Intel ULP SoCs

Actually it won't.

http://arstechnica.com/gadgets/2014/11/yoga-pro-3-review-broadwell-is-a-mixed-blessing/

GFXBench 3.0 T-Rex Offscreen:
HD 4400: 68
HD 5300: 45

68 fps on the HD 4400 is what the iPad Air 2 and Nvidia Shield does right now. HD 5300 has 24 EUs with 4.5W TDP while Bay Trail/Cherry Trail is at 16 EUs with probably 3W TDP. Recent Intel presentations say Cherry Trail is ~2x Bay Trail(Early leaks: 35 fps). It looks like the competitive advantage for CT isn't any better than BT was, if anything it might be worse.

 

Something interesting about Baytrail is that the Z3770 has significantly lower GPU performance than some lower end models like Z3740D. I have Dell Venue 8 Pro and HP Omni 10. I felt like exploring GPU performance differences between the two and ran 3DMark2001 and 3DMark2003 on both. I expected the Z3770 with its dual channel memory and inherent "high end model" status to win by a good bit but instead it posted scores in both benchmarks that were about 17% lower. I looked at some online reviews of Z3770 machines and saw a few noting this behavior.

My guess is the Z3770's higher CPU clocks are significantly reducing GPU turbo. These SOCs have the same SDP rating. I can't figure out a way to control CPU clocks to test this.

GPU performance of tablets is complex. There also tend to be thermal throttling issues with long term gaming with many tablets.

 

Power Options -> Change Plan Settings -> Change advanced power settings -> Processor Power Management -> Maximum Processor State
is unavailable?
To test which states (percentages) have a meaning, I usually just run CPUZ, load a single core to 100%, and watch the clock while decreasing maximum state in 5% steps. Minimum state may not increase the clock because of C-states. Also, there probably is a better way to retrieve those percentages.

 

Normally the Windows tablets hide most of the power options and schemes but I found a registry setting to enable the maximum processor state control. However testing with 90% CPU (1333MHz is the result) just slowed the CPU score results and did nothing for the 3D tests.

I also tried running HDMI at 1024x768. The tablet screen doesn't allow centered, non-scaled output. I wanted to see if scaling 1024x768 to 1920x1200 was imparting overhead but that didn't affect the score either.

I also tried the original HP Intel HD driver and the latest Intel HD driver from Intel's site. No change.

So I haven't a clue why the IGP is slower. Maybe it's firmware related.

For reference the Venue 8 Pro scores around 4700.

 

Another idea that probably won't work
http://www.cpuid.com/softwares/hwmonitor.html
On most Sandy/Ivy/Haswell systems, if MEI drivers are installed, this is usually capable of showing the current CPU/GPU/Uncore power. However, I don't know if it works on BayTrail.

 

HWMonitor works. Interesting results too.

-The DV8P package power limit hits up to 2.4W sometimes. The HP Omni stays very close to 2.0W.
-The DV8P has two cores locked to 1.33GHz and two turboing up to 1866MHz. This is a change Dell made in later BIOS revs perhaps to improve battery. Max IA core power consumption is a mere 0.14W in Prime95!
-The Omni10 turbo all cores to 2.4GHz and IA core power pushes the package power to the limit of 2.0W.
-With ATITOOL's spinning fur cube running alone, both tablets display a similar average framerate at maximum GPU utilization. Z3740D runs 660MHz GPU while Z3770 runs 666MHz.
-Combining ATITOOL with Prime95 shows 2.30-2.40W with 0.14W(IA) and 68fps avg on DV8P. 2.0W with 0.96W(IA) and 48fps avg on Omni.

 

Lots of news today:
http://www.anandtech.com/show/9026/...rated-lte-atom-renaming-and-14nm-cherry-trail


At the bottom of the Atom line they now have SoCs with and Intel CPU and.. an ARM GPU. I guess these are going against the rock-bottom Cortex A7 SoCs from MediaTek, since the GPUs are pretty low end. There's a Rockchips model too.

 

How bad must the Intel/IMG relationship become, that it results in Intel looking to ARM for graphics for its cost sensitive products. The anandtech piece implies that only one of these chips is a rockchip chip. Is that a branding thing, i.e. intel sells two of them, and rockchip sells 1, and if so, have they all fundamentally been designed mostly by rockchip ?

 

Well, all we do know as fact is that none of the Atom line announced today have any IMG graphics, and 3 of these "atoms" have Mali graphics.

Assuming the anandtech is correct and Intel will market 2 of the 3 chips as Intel chips, then from a PR point of view it's pretty disastrous as it effectively appears as intel dumping IMG and replacing with Mali. Having a policy of increasingly using your own graphics instead of 3rd party is one thing, switching for another 3rd party supplier after 10 years has an entirely different "smell".

 

FWIW, there's a huge gap in Intel's offerings from today's news. They only have rock-bottom to upper-bottom Atom X3 SoCs (the fastest Atom X3 with LTE has a worse GPU than MediaTek's MT6732).

That said, Intel hasn't shown or mentioned anything between the low-end smartphone SoCs and the BayTrail successors for tablets and low-cost PCs. This Atom X3 platform is a new market for Intel (ultra-low cost?) and the X5/X7 will replace the current BayTrail Z37xx SoCs with Gen8 GPUs.
They still haven't said anything about what will replace the Atom Z34xx and Z35xx lines for mid and high-end smartphones, and these are the ones who have PowerVR GPUs inside.

 

Are you kidding me with mali 4xx?!? We will never escape etc1...

 

Asus Zenfone 2 Review: http://www.anandtech.com/show/9251/the-asus-zenfone-2-Review

 

But poor battery life.

That might keep Moto G more popular.

At those price points, there probably isn't huge expectations of performance anyways. It doesn't take much power to run Facebook, messaging apps and cutie games.

Is Intel subsidizing these kinds of devices though, the way they were giving away or had negative margins for cheap tablets?

 

I don't get you guys. Is web browsing all you do with your smartphones or do you just happen to stop reading the first two paragraphs of each page in the review?

The battery life is only bad in the Web Browsing tests, which could be a software optimization problem. All the other battery life tests look like this:



Here's GSMArena's take on the subject:




And charging times are spectacular:

 

Contra revenue is still a reality. Intel is effectively buying marketshare and denying revenue to competitors. For instance, this shuts AMD completely out of the market, and puts pressure on Qualcomm, Mediatek, and the rest. Whether this will allow Intel to corner the market and increase pricing to nicely profitable levels is another matter given that both Apple and Samsung are likely to stay with their own solutions.