Kabini Rumors

[monstercameron]

Correct, turbo mode has been stated as a higher power mode.

http://news.softpedia.com/news/MWC-2013-AMD-Turbo-Dock-Demoed-on-Temash-Tablet-332291.shtml

What's he currently demoing, 1GHz or 1.4GHz?

he was doing 1GHz at first then he did something (I could figure out what he did) and he was able to keep the clockspd at 1.4GHz both on a charger and off...

 

[mikk]

Correct, turbo mode has been stated as a higher power mode.

http://news.softpedia.com/news/MWC-2013-AMD-Turbo-Dock-Demoed-on-Temash-Tablet-332291.shtml

Of course. The point is we thought months ago the Temash Quadcore SoC has a 5.9W TDP in nominal mode and 8W in turbo mode. The last infos we got stated for Temash Quadcore 8W in nominal mode and 15W in turbo mode. From the power numbers we got today it seems unlikely that the SoC in turbo mode has a 8W TDP. With 100% CPU load on CPU and GPU it could go up over 20W which makes 15W for the SoC more plausible.

 

[monstercameron]

Of course. The point is we thought months ago the Temash Quadcore SoC has a 5.9W TDP in nominal mode and 8W in turbo mode. The last infos we got stated for Temash Quadcore 8W in nominal mode and 15W in turbo mode. From the power numbers we got today it seems unlikely that the SoC in turbo mode has a 8W TDP. With 100% CPU load on CPU and GPU it could go up over 20W which makes 15W for the SoC more plausible.

arent tdp and power draw different things?

 

[inf64]

arent tdp and power draw different things?

Yes they are. Also the whole laptop is drawing said amount so APU itself although the greatest contributor is nowhere near that number.

 

[mikk]

arent tdp and power draw different things?

Yes and No. AMD could specify TDP for Temash at 1W while it requires 20W. Everything is possible in theory. But it is highly unlikely that AMDs TDP differ that much.

 

[Vesku]

In this netbook form factor there seemed to be no arbitrary power limit on the turbo mode. He said the clocks held regardless of what amount of load he was putting on it in terms of CPU + GPU. Might be a different case with this same SoC in a tablet+dock.

Sunspider score was 610ms (at 1.4GHz I believe) puts it in between the Exynos 5 @ 1.7 GHz (Score: 690.5 ms) and the Celeron 847 (Sandybridge) @ 800MHz (Score: 527.1 ms). That's a single threaded test, correct?

Summary of Live session with the Temash netbook: http://ultrabooknews.com/2013/05/10/live-now-acer-aspire-v5-and-amd-temash-testing/
Sunspider results for other low cost low power chips: http://www.anandtech.com/show/6476/acer-c7-chromebook-review/3

 

[monstercameron]

In this netbook form factor there seemed to be no arbitrary power limit on the turbo mode. He said the clocks held regardless of what amount of load he was putting on it in terms of CPU + GPU. Might be a different case with this same SoC in a tablet+dock.

Sunspider score was 610ms (at 1.4GHz I believe) puts it in between the Exynos 5 @ 1.7 GHz (Score: 690.5 ms) and the Celeron 847 (Sandybridge) @ 800MHz (Score: 527.1 ms). That's a single threaded test, correct?

Summary of Live session with the Temash netbook: http://ultrabooknews.com/2013/05/10/live-now-acer-aspire-v5-and-amd-temash-testing/
Sunspider results for other low cost low power chips: http://www.anandtech.com/show/6476/acer-c7-chromebook-review/3

yeah sunspier is single threaded

 

[sontin]

Tegra 4 with 1,9GHz has a score of 500 in Sunspider. Yeah it says nothing about IPC but a 1GHz Temash is not really competitive...

 

[Vesku]

How does Tegra 4 do in Cinebench 11.5?

 

[mikk]

Sunspider is highly dependent on OS/Browser, difficult to compare.

 

[sontin]

How does Tegra 4 do in Cinebench 11.5?

Hope for a RT version.

 

[Vesku]

Sunspider is highly dependent on OS/Browser, difficult to compare.

Good point, the more I read up on Sunspider the more caveats I saw saying it's murky outside of same OS same exact browser version.

 

[SiliconWars]

The given TDP is of the chip only. It will not exceed 5.9 Watts at 1.0 GHz but will obviously be a lot higher at 1.4 GHz on all four cores.

The system power draw is a lot worse under graphics load because the memory is being pushed a lot harder. We don't know the quality of the memory but we can guess it will be low quality 1.65V bulk RAM at best.

It has a mechanical hard drive, an SSD will help to save 1-2 Watts. If you start adding all these little things up you can see why the system pulls 15W or more - it's got nothing to do with the actual SoC though.

Overall it's quite encouraging that they could get close to AMD's numbers of 0.99 and 1.39 for Cinebench on a lower end system. The difference in ultrabook performance for the same intel cpu also varies depending on the quality of the rest of the components.

 

[monstercameron]

^a ssd was installed in this case but the display was probably eating the most power.

 

[SiliconWars]

Ok I missed that, thanks.

Here is the thing that's worth remembering - no matter how low power your SoC is, the other components aren't going to be much different from system to system.

Silvermont for example might have a very low core power draw (This is why Intel is trumpeting core power instead of system power), but the rest of the system is going to be the same as anything else.

Imagine a 3W SoC vs a 6W SoC, then the rest of the components can pull 10W at max load for 13W and 16W overall. Something that appears to be drawing half the power is actually only 23% better on battery life, and at a high performance cost. This is why current Atom's are terrible - the low power draw is due to poor performance and the power savings are generally negated in overall system power. Would you rather have 6 hours of (current) Atom performance or 5 hours of Jaguar performance? It's no contest.

 

[Exophase]

We don't know the quality of the memory but we can guess it will be low quality 1.65V bulk RAM at best.

Don't know why you would assume that, commodity DDR3 is 1.5V and lots of even cheaper low power devices use 1.35 DDR3L. For instance my Exynos 5250 based Chromebook does. I don't even know where this 1.65V DDR3 comes from but I assume it has to do with DIMMs clocking beyond the spec, and some OEM Temash laptop isn't going to be using DDR3 2400.

 

[SiliconWars]

There is plenty of 1600 MHz 1.65V DDR3 around at the OEM level.

 

[Enigmoid]

it uses a DDR3-1066...
might be a bottleneck even to the cpu

Considering there is almost no difference in a 3770k between 1066 mhz and 1600 mhz ram in most apps (less than 5%) there is no chance 1066 mhz ram is hurting cpu performance in any way.

 

[wlee15]

Considering there is almost no difference in a 3770k between 1066 mhz and 1600 mhz ram in most apps (less than 5%) there is no chance 1066 mhz ram is hurting cpu performance in any way.

Remember that Kabini and Temash only have a single 64-bit memory channel.

 

[MightyMalus]

Considering there is almost no difference in a 3770k between 1066 mhz and 1600 mhz ram in most apps (less than 5%) there is no chance 1066 mhz ram is hurting cpu performance in any way.

Well, if it behaves like Trinity...

 

[Exophase]

Remember that Kabini and Temash only have a single 64-bit memory channel.

Even the 1.4GHz Temash is well under half as powerful as an i7-3770K, outweighing the halved memory controller width. Not counting GPU use of course.

 

[itsmydamnation]

Even the 1.4GHz Temash is well under half as powerful as an i7-3770K, outweighing the halved memory controller width. Not counting GPU use of course.

With a whole lot less cache as well, how much of a bandwidth multiplier is that extra 7 odd Mb of cache.

 

[Exophase]

With a whole lot less cache as well, how much of a bandwidth multiplier is that extra 7 odd Mb of cache.

It's closer to an extra 6.125MB of cache since it's strictly inclusive, and the difference is probably not that big when it comes to external bandwidth pressure. It's not usual for a program to be balanced such that it's simultaneously heavily accessing an 8MB hot set and saturating external bandwidth. Being bandwidth bound tends to leave you with a more narrow working set in cache.

Working sets that fit with a good hit rate within 8MB but not 2MB will tend to be latency bound to main memory instead of bandwidth bound, so Jaguar will probably be screwed regardless of its external bandwidth.

 

[itsmydamnation]

what about workloads that fit in 512k vs 2mb when you have 4 of them concurrently.

 

[mikk]

There is plenty of 1600 MHz 1.65V DDR3 around at the OEM level.

Do you have a source? I never seen 1.65V in a notebook. JEDEC standard voltage is either 1.35V or 1.5V. I just had a look on Sodimm on geizhals

http://geizhals.at/de/?cat=ramsoddr3

There is no 1.65V sodimm.