The Intel Atom Thread

[jdubs03]

Why exactly does such require Airmont to improve? What's known about A57 thus far implies it will only have a ~30% performance boost over A15, which would be quite good if it maintained the same power consumption. However as A15 demonstrated ARM isn't necessarily great at efficiency when they're pursuing higher levels of performance, so A57 may well continue that trend.

I say that because, it isn't just the switch from A15 to A57 that Cherry Trail will be competing with but the top-tier QCOMM + Samsung offerings that are at 20nm too. I agree that A57 could push that power-envelope, but that helps when the node is decreased.

New mirco-arch plus 14nm will likely give CPU and GPU dominance. I'm not sure that just 14nm will be enough. It should be pretty interesting to see how it plays out.

 

[cbn]

Still, if there is anything good from all this is confirmation that 2 lanes can be paired to make x2, I hope OEM choose to use it if planning to add a 820M on a BT device.

I think the Bay Trail iGPU would be adequate for me.

If adding a small discrete GPU, then I would prefer it go on something like Avoton. This part doesn't come with an iGPU.....but it does have some features Bay Trail-D is missing such as AES-NI and ECC RAM capability. (Avoton also comes with 2+ Intel LAN integrated into SOC).

 

[Infraction Jack]

I have also been considering the Avoton C2750. 2.4GHz Octa-core, seems like it would make for a good game sever and a light gaming machine with a hd 7750.

http://www.asrock.com/server/overview.asp?Model=C2750D4I

 

[shady28]

I have also been considering the Avoton C2750. 2.4GHz Octa-core, seems like it would make for a good game sever and a light gaming machine with a hd 7750.

http://www.asrock.com/server/overview.asp?Model=C2750D4I

Not sure why you would do that given the price. That board runs ~ $350. Looking at its specs, its clearly meant to drive a NAS (lots of drive ports, dual network ports, only an x8 PCIe slot).

An A10-7850K will cost a lot less and benchmarks better - even in heavily threaded loads. Looking at geekbench 3 - the Avoton runs around 7300 which is slightly below A10-6800k levels.

Would have been a lot more interesting chip if someone had put it into a consumer oriented board.

 

[Infraction Jack]

The price is not that bad because it is a board+8 core cpu in MITX and I will be able to run it all on a 250 watt PSU.
a A10-7850K and a GIGABYTE GA-F2A88XN-WIFI FM2+ would cost me $281.68 and would be a power hog with only 4 cores to work with.

 

[Sweepr]

Not sure why you would do that given the price. That board runs ~ $350. Looking at its specs, its clearly meant to drive a NAS (lots of drive ports, dual network ports, only an x8 PCIe slot).

An A10-7850K will cost a lot less and benchmarks better - even in heavily threaded loads. Looking at geekbench 3 - the Avoton runs around 7300 which is slightly below A10-6800k levels.

Would have been a lot more interesting chip if someone had put it into a consumer oriented board.

I think Avoton would have a slight advantage in many MT loads. That's an awkward comparison ($179 Core i5-4440 would smash both) but Avoton packs respectable MT performance for a 20W chip (the other two have 77-95W TDPs). IMHO it would make an interesting (niche) desktop chip.

Cinebench 11.5 - 3.77
7-zip total rating - 13509 MIPS
TrueCrypt AES - 2.0 GB/s

http://forums.servethehome.com/inde...-c2750-benchmarks-supermicro-a1sai-2750f.2444

A10-7850K
Cinebench 11.5: 3.59
7-zip total rating: 11295 MIPS
TrueCrypt AES - 2.1 GB/s

http://new.sliven.net/res/news/129364/AMD 4th Generation APU Reviewers Guide.pdf

 

[Sweepr]

Intel's Back-To-School Windows/Android Platform Leaked

Following recent hints that new versions of the Baytrail-M processors would be launched Intel have confirmed by publishing details of the new student-focused platform ahead of IDF. It’s called BTS’14 and it’s aimed at lowering the cost of Android and Windows tablet and 2-in-1 products.

A number of Pentium N3000 CPUs will be launched which include improved CPU and GPU frequency over previous Baytrail-M parts and support for Intel Quick Sync, the hardware video encoder usable through the Intel Media SDK. A set of Celeron N2000-series CPUs will also feature Intel Quick Sync support, increased memory speed over previous models and in some cases a CPU clock boost.

CPUs known to be under the new BTS’14 banner are:

Intel Pentium N3530 processor. 2.17Ghz-2.58Ghz. GPU 896Mhz. 7.5W TDP
Intel Celeron N2830 processor. 2.17Ghz-2.42Ghz. GPU 750Mhz. 7.5W TDP
Intel Celeron N2930 processor, 1.83Ghz-2.16Ghz. GPU 854Mhz. 7.5W TDP
Intel Celeron N2807 processor. 1.58Ghz-2.16Ghz. GPU 750Mhz. 4.3W TDP

Support is also being added for lower-cost eMMC SSDs and SDIO-connected WiFi modules.

www.umpcportal.com/2014/04/intels-back-to-school-windowsandroid-platform-leaked

 

[sm625]

Atom is too underpowered for the next process. At 14nm you can fit 2c4t broadwell into a tablet and have similar power characteristics as an ipad. But atom is just too slow. A Z3770 scores roughly the same as an AMD E450 in single threaded operations, which is what sets the tone of a device's general performance. It is slower than a Celeron SU2300 in terms of pure integer performance. It is basically just a pentium 4 2.0GHz with AES, AVX, etc added to it. Its just not enough. Javascript performance will be atrocious, leading most users to be frustrated with it. If intel doesnt ditch atom and start promoting its Y series (pentium 3560Y for example), then it is simply going to be eliminated from the mobile market. It makes much more sense to shrink and integrate something like a 3560Y than it does to try and make the Z3770 faster.

 

[Sweepr]

I think they're doing the right thing. They probably want to keep their Core ix line ''fat margins'' with Broadwell-Y, while Atom aims at cheaper ARM-competitors. Lets give them some credit, we've seem huge gains in the Clover Trail -> Bay Trail transition and who knows what comes next? Perhaps this year's Cherry Trail is more than a 14nm Bay Trail (CPU performance) and there's a new architecture which we still dont know nothing about coming around mid-2015, Goldmont. Now that they've released a proper quad-core (not 2C + HT) I expect them to focus on per-core performance improvements.

 

[shady28]

I think they're doing the right thing. They probably want to keep their Core ix line ''fat margins'' with Broadwell-Y, while Atom aims at cheaper ARM-competitors. Lets give them some credit, we've seem huge gains in the Clover Trail -> Bay Trail transition and who knows what comes next?

I have to agree. Bay Trail is a great product for tablets, which is clearly what it's aimed at. From what I can tell, it's reinvigorated the Win 8.1 tablet platform to some degree. If bay trail had been around for the Win 8 launch, things might be going differently at MS.

If they can pull out another 20% - 30% in CPU performance and maybe 50% in iGPU from Cherry Trail, and keep the same or lower power profile, then I think it will make Win 8 tablets a reasonable desktop replacement for the average consumer.

I don't think that will hurt overall sales as much as some believe. Desktops tend to get upgraded and last a long time. Laptops, less so. Tablets, even less. So they sell a cheaper chip, but people turn over their machines more frequently.

This is the way the industry is going and they really have no choice but to make their best shot; anything less and they risk being relegated to server farms / data centers, and eventually not even that.

 

[cbn]

Atom is too underpowered for the next process. At 14nm you can fit 2c4t broadwell into a tablet and have similar power characteristics as an ipad. But atom is just too slow. A Z3770 scores roughly the same as an AMD E450 in single threaded operations, which is what sets the tone of a device's general performance. It is slower than a Celeron SU2300 in terms of pure integer performance. It is basically just a pentium 4 2.0GHz with AES, AVX, etc added to it. Its just not enough. Javascript performance will be atrocious, leading most users to be frustrated with it. If intel doesnt ditch atom and start promoting its Y series (pentium 3560Y for example), then it is simply going to be eliminated from the mobile market. It makes much more sense to shrink and integrate something like a 3560Y than it does to try and make the Z3770 faster.

11.5 watt Pentium 3560Y (Haswell dual core):

http://www.passmark.com/cpubenchmark/cpu.php?cpu=Intel+Pentium+3560Y+@+1.20GHz&id=2078

1093 multi-thread cpu marks
581 single thread cpu marks

10 watt Pentium J2900 (Bay Trail-D quad core):

http://cpubenchmark.net/cpu.php?cpu=Intel+Pentium+J2900+@+2.41GHz&id=2173

2169 multi-thread cpu marks
611 single thread cpu marks

The Bay Trail-D actually beats the Y-series Pentium in single thread, while having a lower TDP.

Is there any reason to believe this situation will change at 14nm? Is 14nm node for mainstream desktop going to be more focused on lowering leakage (ie, mobile) so the Core series parts get more clockspeed out of lower voltage?

 

[VengenceIsMine]

I say that because, it isn't just the switch from A15 to A57 that Cherry Trail will be competing with but the top-tier QCOMM + Samsung offerings that are at 20nm too. I agree that A57 could push that power-envelope, but that helps when the node is decreased.

New mirco-arch plus 14nm will likely give CPU and GPU dominance. I'm not sure that just 14nm will be enough. It should be pretty interesting to see how it plays out.

It's going to be a combination of 4 things on the CPU side on the mid-higher end Atom SKU's: slight clock bump allowed by the process node, jump in memory bandwidth & latency, small micro arch improvements & the last and largest change will be significantly increased power efficiency. Individually, none of them are game changers but they keep Intel in the game of the CPU side vs A57, Qualcomm & other custom cores. The big jump is on the GPU side where they go from middling to very good by going from 4 ivy bridge gen cores to 16 Haswell gen cores.

The big problems are that in and off itself, Cherry doesn't drop the platform cost much which is still higher than that of the ARM commodity whitebox guys and that Intel still lags on how much they integrate on their SOC vs ARM so their PCB's are still physically larger and are more expensive.

 

[witeken]

Intel told the BOM of Bay Trail would drop significantly in 2014, so I don't think Cherry Trail will have any BOM issues, and I haven't heard Intel talking about contra-revenue for other platforms than Bay Trail. We'll see.

 

[Khato]

The big jump is on the GPU side where they go from middling to very good by going from 4 ivy bridge gen cores to 16 Haswell gen cores.

According to all the information I've seen Cherry Trail is actually going to be based off Broadwell graphics, not Haswell. Which is another indication of how seriously Intel is taking this market now eh? Sadly we still have to wait quite some time before we get to see how much of an improvement Gen8 graphics are compared to Haswell's Gen7.5.

 

[cbn]

The big problems are that in and off itself, Cherry doesn't drop the platform cost much which is still higher than that of the ARM commodity whitebox guys and that Intel still lags on how much they integrate on their SOC vs ARM so their PCB's are still physically larger and are more expensive.

I've been wondering about the extra die size of Cherry Trail's IGPU vs....what Intel could have done with increasing integration?

16 EUs.....I am skeptical the GPU really needs to be that big, but maybe I am wrong?

 

[VengenceIsMine]

I have to agree. Bay Trail is a great product for tablets, which is clearly what it's aimed at. From what I can tell, it's reinvigorated the Win 8.1 tablet platform to some degree. If bay trail had been around for the Win 8 launch, things might be going differently at MS.

If they can pull out another 20% - 30% in CPU performance and maybe 50% in iGPU from Cherry Trail, and keep the same or lower power profile, then I think it will make Win 8 tablets a reasonable desktop replacement for the average consumer.

I don't think that will hurt overall sales as much as some believe. Desktops tend to get upgraded and last a long time. Laptops, less so. Tablets, even less. So they sell a cheaper chip, but people turn over their machines more frequently.

This is the way the industry is going and they really have no choice but to make their best shot; anything less and they risk being relegated to server farms / data centers, and eventually not even that.

It's not going to be 20-30% on the CPU side unless it's an incredibly memory centric measurement. It will be a substantial gain on the power side however. The GPU side is going to be easily more than 50% in most scenarios, Bay Trail only does ok with 1080p while Cherry Trail should be seriously faster even at higher resolutions 2560x1600 and 3200x1800. Besides GPU the largest gains will be in power unless that all gets wasted on high res screens that are power hogs. (Galaxy Note 10.1 2014 Edition, I'm looking at you)

Bay Trail would have helped 8.0 certainly but there were just so many gaping holes in the OS and ecosystem, lack of decent touch screens, high ram and storage requirements for the OS and just meh designs from the OEM's I am not sure it would have helped the 8.0 launch that much.

 

[VengenceIsMine]

I've been wondering about the extra die size of Cherry Trail's IGPU vs....what Intel could have done with increasing integration?

16 EUs.....I am skeptical the GPU really needs to be that big, but maybe I am wrong?

It's a pixel race, it's simply easy to market that way and consumers respond to it. Personally, beyond 1080p in the under 10.6 size, I think it's diminishing returns somewhat but that isn't where the industry is headed.

As for integration, take a good look at the PCB of something like a Nexus 7 2013 vs a Bay Trail tablet like the Toshiba Encore or Dell Venue Pro 8. It's nothing like as bad as a Surface (which is good for a Haswell board but still monstrous vs a Qualcomm based or even more so with Apple) but it's still substantially larger than the ARM based ones and there are several sensors, DSP's and other things that Intel needs to either integrate or work with the supporting suppliers to improve packaging or something, they are still just too big, complicated, expensive and less power efficient.

 

[VengenceIsMine]

According to all the information I've seen Cherry Trail is actually going to be based off Broadwell graphics, not Haswell. Which is another indication of how seriously Intel is taking this market now eh? Sadly we still have to wait quite some time before we get to see how much of an improvement Gen8 graphics are compared to Haswell's Gen7.5.

You might be right, Intel I never pay enough attention to the the way Intel switch referring to generations of graphics cores with x generation of HD Graphics cores... Even if it is Haswell instead for Broadwell, it's going to go from OK at 1080p to quite zippy even at 32x18.

 

[VengenceIsMine]

Intel told the BOM of Bay Trail would drop significantly in 2014, so I don't think Cherry Trail will have any BOM issues, and I haven't heard Intel talking about contra-revenue for other platforms than Bay Trail. We'll see.

Hm. What I saw/heard was that the BOM was the smallest of gains. Smaller than CPU, smaller than power efficiency, smaller than GPU performance etc. While contra revenue helps, it's a band aid vs having a lower BOM in the first place. You have to throw a lot of $ on the table to make an OEM/ODM take notice because they know that isn't a long term fix.

 

[IntelUser2000]

It's not going to be 20-30% on the CPU side unless it's an incredibly memory centric measurement. It will be a substantial gain on the power side however.

The leaked presentations put Cherry Trail at 2.7GHz with Burst which is a 12.5% increase in frequency. 7-8% IPC gains(compensating for the fact that frequency/perf increase is not linear) should get 20% on the single thread side.

On multi-thread side 30% would be possible if the sustained frequency running at 4 cores is higher. Multi-thread is only 3x the speed on Bay Trail vs previous Atoms when IPC gains + frequency + double the cores should result in near 4x.

Of course it would be a sore disappointment if they don't achieve 20-30%. Ideally they should be 50% faster but whatever. They think ARM vendors would be incompetent as AMD and would eventually run into "limits"(like power and process...). Not that they won't, but the competitors seem to be doing pretty damn well for being nearly 2 process generations behind don't they?(which also puts into question whether Intel is being true about their "process advantage").

Hm. What I saw/heard was that the BOM was the smallest of gains.

That's supposedly the point contra-revenue addresses, and something they want to gradually eliminate it, the point of nil by 14nm(Cherry Trail+) timeframe.

If you for example want an platform that needs to be fit for both PCs(like Netbooks/Notebooks), and Tablets you'd add necessary adjustments for both SATA and eMMC interface. Their argument was that it was a cost adder, and making it Tablet-only(eMMC) would reduce the cost. And there's other things like most laptops having somewhat user serviceable parts, while on Tablets nearly everything has to be done on a manufacturer level.

Flexibility = cost

The GPU side is going to be easily more than 50% in most scenarios,

I think anything less than HD 4400(which is on 15W U Haswell chips) will be a disappointment. Even with that, Nvidia with TK1 won't be terribly far behind. With rumors that Broadwell class GPUs and 14nm should bring 4x perf/watt improvements.

 

[North01]

Intel Braswell

Source

 

[Sweepr]

Interesting name choice. 🙂

Intel outs 64-bit KitKat Android, 'Braswell' chip for Chromebooks

Kirk Skaugen, senior vice president at Intel, disclosed that a chip code-named Braswell will be the follow-on to a version of the Bay Trail processor now found in low-cost devices like Chromebooks and sub-$500 Windows PCs.

Braswell will be a system-on-a-chip, aka SoC, built on the company's next-generation 14-nanometer manufacturing process. SoCs typically squeeze most of a device's core electronics onto one piece of silicon.

www.cnet.com/news/intel-outs-baswell-next-chip-for-chromebooks/

 

[erunion]

so, is Braswell a unique die for celeron/pentium branded atoms?

 

[bullzz]

is this based on broadwell? what is the point of cherrytrail then

 

[CryNet]

A question. Cherry Trail tablets is expect to this year?

Cherry is braswell 64?