Szdani88

4K times 2K resolution too

Braswell 14nm Atom is coming to notebooks / netbooks in Q3 2015, with a new iGPU in tow. 

Atom will finally get to 14nm, moving from the very successful but aging Bay-Trail 22nm quad-core platform, and Intel will start shipping products based on Braswell Pentium and Celeron branded 14nm SoCs later this year. There will be quad-core,  four-thread parts,  as well as dual-core chips with two threads. Both will use the Type 3 BGA package, measuring 25mm x 27mm. It has 1170 pins with a pitch of 0.593 mm.

The platform supports DDR3L 1600 MHz in SODIMM form factor and is capable of handling a maximum of 8GB, which we believe is fine for this market segment. The Braswell platform has 4x 1PCIe 2.0, 2 SATA 3.0. eMMC 4.51 and SD Card 3.01 support. It supports SDIO, UAART and 8 I2C connectors. It comes with total of five USB ports, where four are USB 3.0 and one is USB 2.0. The platform supports HD audio and the integrated graphics of choice is based on Intel's "Low power Gen 8 graphics" with up to 16 EUs. The graphics part supports DirectX 12, as long as you use Windows 10 and Open GL 4.2.

You can use up to three displays and they support maximum of 4K x 2K resolution. Internally the platform supports eDP 1.4 with up to 2560x1440 resolution, while for external displays you can use both HDMI or Display Port. The maximum resolution is once again 4K x 2K.

Braswell supports Chrome, Linux, Windows 7 64-bit, Windows 8.1 and Windows 10 only in 64 bit mode. Both Windows 10.1 and Windows 8.1 64-bit support comes without connected standby support, making us wonder if that was the cause for the Braswell delay.

This Out-of-Order platform in 14nm supports MIPI CSI 2.0, with integrated ISP and up to two HD cameras. You can take still pictures up to 5 megapixels and video up to 1080p at 30FPS. It is expected in Q3 2015, hopefully just in time for the back-to-school summer shopping spree, as Braswell targets the $199 to $399 notebook and Chromebook market. 

Laptops have 35 percent longer 1080 HD playback

Fudzilla has managed to get its paws on Intel data about the performance of the Skylake in detachable notebooks. 

Data regards the Skylake-Y the 5W TDP variant of the core which Chipzilla has compared the early silicon with the Broadwell-Y or 5W TDP version of Core M 5Y10 CPU.

According to Intel, the Skylake mobile platform has 60 percent lower SoC power and can sustain up to 35 percent longer 1080p HD playback than the Core M 5Y10.  If this is true, this is a great achievement. Intel used the same battery capacity and  the system configuration on the Skylake-Y and 5th Generation Core Y processor test beds. You can expect the same 60 percent lower SoC power and up to 35 percent longer HD playback from the Skylake-U versions 5th generation Broadwell-U processor.

This is means huge savings considering that Skylake is based on the same 14nm manufacturing node. Intel realises that battery life is one of the key factor for notebook users and since the line between notebooks and tablets is blurring, having a longer battery life in watching video really counts.

These are the preliminary results of the early silicon and the power consumption looks good. We don’t expect that Intel will start Skylake manufacturing until summer 2015 considering that Q4 2015 is the time when it should start shipping.

We are not sure how much can Skylake platform improve the overall battery life, but since it is brining "No Wires" strategy and wireless changing, this is the platform to be excited about, as it is bringing some significant changes.    

Cannot fit more on 2.5D interposer

In early January we heard a thing or two about an AMD card codenamed Fiji, the one that comes with High Bandwidth Memory (HBM). 

The card is expected in the second part of 2015, and it will beat Nvidia Pascal to market. Nvidia is also using High Bandwidth Memory (HBM) for its 2016, 16nm GPU, but the company is using another technique, quite different compared to AMD's Fiji. Nvidia also announced Pascal Unified memory with 3D memory, an NVLink GPU in 2016 pointing to a Stacked DRAM powered Volta, the successor of Maxwell. Pascal is next year, as Nvidia needs to have a 16nm TSMC node operational and ready before it goes after this next big thing.

AMD is using what is calls a 2.5D-IC silicon interposer, which means that there will be two separate chips on the same silicon interposer and package substrate. Fiji in 28nm will be one of these chips, and the second batch of chips will be the High Bandwidth Memory (HBM) memory designs. However, there is a catch with AMD's approach.

From what we've learned, Fiji is limited to 4GB memory. With the current memory technology the GPU would simply be too big to put on an interposer and package. The interposer should be viewed as a stack of conductors that lets the GPU and HBM memory communicate at much higher speeds than ever before. The interposer then gets into the package that goes on PCB. You could say the interposer is the middle-man that makes things faster.

Hynix has HBM memory with 1024-bit wide interface and 1Gb/s per pin data-rate. This results with 128GB/s bandwidth per memory chip. In case of four 8 Gb chips (1GB) with a 1000MHz core clock you can end up with total bandwidth of 512GB/s. There are indications that HBM memory on Fiji might work at 1.25GHz, which would result in 640 GB/s. The Geforce GTX 980 has 224GB/s bandwidth, while Geforce Titan Back has 336 GB/s.

SK Hynix has listed that 1GB, 128GB/s chips with 1.0Gbps speed packaged in 5mKGSD are available now, January of 2015. These HBM chips feature 4Hi stack VDD/VDDQ 1.2V. The old GDDR5 needs 1.5V to work, meaning that HBM is not only faster, it is more power efficient as well. Fiji could end up having more than twice the bandwidth or Nvidia's current and upcoming Titan cards.

Nvidia on the other hand is using what is called Vertical stacking 3D, or on-package stacked DRAM for its Pascal 2016 GPUs. Nvidia gives a straightforward explanation of the meaning of 3D memory on Pascal: "3D memory: Stacks DRAM chips into dense modules with wide interfaces, and brings them inside the same package as the GPU."

The clear benefits are a massive increase in bandwidth and quadruple energy efficiency. Nvidia is waiting for 16nm to make such a chip possible, and 3D memory is better approach than the interposer.

Back to the AMD Fiji approach with 2.5 stacking and interposer. Our sources claim that you cannot put more than four memory chips on the interposer, meaning that Fiji is limited to 4GB of memory. Eight chips next to a massive GPU would result in massively big chip (remember the GPU and memory chips are on the same board - interposer) and then put on the package. This is why we don’t think 8GB HBM Fiji will happen with this generation, but with time Hynix will come with more dense HBM memory chips making 8GB 2.5 D cards possible.

The only thing that comes to mind is that AMD could be using 4GB HBM memory on the interposer and then put some additional GDDR5 chips on the PCB. Think of it as L2 and L3 cache with some older CPUs. Level two cache would be really faster, while Level tree cache would be slower, but would be able to get some important things done.

We don’t think that this would happen as GDDR5 memory would be significantly slower than the HBM part, making the similar case than with Geforce GTX 970 memory. Fiji has a good chance to beat whatever Nvidia comes up with in 2015, but it will probably have a hard time fighting against Pascal. Pascal is coming in 2016 and one can only hope that it will happen in early 2016.

We believe Fiji and Pascal will enable single GPU 4K gaming at acceptable frame rates and prices.