CherryTrail-T information.

[ShintaiDK]

http://chinese.vr-zone.com/90655/in...ith-14nm-process-at-idf-2014-autumn-11142013/

Enhanced core, Gen8 GPU, higher speed, faster memory, more memory, 14nm.

 

[liahos1]

it is also my understanding that Airmont represents both a Tick and Tock. so we should see some performance gains in IPC....

 

[Homeles]

it is also my understanding that Airmont represents both a Tick and Tock. so we should see some performance gains in IPC....

It's a tick.

 

[Arachnotronic]

Spoke with Silvermont's lead architect at IDF. Do not make the mistake of applying "tick" and "tock" here...Airmont will bring non-trivial micro-architectural enhancements, although Goldmont should be a bigger bang.

 

[liahos1]

i guess its a question of intel's priorities with cherry trail. If you assume the 30% baseline reduction in perf/watt when comparing haswell to broadwell applies to baytrail --> cherrytrail, does intel increase the perf keeping watt constant, do they keep performance constant and lower power or a mix inbetween. I just wonder if broadwell is going to be hitting thermal envelopes appropriate for tablets next year, what does that mean for atom? Maybe broadwell offers a much superior perf/watt solution relative to cherrytrail and you just end up paying a premium for it

 

[Arachnotronic]

i guess its a question of intel's priorities with cherry trail. If you assume the 30% baseline reduction in perf/watt when comparing haswell to broadwell applies to baytrail --> cherrytrail, does intel increase the perf keeping watt constant, do they keep performance constant and lower power or a mix inbetween. I just wonder if broadwell is going to be hitting thermal envelopes appropriate for tablets next year, what does that mean for atom? Maybe broadwell offers a much superior perf/watt solution relative to cherrytrail and you just end up paying a premium for it

Broadwell will offer better performance in 10" tablets, but anything below that is Atom's turf. Atom is designed to not exceed ~2.5W even at full load in tablets. Broadwell-ULT is probably more like 5-6W.

Also, note that the Atoms have IP blocks that Broadwell doesn't like an image signal processor, camera, audio, etc. Much more suitable for real tablets.

 

[Lepton87]

The increase in max burst speed isn't impressive, let's hope it brings IPC improvements. 10% or higher would be good. Combined that would bring 20% performance improvement and that's certainly noticeable.

 

[liahos1]

The increase in max burst speed isn't impressive, let's hope it brings IPC improvements. 10% or higher would be good. Combined that would bring 20% performance improvement and that's certainly noticeable.

Do you think a 20% improvement is going to cut it? I would think they'd need more than that to keep a lead on the v8 soc's coming out in late 2014(?) or in 2015.

 

[escrow4]

Battery life? Currently the Clover Trail tablet I have hits 8hrs or so reading ebooks via Calibre (and no, I hate slow gimped ebook readers). Worth upgrading to Cherry Trail if it cracks 20hrs.

 

[Sweepr]

Impressive bandwidth increase. I smell substantial GPU performance improvements coming.

 

[sm625]

The fastest 4 core Z37xx atom only scores < 1200 in passmark. A Q6600 underclocked to 1600MHz would still blow it away. It's not even close to competitive. Two 1.2GHz haswell cores would also destroy it. So it would be pointless for them to reduce power further. They need to increase max power, reduce idle power, and aim for increased performance overall rather than reduced power overall.

They're not going to catch up to apple so they are going to get their chips in any apple products. And android is a sea of the lowest bidder so they arent going to break any major ground there. And x86 is dead because windows is so ancient and antiquated and... just bad.

 

[SlimFan]

The fastest 4 core Z37xx atom only scores < 1200 in passmark. A Q6600 underclocked to 1600MHz would still blow it away. It's not even close to competitive. Two 1.2GHz haswell cores would also destroy it. So it would be pointless for them to reduce power further. They need to increase max power, reduce idle power, and aim for increased performance overall rather than reduced power overall.

They're not going to catch up to apple so they are going to get their chips in any apple products. And android is a sea of the lowest bidder so they arent going to break any major ground there. And x86 is dead because windows is so ancient and antiquated and... just bad.

...Passmark? Would that be the end-all, be-all of benchmarks? "catch up to Apple" ... what do you mean? Are you implying that the Apple A7 is faster than a Bay Trail? I'm not sure I've seen a set of comparable benchmarks that say that. Is there a "passmark" score for Apple's CPU?

x86 can run Android too, right? Why does the age of Windows make x86 dead?

 

[Nothingness]

...Passmark? Would that be the end-all, be-all of benchmarks? "catch up to Apple" ... what do you mean? Are you implying that the Apple A7 is faster than a Bay Trail? I'm not sure I've seen a set of comparable benchmarks that say that. Is there a "passmark" score for Apple's CPU?

Did you miss Geekbench or do you dismiss it for some reason?

http://browser.primatelabs.com/geekbench3/compare/207541?baseline=177323

 

[Khato]

Did you miss Geekbench or do you dismiss it for some reason?

http://browser.primatelabs.com/geekbench3/compare/207541?baseline=177323

Indeed. Even once the AES, SHA1, and SHA2 hardware-accelerated results are ignored the 1.4 GHz A7 is anywhere from comparable to 1.5x faster than Baytrail in the integer tests. Then it's typically more like 1.5-2x faster when it comes to floating point.

But what was everyone expecting when you consider that a single A7 core draws as much power as four Silvermont cores?

 

[mikk]

Did you miss Geekbench or do you dismiss it for some reason?

http://browser.primatelabs.com/geekbench3/compare/207541?baseline=177323

Different OS, not comparable.

 

[Nothingness]

But what was everyone expecting when you consider that a single A7 core draws as much power as four Silvermont cores?

I'd like someone to run a program similar to what Anand ran on A7. It's close to a power virus. I bet BT will get way above 2.5W

 

[Nothingness]

Different OS, not comparable.

So one never will be able compare A7 against Bay Trail, how convenient

 

[Khato]

I'd like someone to run a program similar to what Anand ran on A7. It's close to a power virus. I bet BT will get way above 2.5W

I'm talking about the numbers he obtained running Kraken. That's a single-core well-known benchmark that's meant to emulate usage, not a power virus... and the iPad Air was still showing at least a 3W delta compared to idle. (The power virus was closer to a 3.75W delta.)

But I'm definitely with you in wanting to see comparable measurements done on Baytrail-T. Commented as such on the iPad Air review. And I wouldn't be too surprised if we do see such measurements at some point in the future.

 

[Nothingness]

I'm talking about the numbers he obtained running Kraken. That's a single-core well-known benchmark that's meant to emulate usage, not a power virus... and the iPad Air was still showing at least a 3W delta compared to idle. (The power virus was closer to a 3.75W delta.)

Ha right I had missed that Kraken graph Thanks!

But I'm definitely with you in wanting to see comparable measurements done on Baytrail-T. Commented as such on the iPad Air review. And I wouldn't be too surprised if we do see such measurements at some point in the future.

As long as it's not setup by Intel...

 

[dahorns]

Indeed. Even once the AES, SHA1, and SHA2 hardware-accelerated results are ignored the 1.4 GHz A7 is anywhere from comparable to 1.5x faster than Baytrail in the integer tests. Then it's typically more like 1.5-2x faster when it comes to floating point.

But what was everyone expecting when you consider that a single A7 core draws as much power as four Silvermont cores?

Also keep in mind that this particular test is comparing the 64 bit A7 and the 32-bit Bay Trail. The A7 received a huge boost at 64-bit compared to its 32-bit scores, ~%50 increase. We haven't had the opportunity to see Bay Trail at 64-bit yet, even though the chip is capable. (Check Anand's review of the 5s for the numbers for the A7).

I'm not sure there is any reason to expect a 50% increase in the scores, but theoretically some increase should happen.

 

[Ajay]

http://chinese.vr-zone.com/90655/in...ith-14nm-process-at-idf-2014-autumn-11142013/


Enhanced core, Gen8 GPU, higher speed, faster memory, more memory, 14nm.

When is it due in shipping products?

The increase in max burst speed isn't impressive, let's hope it brings IPC improvements. 10% or higher would be good. Combined that would bring 20% performance improvement and that's certainly noticeable.

That's what I thought, especially when, in another thread, it looked like Airmont would have 16 EU - lots of hungry mouths to feed (compared to 4 in Silvermont).

 

[Nothingness]

Also keep in mind that this particular test is comparing the 64 bit A7 at and the 32-bit Bay Trail. The A7 received a huge boost at 64-bit compared to its 32-bit scores, ~%50 increase. We haven't had the opportunity to see Bay Trail at 64-bit yet, even though the chip is capable. (Check Anand's review of the 5s for the numbers for the A7).

I'm not sure there is any reason to expect a 50% increase in the scores, but theoretically some increase should happen.

If you remove the AES and SHA scores the speedup over 32-bit for the A7 is due to vectorization which already happens in 32-bit for Bay Trail. It looks like for 4770K the speedup is about 10%: http://browser.primatelabs.com/geekbench3/compare/204522?baseline=204623

Also more generally, going to 64-bit often has a negative impact due to increase in pointer sizes which impacts cache footprint. This can be seen in some SPEC benchmark results where some of the tests are compiled as 32-bit executables.

 

[Khato]

As long as it's not setup by Intel...

Haha, while I don't mind the Intel-supplied setups I can understand why some would. The impression I got from little tidbits Anand has said (I think mostly on the recent video) was that he was going to dig into it a bit more once he received an iPad Air that he could take apart to instrument.

Either way, it's only going to get more interesting from here. Baytrail is arguably the most efficient low-power processor available while beating or matching its competition. The A7 bests it, but I'd argue that it's straddling the divide between Baytrail and Haswell. (Take a look at notebookcheck's review of the HP Spectre 13 and note that it's i5-4202Y is quite content in a fanless chassis, has power consumption figures that appear to be comparable to or a touch higher than A7, and given that its performance is roughly 75% that of the i5-4200U would still handily beat the A7 in geekbench.) With Cherrytrail we're going to get a marked bump in CPU performance and 2.5x the GPU performance or more... while Broadwell will bring a 30% reduction in power compared to Haswell at the same performance. Intel certainly appears to be setting a pretty decent bar for the competition.

 

[Enigmoid]

Did you miss Geekbench or do you dismiss it for some reason?

http://browser.primatelabs.com/geekbench3/compare/207541?baseline=177323

Indeed. Even once the AES, SHA1, and SHA2 hardware-accelerated results are ignored the 1.4 GHz A7 is anywhere from comparable to 1.5x faster than Baytrail in the integer tests. Then it's typically more like 1.5-2x faster when it comes to floating point.

But what was everyone expecting when you consider that a single A7 core draws as much power as four Silvermont cores?

Geekbench is crap.

Anyway, the A7 is nowhere near as good as its believed to be. From the 3dmark physics problems (unable to handle random memory access any faster than A6) to Apple's software optimization.

Lets look at the iphone 4s for example.

4s- Dual core A9 at 800 mhz.
Pick an android phone for example, samsung galaxy S2 - dual core A9 at 1.2 ghz.

You would expect the S2 to be ~50% faster would you not.

Nope.

Iphone 4s is 66% faster. Iphone 4 (single core A8 at 800 mhz) is almost as fast.

Iphone 4s is 59% faster. Again iphone 4 holds strong.

Look at how well the 3GS is doing. JS benchmarks are horrible indications of performance.

Hell, lets even look at the iphone 5

Iphone 5 IOS6, release date.

1 year later, IOS 7

Hmm... Somehow performance has improved 71% as a function of software.

On the GPU side while the iphone 4/4s commanded an impressive lead over competeting android SOCs (often 2x). The S800 is pretty much dead even with the A7 rogue GPU.

The differing Ecosystem makes any comparison difficult to do. Even so past tests have shown that apple can somehow improve performance on its SOC's be almost a factor of two (50%+ performance gain at only 66% the clockspeed) on the same architecture. Is the A7 a boss? Maybe, but probably not. IMO, a S800 in an iphone would probably demolish the A7.

 

[Nothingness]

4202Y indeed is faster: http://browser.primatelabs.com/geekbench3/compare/207653?baseline=190269

But that's what I'd expect for a 2 GHz Haswell I find the battery results good but not great compared to iPad. Of course that's a x86 and hence it runs Windows for those who need it. Still too expensive for me