Snapdragon 810 Performance Preview (AnandTech)

[Sweepr]

. In most cases, it appears that CPU performance is about what I’d expect from a cluster of four Cortex A57s at 2 GHz, although there are a few anomalous results that could be a concern. If anything, it’s clear to me that CPU isn’t really an area of weakness on the Snapdragon 810, especially with all of the work that Qualcomm has done for an energy aware scheduler to maximize the performance and efficiency of their big.LITTLE implementation.

Overall then the performance gains for the Adreno 430 and Snapdragon 810 seem to be almost exclusively focused on shader performance, but in those cases where rendering workloads are shader bound, Qualcomm's 30% estimate is on the mark. Real-word performance gains meanwhile are going to depend on the nature of the workload; games and applications that are similarly shader-bound should see similar performance gains, while anything that's bottlenecked by pixel throughput, texturing, or front-end performance will see much narrower gains.

While my time with the Snapdragon 810 hasn’t revealed any significant issues, the real concern here seems to be more along the lines of the GPU performance. While ALU performance and compute performance in general are significantly improved with the Adreno 430, the performance uplift doesn’t really seem to be as large as one might hope. Although Qualcomm is trying to sell the idea of a 4K tablet with the Snapdragon 810, it feels as if it’s too early to try and drive such high resolutions if the GPU can’t handle it. In order to see an appreciable increase in performance this year, it’s likely that OEMs will need to stay with 1080p display resolutions to really deliver improved graphics performance for gaming and other GPU-intensive use cases.

www.anandtech.com/show/8933/snapdragon-810-performance-preview

Interesting read, thanks to Joshua and Andrei. Performance appears to be a mixed bag, I'd expect a reference platform 64-bit enabled 2GHz Cortex A57 with LPDDR4 @ Lollipop to perform a bit better. GPU performance is moderately improved but still inferior to the best tablet SoCs (Apple A8X/NVIDIA Tegra K1 Denver).

I will update Part 2 and Part 3 of my new Exynos 5433 benchmark results with Snapdragon S810 numbers from this preview.

 

[jdubs03]

always on top of your game. thanks for the post. from what i can see here i'm not impressed, qcomm needs to get their shi together and make progress on their custom solution (next year no less). samsung has the advantage. i'd imagine nvidia does too, especially in their gpu...by a long long way; in kraken its only 10% faster than the k1-32bit in the shield tablet.

 

[Sweepr]

Tom's Hardware Snapdragon 810 Preview

It's important to note that both AnandTech and Tom's Hardware tested the tablet reference platform, which probably has more thermal headroom than future S810-based phones.

www.tomshardware.com/reviews/snapdragon-810-benchmarks,4053.html

 

[soccerballtux]

Tegra K1 doesn't seem so bad now.

I'll still stay away. NVidia never released updated graphics libraries for the Motorola Atrix 4G (had the Tegra 2). Note, the Atrix 4G was a phone billed as the most powerful smartphone upon release (released with Froyo 2.2 when 2.3 was already out, which Motorola 'kindly' updated to 2.3 a year later, promising 4.0 a year after that, which oh by the way was 2 years after release-- the date came around and they said 'oh, well, it's 2 years old, so no'). The drivers though...that mostly ended the modding community's efforts. Never liked NVidia's marketing; genius and business driven as they are, they leave a sour taste in my mouth every time.

 

[dawheat]

Tom's Hardware Snapdragon 810 Preview

It's important to note that both AnandTech and Tom's Hardware tested the tablet reference platform, which probably has more thermal headroom than future S810-based phones.

This just adds to the underwhelming reaction to the S810. Yes it's a reference platform and things will likely improve, but this is also close to best case performance compared to the thermal constraints in a normal 5" sized phone.

 

[jdubs03]

here are some more sites:

http://www.ubergizmo.com/2015/02/snapdragon-810-fully-revealed-benchmarks-architecture/

interestingly the geekbench single-core figure here is ~1400, considerably better, but strange how its 100pts higher here.

http://www.slashgear.com/snapdragon-810-benchmarked-5-things-you-need-to-know-12368604/

and just to show performance on a phone from geekbench:

http://www.gsmarena.com/sony_xperia_z4_visits_geekbench_with_its_snapdragon_810_chipset-news-11134.php

 

[Roland00Address]

Give me final device performance numbers not reference platforms. Last time we had a reference platform the numbers were all over the place from final shipping devices

http://www.anandtech.com/show/7251/lg-g2-and-msm8974-snapdragon-800-a-quick-look/4

See some numbers the chips are neck and neck and some the reference platform is 33% faster than the actual LG G2 phones using the same chip, some tests are neck and neck and some are a 10 to 15% difference. Now this may be a coincidence or it may not be but Anandtech says LG G2 only cheated in two tests Antutu which Anandtech does not benchmark with and Vellamo which was only 1% faster on the reference platform for with Vellamo LG removes all throttling caps.

But whatever even 33% higher performance is not worth less battery life in a phone. Same battery life I am all for 33% faster but not at the expense of my battery.

 

[Sweepr]

Snapdragon 805 provided about the same level of performance at Quad HD as Snapdragon 801 running Full HD. Bad news is S801 @ 1080p and S805 @ 1440p are 40-50% faster than S810 pushing 4K.

 

[krumme]

What is most striking is how much advance Samsung have to TSMC on proces development atm. and how that hurts Qualcomm perf.

The 14nm samsung a57.. is going to kill this soc on cpu and gpu (lpddr4 on s6?) perf. and perf/watt in an actual phone and they are going to be on the market about the same time. S6 april is an extremely high volume product, so Samsung was telling the truth about running 14nm production in november.

Its sad TSMC process is behind because the rest of the soc is as usual from QQ impressively strong. Be it DSP, media decoding, modem whatever. Looking forward to reviews of phones using the different socs, eg. actual camera performance and functionality, 3g/lte power consumption ...

Now we might even see Samsung own new design for the Note 4 or even if its not ready - its going to be 3 iteration for SS on a57, and second rev. on 14nm ! - damn. we could use that kind of progress on the desktop GPU side lol.

 

[dawheat]

http://cdn.slashgear.com/wp-content/uploads/2015/02/qualcomm-snapdragon-810-mdp-sg-9.jpg

The reference smartphone is kind of hilarious BTW. Dat thickness

 

[ShintaiDK]

http://cdn.slashgear.com/wp-content/uploads/2015/02/qualcomm-snapdragon-810-mdp-sg-9.jpg

The reference smartphone is kind of hilarious BTW. Dat thickness

Haha, holy crap. So thats how they fixed it. First phone with watercooling.

http://www.technobuffalo.com/2015/0...xes-its-overheating-snapdragon-810-processor/

 

[jhu]

http://cdn.slashgear.com/wp-content/uploads/2015/02/qualcomm-snapdragon-810-mdp-sg-9.jpg

The reference smartphone is kind of hilarious BTW. Dat thickness

It's more of a reference smartphone form factor. I don't think the modem is enabled and/or there isn't a sim card slot.

Aha!

This product is intended for software development. It has not been certified to operate on a carrier network and cannot make phone calls or transmit data over a cellular network.

I remember looking into one of these things last year. The MDP phone is only slightly more expensive than no-contract flagship phones, so why not be the first one on the block to have a 64-bit Android phone? Not being able to make calls was a deal breaker so I didn't buy one.

 

[Sweepr]

Here's real world performance inside a 5.5'' phone (LG G Flex 2):

Geekbench 3 @ 64-bit:
Single-core: 1222
Multi-core: 3965

AnTuTu v5.6.1 @ 64-bit: 54251

3DMark Ice Storm Unlimited: 22048

www.youtube.com/watch?v=W9IJ6wyjdSM

 

[Arachnotronic]

Here's real world performance inside a 5.5'' phone (LG G Flex 2):

Geekbench 3 @ 64-bit:
Single-core: 1222
Multi-core: 3965

AnTuTu v5.6.1 @ 64-bit: 54251

3DMark Ice Storm Unlimited: 22048

www.youtube.com/watch?v=W9IJ6wyjdSM

Not bad. Not great, but not bad.

 

[kpkp]

And than there is this:

Example Geekbench runs, one after the other (single core / multicore score)

1. 1215 / 3683
2. 1112 / 3249
3. 895 / 2931
4. 672 / 2414
5. 692 / 2465
6. 675 / 2421
7. 663 / 2356
8. 617 / 2263
9. 558 / 2065

 

[Sweepr]

And than there is this:

Full quote.

I can't tell you what Snapdragon 810 device I'm using, but I'm using one, and I will say that this thing categorically throttles like a son of a bitch. It can't even run Geekbench 3 three times in a row without having to throttle down performance a solid 20-30%. I can see why Samsung ended up avoiding this chip. Also, it gets hot. hothothot.

My results at this point show a 35+% peak drop in multicore performance and a staggering 50+% drop in single core performance as thermal load builds. That's pretty fucking terrible.

Example Geekbench runs, one after the other (single core / multicore score)

1. 1215 / 3683
2. 1112 / 3249
3. 895 / 2931
4. 672 / 2414
5. 692 / 2465
6. 675 / 2421
7. 663 / 2356
8. 617 / 2263
9. 558 / 2065

I've never noticed that much throttling with the 20nm Exynos 5433. My scores usually improve after the 1st run and then start to drop a bit after the 8-10th Geekbench run but never below 4000/1200 @ MT/ST (Geekbench 3 @ 32-bit). I can see why Samsung is ditching Qualcomm this round.

 

[krumme]

Full quote.



I've never noticed that much throttling with the 20nm Exynos 5433. My scores usually improve after the 1st run and then start to drop a bit after the 8-10th Geekbench run but never below 4000/1200 @ MT/ST (Geekbench 3 @ 32-bit). I can see why Samsung is ditching Qualcomm this round.

Arggg. That sad news.
The problem for us is Samsung process will effectively dominate the mobile arm high end market for the next year with such lack of competition meaning more expensive products eg s6 note 5....
On the positive side Samsung is providing excactly stellar perf and gold yields it looks. Perhaps 10nm and euv is more advanced than we could hope for?

Btw keep up the good work with the bm ! thanx

 

[jhu]

Arggg. That sad news.
The problem for us is Samsung process will effectively dominate the mobile arm high end market for the next year with such lack of competition meaning more expensive products eg s6 note 5....
On the positive side Samsung is providing excactly stellar perf and gold yields it looks. Perhaps 10nm and euv is more advanced than we could hope for?

Btw keep up the good work with the bm ! thanx

I don't agree. There really aren't any non-Samsung Exynos products out there (there's a few, but not in the USA). Everyone else will be using Qualcomm because they don't have any good choices. With the same logic, you could argue that Apple will dominate the ARM high end market due to lack of competition, but that's clearly not the case. Phones are an area where there's a large margin of acceptability for performance.

 

[Sweepr]

Arggg. That sad news.
The problem for us is Samsung process will effectively dominate the mobile arm high end market for the next year with such lack of competition meaning more expensive products eg s6 note 5....
On the positive side Samsung is providing excactly stellar perf and gold yields it looks. Perhaps 10nm and euv is more advanced than we could hope for?

I'm glad they are going Exynos only this time, had they done this with the Note 4 and they would have the first Android 64-bit flagship phone half a year before the first S810 phones. This move will not only reduce their reliance on Qualcomm but also create differentiation among the Android space (2014 Android flagships had pretty much the same ''boring'' specs). If rumours are true then Samsung doesn't need to care about performance parity with Qualcomm's offerings anymore, they can finally take on Apple's successful SoCs with the best they can offer.

Btw keep up the good work with the bm ! thanx

Thanks, I will.

 

[jdubs03]

here is another source. antutu score tops the exynos 7420. but more numbers corroborating the geekbench multi-core number.

http://www.trustedreviews.com/opinions/snapdragon-810-performance-shootout-qualcomm-s-super-chip-vs-mobile-processor-rivals

 

[dawheat]

http://www.androidpolice.com/2015/02/18/lg-g-flex2-review-practical-less-interesting/

S810 performance in the first shipping phone is as rumored - decent initial performance followed up with aggressive throttling. Though to be fair, this could be on LG as their phones have had a history of throttling more than competitors.

Honestly, I thought the author was a bit easy on the Flex 2 - stuttering and jankiness in a 2015 flagship level phone shouldn't be acceptable, especially when running Lollipop out of the box.

 

[Sweepr]

Doesn't look good.

The Snapdragon 810 in the G Flex 2 just does not feel all that snappy. It's not especially slow, but it experiences jank, app refreshes, and gets noticeably warm while doing things like web browsing. Benchmarks confirm there's some aggressive thermal throttling at play.

The other thing to note is that LG has consistently released phones that regularly engage in some level of throttling behavior. The G3, while it doesn't throttle as heavily or often as the G Flex 2, does experience a pretty consistent 15-25% drop in single core CPU performance if you run Geekbench 3 ten or fifteen times. The G Flex 2, by comparison, sees more like a 50-60% drop after so many runs, and by then is actually benchmarking substantially lower than the already-throttled Snapdragon 801 chip, instead yielding scores more like a Snapdragon 615 (I have a 615 device, and it will throw out the same scores basically all day with maybe a 10-15% variation).

Either LG screwed up or S810 is not an ideal choice in its current state.

 

[oobydoobydoo]

Doesn't look good.





Either LG screwed up or S810 is not an ideal choice in its current state.

I wonder how this will affect the whole android market? If every single device that isn't made by samsung throttles like nobody's business, that might actually serve to push even more people over to iOS. It will certainly push people over to Samsung.


If I was planning to buy an Android device in 2015, it would be an Exynos Device.

 

[exar333]

I wonder how this will affect the whole android market? If every single device that isn't made by samsung throttles like nobody's business, that might actually serve to push even more people over to iOS. It will certainly push people over to Samsung.


If I was planning to buy an Android device in 2015, it would be an Exynos Device.

Seriously, who other than 1-2% of Android users would even notice the throttling? And of that population, they would likely opt for Samsung vs. iOS anyways.

 

[jdubs03]

Doesn't look good.

Either LG screwed up or S810 is not an ideal choice in its current state.

I think its a bit of both, I have a G3 and it does throttle, though it never gets noticeably warm. Sometimes I do get some lags but not often.
I would've thought that naturally the next iteration of a uArch would enable cooler and faster operation, but that doesn't appear to be the case with the S810. Though, with more devices and repeated benchmarks we will know the extent of the S810 issues. The fact that after 10 tries and it benches underneath the S801 in the G3 is a bad bad sign.