Surface 3 vs. BayTrail the comparison that everybody wanted & nobody did w Bonus Snapdragon 810

[GreyFox7]

Well that's not the only one but the pieces are starting to fall into place.

Source Reddit: http://www.reddit.com/r/Surface/comments/32s5n0/surface_3_vs_bay_trail_the_comparison_that/

UPDATE: PC World and Paul Thurrot reported that S3 scores over 25000 on Ice Storm Unlimited, which makes the GPU side of Surface 3 a HUGE improvement over any Bay Trail devices. Thanks product342!

With the Microsoft Surface 3 getting somewhat expected (positive) reviews so far, one thing has made potential buyers wonder - Why compare it to iPad or Core i5 devices, instead of Windows tablets using Bay Trail? We know that the direct competitor for Surface 3 is cheap 8 to 11-inch Bay Trail tablets, something that many reviewers failed to acknowledge.

Therefore, based on what we know so far, I have decided to compare Surface 3 to several popular Bay Trail devices, ones that are using either Z3740 or Z3770.

First, let’s look at the processor Surface 3 uses. It’s Atom x7-Z8700, a high-end Cherry Trail processor:

http://ark.intel.com/products/85475/Intel-Atom-x7-Z8700-Processor-2M-Cache-up-to-2_40-GHz

It’s quad-core Airmont, a 14nm-version of Bay Trail’s Silvermont which was 22nm. It’s basically the same architecture, though smaller and more efficient. This quad-core CPU is clocked between 1.6 to 2.4 Ghz. It has a new 8th generation HD Graphics GPU with 16 “Execution Unit”. Bay Trail had only 4 EUs in their GPU, so 16EU is supposed to be a huge upgrade.

The most popular Bay Trail chipset in the market is Z3740. It is used by Lenovo Miix 2, Toshiba Encore, and most importantly, Asus Transformer Book T100 (at least in the first iteration). The D variant is used by popular Dell Venue 8 Pro, which I’m going to explain below.

http://ark.intel.com/products/76759/Intel-Atom-Processor-Z3740-2M-Cache-up-to-1_86-GHz

http://ark.intel.com/products/78416/Intel-Atom-Processor-Z3740D-2M-Cache-up-to-1_83-GHz

The first thing you are going to notice is that the clock is lower than Z8700. It starts at 1.33Ghz and goes up to 1.86Ghz, or 1.83Ghz with Z3740D. Z3740D is also “single channel”, meaning the memory bandwidth is lower than Z3740, but this has been a trivial issue. GPU is clocked higher, but as I said, it has fewer EUs.

http://arstechnica.com/gadgets/2015/04/surface-3-review-smaller-slower-cheaper-better/2/

http://browser.primatelabs.com/geekbench3/search?utf8=%E2%9C%93&q=Z8700

Now, as proved by Ars Technica review, Surface 3 gets a little over 3400 in Geekbench 3’s multicore benchmark. This is verified by Geekbench’s own database, where 64-bit benchmark ends up at a little over 3400. Single core is generally a little below 1000, but the multi core performance adds up pretty well.

Now, what about Z3740? Well if you search for Z3740…

http://browser.primatelabs.com/geekbench3/search?utf8=%E2%9C%93&q=Z3740

You’ll see that most of the results are from Asus Transformer Book T100, and there is a quite a bit of performance gap between T100 and S3. T100 maxes at around 2600 at best. While some might argue that T100’s score is from 32-bit one and S3 64-bit, keep in mind that T100 is a 32-bit only device while S3 is 64-bit only.

Now, what about Z3740D and DV8P? The gap widens:

http://browser.primatelabs.com/geekbench3/search?utf8=%E2%9C%93&q=Z3740D

With the boost clock being 30Mhz lower and having a single-channel memory interface, DV8P and Z3740D devices maxes at around 2200. It’s a pretty huge gap of more than 1000 when compared to S3’s 3400.

Therefore, Result 1: If you are coming from Z3740, Surface 3 will clearly be an upgrade CPU-wise.

(The GPU comparison is edited to address previous mistakes) Now let’s turn to GPU side. Spec sheet made us anticipate much more enhanced GPU performance from Bay Trail to Cherry Trail, but is the hype real? To check that we turn to 3D Mark’s database. When reviews state they did a 3D Mark Ice Storm test, I was not sure which Ice Storm they were talking about, since there are 3 types – normal, Extreme, Unlimited. Comparing Ars Technica’s result for Surface Pro 3 to 3D Mark’s database, I presume that reviewers are talking about Unlimited most of the time.

So, according to Ars, Surface 3 got 18301 in Ice Storm Unlimited. This is where things go a little ugly…

http://www.futuremark.com/hardware/mobile/Dell+Venue+8+Pro+5830/review

Dell Venue 8 Pro gets an average of 14990. Where’s the four times more EUs? At the difference of just 3000, Surface 3 does not seem to be a huge upgrade over DV8P in terms of graphical performance.

BTW, in case of T100…

http://www.futuremark.com/hardware/mobile/Asus+Transformer+Book+T100TA/review

Score is lower at around 13000, but still the point kind of stands. In case you were wondering, T100 has the resolution of 1280 x 800, the same as DV8P.

http://www.pcworld.com/article/2906498/hands-on-with-microsofts-surface-3-full-windows-and-a-new-cpu-cut-the-compromises.html

https://www.thurrott.com/mobile/microsoft-surface/2849/surface-3-desktop-applications

However, PC World, and Paul Thurrott in particular, seem to indicate that something went wrong with Ars Technica's benchmark. Thurrott originally got a lowly 12000 - 13000 on Ice Storm Unlimited, but once he rebooted his Surface 3, it was getting a whopping score of 25590. PC World's benchmark proved the same, this time exceeding the score of iPad Air 2.

Benchmark score can go down due to throttling and other tasks, but it does not simply go up. In this case, it's safe to assume that 25000+ is the real score for Surface 3, considering its GPU is supposed to be 4 times bigger than Bay Trail's, and we have two results that state 25000+ opposed to one 18000ish. By the way, all Bay Trail processors have the same GPU, all of them clocking at around 15000.

Result 2: In any case, Surface 3 is a huge improvement over Bay Trail in terms of GPU

Now, the last point: A more expensive and potent, yet less popular alternative to cheap 8” Windows tablets have been Full HD tablets with Z3770.

http://ark.intel.com/products/76760/Intel-Atom-Processor-Z3770-2M-Cache-up-to-2_39-GHz

3770 is higher clocked, between 1.46 to 2.4 Ghz, something more in line with Z8700. It is used in multiple higher-end Bay Trail devices, namely Dell Venue 11 Pro and Lenovo Thinkpad 8.

http://browser.primatelabs.com/geekbench3/search?utf8=%E2%9C%93&q=Z3770

On Geekbench, it has gotten somewhere between 2900 to 3000 in multi core score. Compared to 3400 of Surface 3, I’d say it’s pretty close, with single core performance being almost identical.

http://browser.primatelabs.com/geekbench3/search?utf8=%E2%9C%93&q=Z3795

There is also Z3795, which is clocked just like Z8700, and is used in the most expensive Bay Trail devices. They go all the way to 3185, being identical to Surface 3 in terms of 32-bit score.

http://www.futuremark.com/hardware/mobile/Dell+Venue+11+Pro+5130/review

On GPU side, the case remains similar to Z3740. At the average of 15718, Z3770-powered Dell Venue 11 Pro not very far off from Surface 3’s 18000+. There is no Ice Storm result for Z3795, but since all Bay Trail processors have the same GPU, I’d assume it is going to be the same thing.

Result 3: CPU-wise, if you are coming from Z3770 or Z3795, Surface 3 is not a performance upgrade.

However, keep in mind that Z3770 or Z3795 devices are much more expensive than run-of-the-mill Z3740 ones, even more so with currently flooding Z3735 ones. They are basically in the same price point with Surface 3, maybe even more expensive with some others.

So there you go. I hope this review was enough in helping those who want to upgrade from Bay Trail. As for me, as a Dell Venue 8 Pro user, not only I want a better performance, I just want a much bigger screen than I have right now, so I’m definitely going to buy Surface 3. Even if you have Z3770 or Z3795 devices, pen support and many other pros as a form factor might encourage you to buy Surface 3, so choose wisely before you consider buying Surface 3.

 

[compnovo]

Nice work. Thanks!
I can't afford it, can't justify it, but I really want it.
If the Surface 3 had been available when I made the move from Surface 2 to the SP3 it would have been a really tough choice.

 

[revmike]

I agree with compnovo. My SP2 still meets my needs, but I would love to have the 3:2 ratio for reading my PDF Sermons.

 

[jrioux]

Why compare it to iPad or Core i5 devices, instead of Windows tablets using Bay Trail? We know that the direct competitor for Surface 3 is cheap 8 to 11-inch Bay Trail tablets, something that many reviewers failed to acknowledge.

Thanks for the extensive analysis. However, I'm not sure that I agree with you about the S3's competition being the CHEAP 8 to 11-inch tablets. I don't think many people will be considering either the HP Stream 8 at $150-$170 OR the S3 at $500.

IMHO, the competition includes the mid- to higher range Window 2-in-1s like the Asus T100 CHI and T300 CHI, which start at $400 and $700, respectively. The T100 offers a little less than the S3 at a considerable cost savings when you consider the keyboard. For $70 more than the base S3 plus keyboard, the T300 offers the Core-M CPU, 4GB RAM and 128GB storage. When you up the S3 to 4GB RAM and 128GB storage, the T300 is $30 cheaper.

I expect a number of new Core-M machine will be introduced by the end of the year to increase the competition.

 

[GreyFox7]

OK, but What about the Snapdragon 810?
You know because that's the bees knees in the Android World that everybody uses.

Well here's a couple data points:
.............. .................... Z8700 - Snapdragon 810
Google Octane ............ 8098 - 7907
3DMark Ice Storm U . 25429 - 24372

Related:...................... Z3580* - Snapdragon 810
Antutu ....................... 50401 - 52102
*Z3580 is an older BayTrail SoC clocked at up to 2.3 ghz vs 2.4 in Z8700, with 12.8 GB/s memory bandwidth vs. 25.6 GB/s in the Z8700 and of course lower graphics performance than Z8700.

Surface 3 built using Intel x7-Z8700 will be holding it's own on the same playing field with the 2015 Android devices.

 

[GoodBytes]

Neowin upgraded the Surface 3 review unit they have to Windows 10 Build 10061. It was painful in terms of wait. It took 2h to install, and 45min to update to the latest build. They think that the problem is that CPU overheats due to lack of cooling and throttles. So this is something to keep in mind.

Once installed, Windows 10 runs perfectly smoothly on the Surface 3.

 

[GreyFox7]

Neowin upgraded the Surface 3 review unit they have to Windows 10 Build 10061. It was painful in terms of wait. It took 2h to install, and 45min to update to the latest build. They think that the problem is that CPU overheats due to lack of cooling and throttles. So this is something to keep in mind.

Once installed, Windows 10 runs perfectly smoothly on the Surface 3.

IDK if W10 changes anything with respect to setup but this has been notoriously slow with Windows. It's like setup uses an old 8bit 2bit file copy routine and some other inefficient processes. Perhaps best described as one potato, two potato... coupled with comparatively slow eMMC storage and you have the perfect bottlenecked process.

Then there's the Registry, good lord what a flippin mess that is... of all the Things Microsoft changes from on OS iteration to the next, oftentimes needlessly, the Registry has just gotten to be a more and more convoluted pile of garbage. It's in even more dire need of overhaul than IE.
Bring on the SPARTAN REGISTRY!

 

[GoodBytes]

Registry doesn't impact anything. It is a database.
Accessing, reading, writing stuff on the registry is the space speed if you have only 1 item or a billion. Much like this forum. It doesn't go slower as more posts are added.

 

[GreyFox7]

Registry doesn't impact anything. It is a database.
Accessing, reading, writing stuff on the registry is the space speed if you have only 1 item or a billion. Much like this forum. It doesn't go slower as more posts are added.

Illogical response. Adding entries to a database does have an impact.
Adding tens of thousands of entries in a hierarchical cross linked fashion most certainly will have an impact.

 

[GoodBytes]

Nope. Well technically YES, like all databases, but the performance margin difference is small, and nearly plateaus. That is the power of databases. Else we would just use a text file, I mean, might as well.

And the registry is fairly clean in any case. Doesn't mean you don't understand how it works, that it is a mess.