[AnandTech] Atom Z3770's Cinebench performance

[Blandge]

I'm guessing TuxDave was referring to the fact that it's a 15 Wh (watt-hour) battery, not 15 watt.

I can't tell if Enigmoid truly doesn't understand, or if he's just being lazy with terminology. In any case, 1Wh means it can supply one W of power for an hour, so despite his incorrect terminology, the math roughly works out to 1.25W average power.

 

[SiliconWars]

If AMD could made Temash/Kabini in Intel's 22nm: unfortunately, we know it's not possible.

It's got more to do with designing a chip for a phone than it has with 22nm or 28nm. After watching Intel's and Nvidia's struggles to break into that market, AMD is unlikely to bother with a phone chip for a long time.

 

[Dresdenboy]

Not to mention that cinebench is FP and Atom generally is much better at integer loads. AFAIK even silvermont isn't fully OoO in FP but I might be wrong.

Of course TDP can't be compared to AMD A4 because baytrail-t GPU is worse and hence will use a lot less power.

Andreas Stiller (with c't mag/heise) also believes, that Silvermont has an in order FPU. But those high CB scores, achieved w/o HT to fill the pipeline bubbles, suggest OoO FPU.

 

[Dresdenboy]

It's got more to do with designing a chip for a phone than it has with 22nm or 28nm. After watching Intel's and Nvidia's struggles to break into that market, AMD is unlikely to bother with a phone chip for a long time.

At least not for its own business. But its SoC division is looking for designers to work on 1-2W SoCs for handheld devices and tablets.

 

[jpiniero]

It's got more to do with designing a chip for a phone than it has with 22nm or 28nm. After watching Intel's and Nvidia's struggles to break into that market, AMD is unlikely to bother with a phone chip for a long time.

I'm sure AMD has plans to release a phone chip once they have the ARM business up and running.

 

[rainy]

It's got more to do with designing a chip for a phone than it has with 22nm or 28nm. After watching Intel's and Nvidia's struggles to break into that market, AMD is unlikely to bother with a phone chip for a long time.

I'm looking just from power consumption perspective, not any other: is good and could be excellent at Intel's 22nm.

 

[wlee15]

Andreas Stiller (with c't mag/heise) also believes, that Silvermont has an in order FPU. But those high CB scores, achieved w/o HT to fill the pipeline bubbles, suggest OoO FPU.

The Intel Optimization guide clearly states that instructions within each FP Reservation stations are schedule in-order, but instruction in different Reservation stations can be schedule out of order. i.e an later FADD instruction can go in front and earlier FMUL instruction.

 

[TuxDave]

I'm guessing TuxDave was referring to the fact that it's a 15 Wh (watt-hour) battery, not 15 watt.

Yeah, mostly that. And that a product consumes 1.25 "watts per hour"

 

[Enigmoid]

I can't tell if Enigmoid truly doesn't understand, or if he's just being lazy with terminology. In any case, 1Wh means it can supply one W of power for an hour, so despite his incorrect terminology, the math roughly works out to 1.25W average power.

Sorry for the confusion was kinda rushed.

 

[Arachnotronic]

Andreas Stiller (with c't mag/heise) also believes, that Silvermont has an in order FPU. But those high CB scores, achieved w/o HT to fill the pipeline bubbles, suggest OoO FPU.

According to the IA optimization manual, Silvermont's FPU is in-order.

 

[Arachnotronic]

I'm sure AMD has plans to release a phone chip once they have the ARM business up and running.

Yeah, they've got to get themselves a lot of IP that they can't afford right now. Cellular baseband, WiFi, GPS, Bluetooth, and so on. Do you really think Qualcomm is "winning" because it can do a particularly better CPU or GPU than anybody else?

No! It's all about the modem/connectivity which Qualcomm currently dominates. Unless AMD wants to play the modem R&D game, too, it's going to be shut out of the phone market. That said, it's not as though AMD needs it to be successful.

 

[bullzz]

http://www.anandtech.com/show/7251/lg-g2-and-msm8974-snapdragon-800-a-quick-look/5

s800's GPU says hello. its going to be bad when we see baytrail in those benchmarks

 

[Exophase]

Andreas Stiller (with c't mag/heise) also believes, that Silvermont has an in order FPU. But those high CB scores, achieved w/o HT to fill the pipeline bubbles, suggest OoO FPU.

Intel's software optimization guide is very clear on the design of Silvermont's FPU/SIMD units.

EDIT: Urgh beat by two people >_<

 

[Arachnotronic]

Intel's software optimization guide is very clear on the design of Silvermont's FPU/SIMD units.

EDIT: Urgh beat by two people >_<

 

[liahos1]

http://www.anandtech.com/show/7251/lg-g2-and-msm8974-snapdragon-800-a-quick-look/5

s800's GPU says hello. its going to be bad when we see baytrail in those benchmarks

its going to be worse than s800 but better or equal to s600. likely to do very well on the cpu side based on early leaks. incidentally the lg2's cpu/gpu performance doesnt seem to be matching up well with the reference platform that qcom put out when anand did his performance preview.

 

[Khato]

incidentally the lg2's cpu/gpu performance doesnt seem to be matching up well with the reference platform that qcom put out when anand did his performance preview.

Actual products almost never live up to Qualcomm's MDP previews. Mostly for the fact that only performance is reported on the MDP, not battery life Though I do expect tablet products to be a bit closer to the MDP performance levels.

Regardless, it's going to be quite interesting to see how they compare in actual tablets. Intel losing on graphics is basically a foregone conclusion, but there's the important question of by how much. Because if it's not a knockout blow, then advantages in CPU performance and power consumption can easily put it in the lead over all.

 

[liahos1]

Actual products almost never live up to Qualcomm's MDP previews. Mostly for the fact that only performance is reported on the MDP, not battery life Though I do expect tablet products to be a bit closer to the MDP performance levels.

Regardless, it's going to be quite interesting to see how they compare in actual tablets. Intel losing on graphics is basically a foregone conclusion, but there's the important question of by how much. Because if it's not a knockout blow, then advantages in CPU performance and power consumption can easily put it in the lead over all.

yes i think thats sort of how i think about it. beyond that, the incremental die budget intel gets going to 14 i think gives them more of an opportunity to dedicate that budget to increased cpu and/or gpu, power savings per given level of performance or mix of all three vs competition.

honestly i have nexus 7, ipad 4, and kindle fire hd. i get these gadgets to test out new product (diligence for my firms investments). I also game a TON. I think a lot of ppl saying intel is dead this round bc of gpu are short sighted. singled threaded performance matters a lot for your average use case on a tablet. games are going to run fine on this soc. most games are games arent going to be made to only run well on s800 since most volume shipments next year are going to be on s600 and s400 or other midrange and low end products.

but for some reason having a soc built to maximize performance for use cases that matter most on a daily basis (cpu performance) is irrelevant and intel is somehow doomed.

 

[krumme]

Yeah, they've got to get themselves a lot of IP that they can't afford right now. Cellular baseband, WiFi, GPS, Bluetooth, and so on. Do you really think Qualcomm is "winning" because it can do a particularly better CPU or GPU than anybody else?

No! It's all about the modem/connectivity which Qualcomm currently dominates. Unless AMD wants to play the modem R&D game, too, it's going to be shut out of the phone market. That said, it's not as though AMD needs it to be successful.

What if they make something for Qualcomm, Samsung or Sony ?

Aquiring IP is cheap unless you use it. What remains to be seen outside of consoles is if they can use it.

Amd as a traditional competitor as we have seen will become less and less if rr can move it.

http://www.anandtech.com/show/7281/understanding-amd-semi-custom-strategy

 

[Dresdenboy]

The Intel Optimization guide clearly states that instructions within each FP Reservation stations are schedule in-order, but instruction in different Reservation stations can be schedule out of order. i.e an later FADD instruction can go in front and earlier FMUL instruction.

Ah, thanks (the others too ). OK, different RS' might have this effect, but need independency between the operands.

In the latest c't issue A. S. still wonders about in-order/OoO.

 

[Dresdenboy]

Yeah, they've got to get themselves a lot of IP that they can't afford right now. Cellular baseband, WiFi, GPS, Bluetooth, and so on. Do you really think Qualcomm is "winning" because it can do a particularly better CPU or GPU than anybody else?

No! It's all about the modem/connectivity which Qualcomm currently dominates. Unless AMD wants to play the modem R&D game, too, it's going to be shut out of the phone market. That said, it's not as though AMD needs it to be successful.

So far it looks like AMD might do phone chips in it's semicustom unit, but not for its own purposes. Besides going into a crowded market, an AMD phone chip also means a lot of necessary support and distractions.

 

[Arachnotronic]

So far it looks like AMD might do phone chips in it's semicustom unit, but not for its own purposes. Besides going into a crowded market, an AMD phone chip also means a lot of necessary support and distractions.

Doubtful that AMD would do a phone chip.

 

[Dresdenboy]

Doubtful that AMD would do a phone chip.

How would you interpret their job offering, where they're looking for a SoC expert to develop 1-2W SoCs targeted at handheld/tablet devices?

 

[mrmt]

What if they make something for Qualcomm, Samsung or Sony ?

The first two can do better than AMD.

 

[monstercameron]

The first two can do better than AMD.

samsung 3d ip isnt as good as amds.

 

[SiliconWars]

It's doubtful that anyone has better low-power 3d capability than AMD. Kepler should help Nvidia a bit but we know they are hard to deal with (too expensive unless desperate).

I wonder if offering semi-custom phone chips is AMD's way of getting around any deal they struck with Qualcomm regarding non-licensing, after Qualcomm bought their handset division.