• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Snapdragon 805 Performance Preview (AnandTech)

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
And Mullins only has a single 64 bit memory interface. I'm surprised AMD has stuck with that because they know the ARM juggernauts are out there unless AMD just has flat out better cache schemes in the CUs.
 
The memory interface is impressive and shows they're thinking ahead. I think this is going to pretty much impossible for K1 to beat, even in GPU.


Its still ARMv7 but it will serve fine into 2015. I think if I were going android I'd wait for the 20nm ARMv8 S810 chipset.

This pretty much floors all the x86 android SoCs and even Apples A7 in GPU at least, and it's memory interface is great but I really wish they'd go 64 bit like apple. I think the people who are thinking cherry trail is going to change anything in GPU should take a look at how bay trail is being demolished by this SoC is pretty much every metric, ESPECIALLY GPU.
 
. I think the people who are thinking cherry trail is going to change anything in GPU should take a look at how bay trail is being demolished by this SoC is pretty much every metric, ESPECIALLY GPU.

And I think the people who are thinking Cherry Trail won't demolish Bay Trail in the iGPU department are in denial. Even if it has lower clocks, we're still talking about 4x the amount of EUs and a new architecture (hopefully with much better perf/watt) on top of that. 😉
Using GFXbench offscreen as reference, Adreno 420 is a little bit over 2x faster than Bay Trail-T C0's HD Graphics (19.5 vs 40.7 FPS). Now when it comes to CPU performance, even 1.86GHz Atom Z3740 beat the 2.7GHz S805 in SunsSpider and Kraken (unfortunately those are the only CPU tests including both S805 and BT-T performed by Anand). Now Atom Z3740 is not even the fastest Bay Trail-T chip, Atom Z3770 & Z3795 are around 20-30% faster, so BT-T is still very competitive here compared to Krait cores.
 
Last edited:
And I think the people who are thinking Cherry Trail won't demolish Bay Trail in the iGPU department are in denial. Even if it has lower clocks, we're still talking about 4x the amount of EUs and a new architecture (hopefully with much better perf/watt) on top of that. 😉
Using GFXbench offscreen as reference, Adreno 420 is a little bit over 2x faster than Bay Trail-T C0's HD Graphics (19.5 vs 40.7 FPS). Now when it comes to CPU performance, even 1.86GHz Atom Z3740 beat the 2.7GHz S805 in SunsSpider and Kraken (unfortunately those are the only CPU tests including both S805 and BT-T performed by Anand). Now Atom Z3740 is not even the fastest Bay Trail-T chip, Atom Z3770 & Z3795 are around 20-30% faster, so BT-T is still very competitive here compared to Krait cores.

Cherry trail won't have the memory bandwidth to compete with this on GPU. You're failing to recognize that the Bay Trail benchmarks are running on a much lower resolution screen. This is all done in 1440p, so while only 40% better GPU is impressive, the truly impressive thing is they've done it at a higher resolution.

This is a phone SoC. Bay Trail is working in a larger power envelope, on lower resolutions, and still posting performance numbers that aren't competitive to this SoC at all in GPU.

Sun Spider and Kraken have more to do with CPU power envelope and platform than the GPU tests because they're browser tests. Doesn't apple absolutely dominate sun spider w/ A7?
 
Cherry trail won't have the memory bandwidth to compete with this on GPU. You're failing to recognize that the Bay Trail benchmarks are running on a much lower resolution screen. This is all done in 1440p, so while only 40% better GPU is impressive, the truly impressive thing is they've done it at a higher resolution.

This is a phone SoC. Bay Trail is working in a larger power envelope, on lower resolutions, and still posting performance numbers that aren't competitive to this SoC at all in GPU.

Sun Spider and Kraken have more to do with CPU power envelope and platform than the GPU tests because they're browser tests. Doesn't apple absolutely dominate sun spider w/ A7?

offscreen eliminates res differences and sunspider is a bad cross uarch and/or os benchmark...
 
Cherry trail won't have the memory bandwidth to compete with this on GPU.

Wrong, Cherry Trail-T tops out at 25.6GB/s just like the S805.

8.jpg


You're failing to recognize that the Bay Trail benchmarks are running on a much lower resolution screen.

GFXbench offscreen forces different devices to run @ same resolution.

This is a phone SoC. Bay Trail is working in a larger power envelope, on lower resolutions, and still posting performance numbers that aren't competitive to this SoC at all in GPU.

S805 doesn't pack an integrated modem, unlike S800/S801, which means that while it might show up in premium >5'' phones/phablets (like the rumoured Galaxy S5 Prime, One ''M8'' Prime, Zperia Z3) it might be better suited for tablets. Perhaps that's why some of Qualcomm's customers will skip the S805.

Qualcomm tells us that some of its customers will choose to stay on Snapdragon 801 until the 810 arrives next year, while some will choose to release products based on 805 in the interim.
 
Wrong, Cherry Trail-T tops out at 25.6GB/s just like the S805.

8.jpg




GFXbench offscreen forces different devices to run @ same resolution.



S805 doesn't pack an integrated modem, unlike S800/S801, which means that while it might show up in premium >5'' phones/phablets (like the rumoured Galaxy S5 Prime, One ''M8'' Prime, Zperia Z3) it might be better suited to tablets. Perhaps that's why some of QUalcomm's customers will skip the S805.

Surprising, for some reason I thought cherry trail was similar to bay trail in bandwidth. I stand corrected on that point.

However, I have yet to see intel deliver on promises made about GPU. I think they're easily capable of beating almost everyone except possibly apple in CPU, but they consistently disappoint on GPU.
 
And Mullins only has a single 64 bit memory interface.

Tegra K1 only has a dual-channel 64-bit mem. interface. Same with the Apple A7 SoC and the upcoming Cherry Trail SoC. So the mem. speed and bandwidth efficiency are important too.
 
Last edited:
Surprising, for some reason I thought cherry trail was similar to bay trail in bandwidth. I stand corrected on that point.

However, I have yet to see intel deliver on promises made about GPU. I think they're easily capable of beating almost everyone except possibly apple in CPU, but they consistently disappoint on GPU.

Come on. With 4x the execution units, a next gen GPU architecture, AND a new fab. process node, Cherry Trail's GPU performance will be miles ahead of Bay Trail's GPU performance.
 
first battery test on Xiaomi MiPad:
http://en.miui.com/thread-22041-1-1.html
looks very good, 5hours gaming on 6700mAH battery means that the SoC consumes around 3W with heavy 3D workload.

While I like the numbers and I am finally starting to get over my cynicism towards Tegra. These tests seem like very short periods that they then extrapolate to a full battery test. I know in my experience phone's battery % is not a super accurate method of measuring. I know in my phone the first few percent go quickly then it goes slowly til the low 30's and finally drops like a rock.

I am not saying that is the case here, but once again I think we should all wait for real reviews before declaring an end all, be all winner.

That being said, if a cheap or Nexus TK1 device comes out I am pretty much sold on buying one at this point. Especially if it can play portal/HL2 and I assume other similar PC games ported.
 
This is a phone SoC. Bay Trail is working in a larger power envelope, on lower resolutions, and still posting performance numbers that aren't competitive to this SoC at all in GPU.

Aside from being wrong that Bay Trail uses more power, it really isn't that surprising that a new GPU is faster than a 2 year old GPU microarchitecture. S805 won't compete with Bay Trail anyway, but with Cherry Trail.
 
However, I have yet to see intel deliver on promises made about GPU. I think they're easily capable of beating almost everyone except possibly apple in CPU, but they consistently disappoint on GPU.

AnandTech i7 3770K review:

"Ivy Bridge is unique in that it gives us the mild CPU bump but combines it with a very significant increase in GPU performance."

It didn't disappoint Anand when it was first released 2 years ago. Since Gen8 is a much bigger update than Gen7, I'm not really worried about Intel disappointing, certainly when they also quadrupled the number of EUs on top of the new architecture.
 
Aside from being wrong that Bay Trail uses more power, it really isn't that surprising that a new GPU is faster than a 2 year old GPU microarchitecture. S805 won't compete with Bay Trail anyway, but with Cherry Trail.

2 years old? You intel guys love to pretend that as soon as intel announces an architecture that somehow means it's on the market.


Bay Trail for android came last month. So basically, Bay Trail is competing with the S805. Cherry Trail will compete with whatever comes long after S810.


Qualcomm is raking in the cash at almost the rate that intel is losing it on mobile.
 
AnandTech i7 3770K review:

"Ivy Bridge is unique in that it gives us the mild CPU bump but combines it with a very significant increase in GPU performance."

It didn't disappoint Anand when it was first released 2 years ago. Since Gen8 is a much bigger update than Gen7, I'm not really worried about Intel disappointing, certainly when they also quadrupled the number of EUs on top of the new architecture.
That's because Anand has been watching and expecting intel to fail at GPU for 20 years. He's not gonna be disappointed by them producing another failed GPU.
 
2 years old? You intel guys love to pretend that as soon as intel announces an architecture that somehow means it's on the market.


Bay Trail for android came last month. So basically, Bay Trail is competing with the S805. Cherry Trail will compete with whatever comes long after S810.


Qualcomm is raking in the cash at almost the rate that intel is losing it on mobile.

Cherry Trail is coming at the end of 2014, per Intel, which means it should be in designs roughly at the same time 808/810 show up. Further, Intel's Moorefield/Merrifield will be in designs long before 808/810 and should offer pretty fantastic GPU performance for Android devices.
 
Last edited:
Typical TreVader anti-Intel post.

Cherry Trail is coming at the end of 2014, per Intel, which means it should be in designs roughly at the same time 808/810 show up. Further, Intel's Moorefield/Merrifield will be in designs long before 808/810 and should offer pretty fantastic GPU performance for Android devices.
Well QC is not the leader anymore in android performance.
if intel comes at S810 timeframe then it will also face Erista, 20nm Nvidia SoC loaded with Maxwell. Of course Cherry-T GPU will be destroyed...
 
Well QC is not the leader anymore in android performance.
if intel comes at S810 timeframe then it will also face Erista, 20nm Nvidia SoC loaded with Maxwell. Of course Cherry-T GPU will be destroyed...

We're close to June and Tegra TK1's first tablet showed up a few days ago. It will probably take some time till Erista-based devices hit the market in 2015, perhaps (late?) Q2. Cherry Trail-T is a late 2014 chip so it will face Tegra K1 competition for months before Erista. Also, Intel is speeding up the Atom roadmap, next-gen Broxton (Goldmont architecture + Gen 9 graphics) should be launched in 2015, sometime in H2. Now Snapdragon 810 is rumoured to be 40% faster than Snapdragon 805. If GFXbench is representative of actual graphics performance, then NVIDIA already delivers 50% better graphics performance than the S805 with Tegra K1 (40 vs 60 FPS). I wonder why it took so long for them to leverage their graphics expertise in the Tegra line.
 
The memory interface is impressive and shows they're thinking ahead. I think this is going to pretty much impossible for K1 to beat, even in GPU.

NVIDIA's Tegra K1 inside the Xiaomi MiPad already scores higher than Snapdragon 805, especially in GPU.
 
Gaming : 5 hrs

Reading : 17 hrs

Now,the Mi Pad has a 6.7Ah battery. Assuming the voltage is 3.7, the battery is ~25Wh.

When used for reading, the battery life is 17hrs, so 25/17 = 1.4 W/h , which mainly will be used by the display.

Gaming = 25/5 = 5 W/h. So, subtracting the display + SoC idle usage, 5-1.4 = 3.6W extra power consumption during load.

So out of that 3.6W, speakers + wireless + NAND will use some, so we're looking at ~3 - 3.5W of SoC power.

Screenshot_1.jpg
[/url][/IMG]

Nvidia stated that TK1 AP + DRAM uses 3.6W when constrainted to Apple A7 power when running GFX Bench 3.0 Manhattan offscreen. The results match, so coud be that were not lying about power.

What is also good is that TK1 can be scaled when needed to be used in say Ouya 2.0
 
Gaming : 5 hrs

Reading : 17 hrs

Now,the Mi Pad has a 6.7Ah battery. Assuming the voltage is 3.7, the battery is ~25Wh.

When used for reading, the battery life is 17hrs, so 25/17 = 1.4 W/h , which mainly will be used by the display.

Gaming = 25/5 = 5 W/h. So, subtracting the display + SoC idle usage, 5-1.4 = 3.6W extra power consumption during load.

So out of that 3.6W, speakers + wireless + NAND will use some, so we're looking at ~3 - 3.5W of SoC power.

Screenshot_1.jpg
[/url][/IMG]

Nvidia stated that TK1 AP + DRAM uses 3.6W when constrainted to Apple A7 power when running GFX Bench 3.0 Manhattan offscreen. The results match, so coud be that were not lying about power.

What is also good is that TK1 can be scaled when needed to be used in say Ouya 2.0

Those battery tests are not in the least bit reliable. Instead of running down the battery completely they let it run for a few minutes, look at the battery level indicator, and estimate how much long the tablet would last based off that. The margin of error is gargantuan.
 
Nvidia stated that TK1 AP + DRAM uses 3.6W when constrainted to Apple A7 power when running GFX Bench 3.0 Manhattan offscreen.

Jetson TK1 is not even a mobile optimized platform. In a mobile optimized platform, the AP + mem. power consumption at ~ 2.6w (ie. iso-power as A7) will provide ~ 1.45x more performance than A7 in GFXBench 3.0
 
Those battery tests are not in the least bit reliable. Instead of running down the battery completely they let it run for a few minutes, look at the battery level indicator, and estimate how much long the tablet would last based off that. The margin of error is gargantuan.

That may be the case, but at the same time, no one in their right mind would spend 86 continuous hours listening to music, or 17 continuous hours watching videos, or 11 continuous hours browsing the web, or 5 continuous hours gaming on a tablet either.
 
Well QC is not the leader anymore in android performance.
if intel comes at S810 timeframe then it will also face Erista, 20nm Nvidia SoC loaded with Maxwell. Of course Cherry-T GPU will be destroyed...

Intel's brand new 14nm Tri-Gate transistors aren't afraid of 3 to 5 years old tech.
 
Tegra K1 only has a dual-channel 64-bit mem. interface. Same with the Apple A7 SoC and the upcoming Cherry Trail SoC. So the mem. speed and bandwidth efficiency are important too.

When you say "dual channel 64 bit" are you referring to two 32 bit channels? Sorry I'm just used to the PC world with it's 64 bit wide memory channels.

And I think after Mullins, it would be wise of AMD to move up to two 64 bit memory channels unless they get a stacked/integrated eDRAM scheme going, which is unlikely, and they need more GCN processors too. Qualcomm, Apple, and Nvidia are not just sitting idle on the graphics front.
 
Back
Top