Bay Trail benchmark appears online, crushes fastest Snapdragon ARM SoC

[ShintaiDK]

In practice you are right. But, I was trying to explain the concept, of profit, not being the only thing a business needs to keep their eye on, in order to succeed in the long term. The $10,000 was not meant as a serious thing (it was to make a point/example).
Some of Intels top chips are not that much less than $10,000. I think you once posted a price sheet, where the top chip was something like $3995 ?

So, that would make a dual cpu chip system, what, $8,000, and a (i.e. a 4 way) quad cpu chip system $16,000, for the Intel chips alone.

I think you competely misunderstand the concept. And please stop you anti Intel FUD. The most expensive Intel chip got a MSRP of 4616$.

If you wish to succeed long term you need profits. And even more importantly in the semiconductor business where the cost is spiralling up to pay for future R&D.

If ARM had a chip they could charge 4000$ for due to its performance, they would. Price have a tendency to match what you get.

 

[Nothingness]

Watch out to not get fooled by the usual benchmark manipulations.
Marketeers won't lie outside the lawyer determined boundaries.
They'll let the journalists and naive fans do all the lying for them.

AnTuTu and Quadrant came into focus when Microsoft choose Nvidia's
Tegra3 for surface tablets as a direct competitor.

Isn't it funny that Intel and people trying to prove Intel CPU are faster are always using AnTuTu?

It's not hard to find the reason why: AnTuTu is using the best benchmark-killer compiler, Intel icc. It's probably the only application compiled with icc on Android

Can't wait to see BayTrail/Merrifield results on Geekbench.

 

[Imouto]

So what?

Intel is releasing a chip that will beat the A15 cores and the old Krait cores. If you believe these benchmarks I don't really know why you wouldn't trust ARM's claim of A57 cores being 3x faster than A15. Hello? Maths anyone?

Intel is sitting at 0% mobile market share while all the big players want a weak Intel. Do you see Apple ditching its super profitable custom SoC to chain itself to Intel again? Do you see Samsung killing its custom SoC and foundry business to allow an Intel comeback? Do you really see any of the other cheap OEMs mounting superexpensive Intel chips? And Qualcomm will sit on its hand while an outsider eats its cake.

Good grief...

 

[mikk]

The same is true for this diagram:

You mean same wrong for you? You say AnTuTu scores are not comparable but you can't prove. You can't prove anything with a different benchmark. Same with this slide. Silvermont should have 50% better IPC than Saltwell if not more and higher singlecore frequency. We have seen a slide with a 2,4 Ghz tablet model. 2x in singlecore applications is far from unrealistic. Multithread benchmarks might gain even more since Silvermont tablet comes with up to 4 cores hence why the slide says up to 2.8x. That surely depends on the benchmark, some will gain more on Silvermont and some won't.

 

[24601]

Someone please take my money and give me a phone with Windows 8.1 on it that also works like a phone like Windows Phone 8!!!

I want my phone to boot to desktop :O.

Is at least 720p.

Is LTE.

And is 100% native x86.

And have 100% native DX11 capability.

And have a start menu (╯°□°）╯︵ ┻━┻ (With the option of metro).

And be able to do every single thing my desktop can do :O.

I hate having to have a laptop and want to replace it with a phone.


Right now only WINTEL can bring this to me, and they refuse!!!!

 

[SiliconWars]

Exophase (guessing it's the same one) wrote this in response to Intel's last "reveal".

http://seekingalpha.com/user/13480892/comments/symbol/armh

a) The x86 target is compiled with ICC
b) The ARM target only supports untweaked NDK GCC - more importantly, just VFP w/16 registers and no NEON path is supported

That's how reliable this AnTuTu benchmark is.

The question you need be asking yourselves is this - if Silvermont is anywhere near as good as what Intel is trying desperately to make us believe, why are they so desperate to use every shenanigan in the book?

 

[Chiropteran]

Also, Intel's last chip didn't massively outperform the competition like this one does.

Huh? 30% in only one benchmark is massive?

http://www.anandtech.com/show/5770/lava-xolo-x900-review-the-first-intel-medfield-phone/4

The medfield phone was 35% faster in the Sunspider javascript test, and in the end it didn't make any difference whatsoever.

 

[beginner99]

So what?

Intel is releasing a chip that will beat the A15 cores and the old Krait cores. If you believe these benchmarks I don't really know why you wouldn't trust ARM's claim of A57 cores being 3x faster than A15. Hello? Maths anyone?

Yeah and do you believe those A57 cores run in the same power envelope as A15 and silvermont? they are for low power server and tablets, not for phones and as I already said in this very thread compete probably with haswell ULV 15W (probably) which obviously is in a different league performance wise and then 3x times A15 doesn't' look that pretty anymore.

Intel is sitting at 0% mobile market share while all the big players want a weak Intel. Do you see Apple ditching its super profitable custom SoC to chain itself to Intel again? Do you see Samsung killing its custom SoC and foundry business to allow an Intel comeback? Do you really see any of the other cheap OEMs mounting superexpensive Intel chips? And Qualcomm will sit on its hand while an outsider eats its cake.

Good grief...

Samsung is releasing several clovertrail+ based devices...like galaxy tab 3. So yeah this is actually happening even before silvermont.

 

[Arachnotronic]

So what?

Intel is releasing a chip that will beat the A15 cores and the old Krait cores. If you believe these benchmarks I don't really know why you wouldn't trust ARM's claim of A57 cores being 3x faster than A15. Hello? Maths anyone?

Intel is sitting at 0% mobile market share while all the big players want a weak Intel. Do you see Apple ditching its super profitable custom SoC to chain itself to Intel again? Do you see Samsung killing its custom SoC and foundry business to allow an Intel comeback? Do you really see any of the other cheap OEMs mounting superexpensive Intel chips? And Qualcomm will sit on its hand while an outsider eats its cake.

Good grief...

A57 3x faster? That's...not right. It's expected to be 20-30% faster than the Cortex A15, according to ARM's own benchmarks,

...and it's not coming until 2015...

...and the 2015 implementations will likely be on TSMC 20nm, rather than the FinFET "16nm", while Intel will be squarely at a "real" 14nm and be ramping 10nm.

 

[SOFTengCOMPelec]

Stuff

Reply DELETED, to avoid causing unnecessary arguments on the forum.
.
.
.
EDIT: I agree, I can be a fan boy at times, and/or antagonistic against Intel, or other companies, sometimes.

 

[Nothingness]

Exophase (guessing it's the same one) wrote this in response to Intel's last "reveal".

http://seekingalpha.com/user/13480892/comments/symbol/armh

That's spot on!

That's how reliable this AnTuTu benchmark is.

Let's have some fun:
- Geekbench: S600 vs Z2580 3218 vs 1426 (even if you pick single-thread results, Z2580 lags far behind except on Stream)
- many benchmarks comparing K900 against other ARM phones: http://www.fonearena.com/blog/73728/lenovo-k900-benchmarks.html

Except for Sunspider and AnTuTu the Z2580-based K900 always is far behind. If that's not enough proof that any AnTuTu result should be taken with a huge grain of salt...

 

[mikk]

That was two years ago, its ancient history since the guys that screwed things up got fired. Oh, Bulldozer!! Fail!!?! So what, it happened. Do you hear saying Pentium 4 utter fail rofl trolololo? No, because its history.

So go back where you came from...

Bulldozer isn't history. Current products like Vishera or Trinity are still based on Bulldozer architecture, albeit on a tweaked one. Pentium 4 on the other side is completely irrevelant today.

 

[Imouto]

A57 3x faster? That's...not right. It's expected to be 20-30% faster than the Cortex A15, according to ARM's own benchmarks,

Compared both as being the same crap.

http://www.arm.com/about/newsroom/a...s-most-energy-efficient-64-bit-processors.php

Now please ctrl+f "delivering up to three times the performance of today’s superphones".

Same stupid void claim as leaking an useless benchmark score.

...and the 2015 implementations will likely be on TSMC 20nm, rather than the FinFET "16nm", while Intel will be squarely at a "real" 14nm and be ramping 10nm.

Making Intel chips even more expensive. Qualcomm won't allow another alpha male in the pack.

 

[Arachnotronic]

Making Intel chips even more expensive. Qualcomm won't allow another alpha male in the pack.

How do you figure on that first part? And the second part - please. Qualcomm doesn't really have a choice.

 

[USER8000]

Making Intel chips even more expensive. Qualcomm won't allow another alpha male in the pack.

Qualcomm also has only $60 million in debt and tens of billions of dollars in cash,which they can spend when the need arises. They are also only one of several ARM licensees too.

TSMC has massive volume too,and hence they can probably afford to charge less than any competitor,as the development costs for their process nodes can be amortalised over more CPUs.

 

[Imouto]

How do you figure on that first part? And the second part - please. Qualcomm doesn't really have a choice.

Maybe because a smaller process is always more expensive?

You're getting it wrong. Qualcomm has the upper hand here. While Intel is bigger its market (I recall 70% of its revenue coming from PC) is in free fall. Qualcomm is growing (like a tumor) and Intel is posting worse YoY results every single quarter. To cream it I'll say it again, Intel is sitting at 0% mobile market share. You're somehow expecting it to barge in, kick everyone's ass, kiss the girl and steal all the money.

The mobile market is locked for Intel and every other player wants it that way.

 

[gofmarat]

Maybe because a smaller process is always more expensive?

You are absolutely wrong! A smaller process is always less expensive,

The smaller the transistors, the smaller they can make each individual die, the more dies they can make from each wafer, the less each die costs.
Each process improvement brings not only power improvements but potential manufacturing cost reduction.

 

[frozentundra123456]

Does this even matter? Intel already had a phone processor, and it supposedly performed really well, and nobody anywhere bought a phone using it.

When you buy a cell phone, there are are about a dozen features which are all more important than CPU performance. I don't see this going anywhere quickly. Maybe a few niche phones will use it, but it won't really be relevant.

If it's not used in the iPhone or some form of Samsung Galaxy it doesn't really matter for a real smartphone enthusiast.

I do see the phone market being much more difficult than the tablet market for Intel (or AMD) to make headway in. Intel and AMD are both late to the game, and ARM/Android is firmly entrenched with a ton of apps and "good enough" performance.

In the tablet space the outlook may be better. With the bigger formfactor, at least personally, I would not buy another that does not run x86. I am willing to accept the limitations of android on a phone, but in a tablet, I want more. BTW, I do have an Android tablet, and like the size and portability, but technically it reminds me of a PC, circa 2000 with a celeron, windows ME, and extremely lousy integrated graphics. In other words, the performance sucks really, really bad.

 

[SOFTengCOMPelec]

You are absolutely wrong! A smaller process is always less expensive,

My understanding is that it depends on a lot of factors. To generalize, and say that it is ALWAYS less expensive is probably too much of a generalisation.

Sometimes, it can actually cost more (I am led to believe). New fab plants can be extremely expensive, and the huge sales in chips needed to pay for it, DON'T always materialize.

Anyway, WELCOME TO THE FORUMS.

You are absolutely wrong! A smaller process is always less expensive,

The smaller the transistors, the smaller they can make each individual die, the more dies they can make from each wafer, the less each die costs.
Each process improvement brings not only power improvements but potential manufacturing cost reduction.

Let me give you one example.

What if going down to the new, tiny transistor size, on the new die shrink process, means that the yields are terrible. How will that effect the prices (until yields are hopefully improved in the future) ?

 

[blackened23]

...and the 2015 implementations will likely be on TSMC 20nm, rather than the FinFET "16nm", while Intel will be squarely at a "real" 14nm and be ramping 10nm.

You know what, I can't wait for Bay Trail to just be released just so all of these folks will then have nothing to say. It will be poetic justice since ARM Holdings have been so excessively arrogant about their parts "never being beat by intel" in terms of efficiency. Yet this is going to happen. It's going to be hilarious. In the meantime, I'll just ignore arguing with the folks who have an extreme hatred of intel for whatever reason.

The main understanding here is that phone and tablet makers haven't used intel because 1) The original atom was a complete joke in terms of performance, therefore not desirable 2) In terms of efficiency, intel has always been behind ARM SOCs in prior years. This is now changing. ULV Haswell beats the best ARM SOCs in terms of performance while having 20x more performance. Now we're getting Bay Trail which will have even better battery life than ULV Haswell (which already beats ARM SOCs) while still having more performance and better efficiency. Again, it's going to be hilarious to watch ARM backtrack on their claim that intel will never beat them efficiency wise. It's going to happen, and the doubters will be able to do nothing but maintain silence.

 

[Khato]

Compared both as being the same crap.

http://www.arm.com/about/newsroom/a...s-most-energy-efficient-64-bit-processors.php

Now please ctrl+f "delivering up to three times the performance of today’s superphones".

Same stupid void claim as leaking an useless benchmark score.

Not really. ARM's claim provides no details on their frame of reference whatsoever - they could easily be comparing a 40nm A9 'superphone' to an A57 produced on a node below 20nm.

The only ambiguity in Intel's marketing slides is the exact weighting of their benchmark suite. We know exactly what the pre-production Silvermont system was compared to - a stock Z2580. (Claiming that the Silvermont system was used as the reference point for the iso-perf and iso-power comparisons is grasping at straws.)

As for the validity of the Antutu benchmark, I'm not surprised by the score, but I suspect that the reported frequency isn't correct. That said, I also expect that it uses quite a bit less power to obtain that score than Snapdragon 800 does for its ~33k. (I'm relatively confident in such because as was mentioned in the most recent Anandtech Podcast 21 at around 1:05 in the ARM SoC's are frequently going up into the 1.25V to 1.3V range in order to hit their maximum frequencies while Intel's 22nm SoC process is tuned for the 0.75V to 1.0V range.)

 

[zephyrprime]

I'm under the impression that big.Little is completely worthless for consumer level products such as tablets and smartphones, while it *can* be useful in highly parallel server environments. This is why Qualcomm and Apple do not, and likely will never have, plans to use big.Little. That's my impression anyway, someone correct me if i'm mistaken.

It seems to me like it'd be the other way around. Server environments will usually have some load so the chips will never switch to "little" cores so having a big/little processor is useless. Also, server's don't care nearly as much about power consumption as mobile devices anyway. I think the prob with big little is that it is too undeveloped right now. It doesn't make sense to have 4 little cores on a 4 big core cpu. It would be much better to have just 1 little core and 4 big cores. However, changing the number of available cores dynamically would require a change to the OS. Having an equal number of big and little cores allows for easy software compatibility.

 

[gofmarat]

WELCOME TO THE FORUMS.[/QUOTE]

Thanks! :thumbsup:

You are probably right, always is too definitive, still my statement stands.

Don't forget that Intel have been investing in improving their process technology (every 2 years) for a lot of time . this means that they have been investing money on their fabs this whole time, nothing has changed except the SOC die size.

 

[SOFTengCOMPelec]

WELCOME TO THE FORUMS.

Thanks! :thumbsup:

You are probably right, always is too definitive, still my statement stands.

Don't forget that Intel have been investing in improving their process technology (every 2 years) for a lot of time . this means that they have been investing money on their fabs this whole time, nothing has changed except the SOC die size.

Completely off-topic, but anyway.

We seem to be ending up (in the not too distant future) that there is only enough money/resources/need for one giant fab plant per planet!. How on 'earth' is that going to happen, with so many competing companies, fighting each other ?

Back on topic, yes, amazingly, one of (if not the first) Intels fabs, made the very first commercially available microprocessor. So, their technical abilities to fend off 'Arm' are immense, both on the fab front (1971..present time = Very long time), and on the latest advanced engineering with cpu architectures.
So, it's going to be an interesting fight between Intel, Arm, AMD and anyone else involved.

 

[ShintaiDK]

Completely off-topic, but anyway.

We seem to be ending up (in the not too distant future) that there is only enough money/resources/need for one giant fab plant per planet!. How on 'earth' is that going to happen, with so many competing companies, fighting each other ?

There is no room for anyone but 1 in the endgame. The IC design might actually be the most expensive and not the fab.