Many users do not run an old system side-by-side with a new one. They also don't want to feel foolish for spending 2011 money to get 2004 desktop performance. So just ask Anandtech Bench! Note that the only Pentium 4 listed, the P4 660, is a fast early 2005 part. This P4 660 wipes the floor with Atom D510 in desktop (AKA single-threaded) applications.

It is simply a statement of fact that the Pentium 4s where faster then the current Atoms. They also used about twenty times the electric power an made twenty times the heat.

I'm not sure what the context of this discussion is, but are there any P4's that weight 3lbs and have battery lives of 6+ hours?

Not that I'm a fan of the Atom. But... just sayin'!

My 2 cents. I run the following 2 PC's at home and a Windows Home Server box. Each has a single Dell 2005FPW (1680x1050) monitor + bluetooth keyboard and mouse. The PC's are literally next to each other which makes comparisons easy. Both are running Kapersky Internet Security Suite 2010 and Windows Home Server client software.

Specs
1) A Shuttle SN95G5v3 bought in late 2005 - Athlon 64 3500+ (single core 2.2ghz undervolted, E3 stepping), 1GB DDR 2-2-2-8, 250GB Seagate 7200.7, XFX 6600GT. Running Windows XP Pro 32bit. The PC is pretty much the same as when it was built.
2) A Shuttle XS35GT bought in Jan 2011 - Atom D510 (dual core 1.66ghz + hyperthreading), 2GB DDR2, 500GB HDD, Nvidia ION2. Running Windows 7 Pro 32bit.

Subjective Conclusion
Both boot in about the same time. The XS35GT is smoother once software is loaded, but can occasionally lag when switching between heavy tasks (compile / video playback). The A64 is faster at a single task, but falls behind as soon as you try to do something else. Gaming performance is the same if not better on the ION2 system (Half Life 2, HoMM5, Blood Bowl, Dawn of War, etc). Neither PC will play 1080p flash videos but both will play a 1080p BR rip to a 1080p output. Neither is able to play and re size the 1080p content to the monitors native resolution without stuttering. The ION2 system should be capable of this but Shuttle refuses to update the drivers and the ones you can download from Nvidia wont install.

Subjective SPCR Conclusion
The SN95G5v3 was modded to be quiet (passive NB), undervolted, PS set to 7v. Its ok, but during gaming the GFX gets load. The XS35GT is completely passive, the machine is effectively silent (in middle of the night you might be able to hear a hard disk seek).

I agree that Atom is amazing in laptops. Under $400, over 5 hours of battery life at around 3 lbs.!

HFat and I were helping someone consider their desktop computer options. Atom on the desktop is of limited appeal, IMHO.

I agree with this opinion. I am typing this reply from a 2004 P4 and also own an Atom netbook.

While I love the portability and long battery life of the Atom, I hate how its so sluggish at times, I consider it to be strictly mobile office.
The P4 in comparison is very smooth and responsive, even when multitasking. I do have a 1GB ATI 1950 Pro AGP+S1 card in there so it can even play decent 3D games occasionally - something unthinkable on the Atom.
Its not really a gaming machine but a decent (power hungry + hot + 5x slow fans ) desktop.

I think Atoms still have their place outside laptops, with HTPC, Netops, Small Home Servers, Download stations, etc.

OK, so the Atom is not an option for a silent desktop system.
Which low-powered modern CPU is?
I'm in the market for one.

I currently have a Pentium-M system and am looking for more speed (notice I didn't say more power).

TIA

Regarding idle power consumption there is little to gain by using a low power cpu. If you want to limit power use at load you could take a regular i3 cpu and underclock/volt until you reach your target.

A core 2 duo based ulv cpu (see link below) is another alternative with no tweaking required. It`s not particularly fast being clocked at only 1.2ghz though it still runs circles around the atom.

http://www.zotacusa.com/zotac-ionitx-o-e-intel-pentium-su4100-1-3ghz-dual-core-mini-itx-intel-motherboard.html

The best way to get good performance out of an old rig would be to upgrade the RAM and to get a decent SSD. The CPU is secondary unless you have a particular need for decent CPU performance.

NFG said "silent", not low power consumption at idle.
So far as I know, no one here is running any Sandy Bridge without any fans or heatpipes designed to use the case as a huge heatsink. I'm sure it can be done but it's hard to say what case and heatsink might be required.
The SU4100 beats current Atoms by about a third. Hardly "running cycles". Even an underclocked dual-core Sandy Bridge woudn't be twice as fast as a dual-core desktop Atom.
The advantage of Atoms is that they're cheap and can be run without any fans out of the box. If you're willing to spend more and work more, you can have something more powerful without any fans... but not a whole lot more unless you spend a lot more and go with an exotic deisgn.

A desktop dual-core Atom would be an upgrade over a Pentium M as far as multi-threaded performance is concerned and would allow the use of modern RAM and drives. There are disadvantages however so it might be worthwhile to get a low-power AMD board with better features and single-threaded performance instead.

Get a cheap Sandy Bridge.... either with 4 real cores or at least with two real and two virtual cores. They all idle at around 4 watts. That's almost nothing. If power utilization is an issue... make sure to use an Intel motherboard.

Perhaps you are referring to a different thread, the post I replied to did mention low power.

As for cooling the SB cpu passively, if it ran at 1.2-1.6 ghz it`s likely that a big shuriken or similar cooler could do it. Of course most people don`t buy such cpus to run them at 1.2ghz which is why we don`t see many passive SB systems.



It could be a good, cost effective solution as long as flash video playback at higher resolutions is not a priority

Look for the word "silent". It's there.
Low power normally doesn't mean low idle power consumption but low maximum power consumption, for the same reason that maximum CPU power consumption (after any disabling, underclocking and/or undervolting done in BIOS) is the relevant variable for designing or choosing a cooling system.

Unless that Pentium M is paired with a newer passive discrete card, any high-res video would be a problem, not just Flash.
Does anyone know which drivers for low-power platforms (such as Ion and Zacate) have Flash acceleration support now by the way?

I disagree with you. For some people that is so. For others that is not so.

Though I have never bothered to try doing it, I have read that it is relatively simple to set maximum power consumption in Windows... so maximum power consumption is no longer a hardware only issue. If you are running windows you can cap it at whatever level you select to cap it.

That, it seems, would make idle power consumption and also amount of work done per watt, more important attributes of CPU performance than maximum power consumption... at least for those of us running windows. Wouldn't you agree?

I would just build arround a i3 2100T or Pentium G620T (if you dont benefit from hyperthreading and quicksync), you can still limit the multiplier on the bios i believe, or as CES also suggested through windows power management, either should be a nice fast low power consuming cpu, i would probably go for a big heatsink with low pwm fan like the Thermalright HR02 Macho or Scythe Mugen 3, and intel mobo for low power consumption + picoPSU, and if you can, and ssd like Samsung 470 which has one of the lowest power consumptions.

I don`t see anyone implying that max power use isn`t important, many of us mentioned undeclocking as a means to control that.



Flash video playback is more likely to be part of everyday use . As for hardware acceleration support, according to Adobe, all recent video adapters should support this, as long as video is encoded in h264. While I take those claims with a grain of salt, I believe that we`ll get there soon enough. There`s still a lot of flash content that doesn`t fall into this category though, flash adds, games and websites being an example. Until all flash content is handled by the gpu, a quick cpu is going to be necessary for a smooth browsing experience.

Really? What websites are you talking about?
I never noticed a slow CPU being a problem with Flash because I don't use Flash for high-res videos (or games for that matter). And I've used CPUs a good bit slower than Atoms (underclocked or not).
Large HTML webpages being displayed or even scrolled a little slow, that I've noticed when running a C2D at 600 Mhz for instance.

Hey HFat

I read that and thought it couldn't be right. I own and/or work with several Atom based systems and Sandy Bridge ones and the SBs are much, much faster even when under clocked. But before making any solid claims I went and tested myself.

Unfortunately I didn't have any SB machines I could run tests on (maybe on the weekend), but I did have an i870. Unless there is some really weird corner case, a SB should be faster per-clock than a i870.

I ran the tests on FreeBSD, with the unixbench program. FreeBSD can lock a CPU to any supported speed-step speed, so I was able to get the clocks as close as possible. Unixbench is completely single threaded, so the i870 being quad core gave it no benefit over the dual core Atom.

So, a 1.66 GHz Atom 510 VS a 1.60 GHz i870 per core:

Integer: 328.7 vs 923.1, 2.8X faster
Floating point: 143.4 vs 387.3, 2.7X faster

So the under clocked i870 is close to 3X the speed of the Atom. A SB under clocked to 1.6 GHz would be even faster.

To put it another way: to match the Atom I had to drop the i870 to approx 600 MHz. That's a serious under clock.

I acknowledge that this is a purely CPU benchmark and that the rest of the system components may make a big difference.

As a point of comparison more relevant to this thread, a 1.5 GHZ P4 Prescott (closest I could get to 1.66) is 371.4 / 151.8, or about 1.1X faster than the atom.

I can`t name a specific site right now, typically it`s sites loaded with flash ads or with a poorly engineered flash user interface. Granted it`s not the cpu`s fault, it`s bad design (I blame both web designers and adobe) but it can be a problem, depending on your browsing habits. Flash games can be even worse offenders, I`ve come across a version of solitaire that struggles to run smoothly even on my core i3.

edit: I can now remember a website that gave me trouble: http://www.evo.co.uk/ Right now the site should work smoothly, but a few weeks ago, there were some ads that really abused the cpu, a core i3! I can only imagine what would happen had I tried to load the site on an atom pc.

I don't have a Sandy Bridge to test anymore so I was using the published results of some kind of benchmark I found with Google. I had assumed 1.2Ghz instead of 1.6 for the SB underclock. With 1.6 and my numbers, the SB would be more than twice as fast, clock for clock. But the difference would still be smaller than yours.
It could be that your test is more representative of real-world performance. But there are two small issues with your test: I was thinking about the D525 which is more recent and the cheaper SBs don't have HT. Atoms do which help a bit with multi-threaded loads. On the other hand I think SBs are faster than your CPU, clock for clock.
These CPU numbers are to be taken with a grain of salt because every architecture has its advantages and disadvantages. Atoms in particular are really, really slow at some instructions and OK at others so performance is going to depend on which instructions are used and a synthetic benchmark can't capture that.

I couldn't quantify the difference between Atoms and Sandy Bridge by looking at my real-world usage because I haven't pushed a SB. And if you don't push the CPUs, there's effectively no difference. On the desktop, if you're not compiling, doing 3D or dealing with high-res media, the kind of task where you could see an unmistakable difference would be sorting a large table or something. And I haven't done that kind of stuff with a SB.
I haven't pushed an Atom for a while either but I remember Atoms being about 20% slower than a Pentium M and 40% slower than a K10 for single-threaded applications, clock for clock. Are SBs that much better? Could be.
At multi-threaded performance, Atoms are of course better than old single-core CPUs.

take a look at this Atom D510 vs i3 2100

http://www.anandtech.com/bench/Product/110?vs=289
hope the link still works.

While I dont have any experinance with underclocking I do have experiace with (currently working) Atom netbook, Pentium 4 3.0 and 2.4, a few Q6600/Q9550 based machines and a few i7 2600s.
It is my experiance that newer Atoms 1.6 are in the range of P4 2.4 performance even though the overall subjective feel of working on a Pentium4 is still better IMO.

It seem near pointless comparing Atom to Core family as the performance (from what i'v seen) is in the X6+ range. (Some multithreaded apps even show a X40 in real world advantages of i7 2600 over an Atom or P4 tested!) . Yeah if you lower the clock enough you might get to ~X3 like washu did, but why on earth would you want to do that? Using most BIOS you can limit TDP and power use of the SNB CPUs and still keep the clock at reasonable speeds.

I'll try a SB at 1.2 when I get the chance.

I do have a D525 system, but it runs windows and can't easily be brought down to benchmark. Subjectively it is no faster than the D510 systems I have. The only reason I have it is that it was the same price as a D510 board.

Of course, but I don't think any task going to make that big of a difference in the Atoms favor. I tried a different unix benchmark (ubench) and it gave a 4X improvement to the i870 over the Atom.

Of course. It should bring your 2.7x down to 2.5x for instance. Hardly something you can detect subjectively!
As I stated above, I even failed to subjectively detect the difference between a SB (not underclocked so about 3-4x faster than a D510) and an Atom because both were fast enough.

Well, you apparently just used two similar benchmarks which give quite different results. So why could such a difference only be in the Atom's disfavor?
I looked at a few measurements done by Anandtech with the old Atom (first thing I stubled on with Google) comparing it to a P4. Clock for clock, a single-core Atom (or a dual-core doing single-threaded work) was:
-8% slower than the P4 on a Java Script benchmark
-34% faster than the P4 at WinRAR compression (about 10 minutes for 300MB so this shouldn't have been dominated by I/O)
The first result is similar to your unixbench. But the second one is indeed in the Atom's favour. Of course a modern CPU would soundly beat both the Atom and the P4 so you might not detect the difference subjectively. It would however affect the margin: would the modern CPU be 1.8x faster or 2.6x faster for instance?
In the real-world whether you have a SSD or a HD, whether you have one or two cores, whether you have hardware-accelerated video decompression and even some minor factors like RAM performance typically make a larger difference than these nuances in CPU performance anyway.

A few days ago I was using a old Toshiba gaming laptop. It had a P4 with a 64 megabyte "geforce" video card, and it was noticeably faster then the netbook at general usage while being noticeably louder. Both systems had Microsoft Windows XP.



That 38% clock-for-clock Atom advantage turns into a 62% Atom disadvantage once we return to the real world and the real shipping clock speeds. (You forget that P4 was sold at approximately DOUBLE the Atom's clock speed.)

P4s were sold over many years at different speeds. I'm not forgetting but making the numbers more useful. Instead of cherry-picking a P4 which suits my argument, I'm telling you values which you can adjust for your P4's clock.

Atoms which are worth using for the desktop have two cores. In the measurements I'm talking about, the old dual-core Atoms beat the P4 in spite of its faster clock at every multi-threaded application from Photoshop to WinRAR. But it wasn't the fastest P4 and the fastest (not available in 2004) would have brute-forced their way to marginal victory.
The single-core Atoms are a different deal, competitive only with the slowest P4s.

Not sure what this is really about beyond the fact that Atom is slow in single threaded apps, especially ones that are slower with in order architectures. We've known this since they came out. Atom should improve once they go to out of order and also make some ipc improvements. Of course for any non-netbook system, a ulv sandy bridge is a multiple times faster choice, while still having amazing battery life. I'd care far more about ulv sandy vs p4.

The best current Atom is the D525 which runs at 1.8 GHz and is a dual core CPU with hyperthreading.
AMD's E-350 looks to be fairly close perormace wise too.
Having used Atom's I would say they don't offer the performance of a proper desktop even my old X2 4200 thumps an Atom heavily for any serious tasks. A fairly fast older single core CPU also feels more robust performance wise..and forget about comparing them to modern budget desktop processors they would lose heavily.

But..it has to be said that most users of these processors are not doing rendering, video encoding or anything remotely that taxing so they suit the target market quiet well. Over time we'll see both makers putting more effort into these types of processors so we should get a nice performance boost in the next year or so.

I'd pick an E-350 for a Zotac type PC..as for a mini desktop you'll have to look at the ITX boards and what they support. It can be done but the cost is likely to be a bit higher esp case wise.