The Intel Atom Thread

Page 155 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
My guess is that eventually MS will allow conventional x86 desktop apps to be run on Windows Phone 10 without needing Continuum.

never gonna happen, not only because it's pointless since they would be unusable on a phone, but also because then it would be easy as cake for crackers to break any kind of security on the phone

If im not mistaken every NEW application made for windows 10 will be compatible for both Desktop and Phone versions.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
Seams that Single Thread performance of Cherrytrail is still lower than AMD Mullins.

Cinebench 11.5 Single Thread

A10 Micro-6700T = 0.55

Z8700 = 0.46


And 3D Mark 11 Performance is equal

A10 Micro-6700T R6 (Mullins) = 582

Z8700 = 562

And just to point out Mullins score is most probable with the old 14.1 Catalyst.

Now although iGPU performance gains are very nice, for a 14nm SoC I strongly believe Intel intentionally castrated CheryTrail to make Broadwell Core M look better.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Not surprising giving that it's running at the same clocks as BT. But MT performance went up due to higher sustained Turbo clocks.

A10 Micro-6700T = 1.49

Z8700 = 1.79

Also Mullins graphics performance seems to differ from preview to preview, I found much lower 3DMark 11 scores here.
It would be interesting to see Mullins performance outside of this AMD reference platform and how much power it's using compared to 14nm CT. Too bad AMD will replace it with K12 ARM parts next year.

Note that CT's drivers could still be unoptimized too, there's reports of some inconsistencies in some benchmarks.
 

scannall

Golden Member
Jan 1, 2012
1,944
1,638
136
Seams that Single Thread performance of Cherrytrail is still lower than AMD Mullins.

Cinebench 11.5 Single Thread

A10 Micro-6700T = 0.55

Z8700 = 0.46


And 3D Mark 11 Performance is equal

A10 Micro-6700T R6 (Mullins) = 582

Z8700 = 562

And just to point out Mullins score is most probable with the old 14.1 Catalyst.

Now although iGPU performance gains are very nice, for a 14nm SoC I strongly believe Intel intentionally castrated CheryTrail to make Broadwell Core M look better.

You could be right about Intel dragging down Cherry Trail. Apple's A8X smokes both of them in the same power envelope.

A10-6700t
Z8700
A8X

The Core-M is over $200 each. And does perform better. But as expensive as they are it makes no sense to use them in tablets for example.

Core-m 5Y71
 

jj109

Senior member
Dec 17, 2013
391
59
91
Seams that Single Thread performance of Cherrytrail is still lower than AMD Mullins.

Cinebench 11.5 Single Thread

A10 Micro-6700T = 0.55

Z8700 = 0.46


And 3D Mark 11 Performance is equal

A10 Micro-6700T R6 (Mullins) = 582

Z8700 = 562

And just to point out Mullins score is most probable with the old 14.1 Catalyst.

Now although iGPU performance gains are very nice, for a 14nm SoC I strongly believe Intel intentionally castrated CheryTrail to make Broadwell Core M look better.

Why are you comparing Mullins to Cherrytrail.

Mullins is basically a boost-based Beema, just like how Core M is a boost based Core ULV. There's a reason why the only Mullins tablets we've seen are metal-backed and 11mm thick.
 

lalla521

Junior Member
Mar 5, 2015
24
0
0
i believe that intel is trying desperately to be relevant in the mainstream/value market and get profits out of it.

i don't think they care in the slightest about actually competing at the high end/winning the benchmark game, that's why atom is still a pretty poor cpu.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
What would be the point of that? Run apps with unusable UI?
It makes more sense to make them run only on a big display with mouse and keyboard.

As I understand it Continuum is way of running runs apps on Windows 10 phone with big screen and keyboard/ mouse, but these apps must come from the MS App store.

Trouble is I don't think every x86 desktop developer is going to put their program in the MS App store. (ARM apps would be different because there are no exiting Windows desktops apps using that ISA)
 
Last edited:

cbn

Lifer
Mar 27, 2009
12,968
221
106
At 1:36 into this Windows 10 Continuum Video Joe Belfiore mentions "and with new parts from Qualcomm we have processors that can drive two screens separately in these new phone devices"

Do we know if Intel Moorefield is able to drive two screens separately?
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Seams that Single Thread performance of Cherrytrail is still lower than AMD Mullins.

Cinebench 11.5 Single Thread

A10 Micro-6700T = 0.55

Z8700 = 0.46


And 3D Mark 11 Performance is equal

A10 Micro-6700T R6 (Mullins) = 582

Z8700 = 562

And just to point out Mullins score is most probable with the old 14.1 Catalyst.

Now although iGPU performance gains are very nice, for a 14nm SoC I strongly believe Intel intentionally castrated CheryTrail to make Broadwell Core M look better.

Thats just the thing, if i compare that to this
http://laptopmedia.com/review/intel-pentium-n3700-review-the-budget-braswell-family-is-here/

It means that the Z8700 igp whould be around the same of an A10-6700T, and thats kinda good.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
Thats just the thing, if i compare that to this
http://laptopmedia.com/review/intel-pentium-n3700-review-the-budget-braswell-family-is-here/

It means that the Z8700 igp whould be around the same of an A10-6700T, and thats kinda good.

Being around Mullins iGPU in synthetics it not the same as real gaming performance.
Haswell HD4600 can keep close to Kaveri 7600@45W TDP in 3D Mark Cloud Gate, but in real games the difference is chaotic.

23mtsoi.jpg


azffrt.jpg
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Intel Bay Trail & Cherry Trail CPUs To Run Faster With Linux 4.1

image.php


While the big ACPI and power management changes were sent in more than one week ago for the Linux 4.1 kernel, another batch of ACPI/PM work was sent in this weekend. One of these last-minute changes should end up significantly benefiting the performance of modern Intel Atom Bay/Cherry Trail SoCs.

Of the few last minute ACPI/PM 4.1 changes is a tweak to the Intel P-State driver that changes the setpoint for Intel Atom hardware. With dropping the setpoint from 97 to 60, the P-State CPU frequency scaling driver will be much more aggressive in changing (increasing) the power/performance states when encountering load. Intel's Kristen Carlson Accardi says this change "improves performance on a variety of workloads with little power penalty." Unfortunately, no benchmarks were published, but I'll work on some new Bay Trail numbers once the Linux 4.1 kernel is further along.

This second ACPI/PM pull request for Linux 4.1 also adds Intel Skylake processor support to the RAPL power-capping driver.

www.phoronix.com/scan.php?page=news_item&px=Intel-Bay-Cherry-Faster-4.1

I interpret the linked article by Alvaraz Research as arguing that mid range Atom phone chips provide a gaming platform superior to that of the competition. http://seekingalpha.com/article/312...2ir:1ccd6a61a68016bf6838aad6ebcc37e6&uprof=46. I might quibble some of his theory, but his in hand perspective on overheating vs the competition during expended gaming is the stuff to move market share in Asia.

As far as performance is concerned Moorefield is a much better mid-range chip than Snapdragon 615. There's a significant single-thread performance advantage compared to the 8-core Cortex A53 SoC and twice the graphics performance (27-30 FPS vs 15 FPS @ GFXBench T-Rex, 12-13 FPS vs 5.7 FPS @ GFXBench Manhattan). ASUS Zenphone 2 and the new Lenovo phone are very competitively priced too.

Things will change in H2-2015 with the new Cortex A72/A53 mainstream SoCs. I hope the new Goldmont cores deliver competitive performance and efficiency next year.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Being around Mullins iGPU in synthetics it not the same as real gaming performance.
Haswell HD4600 can keep close to Kaveri 7600@45W TDP in 3D Mark Cloud Gate, but in real games the difference is chaotic.

Too bad you don't have a single review of an actual A10 Micro-6700T tablet/convertible to compare then. Maybe because OEMs realized that the performance, power consumption and battery life wouldn't look so great inside actual thin and light 7-10'' products we usually find Bay Trail-T (soon Cherry Trail-T) as it does inside the thick and ugly AMD 11.6'' reference platform with hidden power consumption numbers?

Blame contra-revenue 24/7 but there's still a market for better performing and more expensive tablets/2-in-1s, that's why Core-M (Broadwell-Y) exists. Hopefully AMD will do a better job with their future ARM-based tablet APUs.

AMD is pushing ARM Cortex A57 and K12 parts to tablets now, perhaps you should let Mullins RIP. :p
 
Last edited:

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Being around Mullins iGPU in synthetics it not the same as real gaming performance.
Haswell HD4600 can keep close to Kaveri 7600@45W TDP in 3D Mark Cloud Gate, but in real games the difference is chaotic.

23mtsoi.jpg


azffrt.jpg

its far better than being 50% or more behind like BT is.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
its far better than being 50% or more behind like BT is.

Actually BT is more like 100% or more behind but yes it is better than BT. Problem is that its coming 1.5 years later and will have to compete against 20nm ARM products in H2 2015.
 

Enigmoid

Platinum Member
Sep 27, 2012
2,907
31
91
Being around Mullins iGPU in synthetics it not the same as real gaming performance.
Haswell HD4600 can keep close to Kaveri 7600@45W TDP in 3D Mark Cloud Gate, but in real games the difference is chaotic.

23mtsoi.jpg


azffrt.jpg

IMO this shows how badly tuned intel's drivers are. I have no doubt that the actual intel graphics hardware is pretty decent (along the lines of 3d mark scores which all vendors optimize heavily for) but the drivers absolutely suck. Intel doesn't do performance or game optimizations and as a result performance in games tanks.
 

jj109

Senior member
Dec 17, 2013
391
59
91
Too bad you don't have a single review of an actual A10 Micro-6700T tablet/convertible to compare then. Maybe because OEMs realized that the performance, power consumption and battery life wouldn't look so great inside actual thin and light 7-10'' products we usually find Bay Trail-T (soon Cherry Trail-T) as it does inside the thick and ugly AMD 11.6'' reference platform with hidden power consumption numbers?

Blame contra-revenue 24/7 but there's still a market for better performing and more expensive tablets/2-in-1s, that's why Core-M (Broadwell-Y) exists. Hopefully AMD will do a better job with their future ARM-based tablet APUs.

AMD is pushing ARM Cortex A57 and K12 parts to tablets now, perhaps you should let Mullins RIP. :p

Mullins fell into the awkward gap where they had to bin enough Beema to get ones that would outperform BT significantly, but the parametric yields had to be high enough so that Mullins didn't get too close to Core M in price (and avoid to oversupply in Beema as well).
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
IMO this shows how badly tuned intel's drivers are. I have no doubt that the actual intel graphics hardware is pretty decent (along the lines of 3d mark scores which all vendors optimize heavily for) but the drivers absolutely suck. Intel doesn't do performance or game optimizations and as a result performance in games tanks.

Cloud Gate is not that GPU intensive like Skydiver and especially Fire Strike that is why HD4600 can keep up with Kaveri in this one. No matter how optimized Intel drivers could be for games, the performance capability of the Haswell iGPU is way bellow even against Trinity.
 

Enigmoid

Platinum Member
Sep 27, 2012
2,907
31
91
Cloud Gate is not that GPU intensive like Skydiver and especially Fire Strike that is why HD4600 can keep up with Kaveri in this one. No matter how optimized Intel drivers could be for games, the performance capability of the Haswell iGPU is way bellow even against Trinity.

Even in fire strike intel graphics generally does much better than in games.

21n4q68.jpg



When does the i3 OC ever catch llano OC in games?

Rarely.

3dmark generally gets around game optimization problems in intel's drivers. Its more indicative of architectural performance but not indicative of real world performance.
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
OK, but on Fire Strike Graphics they end up about the same, yeah the Z8700 is 100mhz slower so it will probaly end up slightly behind.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
Even in fire strike intel graphics generally does much better than in games.

21n4q68.jpg



When does the i3 OC ever catch llano OC in games?

Rarely.

3dmark generally gets around game optimization problems in intel's drivers. Its more indicative of architectural performance but not indicative of real world performance.

Haswell i3 OC in F1 2014 and Metro Last Light at 720p is very close to Llano OC according to my results.
But Llano is VLIW5 architecture and current software like 3D Mark may not been optimized for those older mArchs.
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Anadtech just did the review of the Surface 3 2GB model.... i think its running SC memory.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Anadtech just did the review of the Surface 3 2GB model.... i think its running SC memory.

Likely, too bad they didn't get to review the 4GB RAM version like NotebookCheck did, that would probably improve the graphics scores. It's still >2x as fast as Bay Trail-T's iGPU though.

A few interesting points:

Cherry Trail-T delivers great battery life (normalized battery capacity). I wish there were power consumption tests.

74145.png


74160.png


I am aware that some of those test results vary a lot depending on the browser but it's still impressive to see how competitive 4 small x86 ~1mm² Airmont cores can be. I'm confident that they will be able to deliver outstanding results with future architectures (Goldmont next year).

74100.png


74101.png


74102.png


74105.png
 
Last edited:

Enigmoid

Platinum Member
Sep 27, 2012
2,907
31
91
Likely, too bad they didn't get to review the 4GB RAM version like NotebookCheck did. That would probably improve the graphics scores.

A few interesting points:

Cherry Trail-T delivers great battery life (normalized battery capacity). I wish there were power consumption tests.

I am aware that some of test results vary a lot depending on the browser but it's still impressive to see how competitive 4 x86 in-order ~1mm² Airmont cores can be. I'm confident that they will be able to deliver an outstanding results with future architectures (Goldmont next year).

CT is out of order.
That said it is a significantly narrower core design than the competition.