The Intel Atom Thread

[SlimFan]

"Not yet optimized with Android" would refer to stuff at a system level, various peripherals on the SoC, power management, maybe some aspects of GPU support and integration. It's not going to be about CPU performance. All that optimization they already did for x86 carries over there. Sunspider scores are so different because the browser is totally different and it's highly dependent on browser, this is nothing very new.

This probably also includes Dalvik optimizations, which could affect alleged "CPU benchmarks."

 

[Exophase]

This probably also includes Dalvik optimizations, which could affect alleged "CPU benchmarks."

Code scheduled well for Saltwell will run well on Silvermont. Dalvik JIT was probably not that scheduled well for Saltwell (or anything in-order), but Silvermont would only benefit minimally from typical compile-time scheduling. There simply aren't a bunch of critical uarch-level optimizations you want to do on Silvermont and whatever there are JITs will rarely do for any CPU.

That said, few benchmarks ran today on Android are Java and therefore Dalvik. Most are Javascript or native code.

 

[SlimFan]

Code scheduled well for Saltwell will run well on Silvermont. Dalvik JIT was probably not that scheduled well for Saltwell (or anything in-order), but Silvermont would only benefit minimally from typical compile-time scheduling. There simply aren't a bunch of critical uarch-level optimizations you want to do on Silvermont and whatever there are JITs will rarely do for any CPU.

That said, few benchmarks ran today on Android are Java and therefore Dalvik. Most are Javascript or native code.

From the Anandtech preview this thread discusses:

AndEBench is a combination native compiled microkernel benchmark (indicative of NDK application performance) that also runs a very similar workload atop Dalvik like a normal Android Java application. Here we can see what Intel was talking about when they said they have more work to do getting Dalvik working properly at dispatching threads to appropriate cores, hopefully the Java number will climb considerably. The native test also shows a lead over the competition.

There are lots of silly benchmarks out there, and I don't think all of them are native x86. Some people dismiss Javascript benchmarks ("they cheated in the browser!!") at which point they look at other so-called benchmarks.

 

[Nothingness]

There are lots of silly benchmarks out there, and I don't think all of them are native x86. Some people dismiss Javascript benchmarks ("they cheated in the browser!!") at which point they look at other so-called benchmarks.

Indeed all benchmarks deserved to be looked at. But one has to dismiss outliers.

AnTuTu is the perfect example: it performed very well on Intel phones, while other benchmarks didn't show the same. It turned out it was a compiler trick.

Another example is Sunspider: it performs very well on IE, but most other Javascript bechmarks run poorly on IE.

IMHO you need a very wide panel of benchmarks:
- browser: it's the app the most used
- native: you need to know how fast the processor is, and anyway there are very important native apps, such as Angry Birds
- Java: after all, most Android apps are written in Java.

This last point raises a concern: why is Dalvik so poor on Intel despite Intel having worked on Android for several years? There are traces of Intel commits in Dalvik starting in 2009. And as Exophase wrote, Silvermont should be much less sensitive to JIT tuning than previous Atom cores.

 

[SlimFan]

This last point raises a concern: why is Dalvik so poor on Intel despite Intel having worked on Android for several years? There are traces of Intel commits in Dalvik starting in 2009. And as Exophase wrote, Silvermont should be much less sensitive to JIT tuning than previous Atom cores.

Davlik is a managed runtime and has many pieces apart from instruction scheduling. I don't think the deltas shown between native and Davlik in the AndEBench results would be explained by instruction scheduling sensitivities. I'd wager a guess that you'd see those deltas on the previous Atom cores.

For whatever reason, things just seem to take Intel a long time...

 

[Nothingness]

For whatever reason, things just seem to take Intel a long time...

Indeed... They are pushing updates in gcc for AVX 512 support and can't seem to push updates in Dalvik to improve performance, that's odd. Or simply there's not much to gain for Silvermont.

 

[Exophase]

From the Anandtech preview this thread discusses:

I said MOST benchmarks, fully aware that there was one exception that is partially Dalvik.

The stuff about Dalvik on Silvermont having a problem dispatching threads to cores sounds very questionable to me, this system doesn't even have HT. There's no reason why Dalvik shouldn't be working with the kernel with business as usual to dispatch threads.

 

[Kallogan]

i will so buy a convertible bay trail sheet !!! Those fair prices and full Windows 8 are exciting

Transformer T100 (64GB model) up for pre-order @ Amazon: www.amazon.com/ASUS-Transformer-T100...ansformer+t100

Launch October 18, 11 hours battery life.

Very cute thing

 

[Kallogan]

Any bay trail announced with 11,6" ou 13,3" screen ???

 

[Shivansps]

Any bay trail announced with 11,6" ou 13,3" screen ???

No that i know off, but Dell Inspiron 11 3000 has a dual core Haswell Celeron at 379...

Ill like to see how BT performs compared to a 2955U... It looks like that a quad BT maybe faster on CPU but slower on gpu.

 

[cbn]

Looks like some new Bay Trail launched --> http://ark.intel.com/products/codename/55844/Bay-Trail

E3845, E3845, E3815, E3826, E3827 are the new SKUs. (A mix of quad core, dual core and single core.)

 

[CHADBOGA]

Looks like some new Bay Trail launched --> http://ark.intel.com/products/codename/55844/Bay-Trail

E3845, E3845, E3815, E3826, E3827 are the new SKUs. (A mix of quad core, dual core and single core.)

For the Embedded market.

 

[CHADBOGA]

That's a <3W SDP for Temash. Bay Trail won't touch it in graphics, and that's on a process node advantage lacking both SATA and PCIe.

What were you saying about Bay Trail, Temash & touching?

Looks like AMD are going to get touched up big time by Intel, again.

So decent CPU performance in some tests as expected combined with embarrassing GPU performance. They probably needed to castrate GPU to get decent power numbers.
Am I missing something or is this SoC missing SATA and PCIe?

Talking about embarrassing.

Bad News for AMD

http://www.fool.com/investing/general/2013/10/10/bad-news-for-amd.aspx

"In 3D Mark Ice Storm, a 3D graphics benchmark, AMD's tablet-oriented Temash scores 12,495, while Bay Trail-T has shown scores in excess of 13,000.

This means that AMD doesn't even have a graphics advantage"​

 

[monstercameron]

What were you saying about Bay Trail, Temash & touching?

Looks like AMD are going to get touched up big time by Intel, again.



Talking about embarrassing.

Bad News for AMD

http://www.fool.com/investing/general/2013/10/10/bad-news-for-amd.aspx

"In 3D Mark Ice Storm, a 3D graphics benchmark, AMD's tablet-oriented Temash scores 12,495, while Bay Trail-T has shown scores in excess of 13,000.

This means that AMD doesn't even have a graphics advantage"​

baytrail t z3770 is faster than a4-1200 in all but one area...gpu perf.

the z3770 scores 11000+ graphics subscore http://www.pcper.com/reviews/Proces...rail-and-Silvermont-Arrive/3DMark-Ice-Storm-a

the c-50 scores 12461 graphics subscore http://www.3dmark.com/is/187221

the a4-1250 [with 300Mhz gpu and 1333ram] scores 20000+ graphics subscore http://www.pcper.com/reviews/Proces...rail-and-Silvermont-Arrive/3DMark-Ice-Storm-a

so the a4-1200 with 275Mhz gpu and 1066Mhz ram will score a bit less than the a4-1250, I am estimating around 15K...

 

[Sweepr]

Wow, this has been discussed like 100x times. A4 1200 has a 225MHz GPU (less than half A4 5000's GPU performance) and two slow 1GHz Jaguar cores. Bay Trail-T has >3x faster CPU performance (beats 4 Jaguar cores @ 1.5GHz) and scores close to 13k @ 3DMark graphics. A massive CPU gap like this could completely offset a small GPU advantage (especially when both GPUs are very slow for Windows gaming) in actual games. It will be fun to compare Temash and Bay Trail-T Windows 8.1 tablet design wins. Just a few days till Windows 8.1 launch.

Interesting Bay Trail-T vs 3.9W Temash comparison @ CHADBOGA's link:

- AMD's Temash does not include an integrated Image Signal Processor, which is an IP block that controls and provides functionality for the built-in camera.
- Temash for tablets includes just two Jaguar processor cores running at 1GHz. Per core/per clock, Jaguar has some advantages over the Silvermont core in Bay Trail-T, yet the difference isn't close enough to make up for half the cores running at 42% of the frequency. Silvermont also has more aggressive power management.
- In 3D Mark Ice Storm, a 3D graphics benchmark, AMD's tablet-oriented Temash scores 12,495, while Bay Trail-T has shown scores in excess of 13,000. This means that AMD doesn't even have a graphics advantage.
- No tablet design wins have been announced in the 10" and smaller form factors.

 

[Arachnotronic]

baytrail t z3770 is faster than a4-1200 in all but one area...gpu perf.

the z3770 scores 11000+ graphics subscore http://www.pcper.com/reviews/Proces...rail-and-Silvermont-Arrive/3DMark-Ice-Storm-a

the c-50 scores 12461 graphics subscore http://www.3dmark.com/is/187221

the a4-1250 [with 300Mhz gpu and 1333ram] scores 20000+ graphics subscore http://www.pcper.com/reviews/Proces...rail-and-Silvermont-Arrive/3DMark-Ice-Storm-a

so the a4-1200 with 275Mhz gpu and 1066Mhz ram will score a bit less than the a4-1250, I am estimating around 15K...

Was that Temash in a tablet form factor, or the notebook form factor that it seems to be showing up exclusively in?

 

[ShintaiDK]

Its obvious when AMD now tries to fill tablets with a future ARM chip. That they know they are nowhere near tablets as it is today. Temash is slower and uses much more power.

The MSI tablet that was announced also seem to have died there.

The numbers from PC perspective is the laptop one.

And AMDs own numbers:

Even ARM seems to be in trouble and desperate over BT. They are sending out YouTube vidios now with ARM vs Clovertrail.
http://www.youtube.com/watch?v=Hs84SYNS7Qg&feature=player_embedded

Guess they know what the clock have struck.

 

[Abwx]

Wow, this has been discussed like 100x times. A4 1200 has a 225MHz GPU (less than half A4 5000's GPU performance) and two slow 1GHz Jaguar cores. Bay Trail-T has >3x faster CPU performance (beats 4 Jaguar cores @ 1.5GHz) and scores close to 13k @ 3DMark graphics. A massive CPU gap like this could completely offset a small GPU advantage (especially when both GPUs are very slow for Windows gaming) in actual games. It will be fun to compare Temash and Bay Trail-T Windows 8.1 tablet design wins. Just a few days till Windows 8.1 launch.

2C Temash is a 3.9W TDP chip , how much for the 3770 whose TDP
is not published , surely because it s so low that Intel better keep
it secret for marketing purposes , first time that a corporate wouldnt
scream in all roofs its chips superior TDP behaviour..

Actualy it is 7W TDP , much closer to a 8W 4C Temash than to a 2C ,
so much for your apple to potatoes comparisons.

 

[ShintaiDK]

Actualy it is 7W TDP

The mandatory request for documentation!

 

[sontin]

Baytrail for tablets is never a 7W TDP SoC. The highest tablet SoC is Tegra 4 (A15) with 5W and even this is maybe to high.

 

[Abwx]

The mandatory request for documentation!

Perhaps you , specialy , could help in this matter.??.

Baytrail for tablets is never a 7W TDP SoC. The highest tablet SoC is Tegra 4 (A15) with 5W and even this is maybe to high.

At base frequency TDP is undoubtly quite low but
at turbo frequencies like 2.2-2.4GHz , moreover with
all cores fired , it s unlikely that it s less than 7W , as
aknowledged by their other BT offerings.

Of course they could use only a single memory channel
and reduce a few frequencies here and there but it wont
cut TDP dramaticaly , not counting that the perfs would
be different from the recent Intel controlled reviews.

 

[ShintaiDK]

Perhaps you , specialy , could help in this matter.??.

So you just make up random fake numbers up again? Why I am not surprised.

 

[Abwx]

So you just make up random fake numbers up again? Why I am not surprised.

If based on existing BT specs it s neither random nor fake ,
but then if TDP was so great why no official number for
a product that is supposed to be already in the market.??.

Would you expect a manufacturer to hide his product specs
if they were that superior to the competition..??.

 

[Sweepr]

2C Temash is a 3.9W TDP chip , how much for the 3770 whose TDP
is not published , surely because it s so low that Intel better keep
it secret for marketing purposes , first time that a corporate wouldnt
scream in all roofs its chips superior TDP behaviour..

Actualy it is 7W TDP , much closer to a 8W 4C Temash than to a 2C ,
so much for your apple to potatoes comparisons.

Bay Trail-M with higher base frequency, GPU clock, PCIe + SATA has 7.5W max TDP. Bay Trail-T is definitely lower than that, also, we already have BT-T power numbers in case you missed: 2.5W with 4 cores @ load. How much power a 15W TDP A4 5000 @ CPU load uses to deliver inferior CPU performance (vs Z3770)? Quite a bit more I guess. And you will never see any of these 8W-15W APUs in the small/thin 8-10'' tablets BT-T will find its place.

 

[Abwx]

Bay Trail-M with higher base frequency, GPU clock, PCIe + SATA has 7.5W max TDP. Bay Trail-T is definitely lower than that, also, we already have BT-T power numbers in case you missed: 2.5W with 4 cores @ load.
How much power a 15W TDP A4 5000 @ CPU load uses to deliver inferior CPU performance (vs Z3770)? Quite a bit more I guess. And you will never see any of these 8W-15W APUs in the small/thin 8-10'' tablets BT-T will find its place.

In wich review of an actual product can we find thoses BT numbers.?.:sneaky:

Actualy nowhere...