Tick Tock Tick Tock Tick Tock Tock
Intel’s Tick Tock clock might have just stuttered.
A few websites have been re-reporting on a leak from BenchLife.info about Kaby Lake, which is supposedly a second 14nm redesign (“Tock”) to be injected between Skylake and Cannonlake.
UPDATE (July 2nd, 3:20pm ET): It has been pointed out that many hoaxes have come out of the same source, and that I should be more clear in my disclaimer. This is an unconfirmed, relatively easy to fake leak that does not have a second, independent source. I reported on it because (apart from being interesting enough) some details were listed on the images, but not highlighted in the leak, such as "GT0" and a lack of Iris Pro on -K. That suggests that the leaker got the images from somewhere, but didn't notice those details, which implies that the original source was hoaxed by an anonymous source, who only seeded the hoax to a single media outlet, or that it was an actual leak.
Either way, enjoy my analysis but realize that this is a single, unconfirmed source who allegedly published hoaxes in the past.
Image Credit: BenchLife.info
If true, this would be a major shift in both Intel's current roadmap as well as how they justify their research strategies. It also includes a rough stack of product categories, from 4.5W up to 91W TDPs, including their planned integrated graphics configurations. This leads to a pair of interesting stories:
How Kaby Lake could affect Intel's processors going forward. Since 2006, Intel has only budgeted a single CPU architecture redesign for any given fabrication process node. Taking two attempts on the 14nm process buys time for 10nm to become viable, but it could also give them more time to build up a better library of circuit elements, allowing them to assemble better processors in the future.
What type of user will be given Iris Pro? Also, will graphics-free options be available in the sub-Enthusiast class? When buying a processor from Intel, the high-end mainstream processors tend to have GT2-class graphics, such as the Intel HD 4600. Enthusiast architectures, such as Haswell-E, cannot be used without discrete graphics — the extra space is used for more cores, I/O lanes, or other features. As we will discuss later, Broadwell took a step into changing the availability of Iris Pro in the high-end mainstream, but it doesn't seem like Kaby Lake will make any more progress. Also, if I am interpreting the table correctly, Kaby Lake might bring iGPU-less CPUs to LGA 1151.
Keeping Your Core Regular
To the first point, Intel has been on a steady tick-tock cycle since the Pentium 4 architecture reached the 65nm process node, which was a “tick”. The “tock” came from the Conroe/Merom architecture that was branded “Core 2”. This new architecture was a severe departure from the high clock, relatively low IPC design that Netburst was built around, which instantaneously changed the processor landscape from a dominant AMD to an Intel runaway lead.
After 65nm and Core 2 started the cycle, every new architecture alternated between shrinking the existing architecture to smaller transistors (tick) and creating a new design on the same fabrication process (tock). Even though Intel has been steadily increasing their R&D budget over time, which is now in the range of $10 to $12 billion USD each year, creating smaller, more intricate designs with new process nodes has been getting harder. For comparison, AMD's total revenue (not just profits) for 2014 was $5.51 billion USD.
Read on to see more about what Kaby Lake could mean for Intel and us.
As I have said before, developing integrated circuits is like a game. You are given a budget in complexity and you need to create an architecture that completes the most tasks within those constraints. If you can add more features within the same die area, you can afford to add more logic. Smaller transistors also take less power, which increases that budget for chip designers.
But you cannot redesign everything, every time. That's too much work.
With Intel's release structure, they separate the benefits of better designs and smaller features into alternating products. This allows them to allocate resources on a regular pattern, which probably smooths out their research and development budget by keeping the problems that their engineers need to solve constant and steady. Broadwell, which brought the Haswell architecture to 14nm, was painful though. The release was delayed so severely that just a few products were launched to end-users. Apart from Core M, even device manufacturers will not have too many Broadwell options before the launch of Skylake in a couple of months, at least compared to what we are used to from Intel. The couple of Broadwell parts that are designed for the LGA, user-swappable socket that we received were lower performance than the top-end Haswell SKUs that preceeded it. It was like they just released them to satisfy their promise of Iris Pro for customers looking to upgrade Haswell-based systems, even though they wouldn't really be upgrades, outside of graphics, in many cases.
Skylake was supposed to take 14nm, brought in by Broadwell, which shrunk Haswell, and create a new architecture that Cannonlake will shrink to 10nm in 2016. Tick. Tock. It now looks like 10nm will be delayed, and we will not just get a Skylake refresh in 2016. Rather, the rumor claims that Intel is planning to inject a new architecture, Kaby Lake, in the gap. What the rumor does not say, however, is how far Kaby Lake will diverge from Skylake.
This is technically a Xeon Phi core, but you know.
What this does provide is an extra release cycle for Intel to rethink their existing circuit elements, which could provide a better vocabulary for them to use later. This time might not be wasted. In the areas that are usually copied blindly from previous architectures, the engineers can inspect whether they are showing their age. Given the chance, engineers will find subtle ways to make existing solutions better. An optimist would look at this situation and focus on the light being cast into the dark corners of previous designs, rather than just doing a two-refresh lifecycle for Skylake.
A pessimist would note that Intel always could have focused on architecture. Smaller fabs were better.
Considering both views, the result is that time, measured in product iterations, is not being wasted. The research done on Kaby Lake, whether vast or miniscule, could help Intel in the long run. Whatever they do better now, will be done better forever. What it might not do is bring Intel as far forward as they would be had they been given a healthy and on schedule transition to 10nm and beyond. Would have, could have, should have.
Good Graphics… I Said Good Graphics!
Like the current stack, Kaby Lake will be available in Y, U, H, and S suffixes.
Repeated for relevance. Image Credit: BenchLife.info
The -Y series will service products at 4.5W with dual-core parts. The -U series will handle the range from 15W to 28W with dual-core solutions, which is popular for mid-range laptops and X-in-one devices. The -H series has TDPs of 35W, 45W, and a “TBD” class, with each product having four cores. "H" products are probably intended for higher-end Ultrabooks. These three suffixes should all be BGA-based, which needs to be soldered to the motherboard and thus mass-produced devices.
The -S suffix will handle user-replaceable components with the LGA-1151 socket. They will be available in TDPs of 25, 35, 45, 65, 80, and 91 watts. Most Kaby Lake-S processors will be quad-core, but dual-core options will be available at the low end. Since they use the same socket as Skylake and what we believe Cannonlake will use, it might be possible to upgrade the CPU across all three generations… might. If so, this would be Intel's largest upgrade path in the last decade (counting by architectures).
In terms of graphics, almost every S-class processor will be paired with GT2, with two exceptions.
Image Credit: Anandtech
At 35W and 65W, GT4 graphics will be available on Kaby Lake-S (Update – July 2nd, 3:40pm: Missed the footnote stating that Iris Pro for -S SKUs will be based on Skylake) with 64MB of eDRAM, which is better known as Iris Pro. Like the recent Broadwell release on LGA-1150, this will allow enthusiasts to build desktops with Iris Pro graphics, which could be useful for home theater PCs that cannot afford a discrete graphics solution for one reason or another. Users who want the best CPU performance will not get Iris Pro, however. This is a bit disappointing, particularly with DirectX 12 and Explicit Unlinked Multiadapter soon taking advantage of the on-chip GPU. With Broadwell on LGA-1150, the lack of high-performance Iris Pro could be written off as Intel not really caring. The trend persisting through Kaby Lake, and we assume Skylake too, shows it as much more deliberate.
Hopefully Intel will at least give options when an on-processor GPU has practical advantages over discrete GPUs. Tasks, like AI path-finding and visibility, could be shredded by a GPU, but the latency of getting the task to an add-in graphics card and back again negates this, even if the software developer could afford the effort. On-processor GPUs are promising for these tasks, but less so if high-performance graphics is only available to a sliver of users that are sort of high end, but not high-end.
And then we get to the other exception…
On the table, there is a row of quad-core processors at 25W, 45W, and 80W that have “GT0” graphics. Since there is no such thing as “GT0” graphics that I know of, and I have not even seen this moniker used before, this might be Intel's way of denoting desktop processors without integrated graphics. As I alluded to in the introduction, I could be misinterpreting the table. If I am not, then we might see 80W TDP processors available from Intel without integrated graphics at all.
80W is below the “high-end” CPUs from Intel at the ~90 watt point, such as Devil's Canyon (88W) and whatever the 91W Kaby Lake processor would be. Still, without needing to account for a GPU, this might have the same performance (or better) as the 91 watt version… or not. It appears as though Intel will not release these 80W CPUs with GT0 unlocked though, as the table only lists the 91W part as being part of the “K” branding. If I had to guess, mainstream enthusiasts will still want the model with the iGPU, even if they don't plan on using it (and DirectX 12 titles don't use it for them). I cannot think of a use case that Intel would be targeting if all this is true.
So there we have it. This was an interesting little leak, which could be totally crap. It sounds reasonable though, because the details mesh together in ways that I doubt a hoax would bother lining up. In many ways, this could be an opportunity for Intel to better position themselves for the future. It takes the place of their regular schedule however, which would imply that whatever they do would be inferior, all things considered, to their original plan succeeding perfectly, since it is by nature a compromise from that.
It may all depend on whether you count 10nm woes as an external factor.
“At 35W and 65W, GT4 graphics
“At 35W and 65W, GT4 graphics will be available on Kaby Lake-S with 64MB of eDRAM”
GT4 Kaby Lake-S don’t exist in the table is a Skylyke part.
See note 4 in the picture.
Whoops! Must have missed
Whoops! Must have missed that. Thanks!
I wonder if this delay is
I wonder if this delay is more Intel responding to future HBM based APUs, and maybe Skylake is going to get some HBM attention. Sure 10nm is going to cost more and have its share of yield issues, and maybe the economics, along with the billions spent on contra revenue have Intel trying to get more than its money’s worth out of the 14nm process node. Process shrinks will hit the economic wall long before the laws of physics run out for shrinking transistor sizes. AMD is at least using its high density libraries for its low power APU parts, and getting more from the 28nm node for its mobile SKUs. That’s not to say that the ZEN based parts will use any high density libraries for the desktop and some server variants, but having a different set of design libraries for low power parts could enable a lot more features(GPU, etc.) on the system on a chip mobile SKUs. 14nm and high density libraries for APUs/SOCs will definitely result in more die area for more GPU resources once the Zen microarchitecture arrives, and AMD may just have a high performance design library for its performance Zen server and desktop SKUs and a denser design tape-out for its 14nm Zen/K12 mobile parts.
Intel basically designs for its server parts, and makes the adjustments to derive its desktop/laptop SKUs, and its ATOM line of parts is at a disadvantage against the power efficient ARM based designs. With the process shrinks giving the tic tock cadence more than a few interruptions lately, Intel may have to give its mobile x86 SKUs more of a Architectural reworking than a simple scale down, and maybe some high density libraries would help, but Intel does not appear to be about the graphics improvements that are necessary to compete with AMD’s graphics, or Nvidia’s, and others’ ARM based mobile parts and graphics. HBM is the big news now, and even bigger if APUs start getting HBM, and even better graphics, to go along with a new ZEN microarchitecture. The economical node shrinks are running out, Interposers with HBM and beefy GPUs are on the horizon for the desktop market, derived from the Zen server SKUs, while the ARM armada continues to control the mobile market. HBM and stacked memory will be quickly adopted for the mobile market, where its lower clock speeds and wider data paths will save more on battery life, and the extra saved mainboard space. Using High density libraries for low power parts will have its advantages for extra features squeezed into the available process node. Intel needs to get more creative than finding more markets to buy its way into, it needs to get more creative layout engineering, and microarchitectural reengineering for its mobile SKUs, and much better graphics for the buck.
AMD’s Zen microarchitecture does not even need to best Intel’s products in IPCs to have AMD’s graphics make the case for its APUs, and if the AMD desktop variants are derived from the server variants with HBM and beefy Greenland graphics the desktop market will have its technology race back. If AMD’s IPC performance is just in the ball park, then the graphics will carry more sales for AMD and Intel needs to pay more attention to its SOC graphics and at a much more affordable cost.
I hope you are right. But it
I hope you are right. But it seems like aa lotof wishful thinking and cheering for AMD to me.
I disagree completely. Intel
I disagree completely. Intel spent BILLIONS of R&D optimizing x86 for the mobile process specifically so it could compete with ARM. They do NOT simply design for server first then derive desktop from that.
This is completely documented if you look a little closer since it was big news recently about Intel merging its financial reports so the R&D loss for mobile research didn’t look as significant to stock holders.
When you optimize for POWER EFFICIENCY it benefits server, desktop, and mobile. When you optimize for desktop vs server those are very different tasks with completely different engineers optimizing those architectures.
Why would intel waste time
Why would intel waste time with HBM when they developed EMIB for HMC and already have it ready for Knights Landing? Altera(which intel just bought) uses EMIB on their 10 TFLOPS FPGAs too.
I think two 14nm CPUs on the same socket is unrelated to the HBM/HMC and is simple due to Cannonlake probably having something that makes using LGA1151 disadvantageous, like allowing for new interconnects between the CPU and PCH, or features integrated to the CPU that are on motherboards now, like silicon photonics lasers or some type of Omnipath or DMI.
Also Kaby Lake would be the third 14nm CPU, and rather than fail to release and have the issues Broadwell had, making it a useless architecture, maybe theyre just delaying it now rather than later.
Oh yeah i meant to add that
Oh yeah i meant to add that if intel was facing real competetion from AMD they would probably accelerate the release of Cannonlake, not delay it and add Kaby Lake.
If anything its the lack of competetion in the desktop and laptop CPU market that allows intel to keep selling chips like Haswell refresh.
Also Haswell refresh was really the first departure from tick tock.
Didn’t they already do that
Didn’t they already do that with devil’s canyon & broadwell difference from devil’s canyon was mneah, get use 2 more of the same. Good luck catching up AMD! Looks like Intel has hit a bottleneck; oops & all that money spent 4 what. Eye, eye, eye; rich bastards. What goes around comes around. It would probably have a few more tricks in the hat though. It will be tick, f#@&!