GPU Results and Conclusion
Although we took only a very brief look at CPU performance from the Snapdragon 820's custom Kryo cores, we were able to test the graphics capability of the SoC's Adreno 530 GPU against some additional comparative results.
GPU performance is an area where previous Snapdragon processors fared well, and the Adreno 530 is said to offer up to 40% higher performance than the previous 430 found in the Snapdragon 810. Equivalent power savings are also prominently featured with the new GPU core, but once again the battery life/thermal aspects of this new SoC can't really be tested until shipping hardware is in the hands of reviewers.
For now we can only take a broad look at the capability of this new core with existing benchmarks, though it must be pointed out that the advanced features of the Adreno 530 (such as OpenGL ES 3.1+ support) won't be taken advantage of with most of the existing games and benchmarks.
GPU Performance
You may notice that we have selected only fixed resolutions for these tests to even the playing field between generations of SoC GPU cores, as the varying native screen resolutions – though more realistic – would skew the results too much to provide an adequate glimpse of the relative performance of this new GPU core.
GFXBench Offscreen – T-Rex
The T-Rex test is based on OpenGL ES 2.0 and includes textures, material, geometry and particle effective that were highly detailed at the time of release. The graphics rendering engine features planar reflection, specular highlights and soft shadows, providing a good workout even for flagship smartphones and tablets. Offscreen tests are run at 1080p, regardless of the device’s native resolution, and are best used to compare the performance between competing silicon, not competing devices.
The Snapdragon 820's Adreno 530 GPU had such a strong showing here compared to the field that I can't help but wonder if this in an aberration. Part of the problem with these limited hands-on sessions is that we can't go back and re-test like we can when we have more time with the hardware. Still, if these results hold up it's a glowing start to the GPU benchmarks for the 820.
GFXBench Offscreen – Manhattan
Manhattan was the first benchmark to utilize OpenGL ES 3.0 features and uses a nighttime setting with a lot of external illumination to stress the GPU. It uses a deferred rending engine with multiple render targets for the geometry pass, includes both diffuse and specular lighting, uses depth shadow maps, bloom, depth of field and quite a bit more.
Here again the Snapdragon 820 and Adreno 530 graphics dominate this offscreen test, and again I have to wonder how these results will hold up with further testing. Once review hardware of a shipping smartphone powered by the 820 SoC is available all of this should become clear. For now we will forego the remaining GFXBench results and move on to 3DMark.
Use 3DMark Ice Storm Unlimited for chip-to-chip comparisons of the hardware inside your device without vertical sync, display resolution scaling and other operating system factors affecting the result. In Unlimited mode the rendering engine uses a fixed time step between frames and renders exactly the same frames in every run on every device. The frames are rendered in 720p resolution "offscreen" while the display is updated with small frame thumbnails every 100 frames to show progress.
Ice Storm Graphics test 1 stresses the hardware’s ability to process lots of vertices while keeping the pixel load relatively light. Hardware on this level may have dedicated capacity for separate vertex and pixel processing. Stressing both capacities individually reveals the hardware’s limitations in both aspects. Pixel load is kept low by excluding expensive post processing steps, and by not rendering particle effects.
Graphics test 2 stresses the hardware’s ability to process lots of pixels. It tests the ability to read textures, do per pixel computations and write to render targets. The additional pixel processing compared to Graphics test 1 comes from including particles and post processing effects such as bloom, streaks and motion blur. The numbers of vertices and triangles are considerably lower than in Graphics test 1 because shadows are not drawn and the processed geometry has a lower number of polygons.
The purpose of the Physics test is to benchmark the hardware’s ability to do gameplay physics simulations on CPU. The GPU load is kept as low as possible to ensure that only the CPU’s capabilities are stressed. The test has four simulated worlds. Each world has two soft bodies and two rigid bodies colliding with each other. One thread per available CPU core is used to run simulations. All physics are computed on the CPU with soft body vertex data updated to the GPU each frame. The background is drawn as a static image for the least possible GPU load. The Ice Storm Physics test uses the Bullet Open Source Physics Library.
Here the Snapdragon 820 is dragged back down to earth, though the results are skewed far enough in the opposite direction that additional questions are raised. Who would have thought that pre-production hardware could have been experiencing inconsistent performance between benchmarks? These results are poor, and I have to wonder if thermals were a factor given the short timeframe of the graphics benchmarks. So how did the 820's CPU fare in the same test?
This time the performance is more in line with expectations, and the Snapdragon 820's four custom Kryo cores manage to improve on the octa-core performance of the previous SD810.
Basemark X is the world’s most popular benchmarking tool for evaluation and cross-platform comparison of gaming and graphics performance between Android, iOS and Windows Phone 8 smartphone and tablets.
Basemark X is the only vendor-independent benchmark that utilizes the real-world game engine Unity which is very popular among game developers. This means that it scores correlate exceptionally well with real-life gaming performance.
Basemark X includes two game-like graphics tests: Dunes and Hangar. Both tests contain heavy graphics content rendered with detail and complexity, thus pushing the measured device to the limit. The polygon counts in test sequences are up to 911,000.
On the high setting the Adreno 530 in the Snapdragon 820 managed the highest score among the devices tested, and finished 4th in testing at medium quality. This is a win for the Adreno 530 – especially if these improved results come with the promised power savings.
Basemark OS II is a system-level All-In-One benchmarking tool designed for measuring overall performance of smartphones and tablets from all platforms, including Android, iOS and Windows phone 8.
The benchmark features a comprehensive suite of tests including system, internal and external memory, graphics, web browsing, camera, battery and CPU consumption.
For our final graphics test we have the 820 reference platform losing out to the previous 810 platform hardware, and after the up and down results from these GPU benchmarks this only seemed appropriate. I can't help but wonder how much thermals affected these results. More time (and a controlled thermal environment) would be needed to verify this.
Conclusion
The results presented here – and particularly the GPU benchmarks on this page – provide a mixed impression of this new SoC. There are certainly some impressive results, but also some that seem either too good to be true, or are too low to trust. The pre-production hardware the press had access to might have something to do with these, as the software on the machines was likely not completely optimized for this new SoC.
But the story with the Snapdragon 820 is far more involved that just some simple CPU/GPU benchmark results. The move to a custom quad-core design and the advancements in DSP from this new SoC promise to provide higher efficiency, and much improved battery life, than prior generations. It was a surprising move by Qualcomm to look past the marketing aspect of CPU core-count and instead champion the improved per-core performance of a quad-core design.
In a marketplace that competes with Samsung's 8-core SoCs the Snapdragon 820 has the potential to provide an efficient alternative that could fare well, but only if it ships in flagship hardware in 2016. The lack of Snapdragon 810 devices in 2015 can't be ignored, and a lot is riding on the success of this followup silicon. Of course there's far, far more to the story than these benchmark numbers, not the least of which is the all-new X12 modem with its advanced wireless features (including Cat 12/13 network speeds and tri-band Wi-Fi with 2×2 MU-MIMO).
It's always fun to try out new hardware before it hits the market, but a quality mobile experience is far more than the benchmark data. I have confidence based on the new emphasis on efficiency and power consumption that we will see the Snapdragon 820 on more devices in the coming year than we saw with its predecessor, and at that time we'll have a far better idea of what kind of real-world experience this new SoC provides. I'll look forward to the chance to re-test the 820 on shipping hardware and explore the additional features not covered here.
Why are you comparing the 820
Why are you comparing the 820 to Apple’s CPU from last year?
Spin, to make the 820 more
Spin, to make the 820 more appealing by leaving out a more proper up to date CPU comparison! This happens when review sites have conflicting interests. The hands that feed!
Comparison is against the
Comparison is against the Snapdragon 810, with the other results provided for context. Didn't have access to an iPhone 6S in time to add results from Apple's newest SoC, certainly can update with those numbers.
Is this a paid Ad?
Is this a paid Ad?
How about letting the readers
How about letting the readers Know If Intel is using any licensed GPU IP in their SOC’s, the Atom Z3580 is using a PowerVR G6430 GPU. All the SOC SKUs listed in the tables should have their CPU core counts listed, and the make and model number of the GPU IP if any. This smacks of an advertisement if the proper SOC hardware specifications are not listed in tabular form for all the SOCs listed in the benchmarks and that includes the CPU core counts, and GPU FP/INT unit counts. It would not be too hard for PCPer to compile a master list of CPU/SOC/GPU specifications in tabular from and link to the master list so readers could at least look at the actual CPU core counts on these SOCs listed in benchmarking charts, and the GPU core/EU/SP/other counts and identifying information, and once the master list is done it’s simply a matter of linking to it with every article, and maintaining a master list would help the reader, and PCPer writers to know what CPU/core resources these devices bring to the market.
Simply listing benchmarks with no CPU/GPU/other hardware information in the tables is not enough to qualify as a thorough review of a SOC and its comparison and contrast with its competing products. Mobile SOC compute is more than just simply the the CPU cores and their processing ability, it also includes the GPU cores and their HSA processing ability, so hopefully there will be benchmarks in the future to test the mobile market’s SOCs for their levels of HSA compliance, a lot of those SOC makers are members of the HSA foundation, and even on the products that are using the Imagination Technologies(PowerVR) IP, ARM IP(CPU, GPU), Qualcomm IP, Samsung IP, etc. Even the users of the licensed IP technology who are not direct members of the HSA foundation will be getting the HSA ability included as part of the licensed IP.
This is a performance
This is a performance preview. We can’t really come to any conclusions until we have shipping devices since mobile parts are so thermally constrained. Also, the CPU core counts and number of internal units, and such are interesting for tech enthusiast, they are mostly irrelevant. The performance of the shipping device is what counts, regardless of the underlying hardware.
No it helps let the readers
No it helps let the readers Know what the hardware is capable of doing without the need for benchmarks, Instruction decoder numbers, Etc. lets the reader see IPC potential, execution unit numbers also help. Benchmarks for the 820 are not even there yet to figure out just what the Hexagon(1) DSP on the 820 is capable of bringing to the SOCs performance and power usage equation relative to SOC SKUs that do not have dedicated DSP like the Hexagon.
Let’s take instruction decoders(ID), the Apple A7(Cyclone) has 6(ID) in the Cyclone’s core, the Arm Holding’s reference design(A53, A57, A72) only have 3(ID) in their respective cores, and the newest Arm holdings reference design A72 has a little bit more execution resources than Arm Holdings’ A53/A57 cores but still that same nomber of IDs. Now what number of instruction decoders does the 820’s core have, it is less that the Apple A7, but more than the Arm Holdings’ reference cores. It’s obvious that the Apple A7 has twice as many instructions decoders than the Arm Holdings Reference cores(A53, A57, A72) so your argument is not valid.
Why is it that Intel, and AMD, are so forthcoming when it comes to supplying the CPU core specifications of their x86 based cores, while the makers of the custom ARMv8A ISA running custom cores are not at all forthcoming with their Custom CPU core’s micro-architectures that are engineered to execute the ARMv8A ISA. That IPC metric is very important even for the custom ARMv8A ISA running custom cores, and if it was not for Anand lal Shimpi the core specs on the A7 Cyclone cores would have never been known, that actual A8/A8X, and A9/A9X core specifications are still not revealed!
Hopefully when AMD gets its Custom K12 ARMv8A ISA running cores to market, AMD will also be more forthcoming with the K12 CPU core specifics, for sure if AMD want to use the K12 cores in server variants the server industry will not accept any lack of proper CPU core specification data-sheets! When the K12 cores get to market the custom ARM based makers will be forced to become more forthcoming, and the professional Server SKU review websites will have more information about AMD’s K12 server core variants.
The numbers of Custom micro-architectures that are engineered to run the ARMv8A ISA has gone up, post Apple introducing the Apple A7 cyclone, and the benchmarking software lacks the ability to stay on top of all the innovating that is happening where the Custom ARMv8A running micro-architectures are concerned. Those CPU core specification down to the smallest detail are needed now more than ever! The Snapdragon 820 is a very HSA aware SOC, being that Qualcomm is a founding member of the HSA foundation along with ARM and others. So that HSA 1.0/close to HSA 1.0 compliance is going to become important going forward.
Vulkan benchmarks will have to be done on a lot of Phone/Tablet platforms when the Vulkan graphics API is released.
(1)
http://www.anandtech.com/show/9552/qualcomm-details-hexagon-680-dsp-in-snapdragon-820-accelerated-imaging
This article is poorly done,
This article is poorly done, it keeps mentioning this cpu has half the cores of others. That is not relevant to its top performance at all. The 810 and samsungs 8 core do not use all 8 cores at once, they only use 4 at most at a time, 4 are high power cores, 4 are low power cores.
So you are saying that it
So you are saying that it doesn't have half the cores? Or is the comment just so poorly written I am not understanding why you are complaining about it being mentioned a mere 2 times in the article?
Functionally it does not have
Functionally it does not have half the cores. If you have 8 cores but only 4 cores can function at any given time then you functionally have 4 cores. Not hard to understand.
All 8 cores of the 810 in the
All 8 cores of the 810 in the Lumia 950xl operate at once. Or at least I think they do. I read that numerous places.
http://www.anandtech.com/show
http://www.anandtech.com/show/7925/qualcomms-snapdragon-808810-20nm-highend-64bit-socs-with-lte-category-67-support-in-2015
810 – all 8 cores can be active at once.
So this preview is a later
So this preview is a later date than that article:
http://www.anandtech.com/show/8933/snapdragon-810-performance-preview/4
To me it seems to state that a task cannot actually run across all 8 cores, that if a task requires tons of processing power, it will only be scheduled on the big 4 cores.
Read and learn
Read and learn time!
Superscalar processor:
https://en.wikipedia.org/wiki/Superscalar_processor
Wow hardly impressive and my
Wow hardly impressive and my god how will it fare against the Apple A10 Soc, given it was only tested against the A8 not the A9. Unless battery life is really improved over 810 seems that in real world use other than some gpu bound apps you’ll see very little difference between 810 and 820. Samsung’s new Exnyos might pown this one too.
I wish smartphones were
I wish smartphones were actually as thick as that Snapdragon 810 reference device – imagine week-long battery life, full-size SD cards, durability, easier to hold, physical keyboards/trackpoints/other inputs, and all the other things they could fit in there if they weren’t crippled by shaving irrelevant sub-millimeters off.
Here Anandtech test with the
Here Anandtech test with the Iphone 6S Plus compared to the 820..no contest, and I’m a android guy.
http://www.anandtech.com/show/9837/snapdragon-820-preview/3
Why doesn’t this include the
Why doesn’t this include the A9? Ignoring the A9 on the graphs is extremely misleading and not something I would expect from Pcper. I’m not sure if you’re being paid to deliberately bias this in favor of the 820 or what’s up, but this is not consistent with your past quality. For anyone who cares to see real comparisons, I guess there’s still anandtech – http://www.anandtech.com/show/9837/snapdragon-820-preview/3 .
How are the graphs
How are the graphs misleading, exactly? The results are what they are. Primarily this is to compare the 810 to the 820, to see how the custom quad-core design compares to the 810's 8-core big.LITTLE configuration.
Naturally I would have added A9 results if I had a 6S on hand. Why didn't I? Here's a peek behind the curtain: I didn't have access to an iPhone 6S, only a 6. Wasn't going to buy a 6S out of pocket to have those results before the article went up. That's it.
For those curious, myself included, I planned to add Apple A9 results when available.
Finally, the accusation of being "paid to deliberately bias this in favor of the 820" is a fine piece of slander from the brave 'anonymous' commenter. It's amusing that this tactic is used whenever possible, and only by those who have never done this job themselves.
I am not the guy that posted
I am not the guy that posted that comment but I hope you can see why that sort of omission raises some suspicion. Your readers are here for information and don’t want to see poor analysis. Being anonymous doesn’t have much to do with this, I imagine most people don’t want to take the second to sign up if they don’t have to.
I believe your explanation but we’re not mind readers. You should have just written the fact that you didn’t have an A9 to test *in the article* and saved yourself some accusations.
Poor research is no excuse,
Poor research is no excuse, get a subscription the the Microprocessor Report, and the other professional trade journals. The Publication(PCPER) is the one responsible for getting the reporters the review units. PCPER needs to get it’s reporters and freelance reporters a company subscription to the Professional Trade journals especially Microprocessor Report, and If the money is not there the local College Library is, for the Microprocessor report, and a lot of other professional journals.
Tech reporters should maintain a master list of Microprocessor specification tables, and a lot of the technical/enthusiasts websites need to get together and pool their resources and get someone to do what Anand Lal Shimpi used to do, the companies are hiring all the good tech reporters to shut them up, so that the online press has no one able to do anything other than benchmarking without doing any actual research!
No articles should be posted without the most up to date CPU/SOC based SKUs to run the benchmarks on for comparsion, it’s that or simply taking the time to get some results from other tech websites, with journalistic permission of course. There is an explosion of new Custom ARMv8A ISA running cores coming to market, some of them that are way more powerful than the refrence Core designs that ARM holdings licenses, and Apple(with the A7 Cyclone) beat ARM Holdings to the market with a custom core that was able to run the ARMv8A 32/64 bit ISA, and ARM holdings was the one that created the ARMv8A ISA.
Trust me when I say this:
Trust me when I say this: it's something we are working on. We don't have active relationships with some of the big guys in the field (Samsung, Apple) but we are working to get those started to do exactly what you are stating: fill a void in this market of tech reporting and journalism. As of the few standing, independent tech websites left, this is my primary goal for 2016.
Hope that helps.
So 14 nm soc market is gona
So 14 nm soc market is gona be crowded in 2016! And i suspect intel will probably be at the top of benchmark lather
Would like to see a review of
Would like to see a review of google’s PIXEL C (X1 tegra), since old K1 does pretty well here.
Is a review coming soon, or do you have benchmarks you can add to the charts?