VIA Nano CPU Benchmarked, Beats Intel Atom 279
Vigile writes "Back in May, when the Isaiah architecture was first disclosed, VIA declared a performance victory over Intel's upcoming Silverthorne technology. Since then, Isaiah has become the VIA Nano processor, and Silverthorne changed to the Intel Atom — and now we can finally see tests comparing the two technologies. The Nano's out-of-order super-scalar design is definitely an architectural leap over the Atom's in-order single-issue design, but with Intel including HyperThreading technology in their CPU the competition is closer than expected. The Nano does win the performance tests by a considerable margin, but what might be more impressive is seeing the Atom use only 4 watts of power under full load!" As reader Mierdaan points out, that's 4 watts more than at idle, for about 60 watts total.
Almost had me going there... (Score:5, Informative)
Uhmmm... measurements? (Score:3, Informative)
It seems they may be measuring the whole system load in comparing the efficiency of the processors, which is more than a little unfair. What sticks out more, though are numbers like "63,434 watts". Uhmm... no? Besides being a clearly invalid measurement, it should probably be expressed at watt-hours. No way either machine drew 63 kilowatt hours either.
TFA is broken.
Re:Misleading title? (Score:5, Informative)
Well yes and no. Did you read the part where they measured total power expended to accomplish various tasks? Burning more power is OK if you can finish the job faster and get back into a low power state. When that is factored in the contest is a bit closer. Of course if the Nano does run for long it is going to bake your lap more than the Atom and drain the battery a lot faster.
Looking at the photos makes it plain where the problem now lies, the northbridge. If Intel can get theirs under control they will totally dominate the low power business. But since the Nano draws so much more power a low power northbridge won't help them as much, which bodes ill for the future. Intel has a lot more room for improvement while Via would have to pull a major rabbit out of their hat to cut much off their current power consumption numbers.
And for small computers that aren't running on batteries but do need to be fairly cool (i.e. quiet) the Nano will be the hands down winner just because Intel is playing marketing games.
Re:The chart shows 60 watts (Score:5, Informative)
They get the 4 watts from observing that when idle the system is at 56 watts, but when at a full load it meanders over to the 60.1 watt range.
60-56 = 4
Re:Misleading title? (Score:5, Informative)
More benchmarks and analysis here - HotHardware (Score:5, Informative)
Another review at arstechnica.com... (Score:5, Informative)
...with the same findings.
http://arstechnica.com/reviews/hardware/atom-nano-review.ars [arstechnica.com]
Performance for full day battery life (Score:3, Informative)
The Atom is a bit on the slow side. Like using ARM chips for desktop computing; so why not simply use an ARM chip?
Still, if the Atom's paired with a super low powered chipset we might just finally have computers with more than 8 hours of battery life (while still being affordable/portable/small). Imagine taking your computer to work, and then leaving it on all day. A small detail, but makes a big difference.
The Nano is faster, but it also use about 8 watts more power (according to HardOCP [hardocp.com]). Those 8 watts is a big deal when it comes to battery life, but OTOH Atom is quite a bit faster than even the fastest Atom. The difference being big enough that HardOCP stated that Vista on Nano was notably more resposive - notable enough to be picked up on in blind tests.
So perhaps Atom trades off too much performace...
Re:Interesting. (Score:4, Informative)
The 4 watts comment (Score:4, Informative)
For those of you interested, the Atom CPU really DOES use just about 4 watts at load. The 60 watts number is for the entire system including power supply, motherboard, DVD-ROM, hard drive, etc. Idle power on both of these parts is measured in milli-watts so you can see how much power each uses under load by looking at the power consumption graphs on page 8:
http://www.pcper.com/article.php?aid=597&type=expert&pid=8 [pcper.com]
The gap between rest (on the far left) and load (middle) is much greater for the VIA Nano processor than the Intel Atom - in fact you can barely tell the Atom processor has changed wattage at all.
Re:The problem (Score:2, Informative)
> while the atom might be 20 watts unloaded and 24 watts loaded
Just about right, the i945 Northbridge has a TDP of 22W, which kills the Atom's power efficiency.
Re:x86-64? (Score:4, Informative)
x86-64 and EM64T are the same instruction set.
explaination of energy efficiency (Score:5, Informative)
The article writers don't seem to be very technical guys.
If we look at the energy efficiency, Atom hands Via it's bilblical arse.
Let's talk about Joules, the things a battery stores. Batteries are important. Especially for the target marget of these two CPUs. In fact, battery life is most likely THE most important factor in anything below a notebook.
Keep in mind a battery only has so many joules between charges, that's obvious, I know.
Now, an efficient architecture would only use as many joules as needed to get the job done in a timely manner. Joules per seconds are Watts, btw.
So lets look at how these two stack up in terms of Joule consumption and Performance based on this data...
The VIA requires about 17W of power to chug through MP3 encode, for about 460 seconds. That means the power supply had to deliver 17 * 460 = 7,820 joules.
Now the Atom crawled along 30% slower, about 600 seconds to complete. But it only needed a delivery rate of 4 J/s, so it ate 2,400 joules.
So for a 30% improvement in performance, VIA had to gobble down MORE THAN THREE TIMES the energy!
That means you could encode 3x as many MP3s on an atom, but it will take 30% longer. Imagine if this was an iPod. Who would trade 3x less battery life for such a tiny bump? That isn't something to brag about when you are targeting a market starving for battery life.
I'll be really surprised if Via goes anywhere other than a few cheap Asian design wins.
Re:More benchmarks and analysis here - HotHardware (Score:2, Informative)
The Vantage AES test doesn't seem to use the AES instructions on the Nano.
That's like having a 3D Graphics test not use a 3D Graphics API. Pretty worthless. If you're a geek and you buy a CPU with padlock, you are going to use padlock-aware encryption an hashing libraries/applications. I know I do.
It's soooo frustrating seeing review after review missing this.
Target's Linux laptops (Score:3, Informative)
Target is getting serious about Linux laptops. Looking under "Laptops", the first screen of "Featured items" has 3 colors of the EE PC, Linux version, and some HP laptop. The XP version of the EE PC costs $100 more.
The biggest problem with the Atom... (Score:3, Informative)
... is the chipset it's paired with. I recently bought an Intel motherboard with an Atom on it. Whilst the CPU is only 4-ish watts, the board draws around 40-50 watts. That's the board, not optical or hard drives.
That northbridge, with the non-power-optimized video card and memory controller, sucks up the juice. The heatsink on the northbridge is 4x larger than the one on the CPU. Furthermore, the heatsink on the northbridge has a fan, where the HS on the CPU has none.
Re:It's a tie (Score:5, Informative)
I'd say it's rather far from a tie when you consider more than simply power consuption.
As the story points out Intel is restraining board designers from using desirable technology on the Atom platform. No PCI Express, no DVI, no second memory slot. Theory is "Intel appears to fear Atom will cannibalize its Celeron sales". Perhaps. I'll bet VIA is more than willing to cannibalize those sales if Intel is going to let them.
The reference board in this review is nice. There are two ethernet phys, one of which must be gigabit [via.com.tw]. Compact Flash, mini-PCI and PCI Express. Damn. I like that board. That is the perfect board for the small, quiet home server.
Re:Almost had me going there... (Score:2, Informative)
They don't break out the CPU power usage - those figures are for the **entire computer**.
Idle Nano=59.2 Atom=56.4
Load Nano=77.5 Atom=60.1
If we assume that the bulk of the load vs idle power difference is due to CPU power usage, then we have the Atom using approx. 4W more under load, and the Nano using 18W more.
Whatever it's total power draw, the Atom is evidently much more miserly.
Re:The 4 watts comment (Score:5, Informative)
Eliminating the hard drive and optical drive, the Atom board still draws 40-50 watts. On my board, I observe a 55-65 watt draw from the wall with just the board, which taking PSU losses into account, is about right.
The part that it entirely irking is that that board alone draws more than my entire laptop, which includes a Core Duo, hard drive and a *display*, for crying out loud. Not such a great way to show off a low-power CPU.
Re:Another review at arstechnica.com... (Score:3, Informative)
Re:Misleading title? (Score:3, Informative)
I hate to actually read the article and then post, but the article makes it clear that the VIA Nano uses less power to perform the benchmark tests than the Intel chip, by taking slightly more power and finishing much faster. Running with 10% less wattage and taking 30% longer to complete is no savings.
[QUOTE]
For our MP3 encoding test, the VIA Nano processor used a total of 37,323 watt-seconds (Joules) of energy while the Intel Atom processor used 38,290 watt-seconds (Joules) of energy.
[/QUOTE]
Re:explaination of energy efficiency (Score:4, Informative)
Re:It's a tie (Score:3, Informative)
The Intel board could be a lot better if they used mobile chipsets. I have an old Shuttle desktop system based on Intel mobile hardware (915GM chipset.) With a 2.13 GHz Pentium M, the system idled around 35w and peaked just over 51w.
Re:explaination of energy efficiency (Score:1, Informative)
For our MP3 encoding test, the VIA Nano processor used a total of 37,323 watt-seconds (Joules) of energy while the Intel Atom processor used 38,290 watt-seconds (Joules) of energy. That is a difference of just 2.5% indicating that even though the Atom processor is slower, it's not that much less efficient than VIA's Nano.
And later on the same page:
Using the same method to gauge the results of our CineBench 10 test, we find that the VIA Nano used 63,434 watt-seconds (Joules) of energy to render the scene while the Intel Atom used 65,893 watt-seconds (Joules) of energy - an advantage of 3.8% to the VIA CPU.
Re:explaination of energy efficiency (Score:2, Informative)
For our MP3 encoding test, the VIA Nano processor used a total of 37,323 watt-seconds (Joules) of energy while the Intel Atom processor used 38,290 watt-seconds (Joules) of energy. That is a difference of just 2.5% indicating that even though the Atom processor is slower, it's not that much less efficient than VIA's Nano.
Via's Nano may use more power under load, but it finished fast enough to actually save energy over the Atom.
Re:explaination of energy efficiency (Score:2, Informative)
Your measurements are correct if you look at the CPU and nothing else.
However, computers aren't made of just CPUs (at least not yet), so you must factor in the entire load of the computers, including the CPU, to get an accurate measure of power usage.
When you do that, guess which one wins? That's right, the VIA does, with about 3k fewer joules on an MP3 encode. AND it did it 30% faster.
Which would you choose? A slower machine that uses insignificantly more battery power? Or a faster machine that uses insignificantly less?
The time when the Atom will be a clear winner is when system power usage at idle drops to 25w or so. At that point the Atom, with its 4 extra watts, will only be using roughly 20% more power under full load, while the VIA, with its 17 extra watts, will be using around 80% more power. The tradeoff of 30% less power for 60-70% more battery life would be much easier to make than the current 30% less power for +-1% battery life.
So, while it's a step in the right direction, if it requires a power-hungry chipset then the advantage of the Atom completely disappears.
Comment removed (Score:4, Informative)
Re:x86-64? (Score:3, Informative)
Nope. Green and red apples would be more like it.
For most intents and purposes, AMD64 and EM64T is the same instruction set. You are probably thinking of IA64 (Itanium).
And for the SI-challenged among us... (Score:3, Informative)
Re:Misleading title? (Score:4, Informative)
That Raptor is not quite that bad - 9W at idle-but-spinning and 10W at read/write. Still, that's 7-8W more than a decent laptop drive.
Not to mention that the i945G chipset in the Atom PC is something like 20+W. The Nano motherboard idle power seems a bit high too.
If the goal of the benchmark was to test the limits of performance per watt on the two platforms, the choices they made seems silly. All the benchmark really shows is that it makes no sense to put an efficient engine in a heavy suv.
Re:In-order hyperthreading? (Score:2, Informative)
I believe you meant to say *simultaneous* multithreading. You probably confused the name with *symmetric* multiprocessing (SMP, the fancy name for multiple identical CPUs). And it's still generically called SMT, Hyperthreading is just Intel's brand for it.
I have no specific knowledge to back this up, but I'd bet that they're doing something similar to the Sun Niagara (aka UltraSPARC T1). The cores in the Niagara are single-issue in-order. However, the hardware supports multiple threads, so it can pick any of them to issue an instruction on any particular cycle - if there's a data dependency or cache miss, it can fill what would otherwise be a pipeline bubble with an instruction from a different thread.
Basically, the Atom can use hyperthreading to keep the pipeline full instead of out-of-order execution. Out-of-order execution, simultaneous multithreading, and superscalar design are all independent design features.
One is here, one is not (Score:2, Informative)
Another worthwhile observation: You can get the Atom motherboard. They're $83 delivered from Newegg. Via has a habit of announcing product and stalling. The are selling this cute little devil [mini-box.com] though. It's called the "pico-ITX" form factor.
The reviewed item though? Not yet. Until it's available we're comparing what's on the shelf to what's not. Of course the one that comes later is targeted at a slightly higher performance level. It seems likely that when it ships Intel could reply with a machine that has dual Dimms, PCIe X16, dual gigabit nics, and then up the ante with a little DVI goodness and dual core. Everybody knows the chipset supports those features already.
Maybe they'll also do a process shrink on the MCH as well to get the power down.
Personally I'll be getting both. I've played with the Atom board and I like it. These are both good enough boards for some purposes I have in mind and if more better stuff comes along later, well, that's just the way of IT isn't it? If you waited until nothing better was going to come out you might as well go back to pencil and paper.