Fudzilla: Snapdragon 835's second cluster is a Cortex A53

[dark zero]

Again, you can't compare architectures just based on the Kirin 960 GPU. And we don't have any recent Power VR on Android so we can't compare. This new G-71 brings around 60-80% more performance per cluster compared to the T880 at same power ( or less(?) because T880 MP8 at 900MHz consumes around 3'5W, doubling the cores would mean double the power and it would be too much for a phone ). So it is more efficient than the Adreno 540. The GPU implementation of the Exynos 8890 is about the same performance/power of the SD820 based on the S7 review long term performance battery test.

We already had a Power VR based device (iPhone 7) and is great performer.
Considering that Mediatek Helio X30 is using a high clocked quad core configuration + 10 nm manufacturing, and considering that it reached the Antutu score of 160K, is likely to have similar performance.

But yeah, I agree with you in what android has the recent Power VR iteration (the last ones were the Asus Zenfone 2 and the Xiaomi Redmi Note 2 with Power VR 6430 and 6200). The nearest one to show up is the Xiaomi Mi 6 (Mediatek edition) which is supposed to land on February. So we have to wait to show up .

 

[dark zero]

Ok, I revive this topic.
http://wccftech.com/qualcomm-snapdragon-835-specifications-design-details/

There is new info related to this.

The original cores on the Snapdragon 820 were a completely custom Qualcomm solution.

This year, the Kryo cores are based on an unspecified ARM design.

Ok... this is based on a ARM design and is not totally custom as before?
So the Fudzilla rumor is not as false as expecting???

 

[lopri]

AT's S835 preview:

http://www.anandtech.com/show/11201/qualcomm-snapdragon-835-performance-preview

It looks like a worthy successor to the S820. While running an identical workload on respective reference platforms, S835 consumes 3.56W and S820 consumes 4.60W, a 23% reduction in power consumption. We will see how this translates to battery life on actual devices.

Qualcomm and Samsung together achieved a milestone of the first mass-produced 10nm SOC. I know these numbers are largely marketing-driven, but even so this seems to mark the first time in a long while Intel is behind in chip design/manufacturing. Competition between Qualcomm/Samsung, Apple/TSMC, and Intel (plus AMD/NVIDIA) will ensure anyone's victory a short-lived one, and as a consumer I welcome the development.

 

[dark zero]

Seems that is time.... to start to put Cache L3 on the processors to improve their performance even more, just look at Apple. With L3 cache, they improved a lot.

 

[BoozeHobo]

It's amazing how the tables turned against QC in just 3 years. "64-bit iz uselezz", 810 space heater, then 14nm custom beaten by a stock budget 28nm ARM and just how many of 835s they think they can ship into devices that isn't branded Samsung this year?

Their only saving grace is their modem, and I doubt that superiority will last long either.

Qualcomm Adreno GPUs have always been far superior to all the competition and are the only true way to game on your phone. The other GPUs just boast about their 2D performance whereas Qualcomm concentrates on what's truly important, 3D performance.

 

[dark zero]

Qualcomm Adreno GPUs have always been far superior to all the competition and are the only true way to game on your phone. The other GPUs just boast about their 2D performance whereas Qualcomm concentrates on what's truly important, 3D performance.

Power VR has similar potential. And is 3D too. However no one is caring deeply on program games on true 3D.

 

[agent00f]

AT's S835 preview:

http://www.anandtech.com/show/11201/qualcomm-snapdragon-835-performance-preview

It looks like a worthy successor to the S820. While running an identical workload on respective reference platforms, S835 consumes 3.56W and S820 consumes 4.60W, a 23% reduction in power consumption. We will see how this translates to battery life on actual devices.

Qualcomm and Samsung together achieved a milestone of the first mass-produced 10nm SOC. I know these numbers are largely marketing-driven, but even so this seems to mark the first time in a long while Intel is behind in chip design/manufacturing. Competition between Qualcomm/Samsung, Apple/TSMC, and Intel (plus AMD/NVIDIA) will ensure anyone's victory a short-lived one, and as a consumer I welcome the development.

If you know that numbers like "10nm" are marketing driven then be aware that they're still significantly behind intel in actual feature/interconnect size.

 

[Lodix]

If you know that numbers like "10nm" are marketing driven then be aware that they're still significantly behind intel in actual feature/interconnect size.

But then you look at actual products in the market with more density compared to Intel's using the "inferior process node" and you wonder where is their "huge superior lead". And right now Samsung and TSMC's 10nm is more dense than Intel's 14nm. So they are ahead.

 

[agent00f]

But then you look at actual products in the market with more density compared to Intel's using the "inferior process node" and you wonder where is their "huge superior lead". And right now Samsung and TSMC's 10nm is more dense than Intel's 14nm. So they are ahead.

Samsung & others are basically ~one process behind even when they claim the same density number, so intel will be a level ahead when their 10nm ships soon.

 

[darkswordsman17]

Arguing about process advantages is meaningless if you don't have end products that are similar enough to compare (after all, look how competitive Ryzen is despite Intel's process advantage). We're not getting Intel's 10nm stuff in phones, and they're not really producing high end ARM SoCs on it (I know they were making some ARM for someone but I thought that was on older process and wasn't a newest ARM design?). And we don't know costs (Intel would likely want a hefty sum to produce on their newest node), making it not worth it for companies already essentially having no profit margin.

I actually don't know why Intel isn't just trying to make a standard ARM design (A73/A53) themselves. Integrating their own modem would offer a pretty compelling alternative to Qualcomm. And then they could do a custom design and maybe offer something premium.

Instead they seem to be trying to just get Apple's production.

Heck, right now they should look at buying up Imagination Technologies since its stock tanked and it'd be worth it for the GPU IP alone. Then offer a single chip solution. I wonder if they could come up with some small and maybe not as bleeding edge version of Optane that might boost performance for some tasks (maybe VR would be a good use case).

 

[agent00f]

Arguing about process advantages is meaningless if you don't have end products that are similar enough to compare (after all, look how competitive Ryzen is despite Intel's process advantage). We're not getting Intel's 10nm stuff in phones, and they're not really producing high end ARM SoCs on it (I know they were making some ARM for someone but I thought that was on older process and wasn't a newest ARM design?). And we don't know costs (Intel would likely want a hefty sum to produce on their newest node), making it not worth it for companies already essentially having no profit margin.

I actually don't know why Intel isn't just trying to make a standard ARM design (A73/A53) themselves. Integrating their own modem would offer a pretty compelling alternative to Qualcomm. And then they could do a custom design and maybe offer something premium.

Instead they seem to be trying to just get Apple's production.

Heck, right now they should look at buying up Imagination Technologies since its stock tanked and it'd be worth it for the GPU IP alone. Then offer a single chip solution. I wonder if they could come up with some small and maybe not as bleeding edge version of Optane that might boost performance for some tasks (maybe VR would be a good use case).

Intel's problem is high costs, in this low margin space. The only phones which can sustain them are flagships, and samsung already runs their own fab.

 

[lopri]

In my defense, I consulted Intel's own marketing material before I made the statement. I did not link to it because it has been linked multiple times on this forum and did not want to clutter the page. I will provide you with a link.

http://www.anandtech.com/show/11115...n-core-on-14nm-data-center-first-to-new-nodes

 

[agent00f]

In my defense, I consulted Intel's own marketing material before I made the statement. I did not link to it because it has been linked multiple times on this forum and did not want to clutter the page. I will provide you with a link.

http://www.anandtech.com/show/11115...n-core-on-14nm-data-center-first-to-new-nodes

As mentioned, others' 10nm are ~ intel's 14nm which has been around a while. Intel will simply be ahead when their "10nm" launches in ~year.

https://www.icknowledge.com/news/Technology and Cost Trends at Advanced Nodes - Revised.pdf
https://www.extremetech.com/computi...0nm-process-wants-change-define-process-nodes

 

[Lodix]

As mentioned, others' 10nm are ~ intel's 14nm which has been around a while. Intel will simply be ahead when their "10nm" launches in ~year.

https://www.icknowledge.com/news/Technology and Cost Trends at Advanced Nodes - Revised.pdf
https://www.extremetech.com/computi...0nm-process-wants-change-define-process-nodes

They are denser than Intel's 14nm and they will be launching 7nm next year.

 

[agent00f]

They are denser than Intel's 14nm and they will be launching 7nm next year.

Only to people who can't make heads or tails of what's linked.

 

[Lodix]

Only to people who can't make heads or tails of what's linked.

??? Your own link says the same.

 

[agent00f]

??? Your own link says the same.

It shows everyone else is ~ a level behind at same claimed density as mentioned.

 

[darkswordsman17]

It shows everyone else is ~ a level behind at same claimed density as mentioned.

I think he's talking about their 10nm. So it wouldn't make sense to call it 14nm either when it is denser than Intel's 14nm.

The entirety of these node numbers are meaningless (Intel's are too, so anyone getting mad about the other foundries need to also take issue with Intel's BS numbers) and just marketing junk at this point. It doesn't really tell you anything other than the smaller number is denser than the higher one for that same manufacturer/foundry.

Trying to distill the complexities of transistor manufacturing into a simple number is just stupid in the first place, and that's why I'm baffled at why Intel is taking such issue, or rather why they're mad and then just further muddying the waters by insisting theirs is so much better in spite of others offering similar number label, but insisting on sticking to such a stupid label. Better yet, stop pushing this when discussing end products (after all, that is what people are actually dealing with). Because that's what actually matters.

Reading the part about Intel proposing a new standard actually does make some sense (and I think their formula is reasonably straightforward and reasonably more sensible), although I think it still is trying to simplify a complex thing for no real point. It just adds pointless nerd arguments that accomplish nothing. Plus it ignores cost, which matters a bunch (basically to anyone that the simplistic nm numebrs are aimed at). Furthermore, how do we determine if a chip design is even maximizing the potential of the process? That mattersmore. Who cares if you have the best process if you don't also have chip design that makes good use of it? Likewise you could have a good/great design but if it performs poorly due to the process it would still be a bad chip.

If you can show you can offer a similar chip with better characteristics, then do it and then for the marketing you can just say that your manufacturing capabilities is what enabled that. Or if say your performance has stagnated you could then also point out that stagnant manufacturing is the reason.

 

[agent00f]

I think he's talking about their 10nm. So it wouldn't make sense to call it 14nm either when it is denser than Intel's 14nm.

The entirety of these node numbers are meaningless (Intel's are too, so anyone getting mad about the other foundries need to also take issue with Intel's BS numbers) and just marketing junk at this point. It doesn't really tell you anything other than the smaller number is denser than the higher one for that same manufacturer/foundry.

Trying to distill the complexities of transistor manufacturing into a simple number is just stupid in the first place, and that's why I'm baffled at why Intel is taking such issue, or rather why they're mad and then just further muddying the waters by insisting theirs is so much better in spite of others offering similar number label, but insisting on sticking to such a stupid label. Better yet, stop pushing this when discussing end products (after all, that is what people are actually dealing with). Because that's what actually matters.

Reading the part about Intel proposing a new standard actually does make some sense (and I think their formula is reasonably straightforward and reasonably more sensible), although I think it still is trying to simplify a complex thing for no real point. It just adds pointless nerd arguments that accomplish nothing. Plus it ignores cost, which matters a bunch (basically to anyone that the simplistic nm numebrs are aimed at). Furthermore, how do we determine if a chip design is even maximizing the potential of the process? That mattersmore. Who cares if you have the best process if you don't also have chip design that makes good use of it? Likewise you could have a good/great design but if it performs poorly due to the process it would still be a bad chip.

If you can show you can offer a similar chip with better characteristics, then do it and then for the marketing you can just say that your manufacturing capabilities is what enabled that. Or if say your performance has stagnated you could then also point out that stagnant manufacturing is the reason.

Everyone else's "10nm" is similar to intel's 14nm as illustrated. They call it "10nm" because dummies evidently believe it.

 

[lopri]

Intel's 10nm is not out yet. And pay attention to where exactly the tip of the arrow is located in that slide, which is apparently produced to fool dummies their investors. Also note that it is Intel that has been engaged in this particular marketing for some time. Rather loudly as of late. Samsung/TSMC simply announce their products.

 

[agent00f]

Intel's 10nm is not out yet. And pay attention to where exactly the tip of the arrow is located in that slide, which is apparently produced to fool dummies their investors. Also note that it is Intel that has been engaged in this particular marketing for some time. Rather loudly as of late. Samsung/TSMC simply announce their products.

It's rather "hurr durr 10nm" that's meant for dummies, in contrast to for instance normalizing nodes w/ ASML's formula that measures nodes based on CPHP and MMHP which results in actual transistor/gate feature size.

 

[Lodix]

Nobody here is saying that competitors 10nm is equal to Intel's 10nm. But it is denser than their 14nm ( even more now with the +/++ variants ). And next year TSMC/Samsung/GF are launching 7nm.

 

[agent00f]

Nobody here is saying that competitors 10nm is equal to Intel's 10nm. But it is denser than their 14nm ( even more now with the +/++ variants ). And next year TSMC/Samsung/GF are launching 7nm.

"Only to people who can't make heads or tails of what's linked."

 

[lopri]

It's rather "hurr durr 10nm" that's meant for dummies, in contrast to for instance normalizing nodes w/ ASML's formula that measures nodes based on CPHP and MMHP which results in actual transistor/gate feature size.

Here is what you are looking for.

The leading edge semiconductor logic landscape has in recent years collapsed to just four companies. The following is a summary of what is currently known about each company's plans and how they compare. ASML has analyzed many logic nodes and developed a formula that normalizes processes to a "standard node". The formula is:

Standard Node = 0.14 x (CPHP x MMHP)0.67

Where CPHP is the contacted polysilicon half-pitch and MMHP is the minimum metal half pitch.

https://www.semiwiki.com/forum/content/6160-2016-leading-edge-semiconductor-landscape.html

From the first table we can see that at a given "node" Intel is the consistent leader, however when we look at density versus year, TSMC and Samsung should both pass Intel this year with their "10nm" node processes and TSMC and Global Foundries should both lead in 2017 with their 7nm processes. Samsung is making a big bet on EUV for 7nm and will be trying to rely on their 10nm process longer than TSMC. If Samsung succeeds in introducing EUV before their competitors it could give them a cost advantage and set them up for a smoother transition to 5nm.

Why you keep resorting to personal attacks over this is beyond me.

 

[lopri]

S835 v. Exynos 8895

https://browser.primatelabs.com/v4/cpu/compare/2595333?baseline=2418073

S835 database: https://browser.primatelabs.com/v4/cpu/search?utf8=✓&q=sm-g955u
Exynos 8895 database: https://browser.primatelabs.com/v4/cpu/search?dir=desc&q=sm-g955f