplished this feat by disabling the L3 cache on our Pentium 4 Extreme Edition 3.4GHz processor, effectively making it a Pentium 4 Northwood running at 3.4GHz. We’ve used the actual Northwood 3.4GHz to verify that our “simulated” Northwood 3.4GHz results were correct, but we didn’t rerun every test, because we’re confident our results are representative.

Finally, watch the line graphs that come after our usual bar graphs to see how each CPU type scales with clock speeds or model number increments. We will be keeping an eye on Northwood versus Prescott in particular, because that’s a key issue. We’re wondering whether Prescott performance gets relatively stronger as clock speeds go up, because Prescott is intended to run at high clock frequencies. In fact, it’s slated to hit 4GHz or so by year’s end, while Northwood will be phased out.

Our testing methods
As ever, we did our best to deliver clean benchmark numbers. Tests were run at least twice, and the results were averaged.

Our test systems were configured like so:

All tests on the Pentium 4 systems were run with Hyper-Threading enabled, except where otherwise noted.

Thanks to Corsair for providing us with memory for our testing. If you’re looking to tweak out your system to the max and maybe overclock it a little, Corsair’s RAM is definitely worth considering.

The test systems’ Windows desktops were set at 1152×864 in 32-bit color at an 85Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests.

We used the following versions of our test applications:

• Cachemem 2.65MMX
• SiSoft Sandra 2004 (9.89)
• Compiled binary of C Linpack port from Ace’s Hardware
• Discreet 3ds max 5.1 SP1
• NewTek Lightwave 7.5
• Cinebench 2003
• POV-Ray for Windows v3.5
• PICCOLOR v4.0 build 472
• SPECviewperf 7.1.1
• ScienceMark 2.0 beta (23SEP03 build)
• Sphinx 3.3
• LAME 3.95.1 (build from mitiok.cjb.net)
• Xmpeg 5.0.3 with DivX Video 5.11
• FutureMark 3DMark03 build 340
• Comanche 4 demo
• Quake III Arena v1.31
• Serious Sam SE v1.07
• Splinter Cell v1.2
• Unreal Tournament 2003 demo v.2206
• Wolfenstein: Enemy Territory v2.55

The tests and methods we employ are generally publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

Memory performance
As always, we start with memory performance, because these benchmarks are synthetic, a little bit different, and not always indicative of real-world performance. They are, however, interesting, and present the opportunity to make all sorts of colorful graphs.

Both of the Pentium 4 3.4GHz processors have essentially the same amount of memory bandwidth available, as the Sandra results show, but the less aggressive cachemem algorithm allows Prescott to show off its improved data prefetching into its larger L2 cache.

Linpack shows the dramatic difference between these two Pentium 4s. Northwood offers much higher peak floating-point performance when working inside its 512K L2 cache, but Prescott’s larger cache and better prefetch gives it higher throughput with larger data matrices.

Prescott shows higher memory access latencies than Northwood, probably in part because Prescott has higher access latencies in its larger L1 data cache. Let’s look at the whole picture..

Memory performance (continued)
Not only are our 3D graphs indulgent, but they’re useful, too. I’ve arranged them manually in a very rough order from worst to best, for what it’s worth. I’ve also colored the data series according to how they correspond to different parts of the memory subsystem. Yellow is L1 cache, light orange is L2 cache, and orange is main memory. The red series on the Extreme Edition graph represents L3 cache. Of course, caches sometimes overlap, so the colors are just an interesting visual guide.

If you look closely, you can see how at the teeny block sizes, the Prescott (Pentium 4 “E”) has higher access latencies to its L1 cache than Northwood. Prescott makes up for it a little with its larger L2 cache, as demonstrated at the 1024K block size.

Compared to any of the Pentiums, the Athlon 64’s memory access latencies are extremely low thanks to the A64’s built-in memory controller.

Unreal Tournament 2003

The 200MHz clock speed boost isn’t nearly enough to help the Pentium 4 overtake the Athlon 64 in UT2003. The Prescott and Northwood scale almost identically here, tracking very close to one another in performance, as well.

Quake III Arena

Wolfenstein: Enemy Territory

The Athlon 64 again outdoes the P4 in Quake III and Wolf: ET. Notice how Prescott seems to pull away, ever so slightly, from Northwood in Quake III as clock speeds ramp up.

Tom Clancy’s Splinter Cell

Northwood and Prescott perform nearly identically in Splinter Cell, but neither can catch even the lowest model Athlon 64.

Comanche 4

Serious Sam SE

Comanche 4 and Serious Sam are not friendly territory for the Pentium 4. Prescott and Northwood scale similarly here, but Prescott is just kind of pokey in Comanche 4.

3DMark03

Prescott comes back strong in 3DMark’s overall test, outpacing both Northwood and the Athlon 64.

3DMark’s two CPU tests produce very different relative performance numbers from the different processor types.

Sphinx speech recognition
Ricky Houghton first brought us the Sphinx benchmark through his association with speech recognition efforts at Carnegie Mellon University. Sphinx is a high-quality speech recognition routine that needs the latest computer hardware to run at speeds close to real-time processing. We use two different versions, built with two different compilers, in an attempt to ensure we’re getting the best possible performance.

There are two goals with Sphinx. The first is to run it faster than real time, so real-time speech recognition is possible. The second, more ambitious goal is to run it at about 0.8 times real time, where additional CPU overhead is available for other sorts of processing, enabling Sphinx-driven real-time applications.

Prescott cranks through our speech recognition test in record time, nearly hitting the 0.5 mark. By the time it hits 4GHz, Prescott should require less than 50% of CPU time to process speech recognition in real time with Sphinx.

LAME MP3 encoding
We used LAME to encode a 101MB 16-bit, 44KHz audio file into a very high-quality MP3. The exact command-line options we used were:

lame –alt-preset extreme file.wav file.mp3

DivX video encoding
This new version of XMPEG includes a benchmark feature, so we’re reporting scores in frames per second now.

The Pentium 4 continues to excel in both audio and video encoding. Northwood is a little quicker at LAME MP3 encoding, but Prescott leads even the P4 Extreme Edition in DivX encoding. The P4 3.4 ‘E’ is roughly 17 frames per second faster than the Athlon 64 3400+ in our video encoding test.

3ds max rendering
We begin our 3D rendering tests with Discreet’s 3ds max, one of the best known 3D animation tools around. 3ds max is both multithreaded and optimized for SSE2. We rendered a couple of different scenes at 1024×751 resolution, including the Island scene shown below. Our testing techniques were very similar to those described in this article by Greg Hess. In all cases, the “Enable SSE” box was checked in the application’s render dialog.

Both flavors of Pentium 4 perform well in 3ds max. However, the Athlon 64 seems to scale very well with clock speed, as the Athlon 64 FX-53’s strong finishes indicate. The trend seems to indicate the Athlon 64 will surpass the Pentium in a speed grade or two.

Lightwave rendering
NewTek’s Lightwave is another popular 3D animation package that includes support for multiple processors and is highly optimized for SSE2. Lightwave can render very complex scenes with realism, as you can see from the sample scene, “A5 Concept,” below.

We’ve tested the Hyper-Threaded processors with one, two, and four rendering threads. For non-Hyper-Threaded processors, we just tested with one and two threads. For the line graphs, we’ve tried to pick results from the optimal number of threads to represent each processor.

Notice how much different the Northwood and Prescott are here. The Northwood doesn’t respond well to increasing thread counts, while Prescott’s deeper pipeline and improved Hyper-Threading does best with four simultaneous render threads.

POV-Ray rendering
POV-Ray is the granddaddy of PC ray-tracing renderers, and it’s not multithreaded in the least. Don’t ask me why—seems crazy to me. POV-Ray also relies more heavily on x87 FPU instructions to do its work, because it contains only minor SIMD optimizations.

Obviously, Intel has moved increasingly toward SSE2 and the like for floating-point math. Prescott is relatively weak here, even compared to Northwood, let alone the Athlons.

Cinebench 2003 rendering and shading
Cinebench is based on Maxon’s Cinema 4D modeling, rendering, and animation app. This revision of Cinebench measures performance in a number of ways, including 3D rendering, software shading, and OpenGL shading with and without hardware acceleration. Cinema 4D’s renderer is multithreaded, so it takes advantage of Hyper-Threading, as you can see in the results.

Prescott is a little slower than Northwood in Cinema 4 rendering performance, but still enjoys a comfortable lead over the Athlon 64.

Prescott does especially well in Cinema 4D shading, while the Athlon 64 excels in the OpenGL shading tests.

SPECviewperf workstation graphics
SPECviewperf simulates the graphics loads generated by various professional design, modeling, and engineering applications.

The Pentium 4 ‘E’ is an absolute monster in workstation graphics, capturing the top spot in three of the six tests and tying for it in one more. In the other two, the Prescott 3.4GHz is second only to the Athlon 64 FX-53.

ScienceMark
I’d like to thank Alex Goodrich for his help working through a few bugs the 2.0 beta version of ScienceMark. Thanks to his diligent work, I was able to complete testing with this impressive new benchmark, which is optimized for SSE, SSE2, 3DNow! and is multithreaded, as well. Unfortunately, we don’t yet have a version of ScienceMark capable of taking advantage of SSE3’s new complex arithmetic instructions.

In the interest of full disclosure, I should mention that Tim Wilkens, one of the originators of ScienceMark, now works at AMD. However, Tim has sought to keep ScienceMark independent by diversifying the development team and by publishing much of the source code for the benchmarks at the ScienceMark website. We are sufficiently satisfied with his efforts, and impressed with the enhancements to the 2.0 beta revision of the application, to continue using ScienceMark in our testing.

The molecular dynamics simulation models “the thermodynamic behaviour of materials using their forces, velocities, and positions”, according to the ScienceMark documentation.

Primordia “calculates the Quantum Mechanical Hartree-Fock Orbitals for each electron in any element of the periodic table.” In our case, we used the default element, Argon.

The Pentium 4 struggles a bit in ScienceMark’s computational tests, and where the Pentium 4 struggles, the P4 Prescott struggles most. Intel has given SSE3 provisions for accelerating scientific computing, and I’m hopeful that once these are implemented in program code, the Prescott’s performance will improve in many such tasks.

These last two tests, SGEMM and DGEMM, measure matrix math performance using several different codepaths optimized with several instruction set extensions, including SSE, SSE2, and 3DNow!

Both of the P4 3.4GHz CPUs crunch through matrix math like mad. Northwood’s faster with single-precision math, but Prescott 3.4GHz beats out everything we tested in the double-precision DGEMM test.

picCOLOR image analysis
We thank Dr. Reinert Muller with the FIBUS Institute for pointing us toward his picCOLOR benchmark. This image analysis and processing tool is partially multithreaded, and it shows us the results of a number of simple image manipulation calculations.

Intel’s enhancements to Prescott have boosted the Pentium 4’s performance in our image processing test significantly.

Breaking things down into individual tests, we can see that Prescott shows strength across the board, but two tests, float and fixed interpolation, are where it gains the most ground on Northwood.

What about the heat?
As you may have heard, the Prescott 3.4GHz isn’t a cool-running chip. Intel estimates its key power load indicator, TDP, at 103W. Northwood, by contrast, is 89W at 3.4GHz. That’s quite a lot of power to dissipate. And TDP, or Thermal Design Power, isn’t even a peak number.

To get some sense of the real-world implications of Prescott’s voracious appetite for power and cooling, I used Abit’s EQ utility to monitor temperatures of the two processors, the Northwood and Prescott at 3.4GHz, under load. To generate that load, I ran two instances of Folding@Home’s command line client and let Hyper-Threading work its magic. Abit’s EQ utility reports monitoring data over time, and you can see how temperatures ramp up when the CPU is occupied.

I used the same cooler on both processors, a unit certified by Intel for use with Prescott. I tried to keep room temperatures reasonably steady during my tests, but this wasn’t exactly an ideal temperature-controlled test environment, so don’t take the following as gospel.

Under load, the Northwood P4 3.4GHz hit temperatures of 64 degrees C (148 degrees Fahreneheit). In the same conditions, the Prescott raced past Northwood’s peak temperature on its way to a steady peak of 78 degrees C (or 173 degrees Fahrenheit).

That, folks, is hot. And this was inside a computer case lying on its side with the side panel removed so the top was open. Things could get much warmer closed up inside a poorly designed case.

I’d have loved to do some overclocking with our 3.4GHz Prescott, but after seeing those heat numbers, I believe I’ll wait until I can pick up a beefier cooler to use with this thing.

Update: Many of you requested that I test the Athlon 64 3400+ under similar conditions and report a peak temperature for it, as well, so I did just that. I fired up the Athlon 64 3400+ test system with its MSI K8T Neo motherboard and installed MSI’s CoreCell monitoring utility. I then ran a pair of Folding@Home clients. No, the Athlon 64 doesn’t have Hyper-Threading, but I figured fair’s fair. AMD’s Cool’n’Quiet mode was disabled for all our testing, including this temperature test. I doubt this mode would have made any difference for temperatures given the load on the CPU. As with the Pentium 4s, I used a stock heatsink supplied by the CPU maker.

After running under load for quite some time, the Athon 64 3400+ reached a steady peak of 52 degrees C. I saw the indicator jump up to 53C once, but it generally stayed at 52. I tried cutting back to a single Folding@Home instance in case context switching was causing some overhead and not keeping the processor fully occupied, but it didn’t seem to have any effect on temps.

Again, my testing methods were far from perfect, so don’t take these numbers as holy writ, please.

Conclusions
Many of you reading this review are gamers, and I will make things simple for you guys. If you want to play games on your PC, the Athlon 64 is the best choice right now. The benchmarks are darn near unanimous on this point, so I see no reason to mince words. The Pentium 4 3.4GHz and 3.4’E’ GHz both list for $417, exactly the same price as the Athlon 64 3400+. Heck, I’d urge you to consider a lower end processor, like the Athlon 64 3000+, and spend the extra cash on a better graphics card instead.

If you’re into other things, well, the picture gets complicated quickly. The Pentium 4 performs nicely in commercial 3D rendering applications and media encoding. The Prescott, in particular, excels at video encoding, speech recognition, workstation-class graphics, and image manipulation. Prescott also offers the promise of improved performance in the future via software tuned to use SSE3.

However, if I had to choose between Northwood and Prescott at 3.4GHz for my own system, I’d be inclined to choose Northwood. The Northwood core runs cooler, and that’s no small thing. Performance results between the two cores are mixed, but Northwood seems to perform better in cases where Prescott is rather weak compared to the whole field (tests like Comanche 4, Cinebench rendering, and ScienceMark). Prescott may be a promising design for the future, when clock speeds ramp up and the heat problem is under control, but it’s not terribly attractive at present.

The future, of course, always figures prominently in the final analysis. Right now, both Intel and AMD are preparing new sockets for their processors, and all the chipset manufacturers are gearing up to deliver PCI Express, more Serial ATA, better audio, DDR2 memory, and a whole range of other enhancements to the PC platform. The graphics guys, too, are looking to make April a big month for new products. Not only that, but Intel has confessed that 64-bit extensions now lie dormant in Prescott, ready to be turned on in future versions of the Pentium. This fall, Microsoft will deliver a 64-bit version of Windows, and both AMD and Intel processors will run it.

You could buy a new PC right now and have a very solid, very fast system that would last you a couple of years, at least, without much hardship. But if I were you, I’d wait about 45 days or so before pulling the trigger on a new PC. I’d hate to miss out on all the new things coming.