Atom architecture: Tiny, silent PCs, here we come...
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Atom architecture: Tiny, silent PCs, here we come...
Here's Anand's review of the initial Atom processors/technology:
http://www.anandtech.com/cpuchipsets/in ... spx?i=3276
Looks very encouraging with more than enough performance already for basic computing and a dual-core version of the technology (Diamondville - to be released later this year) ought to be ideal for a small HTPC, especially as the Poulsbo chipset supports full video decode. Only problem is that the current Poulsbo apparently only supports display of 720p despite having the ability to decode 1080p. I expect an iteration of the chipset could solve this problem relatively easily.
Doesn't bode well for Via's future as a CPU producer, however.
P.S. As expected, the 3D and Video cores in Poulsbo are PowerVR SGX and VXD:
http://www.imgtec.com/News/Release/index.asp?NewsID=712
http://www.anandtech.com/cpuchipsets/in ... spx?i=3276
Looks very encouraging with more than enough performance already for basic computing and a dual-core version of the technology (Diamondville - to be released later this year) ought to be ideal for a small HTPC, especially as the Poulsbo chipset supports full video decode. Only problem is that the current Poulsbo apparently only supports display of 720p despite having the ability to decode 1080p. I expect an iteration of the chipset could solve this problem relatively easily.
Doesn't bode well for Via's future as a CPU producer, however.
P.S. As expected, the 3D and Video cores in Poulsbo are PowerVR SGX and VXD:
http://www.imgtec.com/News/Release/index.asp?NewsID=712
-
- Posts: 1608
- Joined: Tue Jan 04, 2005 4:02 pm
- Location: United States
Neat.What's even more impressive is the Atom test platform I saw running. Intel showed me an Atom test motherboard running Windows Vista and Unreal Tournament 2004 at smooth frame rates, without so much as a heatsink on the CPU. The surface of the CPU was around 35C and it did not even need a heatsink.
http://www.computerbase.de/news/hardwar ... verthorne/
108 seconds for 1M SuperPI, that is basically an athlonXP clocked at around 1GHz, maybe less. Maybe around a 800MHz clocked PentiumM from the first to third generation. This performance level is known due to speedstepping, so maybe we can compare it to that.
As you can see it performs under the celeron 900 used in the Asus Eee laptop, but is better than the P3 1,13GHz.
That performance for only a couple of watts: astonishing!
108 seconds for 1M SuperPI, that is basically an athlonXP clocked at around 1GHz, maybe less. Maybe around a 800MHz clocked PentiumM from the first to third generation. This performance level is known due to speedstepping, so maybe we can compare it to that.
As you can see it performs under the celeron 900 used in the Asus Eee laptop, but is better than the P3 1,13GHz.
That performance for only a couple of watts: astonishing!
Reading Anands review I got this idea that could be interesting for laptops, or silent systems with laptop chipsets:
Why don't solder one of these onto the mobo complimenting a 'normal' processor? That way when you were doing normal things like surfing and checking mail and so on the C2D could be in deep sleep and the atom could do the work and then when the system went to any real level of stress, like opening a new program or starting video-playback the 'real' processor could step in...
This could even be integrated on package or die level with the plans intel has for 'mixing' a few big cores with a lot of small simple cores. The big cores and most of the small cores could be in deep sleep most of the time and 'kick' in when needed.
I figure this would be much more efficient than the normal sleep states of a multi purpose processor. And an 'idle' powerload even down to the levels where the atom is doing a lot of work still sounds great to me.
Obviously this would require some special chipset technology since it would basically mean a multi-socked system, unless integrated on package.
What do you think?
Why don't solder one of these onto the mobo complimenting a 'normal' processor? That way when you were doing normal things like surfing and checking mail and so on the C2D could be in deep sleep and the atom could do the work and then when the system went to any real level of stress, like opening a new program or starting video-playback the 'real' processor could step in...
This could even be integrated on package or die level with the plans intel has for 'mixing' a few big cores with a lot of small simple cores. The big cores and most of the small cores could be in deep sleep most of the time and 'kick' in when needed.
I figure this would be much more efficient than the normal sleep states of a multi purpose processor. And an 'idle' powerload even down to the levels where the atom is doing a lot of work still sounds great to me.
Obviously this would require some special chipset technology since it would basically mean a multi-socked system, unless integrated on package.
What do you think?
-
- Patron of SPCR
- Posts: 744
- Joined: Tue Mar 04, 2008 4:05 am
- Location: London
- Contact:
-
- Posts: 286
- Joined: Sat Dec 09, 2006 9:45 pm
- Location: Montréal, Canada
Intel may be competing against itself but I can't see how it's relevant here. The atom is so tiny, and hence cheap to build, that in any case where they could sell an atom instead of a Core I don't see any reason why they shouldn't use those lower production costs to boost their profitability as prices doesn't have to go down as long as they're JUST competing with themselves. So canibalizing on the Core shouldn't really be a problem. And the core canibalizing on the atom should just be a matter of pricing.Spare Tire wrote:Now intel is so large that it's competing against itself. In turn, it can't allow Atom to become too powerful that it would eat out the Core market, or make the Core too efficient that it would eat the Atom market.
Then these two processsors are so different I don't really see them canibalizing each other. All other things equal (including processing technology) a core at about 47 million transistors will always be less powerful than a core at more than 400 million (the penryn), and obviously use less power.
A fascinating read (I'm always fascinated when someone explains CPU architecture in a semi-meaningful way ).
I'm glad that Intel changed its design philosophy to "+1% performance for +1% heat".
I'm glad that Intel changed its design philosophy to "+1% performance for +1% heat".
Even if they make desktop chips as efficient as an Atom (from my crude calculations, about 4 times more performance per watt), still desktops will have 20 times more watts to spend... or do you think the industry will suddenly decide to set the performance "back" a couple of years to have stock-passive CPUs?Spare Tire wrote:Now intel is so large that it's competing against itself. In turn, it can't allow Atom to become too powerful that it would eat out the Core market, or make the Core too efficient that it would eat the Atom market.
-
- Posts: 286
- Joined: Sat Dec 09, 2006 9:45 pm
- Location: Montréal, Canada
Well, my point is they might possibly have made it even better but they are not, because Atom and Core would be clashing against one another. Now the pricing might make it seem like a smooth transition between the top of Atom to the bottom of Core, but look at the production cost estimated at Tom's hardware. Somewhere between 4-6$. Might be indicative that Atom is quite a lot less sophisticated, than Core, yet top of the line of Atom is priced just under core, which is a big rip-off to us.
I'm waiting for Via Isaiah and see what they have to offer, since Via will be competing with both, their pricing should wedge in between atom and core and cost something fair for the performance.
I'm waiting for Via Isaiah and see what they have to offer, since Via will be competing with both, their pricing should wedge in between atom and core and cost something fair for the performance.
-
- Patron of SPCR
- Posts: 744
- Joined: Tue Mar 04, 2008 4:05 am
- Location: London
- Contact:
Same here. I've had my VIA 533mhz system for quite some time now and even with a pretty efficient linux build on it, it is a lot slower than I am comfortable with. In other words, I hate using it because it's so slow and as a result, it just sits in the corner of my closet, quietly humming away with its laptop hard drive and consuming just 9 watts total according to my Kill-A-Watt.ntavlas wrote:I think that what we need is greated availability of motherboards that can take notebook parts. They are still scarse and no that power efficient. I would also like to see more power supplies like the pico psu with matching power bricks.
As a development web server, it gets the job done and so far, there's not much else that fits in the mini-itx format and uses anywhere close to as little power. It seems the market for 10 watt computers that are no bigger than a dictionary is virtually non-existant. This latest CPU, especially with news of it working even with no heatsink, could finally be a replacement.
The Atom is designed to be more for the embedded market. The fact that the range will cover a range of speeds and power usage is due to the idea that it's scalable. So at the lower end the chip will be aimed at ARM processors that are used in most mobiles, opening up the world of x86 code to run. Scale through to the faster, higher power Atom processors and they can run set top boxes and smaller, low power PCs. It's a processor that can fill multiple markets, therefore if it only succeeds in one it's still not a failure. Embedded is where there's increasingly more money to be made - Intel want a good slice of the pie...
Atom + mini-itx + 3.5 inch HD + 5.25 dvd + 600 W psu = 40 watt with prime full load
http://www.vr-zone.com/articles/Intel_A ... /5721.html
They assume a efficiency of 80%, but looking at the 650W 80+ corsair review, the efficiency is more like 65%.
This would result in 26 Watt draw.
Abstract maybe 9-12 watts for the two desktop drives and you get around 14 - 17 watts for mobo + cpu + memory.
I know what I'm upgrading my 1 ghz via epia with. I just need GbE and 4 sata
http://www.vr-zone.com/articles/Intel_A ... /5721.html
They assume a efficiency of 80%, but looking at the 650W 80+ corsair review, the efficiency is more like 65%.
This would result in 26 Watt draw.
Abstract maybe 9-12 watts for the two desktop drives and you get around 14 - 17 watts for mobo + cpu + memory.
I know what I'm upgrading my 1 ghz via epia with. I just need GbE and 4 sata
Yeah. I know. But it does make it harder to interpret what the power draw in a sensibly configured system would be. Just have to wait for proper tests I guessMoogles wrote:They clearly just grabbed whatever was available off the shelf. It's not like they genuinely think they need 600 watts to power that configuration.