It would seem that the age of “good enough” computing is upon us. In many cases, what folks want most out of a computer isn’t bone-jarring speed or fluid, stunning visuals; it’s low power consumption (with the correspondingly long battery life and quiet operation), low cost, stylish design, and performance that, well, doesn’t get in the way. The system needs to be adequate to the tasks at handtypically things like web surfing, email, office productivity applications, and perhaps lightweight media playback. Beyond that, eh, why pay more?
The success of products like the iconic Eee PC and the coming wave of sub-$400 desktop computers have heralded this trend toward good-enough computing. Part of what’s driving that trend, of course, is consumer demand, both in traditional PC markets and in the developing world. But the efforts of a couple of PC chipmakers have also played a role in making this trend possible, as design teams for both Intel and Via subsidiary Centaur have been cooking up new PC-compatible CPU architectures aimed at low-power, low-cost systems. Intel, of course, made quite a splash with its introduction of the Atom processor, a teeny sliver of a chip intended to extend PC compatibility into much smaller devices, including GPS receivers and eventually smart phones. Meanwhile, Via introduced a sophisticated new CPU, dubbed Nano, that promises more than double the performance of its C7 processors inside of power envelopes as small as 5W.
Comparisons between the Atom and Nano were pretty much inevitable. Beyond the Lilliputianistic names and the shared emphasis on keeping costs and power use low, the chips were both designed in Austin, Texas and first introduced to the public just weeks apart. Yet if you ask Intel or Via about it, they’ll both tell you that the CPUs aren’t really natural rivals on all fronts. The Atom is a smaller, simpler chip that can go places the Nano can’t, thanks to much lower power envelopes for many Atom models. Meanwhile, the Nano looks to break new performance ground for what is essentially a commodity x86 processor.
One place where these two rivals will undoubtedly meet is in low-cost desktop systems, like Asus’ Eee Box. Here, Nano and Atom will square off in their fastest configurations, those least concerned about power use and most oriented toward performance. And, happily, this is the ground on which we have our first glimpse of both CPU architectures, in the form of the Atom 230 and Nano L2100 processors. These poor little chips have been gallantly laboring away under the weight of our semi-appropriate benchmark suite in Damage Labs for the past little while. Keep reading to see how they match up.
The Nano L2100
If you’re not familiar with the basics of the Nano’s microarchitecture, let me suggest you go read our article on the subject. One of the biggest things that distinguishes it from the Atom is the Nano’s use of both out-of-order and speculative execution, techniques commonly employed to extract more instruction-level parallelism out of code being executed. Via’s older processors, including the C7 and its forbears, kept to in-order execution, as does the Atom. As a result, the Nano promises higher clock-for-clock performance than any of them.
The Nano’s clock speeds aren’t bad, either, at least in desktop form. The L2100 chip we’re testing today is clocked at 1.8GHz, with an 800MHz front-side bus. With a 64KB L1 instruction cache, a 64KB L1 data cache, and a 1MB L2 data cache, the L2100 sounds every bit like a modern CPU. That impression is deepened by its support for many of the latest extensions to the x86 instruction set architecture, including SSE/2/3, Supplemental SSE3, and the x86-64 extensions for 64-bit computing.
Via estimates the Nano’s transistor count at 94 million. The chips are manufactured by Fujitsu on a 65nm fabrication process, and they come out to be 63.3 mm² in total die area. Nano models range in clock speed from 1GHz to 1.8GHz and in TDP (thermal design power, their maximum power and thermal envelopes) from 5W to 25W. The L2100 is at the top end of the spectrum, with a TDP of 25W. Still, Via claims an idle power draw of only 500 mW for the L2100, not much greater than the 100 mW rated idle draw of the rest of the Nano line, even though most L2100-based systems will likely spend their days plugged into a wall socket.
Nano processors aren’t available for purchase just yet, but one of their virtues when they arrive will be pin-compatibility with the C7, which should make for easy product upgrades for Via and its partners. Via illustrated this fact by supplying us with a pre-production Nano test setup that’s nothing more than a Nano L2100 soldered onto an EPIA SN motherboard, a product that’s currently selling with a C7 onboard.
The EPIA SN is a Mini-ITX-sized board that crams in a surprising amount of functionality, including dual DDR2 DIMM sockets, four SATA ports, one ATA port, a pair of Ethernet ports (one Fast, one Gig), multichannel audio outputs, a VGA port, and even a PCI Express x16 slot for graphics cards. You can’t see them in the picture above, but on the underside of the board are two more connectors: a Compact Flash slot and a mini-PCI expansion slot. Via likes to point out that such robust expansion options are possible on a Nano motherboard because it enforces no restrictions on component makers with regard to such things. By contrast, Intel has reportedly set limits on how Atom processors may be used in products; among those limitations is the prohibition both of PCI Express slots and of the inclusion of more than one DIMM slot.
The EPIA SN is powered by the core-logic combination of the CN896 north bridge and the now-venerable VT8251 south bridge chip. The CN896 includes an embedded Chrome9 HC graphics processor, based on GPU technology from Via subsidiary S3 Graphics. At present, Via is the only supplier of Nano chipsets, although rumors about a possible Nvidia chipset have been swirling for a while now.
Unfortunately, despite its relative maturity, we ran into several problems with the Via chipset on the EPIA SN. The SATA controller’s AHCI function isn’t really supported properly; Via’s latest Hyperion driver package lacks a Windows XP driver for AHCI, and installing the Hyperion drivers on Vista x86 with AHCI enabled caused our system not to boot. The SATA ports work well enough in IDE mode, but you’ll have to do without support for SATA device hot-swapping and Native Command Queuing. (We’ve seen similar problems with most of AMD’s recent chipsets, which isn’t an excuse, just an observation.) Even more annoyingly, the EPIA SN’s VGA port produces soft, smeary output that makes text difficult to read. Since the board lacks a digital video output like DVI, this is an especially unfortunate problem. This signal quality issue might be more of a motherboard-level problem than a chipset one, so we’re hopeful that it may not affect all CN896-based systems.
Since this is pre-release hardware, of course, some minor snags are probably inevitable. We were still somewhat disappointed to see that the EPIA SN board and BIOS didn’t appear to support the Nano’s dynamic clock speed and voltage scaling capabilities, a la Intel’s SpeedStep. Via assures us that all versions of the Nano, including the L2100, do include this capability, and I’d hope to see this problem ironed out on production hardware.
We ran into another problem while working with the Nano that doesn’t look to be the fault of the chipset or of Via’s hardware at all. As detailed in this Microsoft Knowledgebase article, the 64-bit versions of Windows Vista (and, we found, Windows XP Pro x64) won’t install on a Nano-based computer. The problem, it seems, is that the Windows installer is looking for a CPUID string that says either “GenuineIntel” or “AuthenticAMD”. When it runs into “CentaurHauls”, it keels over and dies in a splash of blue across the screen. Microsoft does have a hotfix available for Vista x64, but it’s tough to install a hotfix prior to installation of the OS. You’ve got to slipstream the hotfix into a custom Vista install image. Given limited time for the preparation of this article, we chose to test with Vista x86 instead.
Via says we can expect the first Nano-based products to hit the market “towards the end of this quarter.” Among the first wave of products will be netbook systems similar to the Eee PC (likely based on Via’s OpenBook reference design), larger laptops in the “thin and light” category, and small form factor desktops.
The Atom 230
We’ve already covered the Atom’s microarchitecture in some detail, so I suggest reading that article if you want to know more about it. The highlights include a simple design with in-order execution. The Atom design team specifically vetted potential performance-enhancing features for efficiency, leaving out some familiar enhancements along the way. One feature that did make the cut and would seem to fit nicely with the Atom’s in-order pipeline is simultaneous multithreading, better known by the Intel marketing name Hyper-Threading. A single Atom core supports two hardware threads, achieving some additional performance via parallelism at the thread level rather than the instruction level.
Like the Nano, the Atom 230 we’re reviewing supports a robust portion of the alphabet soup of extensions to the x86 ISA, including SSE/2/3/SSSE3 and EM64T, Intel’s version of x86-64. It doesn’t support SSE4, nor does the Nano. There has been confusion in some quarters, because some lower-power versions of the Atom lack 64-bit support, but the architecture is fully 64-bit capable. Our Atom 230 had no trouble with installing and running Windows Vista x64 Edition (although we used the x86 version for our comparative testing against the Nano). In fact, one of the reasons Intel chose to create a new chip rather than dusting off an older design was to ensure compatibility with the latest CPU features and extensions.
The 230 is currently the lone desktop variant of the Atom, clocked at 1.6GHz on a 533MHz front-side bus. The chip includes a 32KB L1 instruction cache, 24KB L1 data cache, and a 512KB L2 cache. Intel manufactures this processor on its high-k 45nm fab process, and the Atom’s roughly 47 million transistors are jammed into a die that’s only 24.2 mm²under half the size of the 65nm Nano. Despite having a similar clock speed, the Atom 230 has a TDP rating of just 4W, vastly less than the Via chip. Other Atom models range from 2.5W TDP all the way down to 0.65W, well below the Nano’s lowest power envelope.
Intel enforces the Atom 230’s desktop focus by disabling a key feature needed for mobile chips: SpeedStep, the combination dynamic of voltage and clock speed scaling capabilities that’s de rigueur even in most desktop-class processors these days. The loss of this feature isn’t likely to matter greatly to any desktop PC; the Atom 230 is still going to run very cool and quiet at idle compared to most CPUs. One benefit of the product segmentation is Intel’s willingness to sell the Atom 230 really stinkin’ cheap: they’re priced at only $29, whereas the 1.6GHz mobile variant costs $44. (Via hasn’t disclosed exact Nano pricing to us.)
Such incredibly low CPU prices enable board makers to offer creations like this one: the Intel D945GLCF. This mobo comes complete with a soldered-on Atom 230 processor for about 69 bucks.
No, really. 69 bucks. With built-in audio, 10/100 Ethernet, and video. And the video output is much crisper than on the VIA EPIA board. We are talking about a low-cost desktop here, folks. I’ve gotta admit, just seeing this combination of size and features for this price makes me feel old. I paid how much for my Amiga 3000? Don’t remind me.
As the name hints, this mobo is based on Intel’s 945 chipset, with a 945GC MCH and an ICH7 south bridge. These are the components of the Atom’s “Diamondville” platform, which will be used in both low-cost desktops and in netbooks like the Eee PC 901. (The versions of the Atom intended for smaller, lower-power devices will use the “Poulsbo” core-logic chip and will be imprinted with the Centrino Atom brand.) Note that, in the picture above, the larger heatsink with the fan onboard is used to cool the 945GC MCH chip, while the little, stubby heatsink next to it covers the Atom processor. This arrangement may tell you everything you need to know about the Atom’s current platform situation.
The D945GCLF seems like a compelling bargain, but its limitations are also obvious: the single DIMM slot, the lone PCI expansion slot, and neither an AGP nor a PCIe x16 graphics slot. As you might have guessed from the motherboard’s dimensions, Intel has embraced Via’s Mini-ITX form factor, and will apparently be limiting Atom desktop boards to this template. If this motherboard’s specs do indeed represent the practical ceiling for Atom desktop expandability, that would be a shame. I saw Atom-based systems with discrete graphics cards running recent PC games rather well during a tour of Intel’s qualification labs in Austin. Better things are possible, although perhaps Intel would prefer to counter the more expandable Nano-based systems with a different processor, maybe something from the Celeron range.
Side by side
Here’s a look at the Atom and Nano boards situated next to each other with their heatsinks removed. The Atom is on the left and the Nano’s on the right, with a quarter dropped in between them as a size reference. In both cases, the larger chip visible on the board, not far from the CPU, is the chipset’s north bridge and integrated graphics processor. The Atom is clearly a much smaller chip than the Nano, but the packaging for the two processors is almost the exact same size, interestingly enough.
In addition to the Nano and Atom processors, we’ve included a Pentium M in our comparison for the sake of reference. Right now, as they read these words, marketing types from both Intel and Via are hyperventilating into paper bags, as are small contingents of pedantic fanboys from each side. Our thinking was this: we needed a familiar x86 processor as a performance and power consumption reference, and a Pentium M at 2GHz seemed like the best candidate we had on hand. We tested the Pentium M 760 on a desktop-class motherboard, just as we did the Atom and Nano, though the board is a little older. We’ve also presented the Pentium M’s performance results in a nice “grayed out” checkerboard pattern on our graphs, so you can easily pretend they’re not there, if it really bugs you. The rest of us may find the architectural comparison rather intriguing.
Please do note, though, that the 915 chipset’s IGP wouldn’t support Vista’s Aero look, so we didn’t have that enabled on the Pentium M system, while it was enabled on the other two. I suspect that might impact performance in some of the WorldBench tests, perhaps.
Also, although this Atom motherboard purportedly is capable of running its memory at 667MHz, we had no luck getting it to work at that speed with our 2GB DIMM. The board wouldn’t POST if we set the speed manually to 667MHz, and the BIOS offers zero tweaking options for memory timings and voltage. Fortunately, the motherboard did choose some nice, tight timings for the memory at 533MHz, which is probably just as good as a clock speed bump for most intents, so we just ran with it. Not that we had much choice in the matter.
Our testing methods
As ever, we did our best to deliver clean benchmark numbers. Tests were run at least three times, and the results were averaged.
Our test systems were configured like so:
M 760 2.0GHz
|Chipset drivers||–||INF Update
DDR2 SDRAMat 667MHz
DDR2 SDRAMat 533MHz
DDR2 SDRAMat 400MHz
|CAS latency (CL)||5||4||4|
|RAS to CAS delay (tRCD)||5||4||4|
|RAS precharge (tRP)||5||4||4|
|Cycle time (tRAS)||15||12||18|
with VIA 220.127.116.110 drivers
with Realtek 18.104.22.16867 drivers
with Realtek 22.214.171.12467 drivers
Chrome9 HC IGP
with 126.96.36.199 drivers
with 188.8.131.521 drivers
with 184.108.40.20656 drivers
|Hard drive||WD Caviar SE16 320GB SATA|
|OS||Windows Vista Ultimate
Thanks to Corsair for providing us with memory for our testing. Their products and support are far and away superior to generic, no-name memory.
Our test systems were powered by OCZ GameXStream 700W power supply units. Thanks to OCZ for providing these units for our use in testing.
Also, the folks at NCIXUS.com hooked us up with a nice deal on the WD Caviar SE16 drives used in our test rigs. NCIX now sells to U.S. customers, so check them out.
The test systems’ Windows desktops were set at 1280×1024 in 32-bit color at an 85Hz screen refresh rate. Vertical refresh sync (vsync) was disabled.
We used the following versions of our test applications:
- SiSoft Sandra XII.SP2c
- CPU-Z 1.46
- WorldBench 6 beta 2
- Valve VRAD map build benchmark
- The Panorama Factory 5.2 m32 Edition
- Windows Media Encoder 9 x86 Edition
- LAME MT 3.97a 32-bit
The tests and methods we employ are usually publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.
Memory subsystem performance
We’ll begin with a look at the cache and memory hierarchies of these processors. This test measures throughput at increasing block sizes in order to probe the performance at each level of the memory hierarchy. I’ve plotted the results like so:
The Nano’s 64K L1 data cache appears to be not just larger, but also quite a bit faster than the Atom’s 24KB L1 counterpart, with well over 2X the bandwidth. Once we move into the L2 cache ranges for both chips, though, things nearly even out, with the Atom outperforming the Nano at the block sizes from 128KB to 512KBthat is, until it runs out of cache. The Nano’s faster at 1MB due to its larger L2 cache, and its real-world performance should benefit from the larger cache with programs that have working data sets larger than 512KB.
The story on main memory bandwidth is kind of tough to see in the graph above, so here’s a closer look at the 256MB block size along with a separate main-memory bandwidth test from Sandra.
Despite disparities in front-side bus speeds (800MHz vs. 533MHz) and memory clocks (667MHz vs. 533MHz), the Nano is only slightly faster than the Atom in two of the three memory bandwidth tests represented here. The Pentium M, with a 533MHz bus and single 400MHz DIMM, achieves higher bandwidth than either of the low-cost processors.
The Nano looks to be quite a bit quicker in accessing memory, at least at this block and stride size. Let’s have a look at a fuller picture of cache and memory access latencies. In the graphs below, I’ve colored the data points that correspond to L1 data caches yellow, while L2 cache is light orange and main memory is dark orange, just as a guide.
The initial latency sample we chose kind of put the Atom in a bad light, since it seems to have trouble at the 8MB block size and 512-byte stride. The adjacent points average out to around 104 ns of latency going to main memory. Still, the Nano’s quite a bit quicker to main memory overall, no doubt thanks to its faster bus and RAM clocks.
For the record, CPU geeks, the Atom’s L1 data cache latency appears to be three cycles, and its L2 latency averages out to about 16 cycles. The Nano’s L1 latency is four cycles, and its L2 latency is about 24.
All told, the Nano seems to have the edge here, with larger caches, a much higher-bandwidth L1 data cache, and quicker access to main memory than the Atom. These factors will have an influence on overall performance in real applications, but they are not really all that important on their own. Let’s move on to some true performance tests.
WorldBench’s overall score is a pretty decent indication of general-use performance for desktop computers, and this sort of measure is especially critical in this class of processor, because these are the kinds of tasks low-end desktops (and laptops) will be asked to do.
This benchmark uses scripting to step through a series of tasks in common Windows applications and then produces an overall score for comparison. WorldBench also records individual results for its component application tests, allowing us to compare performance in each. We’ll look at the overall score, and then we’ll show individual application results alongside the results from some of our own application tests.
The Nano beats the Atom pretty soundly in our first indication of relative performance. Interestingly, the Pentium M is quite a bit faster than either of the low-cost CPUs, although the Pentium M’s overall score is boosted somewhat by the fact that it was able to complete a 3ds max rendering test that the Atom and Nano didn’t finish. They both timed out, apparently because they took too long to finish the test.
If you’re curious about how these processors compare with the mainstream dual- and quad-core beasts you’d find in more traditional PC configurations these days, these results are roughly comparable to the those in our Phenom X3 review, where WorldBench scores ranged from 79 to 131. The comparison comes with some caveats, though, because for that review we used Vista x64, without Service Pack 1, with twice as much RAM on each system and a discrete graphics card.
Let’s have a closer look at the individual WorldBench tests, along with some other benchmarks of our own.
Productivity and general use software
MS Office productivity
Firefox web browsing
In some ways, we might be best served by stopping right here with our benchmark results. CPUs in this class will probably spend the vast majority of their time being asked to handle tasks like word processing, spreadsheets, email, and web surfing. In such applications, the Nano L2100 easily outperforms the Atom 230. Most strikingly, the Nano completes the Firefox test over seven minutes before the Atom does. The Office test involves running multiple applications simultaneously, and the Atom may fare relatively better here in part because its Hyper-Threading capability is lending a hand.
With that said, competence is more important in tasks like these than raw performance. I conducted my own casual, seat-of-the-pants evaluation by surfing around in IE7 on both of these systems, and neither one of them felt slow to me. Both were snappy in pulling up and rendering web pages, and neither one was appreciably slower than the high-end quad-core desktop system I use every day. Both are adequate, even if the Nano is measurably faster.
Multitasking – Firefox and Windows Media Encoder
WinZip file compression
As our tests grow more CPU-intensive, the Nano’s advantage remains considerable. These WinZip results have me questioning my decision to use NTFS file compression on the Eee PC 901.
Nero CD authoring
WorldBench’s Nero test relies extensively on the disk controller, and the lack of proper NCQ support in the Via chipset hurts the Nano here.
Vista’s Experience Index
Ok, at some point I probably swore never to do this, but faced with the job of evaluating performance on a couple of processors whose main goal is adequacy, not blazing speed, I caved. Here’s how each of our test systems scored in the components of the Windows Vista Experience Index. These are relatively simple tests, and the value of the results is fairly nebulous, but these numbers are here for comparison if you want ’em.
Vista Experience Index
The Nano leads the Atom in the CPU and memory tests by margins that seem reasonably consistent with our other test results. Interestingly, the Via chipset’s Chrome9 HC graphics processor scores lower than the Intel GMA 950 on the Atom system.
Even on a cheap desktop or “netbook,” folks are going to want to edit their pictures. Here, once again, the Nano L2100 finishes well ahead of the Atom 230. This matters, too, in my book, since some Photoshop filters take time to apply. The Nano will surely offer a better user experience for image editing.
The Panorama Factory photo stitching
The Panorama Factory handles an increasingly popular image processing task: joining together multiple images to create a wide-aspect panorama. This task can require lots of memory and can be computationally intensive, so The Panorama Factory is multithreaded. I asked it to join four pictures, each eight megapixels, into a glorious panorama of the interior of Damage Labs. The program’s timer function captures the amount of time needed to perform each stage of the panorama creation process. I’ve also added up the total operation time to give us an overall measure of performance.
The Atom 230 takes more than a minute longer than the Nano L2100 to stitch together our photo panoramaa clear win for Via. The flip side of the coin: You could nearly shave off another full minute by using a Pentium M 760, instead.
Multimedia encoding and editing
Multimedia extensions throughput
Media processing is one of the big reasons these brand-new CPU architectures exist. Intel and Via built all-new chips to incorporate support for the various incarnations of SSE in order to ensure a good user experience with video playback and the like. Before we get to our media tests proper, here’s a quick test of media processing power. Sandra’s “multimedia” benchmark is intended to show off the benefits of extensions like MMX, SSE, and SSE2. 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.
The integer 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 in parallel.
The fact that both processors perform so well here in comparison to the Pentium M is testament to the media optimizations both firms have put into their new architectures.
LAME MT audio encoding
LAME MT is 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 or the Atom. You can 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. That means this test won’t really use more than two CPU cores.
We have results for two versions of LAME MT from different compilers, one from Microsoft and one from Intel. We are encoding a massive 10-minute, 6-second 101MB WAV file here.
This test gives us a nice opportunity to see how the Atom’s Hyper-Threading feature impacts its performance, and here, the boost is considerable. In fact, without Hyper-Threading, the Atom takes roughly twice as long as the Nano to encode this audio file.
We should note, also, that even the slowest config tested here compresses our audio file at a rate about five times faster than real-time playback. For ripping CDs to MP3s, any of these processors should be reasonably capable.
Windows Media Encoder video encoding
For this test, I asked Windows Media Encoder to transcode a 153MB 1080-line widescreen video into a 720-line WMV using its built-in DVD/Hardware profile. Because the default “High definition quality audio” codec threw some errors in Windows Vista, I instead used the “Multichannel audio” codec. Both audio codecs have a variable bitrate peak of 192Kbps.
These CPUs are simply out of their league here. They all finished the job eventually, but the Atom took nearly an hour to encode our three-minute video clip. Today’s fastest CPUs can handle this encoding task in under seven minutes. Obviously, you’ll want to spend a little bit more on a CPU if you plan to use your computer for this kind of work.
Incidentally, since video playback is probably more important than encoding for these processors, I also tried simply playing back our 1080p WMV source video clip (it’s 1440×1080, 24 FPS, with 384Kbps audio). That job, also, proved to be too much for these new CPUs. The Nano shot up to 100% CPU utilization and played the clip with a smattering of obvious dropped frames throughout. The Atom’s CPU utilization stayed at around 85%, but it dropped many more frames than the Nano and had long, multi-second pauses where the video image remained static.
Which leads to me this remark: I wouldn’t trust the Task Manager CPU utilization readout on a CPU with Hyper-Threadingthe OS sees its two hardware threads as two CPU cores and doesn’t really know what’s happening beneath the covers.
Interestingly enough, the Pentium M was up to the task of playing back the 1080p video clip, with CPU utilization running about 90% and no apparent dropped frames.
Curious to see what these CPUs could handle, I tried playing the compressed 720p version of the video. Surprisingly, both the Nano and Atom struggled with it, as well. I then popped a standard DVD into the drive and tried playing that. On the Nano, CPU utilization ranged between 38 and 57%, with fluid playback. The Atom also had no trouble, with CPU utilization (perhaps deceptively) staying between 32 and 44%.
The HD video playback issues we ran into aren’t necessarily game-ending problems for these CPUs. They should both be able to handle more intensive video codecs and higher definition content using the combination of a GPU with video decode acceleration, the most popular standard video formats, and software smart enough to take advantage. I’d expect most PC makers to bundle their systems with such software, where possible. (At present, both Via and Intel claim some degree of MPEG2 decode assist in the IGPs we’re using here, but nothing for more advanced codecs.) Just don’t expect a single-core Atom or Nano to handle every HD video format flawlessly without assistance.
Oh, by the way, WorldBench has a Windows Media encoder test, as well.
…and the results are fairly similar to what we saw in our own test.
Roxio VideoWave Movie Creator
The Nano beats the Atom in Worldbench’s VideoWave test, as well, although not by quite as much.
3D modeling and rendering
Now we’ve ventured firmly into territory where low-power, low-cost CPUs aren’t really intended to go. The Nano and Atom systems couldn’t complete WorldBench’s 3ds max rendering test due to timeout error.
Neither CPU is all that quick in Cinebench, either. Even with Hyper-Threading and multiple render threads, the Atom can’t keep up with the Nano.
Valve VRAD map compilation
Next up is a test we picked up during a visit to Valve Software, the developers of the Half-Life games. This test processes a map from Half-Life 2 using Valve’s VRAD lighting tool. Valve uses VRAD to precompute lighting that goes into games like Half-Life 2. This isn’t a real-time process, and it doesn’t reflect the performance one would experience while playing a game. Instead, it shows how a faster CPU can speed up game development.
Valve’s VRAD is the same story. To give you a point of reference, even the slowest dual-core desktop processor we benchmarked in our last round of tests, the Athlon 64 X2 5600+, finished this job in 326 seconds.
Power consumption and efficiency
Now that we’ve had a look at performance in various applications, let’s bring power efficiency into the picture. Our Extech 380803 power meter has the ability to log data, so we can capture power use over a span of time. The meter reads power use at the wall socket, so it incorporates power use from the entire systemthe CPU, motherboard, memory, graphics solution, hard drives, and anything else plugged into the power supply unit. (We plugged the computer monitor into a separate outlet, though.)
We measured how each of our test systems used power across a set time period, during which time we ran our LAME MT multithreaded MP3 encoding test (using the executable from the Microsoft compiler).
Interesting! Let’s slice up these data in various ways in order to better understand them. We’ll start with a look at idle power, taken from the trailing edge of our test period, after all CPUs have completed the encoding task.
The overall platform power use of the Atom 230 and Nano L2100 are very similar to one another, with less than a watt separating them.
Next, we can look at peak power draw by taking an average from the 15-second span from 15 to 30 seconds into our test period, during which the processors were all busy.
The Atom and Nano are both true to their TDP ratings. The Nano system’s power use shoots up by over 20W when the CPU is busy, while the Atom system’s power draw increases by less than two watts. Intriguingly, the Pentium M 760’s peak power consumption turns out to be lower than the Nano L2100’s. Of course, the L2100 is aimed explicitly at desktops, but I still hadn’t entirely expected this outcome.
Another way to gauge power efficiency is to look at total energy use over our time span. This method takes into account power use both during the encoding work and during the idle time. We can express the result in terms of watt-seconds, also known as joules.
We can quantify efficiency even better by considering the amount of energy used to complete this task. Since the different systems completed the encode at different speeds, we’ve isolated the work period for each system. We’ve then computed the amount of energy used by each system to encode the file. This method should account for both power use and, to some degree, performance, because shorter encode times may lead to less energy consumption.
The Nano L2100 and Atom 230 take very different paths to completing the task with almost the same amount of energy consumed. The Atom takes quite a bit longer finishing, but keeps its power draw vastly lower as it works. The Nano consumes more power, yet finishes the work over a shorter period of time.
Looking at these results, one can’t help but think that the Atom could be an astoundingly power-efficient processor when coupled with a chipset and platform with a lower power use floor. Intel, of course, has such things in the works for other markets. In the same vein, we’re definitely at the ugly end of the clock frequency/voltage curve with the Nano L2100. The Nano U2400, which runs at 1.3GHz and has an 8W TDP, ought to offer much better performance per joule.
As it stands, though, the Pentium M 760, an older chip manufactured on a 90nm fab process, used markedly less energy to encode an MP3 than either of the low-cost platforms we’re testing todaya testament to the remarkable energy-efficient performance of the Dothan Pentium M design.
True to their billings, the Atom 230 and Nano L2100 showed themselves to be very different animals through the course of our testing. The Atom 230 combines admirably low peak power draw with largely adequate performance for basic productivity tasks, although it’s a little plodding when asked to handle more CPU-intensive chores like photo editing. The Nano L2100, meanwhile, breaks new performance ground in this class of processora very welcome development that should make the more ambitious forms of low-cost computing tolerable. Those gains do come at the expense of about 20W higher power draw under load than the Atom.
Thus, the Atom 230’s biggest advantage might be its ability to slip into smaller, thinner enclosures than the Nano L2100. Given that our Nano-based test system only pulled 63.6W peak at the wall socket, the additional power draw should be of little concern for the low-cost desktop PCs at which both of these products are targeted. As a result, the L2100’s superior performance wins the day: the Nano is simply better suited for low-cost desktops.
That verdict is accentuated by the possibility that PC makers could offer Nano-based systems with discrete GPUs and decent expandability options. I have to admit, I haven’t yet plugged a graphics card into the EPIA SN’s PCIe x16 slot and tried gaming with the Nano, but I’m intrigued by the prospect of an Econobox-style system with prices in the sub-$400 range that’s actually powerful enough to play the latest games.
My enthusiasm for such possibilities, though, is tempered by some cold realities: although the C7 and its forebears have been around for years, products based on Via CPUs have been hard to come by in North America. We’ve seen the occasional success, like the C7-based HP 2133 Mini-Note, but those are the exception rather than the rule. We’ve not even heard any rumblings of a Nano-based Mini-Note from HP, in spite of the fact that it would require only a CPU swap. The EPIA SN board we tested here is selling at a few places online now (with a C7 processor onboard), but it costs around 300 bucks, because it’s a niche product intended for embedded systems. Meanwhile, this Atom motherboard is selling today for $69 (Did I mention that? 69 bucks!), and Atom CPUs are already populating Asus’ lineup of Eee PC netbooks and cheap desktops. Via has to win more of these product battles in order to be relevant. The Nano seems up to the task, if it gets a chance.