Kabini Rumors

[AtenRa]

Sure, you will see ARM SoCs in the same products.
The problem for AMD is: Kabini needs to much power to be faster. And it's to slow while using the same power.

ARM SoCs are not x86, they cannot use Windows.

Kabini will be very competitive against 32nm ATOM even in low power (5-6W).

 

[Arachnotronic]

ARM SoCs are not x86, they cannot use Windows.

Kabini will be very competitive against 32nm ATOM even in low power (5-6W).

Kabini won't be fighting 32nm Atom for very long. 22nm "Bay Trail" for tablet, notebook, and desktop coming by holiday 2013. And these will be clocked up much more liberally than the tablet edition.

 

[AtenRa]

Kabini won't be fighting 32nm Atom for very long. 22nm "Bay Trail" for tablet, notebook, and desktop coming by holiday 2013. And these will be clocked up much more liberally than the tablet edition.

Don't expect any 22nm ATOM products in 2013.

 

[mikk]

Don't expect any 22nm ATOM products in 2013.

Why?

 

[sontin]

ARM SoCs are not x86, they cannot use Windows.

And yet they are outselling x86 processors. So?
And what is Windows RT? It's Windows.

Kabini will be very competitive against 32nm ATOM even in low power (5-6W).

Baytrail comes in the 2H of 2013. Temash has only a very short time before the first 22nm Atom SoCs will hit them.

 

[Arachnotronic]

Don't expect any 22nm ATOM products in 2013.

Intel has repeatedly gone on record that 22nm Atoms will be here this year, from microserver, to comms, to tablets, to PCs. Desperate times call for desperate measures, and you will indeed see 22nm Atom this year in full force.

 

[lagokc]

And yet they are outselling x86 processors. So?
And what is Windows RT? It's Windows.



Baytrail comes in the 2H of 2013. Temash has only a very short time before the first 22nm Atom SoCs will hit them.

While Windows RT might technically be "Windows", it is missing the one feature that Windows has that nothing else does - compatibility with x86 Windows applications (without WINE). Without that it doesn't really have any advantage over Linux/Android/iOS/everything else.

 

[Arachnotronic]

While Windows RT might technically be "Windows", it is missing the one feature that Windows has that nothing else does - compatibility with x86 Windows applications (without WINE). Without that it doesn't really have any advantage over Linux/Android/iOS/everything else.

It is at a disadvantage to Android/iOS actually since at least those platforms have rich, established ecosystems.

Microsoft needs X86, and as long as Intel delivers quality silicon, Windows RT will quietly fade into nothingness.

 

[Arkadrel]

Saw this from SA.

Maybe those AMD (Temash?) based Surface rumors are true after all.

As some have pointed, the "wintel" days are over and both of them are now trying to save their asses, Intel going Android + Linux (now they lunch compilers and drivers weekly), and MS going ARM.
Who would thought of that just a few years back?

Intel promotes Android convertible notebooks, say vendors
s: http://www.digitimes.com/news/a20130419PD208.html

Viewing that Windows 8 has been unable to stimulate global demand for notebooks, and global sales of Android tablets have been increasing, Intel has begun to promote Android tablet-convertible notebooks,...

If Intel is promoteing Android/linux in future phones/tablets, to lower prices (of devices) to keep profits up at lower sales prices.....

Maybe it doesnt seem so far out, that AMD ends up in the Mircosoft Surface tablets.

 

[sontin]

While Windows RT might technically be "Windows", it is missing the one feature that Windows has that nothing else does - compatibility with x86 Windows applications (without WINE). Without that it doesn't really have any advantage over Linux/Android/iOS/everything else.

But yet legacy software is not running well on Atom, Temash and co.
So i dont see it as an advantage.

 

[Arkadrel]

But yet legacy software is not running well on Atom, Temash and co.
So i dont see it as an advantage.

being ABLE to run something > not being able to run something.

Also alot of things run well, even on something as slow as E-350:

watch this video:
http://www.youtube.com/watch?v=wsbKYr7sD8o


@Tralalak
http://uk.hardware.info/reviews/270...vs-amd-e-450-benchmarks-integrated-gpu-3dmark

A E-450 (CPU:1.65ghz x2, GPU: 80 x 508-600mhz, DDR3-1333) (18w TPD)
scores around ~2430 in 3Dmark 06.

The E-450 is a bobcat at 40nm.
Temash is gonna up GPU performance by alot.

 

[Vesku]

I was pretty impressed with ~$250 C-50 netbooks, much more responsive than I expected prior to use. With Intel's aggressive "$200" pricepoint declaration I'm going to keep my eyes out for the kabini versions that should be showing up. Assuming Kabini launch is much improved over Trinity launch and we actually see decent options.

 

[Olikan]

AMD Jaguar (16h) Software Optimization Guide is now publicly available:

http://support.amd.com/us/Processor_TechDocs/52128_16h_Software_Opt_Guide.zip

i took some time to read it...

1st thing, THIS IS THE BEST GUIDE EVER... so clear, easy to read...

well, one think to note....
>L$2 cache latency is now a shitty 25 cycles (was 17), but bandwidth skyrocked and is larger 2 mb....seems like a L$2.5 cache :awe:

 

[NTMBK]

i took some time to read it...

1st thing, THIS IS THE BEST GUIDE EVER... so clear, easy to read...

well, one think to note....
>L$2 cache latency is now a shitty 25 cycles (was 17), but bandwidth skyrocked and is larger 2 mb....seems like a L$2.5 cache :awe:

L2.5$ sounds exactly right to me. Shared between cores like L3, L3 like size...

 

[Olikan]

L2.5$ sounds exactly right to me. Shared between cores like L3, L3 like size...

mmm....latencyes get lower with die shrinks?

if no... amd almost redone this chip, every damm latency seems lower

 

[Arkadrel]

http://www.amd.com/us/press-releases/Pages/amd_unveils_new_apus.aspx

The AMD A6-1450 APU-based system delivers scores of 981 and 3123 in 3DMark and PCMark respectively......
...the “Larne” reference platform with an AMD A6-1450 Quad Core 1.0GHz APU, AMD Radeon™ HD 8280 Series graphics, 2GB DDR3-1066 system memory and Microsoft Windows 7.

AMD A6-1450 (~9watt?) = 981 in 3Dmark Vantage.
AMD E-450 (~18watts TPD + 4watts FCH) = 904 in 3Dmark Vantage.


AMD managed to cut power in half (compaired to the bobcats ~18watts),
with the Temash and still beat the bobcat's score in 3D performance.

 

[sontin]

Nothing new. TSMC's 28nm process allows for much higher clocks than the 40nm process.

 

[AtenRa]

Why?

Intel has repeatedly gone on record that 22nm Atoms will be here this year, from microserver, to comms, to tablets, to PCs. Desperate times call for desperate measures, and you will indeed see 22nm Atom this year in full force.

http://www.dailytech.com/Leaked+Int...phone+Chip+Delayed+Till+2014/article29545.htm

Production Nov13-Jan 14
RTS Feb-April 14

I wouldn't expect any 22nm ATOM products in 2013 and even Q1 2014 will be far fetched if this road-map is valid.

 

[AtenRa]

Baytrail comes in the 2H of 2013. Temash has only a very short time before the first 22nm Atom SoCs will hit them.

See my post above,

Temash will have 6 to 9 months head start.

 

[mikk]

http://www.dailytech.com/Leaked+Int...phone+Chip+Delayed+Till+2014/article29545.htm

This Roadmap is for Valleyview-M only (and not confirmed by Intel). Valleyview-T launch window is between August-October 2013. For Temash, there is only a very short window before Valleyview-T is coming into tablets. We can't expect Temash tablets before August-September.

 

[Exophase]

Intel states 50-60% improvement with Silvermont (OoO,4CT/4T 2.1Ghz listed in slides) vs previous gen Atoms (in-order,2C/4T 1.5Ghz listed in slides). So that gives it about 7-15% per thread improvement once you normalize the clockspeed.

The cell in the chart is extremely ambiguous. The context is moving from 2C/4T to 4C/4T. To me this implies that the performance due to core scaling is 50-60% better on Silvermont, which is in-line with expectations of scaling due to HT on Saltwell and traditionally core scaling. Again, the context of the figure makes it

50-60% higher overall performance with a 40% increase in clock speed and a move from 2 cores with HT to 4 cores without would actually suggest a major drop in IPC. If we're talking about typical performance scenarios, HT on Atom will tend to give you somewhere around 40% better performance with two fully loaded threads where two full cores will have given you 100% better performance.

I think you'll agree that a substantial IPC decrease goes against everything else we've heard about it.

 

[Exophase]

L2.5$ sounds exactly right to me. Shared between cores like L3, L3 like size...

It doesn't matter, it's an L2 cache because it's the thing that comes immediately after the L1 cache. The name doesn't imply anything about how big it is or how much it's shared.. multi-MB shared L2 is pretty prevalent, all the ARM SoCs are doing it too.

Not sure what the 17 cycles for Bobcat referred to exactly, but it's probably only the L2's contribution to the miss - if you look at measurements you'll find the "full" latency is 20 cycles: http://www.7-cpu.com/cpu/Bobcat.html

For Jaguar AMD explicitly called out the 25 cycles as load-to-use so we know it includes L1 miss penalty, so the real comparison is 25 vs 20. And Jaguar's top clock speed went up (2GHz vs 1.7GHz) so the absolute latency is 12.5ns vs 11.76ns, not nearly as big of a difference. Still, it's a reasonable assumption that the latency got worse due to having larger shared caches vs smaller dedicated caches.

 

[inf64]

Of course, but I just linked the official intel slide presentation for new Atoms. I have no clue what that 50-60% figure actually is. It can be , for all we know, single thread performance jump (although it's a bit high for that;I'd reckon 20-25% ST IPC improvement is realistic).

 

[ShintaiDK]

http://www.dailytech.com/Leaked+Int...phone+Chip+Delayed+Till+2014/article29545.htm


Production Nov13-Jan 14
RTS Feb-April 14

I wouldn't expect any 22nm ATOM products in 2013 and even Q1 2014 will be far fetched if this road-map is valid.

Small hint, there are 3 types(actually 4 types.). That roadmap is only for 1 type.

 

[mikk]

Of course, but I just linked the official intel slide presentation for new Atoms. I have no clue what that 50-60% figure actually is. It can be , for all we know, single thread performance jump (although it's a bit high for that;I'd reckon 20-25% ST IPC improvement is realistic).

+20-25% IPC +40% higher clocks= easily within 50-60% per thread performance.