AMD Ontario APU pictured,die size ~77mm^2

[savantu]

I bolded the relevant part.

Savantu, the CPU in Ontario is really low power, but ontario is CPU+Northbridge+GPU. Give me the power consumption of Atom + Northbridge + integrated GPU and then compare the performance of Ontario vs Atom.

800MHz Ontario without GPU is 5w. That is anything but low power.Even the old Atom, the 2008 generation, did 2w at 2GHz under a thermal virus. That is low power. Ontario could very well outperform it 2 or 3x. But it gobbles more power by the same ratio.

T56N - 18W, 1.6 GHz, dual core, GPU, DDR3
T48N - 18W, 1.4 GHz, dual core, GPU, DDR3
T40N - 9W, 1.0 GHz, dual core, GPU, LVDDR3

T52R - 18W, 1.5 GHz, single core, GPU, DDR3
T23R - 9W, 1.2 GHz, single core, GPU, LVDDR3

T48L - 18W, 1.4 GHz, dual core, no GPU, DDR3
T24L - 5W, 800 MHz, single core, no GPU, LVDDR3



ATOM with IMC and GPU :

D525 - 13W , 1.80 GHz, dual core, GPU
D425 - 10W , 1.80 GHz, single core, GPU

N550 - 8.5W, 1.50 GHz, dual core, GPU

N475 - 6.5W, 1.83 GHz, single core, GPU
N450 - 5.5W, 1.66 GHz, single core, GPU

The new Z6xx series isn't shown on Intel's site.
Even so , some comparisons :
High power -
T56N - 18W, 1.6 GHz, dual core, GPU, DDR3
T48N - 18W, 1.4 GHz, dual core, GPU, DDR3
vs.
D525 - 13W , 1.80 GHz, dual core, GPU
N550 - 8.5W, 1.50 GHz, dual core, GPU

Mid -
T52R - 18W, 1.5 GHz, single core, GPU, DDR3
T23R - 9W, 1.2 GHz, single core, GPU, LVDDR3
vs.
N475 - 6.5W, 1.83 GHz, single core, GPU

Low-
T24L - 5W, 800 MHz, single core, no GPU, LVDDR3
vs.
N450 - 5.5W, 1.66 GHz, single core, GPU

Assuming 2x better performance per clock from Bobcat vs. Atom and better GPU performance, it is still way too high.

Ontario is close to a SOC design, while Atom is the old way. Learn to make a proper comparison.

Before telling others to "learn" you should be sure you aren't telling nonsense yourself.
The current, second generation Atom, is a SoC design with CPU,GPU and NB on the same die. It's spreadsheets aren't on Intel's site ( or at least I couldn't find the Z6xx series ) but Intel claims it can do 1.5GHz in smartphones.
The 32nm generation should be at ARM power levels. Then the battle gets interesting.

Savantu, is he Paul Demone? Because if they are the same person, much would be explained.

No.

[Mechanical Man]

T40N - 9W, 1.0 GHz, dual core, GPU, LVDDR3

ATOM with IMC and GPU :

D525 - 13W , 1.80 GHz, dual core, GPU

Lets take this example of yours and put it in different light. If Bobcat delivers performance expected by BOINC stats( +90% vs atom according to some). Then this 1.0GHz Bobcat will mach 1,9GHz atom in cpu power + has massively better gpu..

[kl0012]

Lets take this example of yours and put it in different light. If Bobcat delivers performance expected by BOINC stats( +90% vs atom according to some). Then this 1.0GHz Bobcat will mach 1,9GHz atom in cpu power + has massively better gpu..

Bonic does not measure memory & multithreading performance.
But you can get some impression about the perf diff. between Atom D510 and Athlon x2 3250e.
http://www.anandtech.com/bench/Product/110?vs=116
As you can see, the average perf. diff. in the real world is much less then 2x.

[flippin_waffles]

Bonic does not measure memory & multithreading performance.
But you can get some impression about the perf diff. between Atom D510 and Athlon x2 3250e.
http://www.anandtech.com/bench/Product/110?vs=116
As you can see, the average perf. diff. in the real world is much less then 2x.

What's real world got to do with anything? AMD has been promoting real world testing and usage scenarios for years now, yet has fallen on deaf ears for the most part.

-journalists/reviewers still use 800x600 gaming resolutions to test CPU performance.
-journalists/reviewers still use SPi to test CPU performance.
-journalists/reviewers still use $1000 CPU's to to test $100 GPU's
-journalists/reviewers still compare $1000 CPU's to $100 CPU's.
-journalists/reviewers still test notebook battery life at the lowest and highest extremes. Nothing in between.

And the list goes on and on...

Boinc is a lot more real world than anything listed above.

[Hornet331]

What's real world got to do with anything? AMD has been promoting real world testing and usage scenarios for years now, yet has fallen on deaf ears for the most part.

-journalists/reviewers still use 800x600 gaming resolutions to test CPU performance.
-journalists/reviewers still use SPi to test CPU performance.
-journalists/reviewers still use $1000 CPU's to to test $100 GPU's
-journalists/reviewers still compare $1000 CPU's to $100 CPU's.
-journalists/reviewers still test notebook battery life at the lowest and highest extremes. Nothing in between.

And the list goes on and on...

Boinc is a lot more real world than anything listed above.

-valid (do you want to test gpu performance on a cpu review?, also tehy test more realworld resolution)
-hardly anymore, show me a recent review who does.
-valid, since with flexible multis you get the exact same performance a lower cpu would offer
-valid, since you want to see what performance you get for what money
-valid, since they cover both extremas so you know you can expect a runtime inbetween those two points. Also your implying they don't test normal usage scenarios.

Boinc cpu test falls in the same categorie as sandra or spi, pure synth test. If they would have provided actual WU/day figures it would be far more usefull.

[flippin_waffles]

-valid (do you want to test gpu performance on a cpu review?, also tehy test more realworld resolution)
-hardly anymore, show me a recent review who does.
-valid, since with flexible multis you get the exact same performance a lower cpu would offer
-valid, since you want to see what performance you get for what money
-valid, since they cover both extremas so you know you can expect a runtime inbetween those two points. Also your implying they don't test normal usage scenarios.

Boinc cpu test falls in the same categorie as sandra or spi, pure synth test. If they would have provided actual WU/day figures it would be far more usefull.

-Not valid. The premise is real world scenarios. How does gaming at 800x600 reflect CPU performance in the real world? How about CPU usage scenarios that reflect what customers do with their computer.
-Granted SPi is used in limited cases today. But is real world different today than it was 3 years ago when it was pervasive across almost every review? I suppose I should have said sandra and other synthetic tests.
-Not valid, not real world, which means something a customer uses in the real world.
-Valid. Doesn't belong in the context of a real world discussion. My point here is journalist/reviewers comparing $100 to $1000 CPU's in pure performance alone, rather than a performance/price comparison which happens a lot.
-Not valid. The runtime vary so much between fully idle and fully loaded, that it tells buyers exactly zero. How about some real world testing with something customers actually do with their notebooks. Have you ever bought one to let it sit in the corner and idle? Or buy one to fold until the battery runs out, recharge it and start all over again?

"Boinc falls in the same category as sandra and spi, pure synthetic". One is as valid as the other.

The Phoronix Test Suite is probably the best software a reviewer could use to indicate real world performance.

http://www.phoronix-test-suite.com/

And it runs on Win7.

[Manicdan]

-Not valid. The runtime vary so much between fully idle and fully loaded, that it tells buyers exactly zero. How about some real world testing with something customers actually do with their notebooks. Have you ever bought one to let it sit in the corner and idle? Or buy one to fold until the battery runs out, recharge it and start all over again?

this is a very important thing that needs to be tested carefully. if your using your laptop to watch a movie, how many can you watch before it dies. it factors in cpu efficiency and screen brightness (which is important that they be set to a similar standard if you test multiple, or atleast mention the brightness used and how it felt). or if you work with excel all day, it might be idle 90%, but when it is loaded does it use half power or full power.

i think some macros need to be made to mimic light use for facebooking and word processing and see how many hours can be obtained. and again, the screen can play a big part and needs to be mentioned but is probably often forgotten about.

when testing a system, all factors are affecting the result, and a good reviewer knows know to isolate certain parts and play fairly

[mAJORD]

800MHz Ontario without GPU is 5w. That is anything but low power.Even the old Atom, the 2008 generation, did 2w at 2GHz under a thermal virus. That is low power. Ontario could very well outperform it 2 or 3x. But it gobbles more power by the same ratio.

T56N - 18W, 1.6 GHz, dual core, GPU, DDR3
T48N - 18W, 1.4 GHz, dual core, GPU, DDR3
T40N - 9W, 1.0 GHz, dual core, GPU, LVDDR3

T52R - 18W, 1.5 GHz, single core, GPU, DDR3
T23R - 9W, 1.2 GHz, single core, GPU, LVDDR3

T48L - 18W, 1.4 GHz, dual core, no GPU, DDR3
T24L - 5W, 800 MHz, single core, no GPU, LVDDR3



ATOM with IMC and GPU :

D525 - 13W , 1.80 GHz, dual core, GPU
D425 - 10W , 1.80 GHz, single core, GPU

N550 - 8.5W, 1.50 GHz, dual core, GPU

N475 - 6.5W, 1.83 GHz, single core, GPU
N450 - 5.5W, 1.66 GHz, single core, GPU

The new Z6xx series isn't shown on Intel's site.
Even so , some comparisons :
High power -
T56N - 18W, 1.6 GHz, dual core, GPU, DDR3
T48N - 18W, 1.4 GHz, dual core, GPU, DDR3
vs.
D525 - 13W , 1.80 GHz, dual core, GPU
N550 - 8.5W, 1.50 GHz, dual core, GPU

Mid -
T52R - 18W, 1.5 GHz, single core, GPU, DDR3
T23R - 9W, 1.2 GHz, single core, GPU, LVDDR3
vs.
N475 - 6.5W, 1.83 GHz, single core, GPU

Low-
T24L - 5W, 800 MHz, single core, no GPU, LVDDR3
vs.
N450 - 5.5W, 1.66 GHz, single core, GPU

Assuming 2x better performance per clock from Bobcat vs. Atom and better GPU performance, it is still way too high.



Before telling others to "learn" you should be sure you aren't telling nonsense yourself.
The current, second generation Atom, is a SoC design with CPU,GPU and NB on the same die. It's spreadsheets aren't on Intel's site ( or at least I couldn't find the Z6xx series ) but Intel claims it can do 1.5GHz in smartphones.
The 32nm generation should be at ARM power levels. Then the battle gets interesting.


No.

How is it still way to high?

All the competing Atom SoCs have virtually the same TDP, with a considerably weaker GPU, and likely still lower CPU performance.

The 18w Models are in a different segment.

You've so far failed to prove the actual cores aren't capable of sub-1w. We simply don't know the Frequency vs Consumption relationship, nor the IPC of the chip yet.

[savantu]

Lets take this example of yours and put it in different light. If Bobcat delivers performance expected by BOINC stats( +90% vs atom according to some). Then this 1.0GHz Bobcat will mach 1,9GHz atom in cpu power + has massively better gpu..

You're putting all the eggs in the BOINC basket. I do not think they are relevant to infer Bobcat's advantage over Atom across the line.
From the TDP it is likely Atom will have 1.6-2x frequency advantage at the same TDP. OoO brings you advantages, but not that much.

As for GPU being "massively better", if the hype/reality follows the same pattern as the CPU, it's clear.

How is it still way to high?

All the competing Atom SoCs have virtually the same TDP, with a considerably weaker GPU, and likely still lower CPU performance.

One of us doesn't see well, that's clear.

You've so far failed to prove the actual cores are capable of sub-1w. We simply don't know the Frequency vs Consumption relationship, nor the IPC of the chip yet.

Are you talking about Atom ? Or what ?
If it's Atom you should check this : http://ark.intel.com/Product.aspx?id...ec-codes=SLB6Q.

[mAJORD]

One of us doesn't see well, that's clear.

?
Atom SoC TDP range: 5.5-13w
Ontario SoC TDP range: 5-9w, 18w (Different market segment)

The only Ontario SKU that has what appears to be a high TDP for what it offersis the single core 800Mhz @ 5w.

Clearly this is not a good indication of the core itself's efficency, otherwise a Dual core 1Ghz part with GPU would have a much higher TDP than 9w, wouldn't it. Perhaps the disabled GPU still consumes some power, Perhaps the IMC is a bit more power hungry compared to Intel's, perhaps the TDP is conservative, or perhaps a combination of things, who knows.

At any rate, neither company has an offering below ~5w in this segment

Are you talking about Atom ? Or what ?
If it's Atom you should check this : http://ark.intel.com/Product.aspx?id...ec-codes=SLB6Q.

Should have read 'Aren't, not 'Are' . My applogies, and fixed.

[savantu]

?
Atom SoC TDP range: 5.5-13w
Ontario SoC TDP range: 5-9w, 18w (Different market segment)

The only Ontario SKU that has what appears to be a high TDP for what it offersis the single core 800Mhz @ 5w.

Clearly this is not a good indication of the core itself's efficency, otherwise a Dual core 1Ghz part with GPU would have a much higher TDP than 9w, wouldn't it. Perhaps the disabled GPU still consumes some power, Perhaps the IMC is a bit more power hungry compared to Intel's, perhaps the TDP is conservative, or perhaps a combination of things, who knows.

Ok, I see your logic. Let's look at what we get at the same TDP ( since they are overlapping in the lower half.)

Dual cores
T40N - 9W, 1.0 GHz, dual core, GPU, LVDDR3
vs.
N550 - 8.5W, 1.50 GHz, dual core, GPU

Evenly matched performance wise and power wise, GPU favours Ontario

Single cores w/ GPU
T23R - 9W, 1.2 GHz, single core, GPU, LVDDR3
vs.
N475 - 6.5W, 1.83 GHz, single core, GPU

Similar performance, 30% lower power for Atom, GPU should favour Ontario

Single cores w/o GPU ( for AMD )
T24L - 5W, 800 MHz, single core, no GPU, LVDDR3
vs.
N450 - 5.5W, 1.66 GHz, single core, GPU

Atom should win on performance and it also has a GPU while Ontario does not. Clear win I would say.

At any rate, neither company has an offering below ~5w in this segment

There are Atom designs in the sub 5w segment when you pair it with the US15 NB.
Also, I don't have specs for Lincroft ( Z6xx series for smartphones and MIDs) , this is the 2nd generation+ based on Moorestown with far better thermals than the original Atom, especially at idle.

Should have read 'Aren't, not 'Are' . My applogies, and fixed.

Well, AMD introduces power gating technology with Ontario, so the GPU being not used should be cut off from the power distribution.

From 800MHz and 5w for a single core, it is virtually impossible to get to sub 1w. Why ? In todays designs leakage is 25-35% of active power. So leakage itself is over 1w.
Maybe 3rd gen Bobcat ( 28/20nm ) will be under 1w without changing the uarch. Right now, at least to me, it doesn't look possible.

[mAJORD]

Ok, I see your logic. Let's look at what we get at the same TDP ( since they are overlapping in the lower half.)

Dual cores
T40N - 9W, 1.0 GHz, dual core, GPU, LVDDR3
vs.
N550 - 8.5W, 1.50 GHz, dual core, GPU

Evenly matched performance wise and power wise, GPU favours Ontario

Single cores w/ GPU
T23R - 9W, 1.2 GHz, single core, GPU, LVDDR3
vs.
N475 - 6.5W, 1.83 GHz, single core, GPU

Similar performance, 30% lower power for Atom, GPU should favour Ontario



There are Atom designs in the sub 5w segment when you pair it with the US15 NB.
Also, I don't have specs for Lincroft ( Z6xx series for smartphones and MIDs) , this is the 2nd generation+ based on Moorestown with far better thermals than the original Atom, especially at idle.

Until we see Performance data for the CPU, it's hard to say if the first two examples will be evenly matched, or not, but for now we can run with that assumption.

Single cores w/o GPU ( for AMD )
T24L - 5W, 800 MHz, single core, no GPU, LVDDR3
vs.
N450 - 5.5W, 1.66 GHz, single core, GPU

Atom should win on performance and it also has a GPU while Ontario does not. Clear win I would say.

Right, and as I said, is the only SKU in the slide that has a high TDP, but again, does not mean their won't be lower (much lower) TDP CPU's.

According to the slides, the T24L is standard votlage DDR3. Which would be another contributor to the higher TDP

Well, AMD introduces power gating technology with Ontario, so the GPU being not used should be cut off from the power distribution.

From 800MHz and 5w for a single core, it is virtually impossible to get to sub 1w. Why ? In todays designs leakage is 25-35% of active power. So leakage itself is over 1w.
Maybe 3rd gen Bobcat ( 28/20nm ) will be under 1w without changing the uarch. Right now, at least to me, it doesn't look possible.

Again.. the 5w figure you're working off, at the least (if gating indeed eliminates GPU consumption) includes the IMC (with standard voltage DDR3 according to the slide) so you can't calculate leakage power from the TDP.

[Oberon]

Anyone want to bet that Savantu owns Intel stock?

As for all of these comparisons with TDP and performance, Out of order execution = die space and power increases, but also leads to a disproportionate increase in performance for the previously mentioned penalties.

[LightSpeed]

Dual cores
T40N - 9W, 1.0 GHz, dual core, GPU, LVDDR3
vs.
N550 - 8.5W, 1.50 GHz, dual core, GPU

Evenly matched performance wise and power wise, GPU favours Ontario

Single cores w/ GPU
T23R - 9W, 1.2 GHz, single core, GPU, LVDDR3
vs.
N475 - 6.5W, 1.83 GHz, single core, GPU

Similar performance, 30% lower power for Atom, GPU should favour Ontario

Single cores w/o GPU ( for AMD )
T24L - 5W, 800 MHz, single core, no GPU, LVDDR3
vs.
N450 - 5.5W, 1.66 GHz, single core, GPU

Atom should win on performance and it also has a GPU while Ontario does not. Clear win I would say.

Why are you predicting Bobcat's performance? How can you be sure that a 1.5Ghz Atom will compare to a 1Ghz Bobcat? My guess is that the Bobcat at 1Ghz will be faster than the 1.5Ghz Atom, so in your Case #1, AMD would have superior performance + superior GPU for the same power. Clear win I would say?

Also, its apparent that the disabled cores and GPU's are sucking power, or they are just worse binned with higher voltages. See the single core Bobcat: 1.5Ghz 18W. Dual core without GPU: 1.4Ghz 18W. Makes no sense when there is a 1.6ghz Bobcat + GPU for 18W.

[MrMojoZ]

Clear win I would say?

Can't win unless Intel makes it, nice try.