AMD Carrizo APU Details Leaked

[Abwx]

Because AMD is incompetent and cannot get products out the door to (literally) save themselves.

Yes, they lack the competence to give their chips for free with tons of money when they have not the upper ground in specs.

Yes, and I'm quite sure BT's GPU is far more than enough for those users.

Oh, the good enough concept is now recycled...opportunisticly, but it didnt apply when you questioned Mullins battery life, anyway you did brought a relevant point because BT perf/watt is mediocre when the GPU is heavily involved and that would negate all power comsumption advantages, if ever it has some, not counting that it s about unplayable generaly, unless it s a 3 years old child that he s using it as gaming device.

 

[III-V]

Yes, they lack the competence to give their chips for free with tons of money when they have not the upper ground in specs.

They're not for free. Why contra revenue is such a difficult concept for people to grasp, I will never know.

But at any rate, I have no objection to the idea that Bay Trail was a disappointment from a BOM perspective. Certainly Mullins isn't much better in this regard, though.

Oh, the good enough concept is now recycled...opportunisticly, but it didnt apply when you questioned Mullins battery life, anyway you did brought a relevant point because BT perf/watt is mediocre when the GPU is heavily involved and that would negate all power comsumption advantages, if ever it has some, not counting that it s about unplayable generaly, unless it s a 3 years old child that he s using it as gaming device.

Man, you just can't take a loss, can you?

And no, BT should have very competitive perf/watt with its IGP. Intel does very well from that perspective.

 

[Abwx]

They're not for free. Why contra revenue is such a difficult concept for people to grasp, I will never know.

But at any rate, I have no objection to the idea that Bay Trail was a disappointment from a BOM perspective. Certainly Mullins isn't much better in this regard, though.

Doing +-/x and = operations is not that difficult to grasp, if you dont know there are other people that know and will help doing the, basic, calculations.

As for thoses products respective BOMs they are identical, the problem is that Intel wants to force its way in the lowish prices and this will not be possible in the short term, X86 users are inherently more perfs greedy than the ARM users, it s two different populations.

Man, you just can't take a loss, can you?

You re not that bad in this respect either, we re just confronting contradicting views and there s no arm in doing so.

And no, BT should have very competitive perf/watt with its IGP. Intel does very well from that perspective.

BT has very good perf/watt CPU wise but his GPU has lower perf/watt than Mullins, to be precise it has about half the perf/watt of Mullins in this matter when tested with Luxmark GPU loading.

http://www.hardware.fr/articles/921-6/consommation.html

 

[AtenRa]

Who games on a tablet? Honestly? And 3D games at that? We've got years to go before that becomes even remotely viable.

And I wasn't suggesting otherwise, in regards to the iGPU.

Just to show you what a Mullins Tablet can do. Now, who doesnt want a windows Tablet like that ??

https://www.youtube.com/watch?v=VXkEtgc1aME

 

[III-V]

Doing +-/x and = operations is not that difficult to grasp, if you dont know there are other people that know and will help doing the, basic, calculations.

As for thoses products respective BOMs they are identical, the problem is that Intel wants to force its way in the lowish prices and this will not be possible in the short term, X86 users are inherently more perfs greedy than the ARM users, it s two different populations.

Intel royally screwed up with Bay Trail's BOM. I'm fairly certain that the die floorplan is poorly optimized, and a 5-10% reduction in die size would be relatively easy to achieve. You can see this in Avoton as well, to a greater extent.

Then there's the whole BOM/extra engineering effort thing. I'd imagine AMD has similar issues here.

But I don't think it's an x86 vs. ARM thing. I don't see why AMD or Intel couldn't compete on cost -- they're just competing in unfamiliar territory. There's undoubtedly some advantage that ARM holds over x86 in this space, but I don't see it as being something that can't be overcome, by Intel at least. AMD has less resources, has to use the same process nodes (and later than everyone else too, unfortunately)... they're in a worse situation for sure. And I'm not so sure that's Intel's doing, as far as AMD's current tablet situation goes. We're likely still seeing the after-effects of Dirk Meyer's hesitance to prioritize mobile.

You re not that bad in this respect either, we re just confronting contradicting views and there s no arm in doing so.

Well, I appreciate that sentiment. I definitely see you in a different light now.

BT has very good perf/watt CPU wise but his GPU has lower perf/watt than Mullins, to be precise it has about half the perf/watt of Mullins in this matter when tested with Luxmark GPU loading.

http://www.hardware.fr/articles/921-6/consommation.html

That's really interesting. I am not sure what to make of that. I wish there were data on the matter, and a wider variety of benchmarks.

I do have a couple of objections though, and I think they're both pretty critical. In that TDP, Kabini/Mullins is much more competitive, and I think Intel would fair much better in the tablet form factor we were discussing. Also, AMD does very, very well compared to both Nvidia and Intel in Luxmark, whereas in gaming workloads, the differences between the three vendors becomes far smaller.

Just to show you what a Mullins Tablet can do. Now, who doesnt want a windows Tablet like that ??

https://www.youtube.com/watch?v=VXkEtgc1aME

I'm from the states, so maybe trying to woo me over with a soccer/football video game isn't the best approach

Definitely impressive, all kidding aside. I hope Cherry Trail makes significant strides on that front.

 

[AtenRa]

I'm from the states, so maybe trying to woo me over with a soccer/football video game isn't the best approach

Definitely impressive, all kidding aside. I hope Cherry Trail makes significant strides on that front.

After the soccer game it plays DIRT.

Also, this Tablet will let you play lots and lots of steam free games or older games you may have on steam account. Not only that, it is a Windows x86 Tablet meaning you can connect to a Monitor and keyboard/mouse when you are home and use it like your Desktop with all your windows apps

AMD should integrate Nolan with an LTE modem, it would be the perfect Windows Tablet SoC, lots of iGPU performance(2x iGPU of Mullins) and good power consumption due to 20nm.

 

[Enigmoid]

BT has very good perf/watt CPU wise but his GPU has lower perf/watt than Mullins, to be precise it has about half the perf/watt of Mullins in this matter when tested with Luxmark GPU loading.

http://www.hardware.fr/articles/921-6/consommation.html

Are you trying to show the architecture is bad?

The celeron uses the same GPU architecture and is miles better from an efficiency standpoint.

That said, the igp is BT's weakness and is not nearly as efficient as other alternatives.

 

[Abwx]

Intel royally screwed up with Bay Trail's BOM. I'm fairly certain that the die floorplan is poorly optimized, and a 5-10% reduction in die size would be relatively easy to achieve. You can see this in Avoton as well, to a greater extent.

Then there's the whole BOM/extra engineering effort thing. I'd imagine AMD has similar issues here.

But I don't think it's an x86 vs. ARM thing. I don't see why AMD or Intel couldn't compete on cost -- they're just competing in unfamiliar territory.

They cant compete because X86 was screwed from the start battling a light OS with W8.1, MS should had used an updated WXP64 as light OS for mobile items, this would had at least reduced the RAM requirements.

I do have a couple of objections though, and I think they're both pretty critical. In that TDP, Kabini/Mullins is much more competitive, and I think Intel would fair much better in the tablet form factor we were discussing.

TDP is a problem once you start chasing the perfs crowns and that comsumption is increased by 40-50% for the sake of 20% more perfs that will be paid cash anyway by more costs, weight and less autonomy, i was always critical of those turboing, it s like the marketing departments completely took over the enginering side, all efforts of the engineers are litteraly crushed by irrational marketing spiners, and if Core M is of any indication it makes no doubt that next gen "low power" chips will be also maxed out up to the current power usages.

The celeron uses the same GPU architecture and is miles better from an efficiency standpoint.

That said, the igp is BT's weakness and is not nearly as efficient as other alternatives.

This review was very sympathetic for Intel, doing the same tests i got 20% higher perf/watt for the GPU and 10%, roughly, for the CPU, indeed they got the worst power deltas of all AM1 reviews despite using a pico PSU, yet the plateform i used had the CPU power drain going through 3 switching mode PSUs while their set up only has 2 to go through.

About the Celeron there s a flaw in their measurement protocol because they use the total power of the plateform, hence the celeron has much more power ratio devoted to computation than both the two low power plateforms, it s not obvious at first glance, if you increase the power comsumption, and hence scores, of the two other plateforms up to Celeron level you ll see that they have much much better perfs/Watt, in the case of BT there s 53% devoted to the plateform and 47% for computation, for the Celeron 1820 it s 33% and 66%.

It s very easy to truncate the numbers, isnt it.?.

 

[Insert_Nickname]

They cant compete because X86 was screwed from the start battling a light OS with W8.1, MS should had used an updated WXP64 as light OS for mobile items, this would had at least reduced the RAM requirements.

The x86 version of 8(.1) actually isn't bad. I had the 8 preview running on an Atom N270-powered netbook with a paltry 1GB RAM. It was totally usable, and "snappy" even on such very low performance hardware.

I can't imagine most modern SoCs can be much worse then that.

 

[Enigmoid]

This review was very sympathetic for Intel, doing the same tests i got 20% higher perf/watt for the GPU and 10%, roughly, for the CPU, indeed they got the worst power deltas of all AM1 reviews despite using a pico PSU, yet the plateform i used had the CPU power drain going through 3 switching mode PSUs while their set up only has 2 to go through.

About the Celeron there s a flaw in their measurement protocol because they use the total power of the plateform, hence the celeron has much more power ratio devoted to computation than both the two low power plateforms, it s not obvious at first glance, if you increase the power comsumption, and hence scores, of the two other plateforms up to Celeron level you ll see that they have much much better perfs/Watt, in the case of BT there s 53% devoted to the plateform and 47% for computation, for the Celeron 1820 it s 33% and 66%.

It s very easy to truncate the numbers, isnt it.?.

Yeah, this is why perf/W on a platform where half of the power is idle power makes for a poor perf/W comparison.

This is why if you go by raw perf/W the 4770k runs away and beats the 5350 in efficiency (generally more than 4x as powerful using 4x the power -- 25W vs. 100W).

 

[Abwx]

This is why if you go by raw perf/W the 4770k runs away and beats the 5350 in efficiency (generally more than 4x as powerful using 4x the power -- 25W vs. 100W).

I measured the 5350 in CB, it use about 12-13W at the CPU level for a score of 2, the 4770K use 54W for a score of 8.16.

At the plateform level the ratios are about the same.

 

[Enigmoid]

I measured the 5350 in CB, it use about 12-13W at the CPU level for a score of 2, the 4770K use 54W for a score of 8.16.

At the plateform level the ratios are about the same.

Yep, and when you add typical platform power (10W for the 5350 platform and 20-25W) for the 1150 platform the 4770k comes out ahead. (Stripped non dgpu platforms).

Ex) 20-25W on the 5350 for 2 points and 75-90W on the 4770k for 8.16 points.

 

[NostaSeronx]

There is no way carizzo will make way into tablets.

A Carrizo SKU with a xW SDP(est; 5-7 watts), will be in a 2-in-1 Convertible. (Tablet with a keyboard)

 

[Abwx]

Yep, and when you add typical platform power (10W for the 5350 platform and 20-25W) for the 1150 platform the 4770k comes out ahead. (Stripped non dgpu platforms).

Ex) 20-25W on the 5350 for 2 points and 75-90W on the 4770k for 8.16 points.

26-27W and 85-90W respectively for a perf ratio of 4, so you re right in this case but in Fritz it s 3.4 , 1.56 in 7zip and 2.8 in Cyberlink media expresso, so it s not a clear cut, FP seems to be owned by the 4770K while integer often see Kabini getting the upper hand.

 

[Enigmoid]

26-27W and 85-90W respectively for a perf ratio of 4, so you re right in this case but in Fritz it s 3.4 , 1.56 in 7zip and 2.8 in Cyberlink media expresso, so it s not a clear cut, FP seems to be owned by the 4770K while integer often see Kabini getting the upper hand.

How are you getting 4 in 7 zip?

http://www.anandtech.com/bench/product/1223?vs=836

Its 5900 vs. 24000.

To get back on point. Low power and efficient chips may be low power and efficient from the chip/SOC standpoint but from a platform view they may not (ie original atoms).

 

[Abwx]

How are you getting 4 in 7 zip?

http://www.anandtech.com/bench/product/1223?vs=836

Its 5900 vs. 24000.

To get back on point. Low power and efficient chips may be low power and efficient from the chip/SOC standpoint but from a platform view they may not (ie original atoms).

I used HFR numbers, they must have messed the things..

On topic, low power chip are in principle more efficient at equal TDP if there s the relevant core count , adding 4 cores to Kabini would increase the drained power by 14W at most at the main level and the score would be doubled to 4 for 40W total power draw, add 8 other cores and the main power increase to 68W for a score comparable to the 4770K, now put everybody on the same node and you ll end with a 16 low power cores that is significantly more efficient even at the plateform level.

 

[Kallogan]

I don't have hopes for amd APUs anymore.

Even Nvidia would have better chances in APUs' world if u look at Tegra K1 perf with denver 64 bits cores, and Tegra is not even maxwell based yet

 

[MisterLilBig]

I don't have hopes for amd APUs anymore.

Even Nvidia would have better chances in APUs' world if u look at Tegra K1 perf with denver 64 bits cores, and Tegra is not even maxwell based yet

No product exists with the Tegra K1 Denver 2.5GHz SoC. Which the scores floating around represent. And we don't even know how the iGPU side of it will do.

 

[NostaSeronx]

I've been doing research on Adaptive Voltage (& Frequency) Scaling.

So, apparently... For AMD to have AVS in Carrizo via per part adaptive voltage.

The critical paths will need to be both Analog and Digital, aka Mixed-Signal, maybe?