Fusion die-shot - 2009 Analyst Day
- Thread starter AnarchX
- Start date
Sub 1W capable was what AMD mentioned in a slide last year. Another slide said scalability
1 to 10W. This was for one single Bobcat-core, not a Ontario chip.
On another note: These slides seems to be about embedded solutions, not about the chip
average Joe will find in his netbook next year. Keeping it in production for 6-8 years is a rather
long time, so in my opinion this is "Geode" territory.
Edit: Point 4 on the first slide is saying that the chips with a 9W TDP has to use
1,35V@1066 memories to stay under 9W.
The question then is, do they give out TDP inclusive the memory, or is it that the higher
voltage/faster memory will make the memory controller pull enough juice to move the
TDP over 9W?
Last edited by a moderator:
Those TDP numbers don't tell much about how they will compare to Atom. E.g. I doubt the 18W part really consumes twice what the 9W part does. We'll need idle consumption, chipset wattage and real measured numbers to say anything certain.
But the numbers are in line with what was expected, right?
But the numbers are in line with what was expected, right?
Indeed, idle power is just as important as TDP, if not more. In the end, that is what determines battery life most of all.
caveman-jim
Regular
sub 1W meant they could get a core running and use less than 1W of power, not that it was intended for production
itsmydamnation
Veteran
what do we really know about Bobcats architecture to determine IPC? we know nothing about the stuff that matters, how clever the branch predictors, prefetch, it can decode and retire two instruction per clock so depending on the scheduler/alu's there is a real chance of higher IPC vs stars. But you then have to add in the overhead of using pointers and clocking up of required cache lines/predictors/etc.
reading the published slides there is alot on power reduction
http://www.anandtech.com/Gallery/Album/753#1
Also i know people are a little caught up on clock speed because you know the 3.2ghz precossots where so good compared to a 2.X conroe.
according to AMD slides bobcat has both clock gating and power gating.
then you have the salty stuff like
http://citavia.blog.de/2010/07/01/some-updates-regarding-ontario-boinc-and-bobcat-8899248/
cheers
Last edited by a moderator:
Blazkowicz
Legend
20% higher volage means 44% higher consumption for instance, so with voltage and clocks I don't doubt the 18W part can use twice the power.
I think these initial SKUs reflect the uncertainty AMD has wrt. to schmoo spread. The TDP numbers are likely set conservatively to ensure that AMD doesn't end up with piles of useless chips.
Down the line, we'll see more bins with better speed/power ratio and commanding a price premium, the same way as we've seen with CPUs (AMD's HE and EE models, Intel's CULV)
Cheers
Down the line, we'll see more bins with better speed/power ratio and commanding a price premium, the same way as we've seen with CPUs (AMD's HE and EE models, Intel's CULV)
Cheers
Well there's a lot we don't know about Bobcat, but we do know that there are only two ALUs, of which only one can do MULs, that the FPUs are only 64-bit wide, that one of the two AGUs is dedicated to stores while the other is dedicated to loads, and that it has a rather long pipeline. And we do know that it's meant to be a small, low-power design.
So I'd be very (pleasantly) surprised if it turned out to be faster than Stars, clock for clock.
The TDPs paint Ontario into a higher-power niche than what Atom can reach.
The lowest TDP Ontario is an 800 MHz single core with GPU off.
Atom can manage 1.6 GHz with a GPU, at a .5W penalty over the lowest Ontario.
Ontario may be good, but a 50% clock deficit is a lot, and that's with no GPU at all.
The top-end Ontario chips have a significantly higher TDP than the nearest Atoms. While it is likely that they will outperform Atom, the TDP difference is potentially more signficant for embedded platforms.
Perhaps Ontario could have been engineered to offer a higher clock rate at 5.5W, though the argument would be that its density figures would be much closer to Atom if its frequency/voltage range was that wide.
The lowest TDP Ontario is an 800 MHz single core with GPU off.
Atom can manage 1.6 GHz with a GPU, at a .5W penalty over the lowest Ontario.
Ontario may be good, but a 50% clock deficit is a lot, and that's with no GPU at all.
The top-end Ontario chips have a significantly higher TDP than the nearest Atoms. While it is likely that they will outperform Atom, the TDP difference is potentially more signficant for embedded platforms.
Perhaps Ontario could have been engineered to offer a higher clock rate at 5.5W, though the argument would be that its density figures would be much closer to Atom if its frequency/voltage range was that wide.
In a small quality notebook the max 18W TDP for CPU/GPU is not much. U can have sub 5W shity CPU-s when a quality high brightness LCD,chipset and desktop hard drive will eat several times more.(and u cant even run aplications fast enough on top of that)
U need to realize also that if your cpu is 2 times faster , than it will run 2 times less time at max TDP. If the thing could idle at 1-2W than the max 18W TDP is even less problem for normal usage.
So in general the 10W difference in CPU is not like OMG its two times more.
The lowest Atom has twice the clock speed and a GPU, however slow it might be to fit the TDP.
Ontario would have 800 MHz and no GPU.
An old rule of thumb is that OoOE yields roughly 50% extra performance, all else being equal.
To break even CPU-wise, an additional 33% frequency would be helpful.
The lack of a GPU at that level does not help power efficiency..
In general, the position for Ontario is one that keeps it slightly above Atom power-wise.
It does not scale quite as far down without losing half its reason for existing.
Ontario would have 800 MHz and no GPU.
An old rule of thumb is that OoOE yields roughly 50% extra performance, all else being equal.
To break even CPU-wise, an additional 33% frequency would be helpful.
The lack of a GPU at that level does not help power efficiency..
In general, the position for Ontario is one that keeps it slightly above Atom power-wise.
It does not scale quite as far down without losing half its reason for existing.
Blazkowicz
Legend
Atom is slower than a 900MHz Celeron-M, so it may end up as fast or slower than a 800MHz Ontario :
http://www.techpinas.com/2009/05/benchmarks-intel-atom-vs-celeron-m-vs.html
http://www.techpinas.com/2009/05/benchmarks-intel-atom-vs-celeron-m-vs.html
http://www.semiaccurate.com/2010/09/13/amd-shows-bobcat-systems/
Zacate running city of heroes smootly, and nothing more
Zacate running city of heroes smootly, and nothing more