What are the possible next steps in the Intel-Rockchip Alliance?

[cbn]

Back in May of this Year, Intel announced an alliance/engagement with Rockchip where two companies would co-launch a quad core SoFIA 3G SOC.

According to the Anandtech article here, Rockchip would more or less license the cpu core IP and perform the integration (with help from Intel).

The idea is that working along side Rockchip would substantially lower costs in this low ASP market.

So I got to wondering what other IP beyond atom Intel might offer Rockchip? I'd imagine Rockchip would want to go beyond atom if at all possible. Maybe they could do something with some of the older big core IP and fab their designs in Intel foundries?

I've even wondering if Rockchip could do something with the newer big core designs as well. This, of course, provided it doesn't conflict with Intel's overall strategy.

 

[NTMBK]

Combine Atom CPUs with Nvidia GPU IP, make a console APU.

 

[cbn]

Combine Atom CPUs with Nvidia GPU IP, make a console APU.

A couple of questions:

1.) How many atom cores?

2.) If going with x86, How about using Westmere in some format? (With or without features disabled?, Maybe 2C/4T?)

 

[Lepton87]

Combine Atom CPUs with Nvidia GPU IP, make a console APU.

Are NV's mobile GPUs even license-able like Mali,PowerVR etc.? I thought it was exclusive to Tegra SOCs.

 

[NTMBK]

Are NV's mobile GPUs even license-able like Mali,PowerVR etc.? I thought it was exclusive to Tegra SOCs.

They were certainly trying to license it out, but I don't know if anyone has taken them up on it yet: http://www.anandtech.com/show/7083/nvidia-to-license-kepler-and-future-gpu-ip-to-3rd-parties

 

[TuxDave]

A couple of questions:

1.) How many atom cores?

2.) If going with x86, How about using Westmere in some format? (With or without features disabled?, Maybe 2C/4T?)

I don't think that'll fly. The Westmere CPU wasn't designed to be IP-friendly, and so unless you're willing to take an existing Westmere complete CPU "as-is", an Atom-based CPU (more SoC friendly) would be easier to license out.

 

[cbn]

Are NV's mobile GPUs even license-able like Mali,PowerVR etc.? I thought it was exclusive to Tegra SOCs.

http://www.anandtech.com/show/7083/nvidia-to-license-kepler-and-future-gpu-ip-to-3rd-parties

Note that what NVIDIA is announcing today is contrary to AMD’s semi-custom approach to SoC production. AMD is offering to build (semi) custom tailored silicon to customer needs, while NVIDIA is taking a more ARM-like approach and offering its GPU IP to 3rd parties for integration on their own. In other words, NVIDIA is looking to compete with ARM and Imagination Technologies rather than AMD or Qualcomm.

I asked NVIDIA about future GPU architectures beyond Kepler, and the answer was pretty awesome: future GPU architectures will be available to licensees at the time of tape out by NVIDIA. Licensees can choose whether or not to adopt an architecture right away or wait for any potential revisions, similar to what ARM does with its cores (e.g. Tegra 4i uses a later revision of the Cortex A9 core). This move has huge implications. Theoretically a licensee could bring an NVIDIA GPU to market before NVIDIA itself, although that does seem pretty unlikely. What we could see however is a licensee introduce a GPU configuration that NVIDIA had no intentions of bringing to market.

 

[cbn]

I don't think that'll fly. The Westmere CPU wasn't designed to be IP-friendly, and so unless you're willing to take an existing Westmere complete CPU "as-is", an Atom-based CPU (more SoC friendly) would be easier to license out.

In my earlier Intel-Rockchip thread NTMBK speculated Intel might have licensed the AMBA bus from ARM for the Goldmont atom. From what I gather the AMBA bus would make integrating 3rd party IP much easier for those Goldmont atom cores (specifically the Broxton SOC).

But lets say (for the sake of argument and to keep things as "simple" as possible) Rockchip was planning to build a 2C/4T big core PC based APU (using Nvidia iGPU). In this scenario. since integration (and third party IP) is not as high as smartphone SOC, would the AMBA bus still be needed?

 

[rgallant]

http://www.anandtech.com/show/7083/nvidia-to-license-kepler-and-future-gpu-ip-to-3rd-parties

only dreaming but intel could shake up the gpu's market by buying the eol 780ti spec's add 1000 cores
build it on 14nm @ 1400 mhz use modded nv drivers and only put out only 1 sku and wait for 3 yrs for nv and amd to catch up. lololol

 

[cbn]

Assuming Rock chip had no plans for integrating IP and just wanted to reuse old Intel CPU designs for desktop (launching in 2015 for the sake of argument), how far back do you think they could go without losing practicality?

Sandy Bridge?

Westmere?

Nehalem?

Penryn?

I'd assume some of these could be freshened up with a die shrink or two depending on the generation of chip. (Eg, Penryn on 32nm or 22nm, Nehalem quad core on 32nm or 22nm, Westmere on 22nm, etc)

 

[cbn]

Regarding that 2C/4T Westmere (81mm2 on 32nm) I mentioned earlier in the thread.....

How about instead of using Intel GMA graphics (45nm) on package, Intel substitutes in a modern 14nm Gen8 graphics chip in its place. (So Intel/Rockchip would then have 32nm 2C/4T and 14nm Gen8 on the same package)

Here are some die shots of that old Clarkdale (Westmere) from the Anandtech review article here:

(Left side, Intel GMA 45nm graphics chip with memory controller. Right side, Westmere 2C/4T processor on 32nm)

And here is what Anand had to say about that original memory controller found on the 45nm GMA graphics die:

http://www.anandtech.com/show/2901/2

Memory Performance - Not Very Nehalem

Let’s start at the obvious place, memory performance. Nehalem moved the memory controller on-die, but Clarkdale pushes it off again and over to an on-package 45nm graphics core.

To make matters worse, the on-package chipset is a derivative of the P45 lineage. It’s optimized for FSB architectures, not the QPI that connects the chipset to Clarkdale. Let’s look at the numbers first:

Processor L1 Latency L2 Latency L3 Latency
Intel Core i7-975 4 clocks 10 clocks 34 clocks
Intel Core i5-750 4 clocks 10 clocks 34 clocks
Intel Core i5-661 4 clocks 10 clocks 39 clocks
AMD Phenom II X4 965 3 clocks 15 clocks 57 clocks
Intel Core 2 Duo E8600 3 clocks 15 clocks



L1 and L2 cache latency is unchanged. Nehalem uses a 4-cycle L1 and a 10-cycle L2, and that’s exactly what we get with Clarkdale. L3 cache is a bit slower than the Core i7 975, which makes sense because the Core i5 661 has a lower un-core clock (2.40GHz vs. 2.66GHz for the high end Core i7s) Intel says that all Clarkdale Core i5s use the same 2.40GHz uncore clock, while the i3s run it at 2.13GHz and the Clarkdale Pentiums run it at 2.0GHz.

Processor Memory Latency Read Bandwidth Write Bandwidth Copy Bandwidth
Intel Core i7-975 45.5 ns 14379 MB/s 15424 MB/s 16291 MB/s
Intel Core i5-750 51.5 ns 15559 MB/s 12432 MB/s 15200 MB/s
Intel Core i5-661 76.4 ns 9796 MB/s 7599 MB/s 9354 MB/s
AMD Phenom II X4 965 52.3 ns 8425 MB/s 6811 MB/s 10145 MB/s
Intel Core 2 Duo E8600 68.6 ns 7975 MB/s 7062 MB/s 7291 MB/s


Here’s where things get disgusting. Memory latency is about 76% higher than on Lynnfield. That’s just abysmal. It’s also reflected in the memory bandwidth scores. While Lynnfield can manage over 15GB/s from its dual-channel memory controller, Clarkdale can’t break 10. Granted this is higher than the Core 2 platforms, but it’s not great.

What we’re looking at is a Nehalem-like CPU architecture coupled with a 45nm P45 chipset on-package. And it doesn’t look very good. If anything was going to hurt Clarkdale’s performance, it’d be memory latency

Opinions?

Would a redesign of Clarkdale's on package graphics/memory controller to GEN 8 and improved memory performance be possible?

 

[cbn]

If 14nm is constrained for a 2C/4T Westmere's Gen8 graphics chip then maybe Rockchip could find another partner? Perhaps Nvidia or Imagination Tech with help from Intel and/or Nvidia on the integration of the graphics chip?

Maybe there could even be two versions of this Westmere 2C/4T + GPU on package chip? A Rock chip version and an Intel version, each differentiated from each in certain way either thru graphics IP or CPU binning or whatever.

Thoughts on this strategy?

 

[cbn]

Any other ideas about Rockchip and Intel with respect to atom cores?

 

[Arachnotronic]

Smartphone chips. Brian K. indicated this at a recent investor conference.

 

[cbn]

In addition to the PC "game console/gaming PC desktop chip" ideas I mentioned in this thread using the Westmere 2C/4T chip, I have to wonder what other ways Intel could collaborate with Rockchip on Big cores?

Take for example what Rockchip is trying to do with the tablet SOC RK3288 as an Android gaming console chip:



Surely there is something much better than what we are seeing above that Intel could offer them.

Maybe if Rockchip proves itself with integrating SoFIA, Intel could grant them access to big core IP with special limitations. (eg, Rockchip must build the big core IP in Intel fabs. Perhaps the big core IP could be limited to certain configurations?)

Personally, I think a 1C/2T big core dedicated die with fast clocks and a decent sized Gen 8 or Gen 9 iGPU would be very interesting provided the volume and distribution were great enough. A extremely cheap gamer and web browser chip that would allow Rockchip users a capability they do not have with four little cores at low clockspeed. (eg, Star Craft II). Personally, I would much rather have something like this than Braswell on the desktop. Furthermore, I have to believe even if the 1C/2T Big core were integrated with 16 EUs and a light amount of I/O, it would be as small as Braswell. If the 1C/2T big core with 16 EUs had a separate PCH, it would be even smaller than Braswell.

 

[Arachnotronic]

cbn

You're not really a fan of integration, are you?

 

[cbn]

cbn

You're not really a fan of integration, are you?

I do think integration is a good thing.

Example: The hypothetical Rockchip-Intel 1C/2T big core 14nm cpu and 16EU Gen 8 or Gen 9 iGPU would be integrated together on one die. They would not exist as separate chips.

My main complaint is using four small cpu cores at low frequency for desktop when for the same silicon die cost a high frequency big core with hyperthreading could be used. The ARM companies have no choice but to use four small cores at low frequency for desktop...that is the only option they have, but with Intel thankfully there are other options because of the existence of Big core. Therefore, I think Intel should look for as many ways to leverage this advantage as much as possible.

 

[cbn]

Smartphone chips. Brian K. indicated this at a recent investor conference.

Some interesting info I found from the Citi 2014 Global Technology Conference:

http://seekingalpha.com/article/246...y-conference-transcript?page=2&p=qanda&l=last

Ehud Gelblum - Citigroup

Two directions I want to go with that. One is the comment you made about phones, are phones as critical part as tablets are to your strategy or are they just an upside, an opportunity because you’ve already created the comps opportunity for tablets.

Brian Krzanich - CEO

I think some hybrid of that. So definitely tablets are closer to our core. They’re much more of a compute oriented product. I think we have -- here's what Intel does best. When we can bring our technology to a product and really use the compute cycles, use the capability inside them -- we’re not going to build commodity tablets for example, but we believe over the next year or so, now that we’re in the market, we’ve established the eco-system, we can bring real innovation to that space and bring some really unique features.

You’ll start to see some of those come out as this holiday season kind of starts to play out and products start to come out. Those are very core. We can take technologies that we’re developing for the PC and move those down into tablets very easily. So it really is able to spin off of that.

Phones are a little bit different and we’re looking at those more opportunistically. We’re trying to figure out where is the best place to play it. Like I said in most of the developed countries there is two big players but you move into China, it is much more of an open market spaces like that. You saw our deal with Rockchip. I think you’ll see them move into phones over the next couple of years. Probably with our architecture you’ll see some other opportunities like that I think over the next couple of years.

 

[Techhog]

A couple of questions:

1.) How many atom cores?

2.) If going with x86, How about using Westmere in some format? (With or without features disabled?, Maybe 2C/4T?)

I really don't understand your obsession with reviving Westmere. It's just not a logical idea.

 

[cbn]

I really don't understand your obsession with reviving Westmere. It's just not a logical idea.

1.) Look at the date the post was written.

2.) How is reviving older processor designs like 2C/4T Westmere not logical when the goal is extremely low cost? (Don't think LGA processor with all the cost adders associated with that. Think BGA at desktop speeds, with soldered on RAM, eMMC 5.0, DC power-in with AC adapter brick, emerging countries like China, etc.)

 

[TuxDave]

(eg, Rockchip must build the big core IP in Intel fabs.

That's pretty much a given. I highly doubt Rockchip would have the talent or capability to reimplement the big core design on another process.

 

[cbn]

So to review this thread so far, there have been three ideas proposed:

1. Combine atom cpu cores with Nvidia graphics IP to make a console APU.

2. Combine Westmere 2C/4T with a new on package graphics/memory controller chip (using various graphics IP), re-using the 65 nm H55 or H57 PCH if at all possible.

3. Create a new big core 1C/2T with 16 EUs on 14nm. (see posts #15 and #17) This either as an SOC or as a CPU/GPU + PCH. (A very simple way of thinking about this would be basically take Braswell and replace the four atom cores with a single Big core with HT for a SOC variant).

 

[cbn]

That's pretty much a given. I highly doubt Rockchip would have the talent or capability to reimplement the big core design on another process.

Regarding a Rockchip and Intel collaboration, I have to wonder if at some point Intel's fabs would be capable of making a 1C/2T big core SOC to be used as phone chip? (This to replace an quad core atom phone chip)

We've seen good progress with products like Core M, so maybe this closer away than we think?

Also, another reason I bring this up is because I am firm believer that a strong advantage Intel has over ARM is the "big core".

P.S. With Intel's big core being a design optimized not only for high IPC, but high frequency as well this ideally requires xtors with high drive current to get the most oit of the design. Crossing my fingers, Intel's 10nm (with a change to a different xtor from FinFET) is the answer to the problem I am proposing.

 

[dealcorn]

The idea that Rockchip recycles obsolete big core technology while the market embraces little cores is a small idea. Rockchip may better understand consumer preferences for mobile devices than Intel. Bayrail-T is Exhibit A for this position. Why not exploit what Rockchip is good at as part of an expanded foundry model?

China is an important market for Intel where the government understands the difference between good and bad. Bad is when Intel's mobile ambitions kill a bunch of Chinese manufacturers. Good is when Intel supplies world class technology to Chinese designers/manufacturers so they may access world markets. Good is better than bad because the government is proactive and it appears they keep score.

Intel's portfolio of foundry design tools is not as refined as what TSMC offers. Rockchip is a design tools pipe cleaner. After Rockchip migrates it's foundry business to Intel, the obvious next step is to make the (now better) design tools and foundry service available to any remaining ARM design house/manufacturers. For Intel to achieve it's mobile ambitions, it needs good process technology, and good components such as a CPU, GPU, modem, camera, etc. Conceptually, there is no reason why Intel should not outsource mobile SoC design activities to many folks with proven competence. It may play better from an anti trust perspective also.

Do note than while Intel is doing SoPHIA dual core in house, Rockchip gets no competition from Intel for quad core SoPHIA. The opportunity is real and Intel has big eyes for this market. Sometimes, sharing is good.

 

[cbn]

With Mark Bohr already working on 7nm according to this thread:

http://forums.anandtech.com/showthread.php?t=2399522

I have to wonder how well dialed in 10nm is going to be and how much this will affect IC design and die configurations?

Are we on the pathway to replacing four atom cores with one big core with HT? If so, I would love to see some of this happen already on 14nm.

If only those four atom cores on Braswell could be replaced with a single fast clocked Big core with HT. How well matched a standard voltage (not ULV) 16EU iGPU would be for classic x86 desktop games with this configuration compared to four atom cores.

Then combine (in some cases) with other BOM reducing components (soldered on RAM, eMMC 5.0 or UFS 2.0 + SATA port, DC in plus AC adapter) and I think it could become an even better value for desktop.

Next after that (hopefully) for Intel-Rockchip would be the 1C/2T big core phone on advanced nodes.