Intel’s Pentium D 820 and Pentium 4 670 processors

Which is better? Well, that depends on a great many things. Let’s have a look at some of them.

Two new Pentiums roll out
In many ways, the two CPUs Intel is introducing today are fundamentally similar. Both are based on the latest version of the Pentium 4’s Netburst microarchitecture, both ride on an 800MHz front-side bus, and both are capable of 64-bit computing via Intel’s EM64T extensions. But beyond these wide areas of overlap are deep differences in approach.

The Pentium 4 670 is arguably the last of an old guard, a CPU intended to extract maximum performance out of a single processor core. The P4 670 extends the familiar Pentium 4 600 series one more speed grade, to 3.8GHz, and like the rest of the 600 series, the P4 670 packs 2MB of onboard cache to further improve performance. For the privilege of owning a CPU that runs at this dizzying speed, you’ll have to pay something close to Intel’s list price of $851.

The cores on the Pentium D 820 pulse along at a relatively leisurely pace of 2.8GHz, but there are two of ’em, so its overall performance in multithreaded applications or when multitasking may be superior to a single-core CPU. We’ve already reviewed the Extreme Edition of Intel’s dual-core desktop chip, code-named Smithfield. The Pentium D is a de-tuned version of the Smithfield core that’s had its Hyper-Threading capabilities disabled and its clock speed dialed back a few notches to 2.8GHz. In fact, its clock speed is sufficiently slow that Intel apparently saw no need for the 820 to have power management features like Enhanced Speedstep, the C1E halt state, or TM2 thermal throttling. Other recent Pentium desktop processors run at 2.8GHz when throttled back, and the Pentium D 820 is already there.

The real virtue of the Pentium D 820 isn’t just its dual processor cores, though; it’s the price. At only $241, the Pentium D 820 signals that Intel is dead serious about bringing dual-core CPUs to desktop PCs. In fact, the Pentium D’s price is low enough to shake up the whole CPU market. Its arrival presents consumers with a series of stark choices between single-core and dual-core processors at roughly comparable prices.

That fact, combined with the proliferation of very different model numbering systems from AMD and Intel, has made head-to-head competitive comparisons of CPUs quite a bit trickier than in the past. Freed from the constraints of model number-clock speed comparisons, Intel and AMD have priced their CPUs at points that don’t entirely correspond to one another. I’ve made an attempt, in the table below, to classify competing CPUs in a reasonably direct manner, but I may have only succeeded in illustrating the problem.

That’s official pricing on the Pentium D line, by the way.

You can see the trouble. Identifying a direct competitor for some of these models, like the Pentium D 830 or the Athlon 64 4000+, is downright befuddling. Fortunately, the two processors we’re reviewing here line up fairly directly against the competition. The Pentium 4 670 is priced pretty close to the Athlon 64 FX-55, and the Pentium D 820 just undercuts the Athlon 64 3500+ by $31. That means, of course, that the Pentium D 820’s two somewhat pokey cores will do battle with a relatively quick single-core competitor from AMD.

This is what I meant by stark choices. AMD doesn’t offer a low-end dual-core processor and apparently doesn’t plan to do so in the near future. For many single-threaded tasks, one of the Pentium D 820’s 2.8GHz cores should be sufficient, but it’s hardly quick by current standards—especially when you consider that the long pipeline of the Prescott/Smithfield cores means lower clock-for-clock performance. Intel also offers the Pentium 4 640 at the same basic price, but there’s no way a single-core Pentium 4 at 3.2GHz can match the Pentium D 820 overall. I’m getting ahead of myself, though. We’ll let the benchmarks tell that tale.

The Pentium D’s companion: Intel’s 945 Express chipsets
Intel’s dual-core chips won’t work without a new core-logic chipset, so naturally, the Pentium D is arriving with a couple of friends, the 945P and 945G Express chipsets. Yes, Intel has retained the dorky “Express” in the name of the chipsets to remind everyone that these puppies support PCI Express. Fortunately, they’ve also retained all the snazzy features of the 915/925 Express series, including dual channels of DDR2 memory, High Definition Audio, and all the rest. We recently looked at the high-end, enthusiast-oriented 955X chipset in our Pentium XE 840 review. The 945 series is a higher volume, lower priced version of the 955X with most of the same features. There are, however, two versions of the 945: the 945G with integrated graphics, and the 945P without.

Like the 955X, the 945 series can handle front-side bus speeds up to 1066MHz and DDR2 memory up to 667MHz, but the 945 is missing what Intel calls Memory Pipeline Technology. Under this fancy marketing name lies a rather commonplace reality: Intel is binning its north bridge chips, picking out the ones that most eagerly run with tighter timings in the memory controller and selling those as 955X chips. Everything else becomes a 945P. This segmentation of north bridges is familiar from the 915/925 series and from the 865/875 products before that. The 955X will perform a little bit better than the 945, and it will cost a little more. The primary speed difference between the two will come in the form of imperceptibly higher memory access latencies on the 945 series. We’ll quantify that difference in our benchmarks.

The 945 series’ ability to support DDR2 667 memory may be most beneficial to the performance of the 945G, whose built-in graphics core will have access more memory bandwidth. In the 945G, Intel has massaged the GMA 900 graphics core from the 915G and given it a new name, the GMA 950. The GMA 950 still has a four-pipe design, but clock speeds are up from 333MHz to 400MHz. These tweaks are sufficient for Intel to claim a 100% performance improvement over the GMA 900 in 3DMark05. The GMA 950 also adds the ability to connect to a media expansion card, useful for such things as adding a TV tuner to one’s system. (The 945P lists for $38 in quantities of 1000, and list price on the 945G in 1K quantities is $42. So the price of a GMA 950 is effective $4. That, folks, is how Intel got the be the world’s #1 graphics supplier in terms of volume.)

The other end of the chipset equation has changed, too, with the introduction of Intel’s ICH7 and ICH7R south bridge chips. The ICH7 series has two more PCI Express lanes than the ICH6, bringing the total up to six. Both ICH7 chips also support Serial ATA’s new 300MB/s transfer rates (or 3Gb/s, if you must), but only the ICH7R has Intel’s Matrix Storage. The Matrix Storage package includes RAID capabilities, new RAID levels 5 and 10, and support for the AHCI specification. Without AHCI, the non-R version of the ICH7 will be devoid of support for Native Command Queuing (NCQ) and hot-plugging of devices. For that reason, I’d expect even some mid-range 945P-based motherboards to use the ICH7R, as Intel’s D945GTP does.

Test notes
We have focused our testing today on the question of thread-level parallelism, in part because we believe that is the most important performance question one can explore in relation to multi-core processors. However, we are excited about the possibilities for better multitasking that may come with dual-core CPUs, and we’d be glad to take your suggestions for testing multitasking scenarios.

Also, we have included results for the Pentium D 840 in our testing, which we obtained by disabling Hyper-Threading on our Extreme Edition 840. Since the Pentium D 840 is just an Extreme Edition 840 sans HT, the numbers should be valid. Similarly, the Athlon 64 3500+ scores you’ll see in the following pages were obtained by underclocking an Athlon 64 3800+ (with the new “Venice” core) to 2.2GHz. The performance should be identical to a “real” 3500+.

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 systems were run with Hyper-Threading enabled, except where otherwise noted.

Thanks to Corsair for providing us with memory for our testing. Their products and support are both are far and away superior to generic, no-name memory.

Also, all of our test systems were powered by OCZ PowerStream power supply units. The PowerStream was one of our Editor’s Choice winners in our latest PSU round-up.

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:

• SiSoft Sandra 2005 SR1 10.50 64-bit
• ScienceMark 2.0 64-bit
• Compiled binary of C Linpack port from Ace’s Hardware
• POV-Ray for Windows 3.6 64-bit
• SMPOV 4.3
• 3ds max 7.0
• Cinebench 2003
• LAME MT 3.97a 64-bit
• Xmpeg 5.0.3 with DivX Video 5.21
• Windows Media Encoder 9
• Sphinx 3.3
• picCOLOR v4.0 build 545 64-bit
• DOOM 3 1.1 with trdelta1 demo
• Far Cry 1.3 with tr3-pier demo
• Unreal Tournament 2004 v3355 with trdemo1
• 3DMark05 v120

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

We’ve tested the Pentium D with the 945 chipset because that’s the platform this CPU will generally use. The rest of the Pentium processors were tested on the 955X chipset. As a result, these synthetic benchmarks give us a fix on the difference between the 945 and 955X chipsets: about 12 nanoseconds of access latency and 200-300 MB/s of memory bandwidth.

Gaming performance
Up next are some gaming tests, which will essentially serve to illustrate the futility of running a dual-core processor in a single-threaded application. Notice that we’ve included above each result a little graph generated by the Windows Task Manager as the benchmark ran on our dual Opteron 275 system (with four total CPU cores.) This should give you some indication of the amount of threading in the application. In some cases with single-threaded apps like the games below, the task will oscillate back and forth between one CPU and the next, but total utilization generally won’t go above 50% for a dual-core or 25% for a quad-core (or quad-front-end, in the case of the XE 840 with Hyper-Threading) system.

Doom 3
We tested performance by playing back a custom-recorded demo that should be fairly representative of most of the single-player gameplay in Doom 3.

Far Cry
Our Far Cry demo takes place on the Pier level, in one of those massive, open outdoor areas so common in this game. Vegetation is dense, and view distances can be very long.

Unreal Tournament 2004
Our UT2004 demo shows yours truly putting the smack down on some bots in an Onslaught game.

The stark choices I’ve been talking about are illustrated vividly here. The Pentium D 820 trails the pack in all three of these games. Sure, Doom 3’s frame rates are still crazy high and Far Cry isn’t bad, but UT2004 is looking a little shaky. I have expressed reservations in the past about playability in UT2004 with lower speed grades of the Prescott core, and those reservations were based on my experience with 3.2GHz processors that achieved higher average frame rates than the Pentium D 840 did here. Those seeking creamy multitasking smoothness from the Pentium D 820 may get it, but they’ll likely sacrifice some smoothness when playing single-threaded games.

Of course, a pokey performance here and there from the Pentium D 820 can be forgiven, given its price. The diamond-studded Pentium 4 670 has no excuse for getting shown up by the Athlon 64 3500+. Yes, the P4 670 is the fastest Pentium 4 for gaming, just edging out the P4 Extreme Edition 3.73GHz, but it’s still not very quick compared to the competition.

3DMark05

Any of these CPUs can pretty much drive a GeForce 6800 Ultra to its limits in 3DMark05’s default resolution, but the multithreaded 3DMark CPU tests are something else. They use the CPU as a software vertex shader, and multi-core CPUs can benefit from that arrangement. The Pentium D 820 nearly manages to match the much more expensive Pentium 4 670 as a result.

POV-Ray rendering
POV-Ray just recently made the move to 64-bit binaries, and thanks to the nifty SMPOV distributed rendering utility, we’ve been able to make it multithreaded, as well. SMPOV spins off any number of instances of the POV-Ray renderer, and it will bisect the scene in several different ways. For this scene, the best choice was to divide the screen up horizontally between the different threads, which provides a fairly even workload.

You might want to get used to seeing the Pentium D 820 outrun the Pentium 4 670 as we step through our rendering benchmarks. Thanks to Hyper-Threading, the Pentium 4 670 benefits from adding a second rendering thread, but not like a dual-core CPU does. The P4 670 trails its competitor, the Athlon 64 FX-55, by over 30 seconds of render time, while the Pentium D 820 trounces the Athlon 64 3500+. Note, though, that with only a single thread, the Pentium D 820 is mighty slow.

3dsmax 7 rendering
We tested 3ds max performance by rendering 20 frames of a sample scene at 320×240 resolution. This particular scene makes use of a motion-blur effect that requires extensive multi-pass rendering. We tried two different renderers: 3ds max’s default scanline renderer and its built-in version of the mental ray renderer.

Both of 3dsmax’s rendering engines produce results relatively similar to those we saw in POV-Ray. The Pentium D 820 outperforms AMD’s fastest single-core CPU, the Athlon 64 FX-55, while the P4 670 finishes in the lower part of the field.

Cinema 4D rendering
Cinema 4D’s rendering engine does a very nice job of distributing the load across multiple processors, as the Task Manager graph shows.

The P4 670 looks relatively stronger in Cinebench, but the Pentium D 820 still leads it.

Cinebench’s shading tests are single-threaded, and it shows. The Pentium 4 670 handles these tests well enough, but the Pentium D 820 is practically outclassed by the rest of the field, including the Athlon 64 3500+.

LAME audio encoding
LAME MT is, as you might have guessed, a multithreaded version of the LAME MP3 encoder. LAME MT was created as a demonstration of the benefits of multithreading specifically on a Hyper-Threaded CPU like the Pentium 4. You can even download a paper (in Word format) describing the programming effort.

Rather than run multiple parallel threads, LAME MT runs the MP3 encoder’s psycho-acoustic analysis function on a separate thread from the rest of the encoder using simple linear pipelining. That is, the psycho-acoustic analysis happens one frame ahead of everything else, and its results are buffered for later use by the second thread. The author notes, “In general, this approach is highly recommended, for it is exponentially harder to debug a parallel application than a linear one.”

We have results for two different 64-bit versions of LAME MT from different compilers, one from Microsoft and one from Intel, doing two different types of encoding, variable bit rate and constant bit rate. We are encoding a massive 10-minute, 6-second 101MB WAV file here, as we have done in our previous CPU reviews.

With but one thread, the Pentium D 820 chugs through the audio encoding process, but with two, it picks up some steam. The P4 670 benefits quite a bit from multithreading, as well.

Xmpeg/DivX video encoding
We used the Xmpeg/DivX combo to convert a DVD .VOB file of a movie trailer into DivX format. Like LAME MT, this application is only dual threaded.

Windows Media Encoder video encoding
We asked Windows Media Encoder to convert a gorgeous 1080-line WMV HD video clip into a 640×460 streaming format using the Windows Media Video 8 Advanced Profile codec.

Both of the new Intel processors outdo their single-core rivals from AMD when it comes to video encoding, and the Pentium D 820 makes a strong argument for favoring two slower cores over one faster one in Windows Media Encoder.

ScienceMark

We’re using the 64-bit beta version of ScienceMark for these tests, and several of its components are multithreaded. ScienceMark author Alexander Goodrich says this about the Molecular Dynamics simulation:

Molecular Dynamics is lightly multithreaded – one thread takes care of U/I aspects, and the other thread takes care of the computation. The computation itself is not multithreaded, though Tim and I were looking into ways of changing the algorithm to support multi-threading programming a couple years ago – it’s a lot of effort, unfortunately. When MD [is] running there [is] a total of 2 threads for the process.

Here are the results:

The Primordia test “calculates the Quantum Mechanical Hartree-Fock Orbitals for each electron in any element of the periodic table.” Alex says this about it:

Primordia is multithreaded. Two main tasks occur which allow this to happen. Essentially, we identified 2 parallel tasks that could be done. We could probably take this a step further and optimize it even more. There is an issue, however, with the Pentium Extreme Edition that we’ve identified. The second computation thread gets executed on the logical HT thread rather than the 2nd core, so performance isn’t as good as it could be. This will be fixed in the next revision. This doesn’t effect [sic] the regular Pentium D. A workaround could include disabling HT on Pentium EE. There are 3 threads for primordia – 2 threads for computation, 1 thread for U/I.

Although the Molecular Dynamics test is lightly multithreaded, the Pentium D 820 can’t overcome its lower clock speed by using its second CPU here. Primordia is another story, though. The Pentiums all perform well in the Primordia test, beating out all of the single-core Athlons.

SiSoft Sandra
Next up is SiSoft’s Sandra system diagnosis program, which includes a number of different benchmarks. The one of interest to us is the “multimedia” benchmark, intended to show off the benefits of “multimedia” extensions like MMX and SSE/2. According to SiSoft’s FAQ, the benchmark actually does a fractal computation:

This benchmark generates a picture (640×480) of the well-known Mandelbrot fractal, using 255 iterations for each data pixel, in 32 colours. It is a real-life benchmark rather than a synthetic benchmark, designed to show the improvements MMX/Enhanced, 3DNow!/Enhanced, SSE(2) bring to such an algorithm.

The benchmark is multi-threaded for up to 64 CPUs maximum on SMP systems. This works by interlacing, i.e. each thread computes the next column not being worked on by other threads. Sandra creates as many threads as there are CPUs in the system and assignes [sic] each thread to a different CPU.

We’re using the 64-bit port of Sandra. The “Integer x16” version of this test uses integer numbers to simulate floating-point math. The floating-point version of the benchmark takes advantage of SSE2 to process up to eight Mandelbrot iterations at once.

This is another place where the Intel processors look quite strong. Once more, though, the dual-core chips steal the show.

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. We use two different versions, built with two different compilers, in an attempt to ensure we’re getting the best possible performance. However, the versions of Sphinx we’re using are only single-threaded.

For once, the P4 Extreme Edition 3.73GHz reaps the benefits of its faster bus to nudge out the P4 670 for the top spot. The Pentium D 820 again grinds its way through a single-threaded app.

picCOLOR
picCOLOR was created by Dr. Reinert H. G. Müller of the FIBUS Institute. This isn’t Photoshop; picCOLOR’s image analysis capabilities can be used for scientific applications like particle flow analysis. Dr. Müller has supplied us with new revisions of his program for some time now, all the while optimizing picCOLOR for new advances in CPU technology, including MMX, SSE2, and Hyper-Threading. Naturally, he’s ported picCOLOR to 64 bits, so we can test performance with the x86-64 ISA.

At our request, Dr. Müller, the program’s author, added larger image sizes to this latest build of picCOLOR. We were concerned that the thread creation overhead on the tests rather small default image size would overshadow the benefits of threading. Dr. Müller has also made picCOLOR multithreading more extensive. Eight of the 12 functions in the test are now multithreaded.

Scores in picCOLOR, by the way, are indexed against a single-processor Pentium III 1GHz system, so that a score of 4.14 works out to 4.14 times the performance of the reference machine.

Dual-core wins again in this multithreaded test, with the Pentium D 820 easily surpassing the Athlon 64 3500+—and the Pentium 4 670, for good measure.

Power consumption
We measured the power consumption of our entire test systems, except for the monitor, at the wall outlet using a Watts Up PRO watt meter. The test rigs were all equipped with OCZ PowerStream 520W power supply units. The idle results were measured at the Windows desktop, and we used SMPOV and the 64-bit version of the POV-Ray renderer to load up the CPUs. In all cases, we asked SMPOV to use the same number of threads as there were CPU front ends in Task Manager—so four for the Pentium XE 840, two for the Opteron 175, and so on.

The graphs below have results for “power management” and “no power management.” That deserves some explanation. By “power management,” we mean SpeedStep or Cool’n’Quiet. (In the case of the Pentium 4 600-series processors and the Pentium D 840 and Pentium XE 840 CPUs, the C1E halt state is always active, even in the “no power management” tests.) Sadly, the beta BIOS for our Asus A8N-SLI Deluxe mobo wouldn’t support Cool’n’Quiet on the Athlon 64 X2 processors. AMD says all of its dual-core chips will support power management once the proper BIOS support becomes available.

The Pentium D 820 doesn’t support power management, as we’ve noted.

The AMD processors simply consume less power, both at idle and under load, than the Pentiums do. That’s been the case for some time now, although Intel has made progress with Speedstep and the like.

Interestingly enough, the Pentium D 820’s pair of 2.8GHz cores consume almost exactly as much power under load as the Pentium 4 670’s single 3.8GHz core. This is one of the main reasons why Intel is moving away from higher clock speeds towards multi-core CPUs. We’ve already seen that the Pentium D 820 can often outperform the P4 670 in multithreaded applications, though they share the same basic power needs.

I sure wish the Pentium D 820 supported lower multipliers for use with Speedstep, though. There’s no good reason why this CPU should consume as much power at idle as it does.

Conclusions
The Pentium 4 670 is, I think, a good example of why the single-core approach wasn’t working so well for Intel. Despite its high clock speed, the P4 670 struggles in some single-threaded applications, particularly games. The comparably priced Athlon 64 FX-55 is much faster, and so is the much cheaper Athlon 64 3500+. Multithreaded apps do take advantage of Hyper-Threading and make the P4 670 more competitive—perhaps on par with the Athlon 64 FX-55 overall when multiple threads are in use. Even so, a Pentium 4 670-based system consumes about 45 more watts under load than an FX-55 system. Intel has managed to tame the Prescott core’s heat and power needs somewhat through power management schemes and better manufacturing, but it’s still rather hungry when going full tilt.

The Pentium D 820 typifies Intel’s new approach, which looks very appealing given the numbers we’ve seen here today. In multithreaded applications, the Pentium D 820 races by the Athlon 64 3500+, which is a more expensive CPU. In fact, the Pentium D 820 frequently outperforms the Athlon 64 FX-55 and the Pentium 4 670, and our Pentium D system consumes no more power under load than our Pentium 4 670 rig.

Still, Pentium D 820’s performance does present some rather bold contrasts. It’s the slowest CPU in the pack whenever we throw a single-threaded test at it. The 820’s gaming performance especially raises some red flags for us, as we’ve noted. Eventually, games will most likely make the conversion to multithreading, but in the interim, I worry that the newest, most intensive game engines may not run terribly well on a Pentium D at 2.8GHz. Many games will work just fine, no doubt, but those that use lots of AI or physics may be a strain. Hard-core gamers will want to stay away, as will others who extensively use one single-threaded application at a time. The Athlon 64 3500+ is the better choice for them.

With that caveat noted, the Pentium D 820 is still a hellvua bargain at $241, and it’s not even the true sweet spot for the Pentium D line. I’d probably pay the $60 or so premium for the extra 200MHz per core offered by the Pentium D 830, personally, just to be sure that my single-threaded performance was snappy enough to satisfy.

I expect the Pentium D processor, teamed up with the 945G chipset, to dominate the mid-range PC market once folks discover its virtues. For everything from corporate desktops to boxes for power users, from video editing workstations to home theater PCs, the Pentium D looks tough to beat. (AMD’s Athlon 64 X2 is an amazing CPU, but with prices starting at over $500, it will be a high-end choice only.) Even casual gamers will want to take a long, hard look at the Pentium D. It fuses the creamy smoothness of true symmetric multiprocessing with the simplicity and affordability of commodity desktop PC components—a combo that’s awfully hard not to like. What’s more, the Pentium D, P4 670, and the 945 chipset should all be available starting today, according to Intel. Make mine twins, please.