The Intel Atom Thread

[liahos1]

Just different OSes reporting things differently.

right but doesnt baytrail turbo during geekbench? Shouldnt geekbench capture that turbo speed in the processor description. the reason i ask is because a7's results are a 1.3 or 1.4ghz (cant remember) but people were speculating it might be turbo-ing as well.

sorry but something like processor speed should be captured in a similar fashion for an soc regardless of os.

 

[beginner99]

Did anyone listen to the new podcast?

Apparently these preview benchmarks were taken in a room controlled by Intel which was incredibly cold- it's described as "literally it was a refrigerator in there". I wonder how much that helped Atom's turbo performance?

weird. But then I guess OEMs sure get a chance to test the hardware themselves in a settign of their choosing before committing to a product?

 

[Nothingness]

right but doesnt baytrail turbo during geekbench? Shouldnt geekbench capture that turbo speed in the processor description. the reason i ask is because a7's results are a 1.3 or 1.4ghz (cant remember) but people were speculating it might be turbo-ing as well.

sorry but something like processor speed should be captured in a similar fashion for an soc regardless of os.

Getting CPU clock is an OS dependent feature. Add to that that depending on the load, when you ask, the same OS might give different answers.

IIRC the only OS where this is "manually" computed is iOS which has no way to get CPU clock.

If John Poole comes here, he'll fix my mistakes 🙂

EDIT: I agree with you that speed should be computed on all OSes, not asked to the OS.

 

[Khato]

right but doesnt baytrail turbo during geekbench? Shouldnt geekbench capture that turbo speed in the processor description. the reason i ask is because a7's results are a 1.3 or 1.4ghz (cant remember) but people were speculating it might be turbo-ing as well.

sorry but something like processor speed should be captured in a similar fashion for an soc regardless of os.

Definitely agree that it should capture SoC speed during the actual benchmark run in a consistent fashion, but that's definitely not the case currently. In the PC space it typically ends up reporting the base frequency rather than turbo/overclocked, which makes finding 'stock' scores rather annoying.

And yeah, with the iPhone 5 it took almost a week for them to discover that the A6 was operating at 1.3 GHz, not the 1 GHz initially reported by all the programs. That's part of the reason I wouldn't be surprised if they're wrong again this time.

 

[Arachnotronic]

Did anyone listen to the new podcast?

Apparently these preview benchmarks were taken in a room controlled by Intel which was incredibly cold- it's described as "literally it was a refrigerator in there". I wonder how much that helped Atom's turbo performance?

It really was cold in there and I was a little suspicious of that also, heh.

However, during the actual power consumption demos, the room was normal temperature.

We'll see what the performance is like when the devices are in the hands of the customers.

 

[Sweepr]

Avoton with 8C @ 2.4GHz scores 3.77, thats quite a bit more than twice Z3770's 4C @ 1.46-2.4GHz score 1.48 pts, even if we take the 32-bit vs 64-bit difference into account. I guess Z3770 doesnt run at max Turbo (2.4GHz) with four cores @ load, maybe 2.0-2.15 GHz? Bay Trail-D (Pentium J2850 - 4C @ 2.4GHz base clock) should perform better than the Z3770 (no surprise really).

http://forums.servethehome.com/proc...-c2750-benchmarks-supermicro-a1sai-2750f.html

 

[Khato]

Avoton with 8C @ 2.4GHz scores 3.77, thats quite a bit more than twice Z3770's 4C @ 1.46-2.4GHz score 1.48 pts, even if we take the 32-bit vs 64-bit difference into account. I guess Z3770 doesnt run at max Turbo (2.4GHz) with four cores @ load, maybe 2.0-2.15 GHz? Bay Trail-D (Pentium J2850 - 4C @ 2.4GHz base clock) should perform better than the Z3770 (no surprise really).

http://forums.servethehome.com/proc...-c2750-benchmarks-supermicro-a1sai-2750f.html

Keep in mind that the C2750 has a 2.6 GHz turbo frequency. Whether or not that was active is up for debate, but could be responsible for a small portion of the difference too.

I definitely agree that the Z3770 wasn't running at maximum turbo on the multi-threaded workloads, otherwise it would've been consuming closer to 4W than something around 2.5W. Unfortunately it's difficult to guess at per-core power consumption from these results as the power supply isn't listed and there's the SSD I/O added in on the 'load' figure. But it is safe to say that under load, most likely at 2.6 GHz, each core was using less than 2W.

 

[SiliconWars]

That's pretty good performance for the server part at least. 8 Jaguars at 1.6 GHz would probably score around 3 in CB11.5 for similar power draw.

 

[Khato]

So, what are thoughts on this little tidbit? The original source apparently being a report from Morgan Stanley.

Market reports suggest that Bay Trail for tablets will be priced at $10 and for notebooks at $20 a chipset.

If true, it'd definitely indicate that Intel's more interested in marketshare than margins, which is likely the correct move for them to make initially. It'd definitely make sense if putting fab capacity to use is a top priority, especially when it comes with the bonus of depriving the competition of sales. Sell the tablet SKUs for minimal profit in order to quickly secure marketshare while keeping the remainder of the SKUs at 50%+ margins so as to not affect margins in their traditional markets.

 

[liahos1]

So, what are thoughts on this little tidbit? The original source apparently being a report from Morgan Stanley.



If true, it'd definitely indicate that Intel's more interested in marketshare than margins, which is likely the correct move for them to make initially. It'd definitely make sense if putting fab capacity to use is a top priority, especially when it comes with the bonus of depriving the competition of sales. Sell the tablet SKUs for minimal profit in order to quickly secure marketshare while keeping the remainder of the SKUs at 50%+ margins so as to not affect margins in their traditional markets.

I spoke with Joe Moore at Morgan Stanley about that and he was pretty confident in his checks. Incidentally, I heard renee james saying the same thing to the evercore analyst at a lunch during IDF. And a friend of mine came back from Taiwan 3 weeks ago with similar numbers.

I think they just need to be in as many sockets as possible, with products that are actually well built. Then the reviews coming in, we know what performance is and hopefully battery life is very good as well. Then all the FUD can start diminishing

 

[Khato]

I spoke with Joe Moore at Morgan Stanley about that and he was pretty confident in his checks. Incidentally, I heard renee james saying the same thing to the evercore analyst at a lunch during IDF. And a friend of mine came back from Taiwan 3 weeks ago with similar numbers.

I think they just need to be in as many sockets as possible, with products that are actually well built. Then the reviews coming in, we know what performance is and hopefully battery life is very good as well. Then all the FUD can start diminishing

Thanks for the further information 🙂 It's great to have some forum members that can provide first-hand information.

And I completely agree about their motivation. It's a play that I'd been expecting Intel to make for quite some time - I wonder if we have the new CEO to thank for the change in direction. Perhaps Paul was too fixated on maintaining the status quo on margins whereas BK with his background on the manufacturing side recognizes that the only way for Intel to leverage its fabs and process technology lead are to actually use them.

For comparison to that $10 figure. A quick search yields that the single-core A8 based Allwinner A10 is something around $7 apparently, and then the isuppli Galaxy S4 BoM lists Qualcomm's Snapdragon 600 at $20... Yeah, if I were any of the ARM smartphone/tablet SoC players I'd be quite worried about an Intel that's fine with taking a hit on margins in order to steamroll its way into that market.

 

[liahos1]

Thanks for the further information 🙂 It's great to have some forum members that can provide first-hand information.

And I completely agree about their motivation. It's a play that I'd been expecting Intel to make for quite some time - I wonder if we have the new CEO to thank for the change in direction. Perhaps Paul was too fixated on maintaining the status quo on margins whereas BK with his background on the manufacturing side recognizes that the only way for Intel to leverage its fabs and process technology lead are to actually use them.

For comparison to that $10 figure. A quick search yields that the single-core A8 based Allwinner A10 is something around $7 apparently, and then the isuppli Galaxy S4 BoM lists Qualcomm's Snapdragon 600 at $20... Yeah, if I were any of the ARM smartphone/tablet SoC players I'd be quite worried about an Intel that's fine with taking a hit on margins in order to steamroll its way into that market.

totally. Hopefully they are aggressive as hell with merrifield as I think greater success in smartphones will be an even bigger wakeup call to ppl thinking intel is going bankrupt

 

[SiliconWars]

I love how being forced into a huge price cut before the chips are cold is being spun as a positive move. 😀 For anyone else that would be a disaster. Isn't that what Nvidia had to do with Tegra?

Like I said weeks ago, Intels numbers just didn't add up. They couldn't break the tablet and phone market while maintaining current margins, something had to give.

 

[blackened23]

Well of course you would say that. And for your mention of "spin", your "spin" is equally speculative and "spin" worthy. Oh wait, you have insiders information at Intel. You're aware of a price war that nobody else is aware of. Sure.

By all means continue dumping on intel in every thread, though.

 

[liahos1]

I love how being forced into a huge price cut before the chips are cold is being spun as a positive move. 😀 For anyone else that would be a disaster. Isn't that what Nvidia had to do with Tegra?

Like I said weeks ago, Intels numbers just didn't add up. They couldn't break the tablet and phone market while maintaining current margins, something had to give.

there margins by definition are going to be higher than anyone else's in the industry. even if lower than their historic margins. for an soc that performs the way it does they should be pricing it between an s800 and s600. instead they are pricing it at mediatek levels. its pretty easy to see why dont you think. they want to get their foot in the door. its funny because if any arm based vendor did the same thing trying to break into the server market. you'd spin it as "intel is doomed...something something something... arm is amazing"

 

[SiliconWars]

there margins by definition are going to be higher than anyone else's in the industry. even if lower than their historic margins. for an soc that performs the way it does they should be pricing it between an s800 and s600. instead they are pricing it at rockchip levels. its pretty easy to see why dont you think. they want to get their foot in the door. its funny because if any arm based vendor did the same thing trying to break into the server market. you'd spin it as "intel is doomed...something something something... arm is amazing"

No what I always said is that nobody is going to pay more for Intel phone/tablet chips just because they are a little bit faster, or because they are Intel.

We know Bay Trail isn't much (if at all) smaller than the competing ARM chips. We know the performance isn't exactly dominating. In short, they are just another chip maker struggling to break the mobile market and being forced to get by on razor-thin margins. None of you believed Bay Trail would be $10 or $20 a month ago - Intel couldn't have believed it either because their Q3 guidance on margins simply doesn't add up otherwise.

 

[Nothingness]

Z3740 vs Z3770: http://browser.primatelabs.com/geekbench3/compare/59070?baseline=53855

Ignoring stream, ST is about 28% faster, MT is only 20%. I wonder if turbo is lower in MT on Z3740 compared to Z3770.

 

[krumme]

And as predicted they will sell it low to get a fot in. They might as well nearly give them away for free it wouldnt matter for their profitability with all the huge fixed cost tied to it.

Intel cost structure can not support this for the long run. But ofcourse servers and b2b market can contine to finance it for perhaps 2 more years. Tops. Look at capex and expected profit comming years.

What will change this situation?
An eartquake destroys tsmc new fabs ?
Suddenly the compilers for armv8 is stolen by aliens?
Win 8 becomes popular?

Give me the perspective on this. The situation just looks worse each day imho.

 

[Khato]

None of you believed Bay Trail would be $10 or $20 a month ago - Intel couldn't have believed it either because their Q3 guidance on margins simply doesn't add up otherwise.

I know I wouldn't have, but I was hoping it would. Under Paul, Intel had been fixated upon margins and there were no indications as to whether or not this would change with the new CEO. Luckily, it appears that it has. And it's questionable how much this kind of move will actually affect margins on the whole given that it doesn't cost Intel a lot more to produce Baytrail chips than it does to have fab equipment sitting idle.

Give me the perspective on this. The situation just looks worse each day imho.

Indeed it does... for TSMC and Intel's ARM competitors. Intel is in no risk whatsoever of being unable to continue their process technology roadmap - even selling products for mobile markets that they're expanding into at foundry-type profits keeps their cost structure quite solid. I still can't understand why some choose to believe that Intel production costs are higher than the competition.

But as I've said before, what happens to TSMC and the ARM competition when an Intel that isn't concerned about margins rolls into the mobile markets? How much revenue does Qualcomm lose when they're both not selling Snapdragon SoCs and missing out on the attached modem sales to Intel as well? Or to TSMC who, as some have delighted in reminding, has increased capex spending to nearly Intel levels in anticipation of future demand when that demand goes to Intel instead? I can guarantee that they aren't going to be able to deal with fab equipment sitting idle anywhere near so gracefully as Intel.

 

[Arachnotronic]

I love how being forced into a huge price cut before the chips are cold is being spun as a positive move. 😀 For anyone else that would be a disaster. Isn't that what Nvidia had to do with Tegra?

Like I said weeks ago, Intels numbers just didn't add up. They couldn't break the tablet and phone market while maintaining current margins, something had to give.

Their current corporate gross margin level is ~60%. Do you realize that unless you're a company like AMD that can't ever turn a profit and literally exists to damage up profitability for the rest of the world, most semiconductor companies - from Atmel that sells $0.50 microcontrollers to Qualcomm that sells mobile SoCs up and down the stack - take in gross margins of 50%+? Intel's margins are good, but they're not as obscene as people like to make them out to be, and I bet a good part of the margin uplift comes from the datacenter group rather than gouging on the PC client side that everyone thinks.

It looks to me that Intel is aggressively pricing these parts to take share now and ask about significant profitability later. As an Intel shareholder, I'm fine with this. Intel needs to gain the trust of its customers and establish itself as a credible supplier here. Also, Bay Trail is a good part, but it's certainly not head-and-shoulders above the competition enough to warrant a premium price.

 

[Enigmoid]

Keep in mind that the C2750 has a 2.6 GHz turbo frequency. Whether or not that was active is up for debate, but could be responsible for a small portion of the difference too.

I definitely agree that the Z3770 wasn't running at maximum turbo on the multi-threaded workloads, otherwise it would've been consuming closer to 4W than something around 2.5W. Unfortunately it's difficult to guess at per-core power consumption from these results as the power supply isn't listed and there's the SSD I/O added in on the 'load' figure. But it is safe to say that under load, most likely at 2.6 GHz, each core was using less than 2W.

I don't think its turbo. Might just be poor scaling on lower power cores.

We know from the AT articles than the a6-5000 ran at 1.5ghz solid.

Jaguar has 3.85x scaling, Baytrail has 3.7x scaling in Cinebench. Realize that we are talking about 0.01 and 0.02 points deviation which is quite possibly error.

Jaguar has 3.41x scaling, Baytrail has 3.47x scaling. Again, this is within error range.

Honestly does not look like turbo problems. Possibly its due to the module nature of baytrail (two cores + cache) not scaling terribly well (when two cores are active they share cache but when one core is active it gets all the cache?). This slightly modular architecture might also be the reason for the non-linear power scaling.

 

[Sweepr]

Dell drops XPS 10, leaving Microsoft's Surface as sole Windows RT tablet

Dell’s XPS 10 tablet models with Windows RT have been removed from the company’s website, which analysts said could leave Microsoft as the only vendor selling ARM-based tablets running versions of Windows RT.

The XPS 10 Web page lists models of the tablet as being “unavailable,” and points users to the Latitude 10 tablet, which runs Windows 8 and has an Intel Atom processor. Microsoft, which sells Surface RT, is now the only device maker selling a tablet with Windows RT.

Dell is holding an event in New York on Oct. 2 where the company will announce new tablets. The company did not comment on whether a new Windows RT 8.1 tablet would be launched, but has showed a new 8-inch Venue tablet with Windows 8 and the Intel Atom processor code-named Bay Trail.

Dell was the only device maker other than Microsoft selling a Windows RT tablet after Lenovo, Asus, and Samsung bailed out on the device. Dell is holding its tablet event ahead of Microsoft’s release of Windows 8.1 RT as a free download for existing Windows RT devices after Oct. 18. Microsoft on Monday announced Surface 2, which is the first tablet based on Windows 8.1 RT.

More here: http://www.pcworld.com/article/2049...et-vendor-as-dells-xps-10-is-unavailable.html

 

[Nothingness]

One can still buy XPS 10 in UK: http://www.dell.com/uk/p/XPS-10-tablet/pd
Same in Dell France and Italy.

 

[ShintaiDK]

One can still buy XPS 10 in UK: http://www.dell.com/uk/p/XPS-10-tablet/pd
Same in Dell France and Italy.

Just a matter of inventory before its gone too. WinRT is dead. All OEMs want x86.

 

[Nothingness]

Just a matter of inventory before its gone too.

Probably yes. I'm just surprised DELL didn't make any announce about that.

WinRT is dead. All OEMs want x86.

Traditional PC OEM yes. I wonder why MS insists...