Intel’s Atom Z3000 ‘Bay Trail’ SoC revealed

Intel is a company famously driven by paranoia, but the threat it now faces requires no special effort to see. Smartphones have grown and morphed into tablets, and together, these two new categories of devices have taken a big bite of out the consumer computing market. Laptop sales in particular have gone soft as tablets have grown in popularity.

The vast majority of these tablets are based on CPU technology from ARM. Intel’s position in the PC market couldn’t be much more commanding right now, but that alone is suddenly looking like a wobbly foundation for the future. The firm has attempted to address the mobile market with successive generations of its Atom processors, but those haven’t seen much success. Now, a rising urgency has prompted Intel to double down on its efforts, much as it did when the Pentium 4 was struggling to keep pace with AMD’s chips. For the first time, the years-old Atom microarchitecture has been completely rebuilt, and Intel has committed to bringing its methodical tick-tock development cadence to the Atom lineup going forward. Architectural refreshes will be interleaved with new process technology in a series of yearly updates.

Each of these updates will produce a host of Atom-based systems on a chip (SoCs) for different markets. The eight-core chip known as Avoton is headed for microservers and storage applications. Its near-identical twin, Rangeley, will be deployed in network and communications devices. An SoC called Merrifield will tackle the smartphone space.

Our concern today is the chip intended for tablets and convertible laptops, which is code-named Bay Trail. We’ve known for a while that Bay Trail was coming and that it would be based on the new Silvermont CPU microarchitecture. Beyond that, details have been scarce. Now, as the Intel Developer Forum opens in San Francisco, Bay Trail is making its first public appearance. We have extensive details on the SoC’s basic architecture—and a first look at its performance.

An overview of the SoC

A shot of the Bay Trail die. Source: Intel.

Bay Trail truly is a system on a chip, with a full complement of the components needed for laptops and tablets. It is in many ways similar to the Avoton chip we covered last week, but Bay Trail is scaled back for a smaller footprint, lower power use, and lower costs.

Like Avoton, Bay Trail is fabricated on a special variant of Intel’s 22-nm manufacturing process that’s been tuned for SoCs. This process has some of the finest geometries in the industry and is the first to implement a tri-gate or FinFET-style transistor structure. The foundries that produce ARM-based SoCs for competitors like Qualcomm and Nvidia are arguably at least a couple of years behind Intel on this front—and Intel is already talking about making the transition to 14 nm next year.

Below is a simplified block diagram of the Bay Trail SoC. We can use it as a starting point and add some detail to get a pretty good picture of what’s included in the silicon.

Logical block diagram of the Bay Trail SoC. Source: Intel.

The four CPU cores probably deserve top billing. The Silvermont architecture clumps two cores together into a single “module” that functions as a basic building block. Each module has 1MB of L2 cache shared between its two cores, and Bay Trail integrates two of these modules. We’ve already covered the Silvermont microarchitecture in some depth. The big change here is a focus on higher per-thread performance. Silvermont does away with symmetric multithreading (aka Hyper-Threading) and instead adds out-of-order execution in an effort to extract more parallelism out of sequential code. This change is eminently sensible given that the performance of one or two main threads often determines the user experience in tablets—and that even a relatively small chip like Bay Trail can execute four hardware threads without SMT.

At the center of the “north complex” above is an orange block known as the Silvermont system agent. The SA is the traffic cop that directs the flow of data between the major functional blocks of Bay Trail’s north complex. We know from our look at Avoton that the system agent employs a crossbar architecture, like a network switch, to ensure high-bandwidth communication from any one component to any other. The SA is linked to the Silvermont modules using a point-to-point interface known as IDI; it’s the same interface Intel has used in its big cores since Nehalem.

Hanging off of the system agent are Bay Trail’s dual memory controllers, each capable of talking to a single channel of DDR3 or DDR3L memory. The chip’s peak possible memory bandwidth is about 17 GB/s when both channels are equipped with 1066 MT/s memory. I assume this relatively low memory speed is due to power considerations. Some variants of Bay Trail will support only a single memory channel, and in that case, the DRAM can run at 1333 MT/s.

With both channels populated, the SoC’s maximum memory capacity is only 4GB, another indication of this platform’s power and size constraints.

One of the largest areas on the Bay Trail die is dedicated to graphics. Interestingly, Intel has jettisoned the Imagination Tech graphics it used in prior generations of mobile SoCs in favor of its own in-house Intel HD Graphics. This is essentially Intel’s latest graphics tech, and it’s the same generation used in larger Ivy Bridge CPUs. (Haswell’s graphics hardware is similar but tweaked for higher efficiency.) Intel says Bay Trail supports the latest graphics APIs, including DirectX 11 and OpenGL ES 3.0. We know from Ivy Bridge that this graphics core supports high-precision datatypes and is capable of general-purpose computing via OpenCL.

Of course, the unit count has been scaled back for this mission. Haswell desktop processors like the Core i7-4770K have 20 graphics executions units, or EUs. Bay Trail has only four EUs. For those keeping score at home, each of those EUs is SIMD32, equivalent to eight “shader processors” in the world of big GPUs, so Bay Trail has the equivalent of 32 SPs. The graphics clock speed varies but can reach as high as 667MHz in optimal conditions. According to Intel, that should add up to three times the performance of Clover Trail, its prior-generation tablet SoC.

Bay Trail’s video processing block can decode an alphabet soup of modern video formats, including H.264, VC1, MPEG2, and MPEG4. The hardware can also encode video in a couple of formats. H.264 encoding is accelerated fully, while MPEG2 encoding is handled via a hybrid approach, with key portions like motion estimation offloaded to hardware and other parts handled in software.

The display controller is fairly beefy, as it will need to be in order to drive the high-density displays now popular in tablets. The chip has two matching display outputs that conform to a range of standards, including HDMI 1.4, DisplayPort 1.2, DVI, and eDP 1.3. Each one can drive a display with a resolution of up to 2560×1600 at a refresh rate of 60Hz.

Oh, and here’s a feature with a classic Intel-style name: Display Power Saving Technology, or DPST for short. This feature combines dynamic backlight dimming with image modification (brightening up the images, basically) in order to reduce the power consumed and the heat produced by the display backlight—presumably without the end user noticing.

The final feature of the north complex is an image signal processor, which controls any onboard cameras and both receives and processes the pixels they capture. Bay Trail’s ISP can connect to dual cameras ranging from eight to 13 megapixels, and it has sufficient throughput to handle 1080p video at 60 FPS. A host of photographic features are included, including auto-exposure, auto-focus, auto-white balance, video stabilization, and a burst shooting mode.

That’s about it for the SoC’s north complex. The lower portion of the diagram above covers the south complex, which handles all sorts of lower-bandwidth I/O. In our briefing, Bay Trail’s chief architect, Rajesh Patel, wasn’t willing to divulge too many details about the nature of the switching fabric that connects the south complex to the system agent. We strongly suspect it’s the same Intel Optimized Switching Fabric, or IOSF, used in Avoton. IOSF is software-compatible with PCI Express, and it’s employed in Haswell and all of Intel’s current discrete platform controller hubs.

This interconnect is noteworthy because it’s an Intel-specific standard that allows the re-use of various logic blocks across a range of chip designs—and because it signals a move within Intel toward the SoC-style methods used to create many ARM-based chips. We learned in our Avoton briefing that “everything being developed now” at Intel makes use of IOSF.

Overview of the Bay Trail platform’s connectivity. Source: Intel.

The south complex offers links to a bunch of tablet- and PC-style interfaces. The biggest highlight may be the inclusion of USB 3.0 for truly speedy external connections. Bay Trail supports the USB “On-The-Go” spec, as well, so systems based on it should be able to act either as hosts (to devices like iPods) or as clients (to desktop PCs and the like.) Conspicuous by their absence are external PCI Express lanes and SATA ports. Primary storage will have to be handled by the eMMC interface, instead.

Cross-chip power-management

Of course, as a tablet chip, Bay Trail makes extensive use of the latest power-management techniques. Both the CPU and graphics cores can scale their clock frequencies and operating voltages in response to varying workloads. In addition, the chip can share its power budget between its major components, allowing additional performance headroom in certain cases.

For example, the image above cycles through several potential scenarios where individual CPU cores or other portions of the chip aren’t in use. These units are powered down, freeing up thermal headroom. In response, the still-busy CPU cores or integrated graphics can temporarily exceed their usual clock speeds.

Oddly, Intel calls this “Burst Technology,” rather than Turbo Boost, in the context of the Atom. The branding is kept separate because the Turbo tech in the Core processor lineup is still more advanced, with a broader dynamic operating range.

Many of the big components of the SoC are in their own power islands, with separate voltage supply rails, power gates to shut off inactive areas, or both. The illustration above offers a quick glimpse of Bay Trail’s dynamic power management in action via thermal imaging. In one case, only two cores appear to be active. In the next, all four CPU cores are active. In another, they all look to be powered down.

The final shot shows an almost completely dark chip.That’s presumably one of Bay Trail’s idle states. Like Haswell, this SoC pursues power savings, even between keystrokes, by dropping into a series of “active idle” states known as S0ix. These power states are managed in hardware and should be transparent to software and the OS. The deeper the sleep, the longer it takes for the chip to wake up. Intel’s Rajesh Patel again wouldn’t offer too many details about S0ix behavior in Bay Trail, but we’d expect much more aggressive pursuit of deeper idle states than in Avoton.

The Atom Z3000 series

Intel is spinning Bay Trail into five different models that make up the Atom Z3000 series, as shown below. These chips have been sampling for months now, as I understand it, and Intel anticipates that products based on them will be available for the holiday season.

The Atom Z3000 series will occupy tablets and convertible systems with screen sizes ranging from 7″ to 11″ and prices up to about $599. There may be some overlap, but generally speaking, systems costing more than that will likely be based on Haswell rather than Bay Trail. Intel estimates that it has 30 different design wins for the Atom 3000 series at present, and it expects that list to grow.

The Atom Z3000 series. Source: Intel.

The Z3000 series will support both the Windows 8.1 and Android operating systems at launch, but the Windows 8.1 situation is a bit tricky. Initially, only the 32-bit version of Win8.1 will be fully supported with connected standby capability, a crucial feature for tablets. Intel has pledged to deliver full 64-bit support with connected standby in the first quarter of next year. Apparently, the hold-up is software; the company claims the hardware is ready.

Obviously, the initial lack of 64-bit support isn’t a deal-breaker in tablets, since the scads of Android and iOS-based tablets in the market are 32-bit devices. Still, being stuck with 3.5GB of usable memory in Windows isn’t ideal, especially for convertible devices that dock with a keyboard and should be more than capable of serious productivity work. The delay in 64-bit support at least mitigates one of the advantages of going with a Wintel device over something else.

I’d like to show you the power specifications of these new Atoms, but Intel has decided not to be too specific on that front. Because of all of the dynamic power management at work, there’s some dispute over the proper power metric for this class of chip. (Intel has two specs, TDP and SDP, that it uses in different contexts.) We will talk a little bit about measured power use shortly, though.

In addition to the Atom Z3000 lineup, Intel will offer Bay Trail chips for traditional notebooks and desktops under the Pentium and Celeron brands. In fact, Intel’s NUC group has Bay Trail-based systems on its roadmap.

The test

We visited Intel’s offices in Santa Clara, California for a first crack at testing Bay Trail’s performance. Although we’re happy to have the chance to test and relay some initial performance information to you, please understand that the information you’re about to see isn’t quite like what you might normally find in one of our reviews.

Although Intel was willing to let us run any test we wished, we were faced with a number of practical constraints, including very limited testing time and a spotty Internet connection. Some of the comparative results you’ll see below come from 64-bit executables in Windows 8, while the Bay Trail tablet we tested was running the 32-bit version of Windows 8.1 and could therefore run only 32-bit executables. We were faced with different OS versions on Android, as well. Some of our cross-platform, browser-based benchmarks are as influenced by the performance of the web browser software as they are by CPU speed, and we weren’t able to use the same browser and revision everywhere, although we did try to stick to Chrome 27.0.1453.94 in Windows.

That said, we think the performance results below ought to be sufficient to give you a general sense of Bay Trail’s competitive standing. Just be aware that the data aren’t as neat and clean as what we’d put into graph form most of the time. We’ll want to follow up with testing of production Bay Trail tablets in our our labs at a later date—and probably with better, more user-experience-focused benchmarks, as well.

Here’s a look at the Bay Trail tablet we used for testing. This is an Intel reference design for 10″ tablets, not something intended for production in exactly this form. The system has an Atom Z3770 processor inside, with clock speeds up to 2.4GHz with Burst. It also has 2GB of DDR3L-1066 memory. The display is ridiculously sumptuous, packing a 2560×1440 array of pixels into a 10″ diagonal rectangle. I stared.

The tablet is fanless and similar in size, thickness, and weight to the current iPad. We used two of these test systems, one running Windows 8.1 32-bit and another running Android 4.2.2. The Win8.1 tablet was in pretty good shape for a pre-release device. We noticed a few quirks here and there, but it generally animated that near-four-megapixel display fluidly. The Android-based systems were more obviously in an early, pre-production state. They had some issues with touch responsiveness and the like, and Intel was very candid about the need for additional tuning before Bay Trail-based Android systems are ready for the market.

Quite honestly, I’d like to spend more time with one of these systems before offering a strong assessment of the general user experience. What I saw of them was good, but obviously marred by little hiccups here and there that wouldn’t likely be an issue in a final, shipping product. I can say that a number of casual games and even some lightweight 3D games in Windows, like Torchlight 2, appear to run well on the Bay Trail reference tablet. That fact is vaguely amazing given where even full-sized Intel laptops were several years ago.

The results

First up is a pair of brower-based benchmarks that, with some caveats about software influencing performance, we can run across a host of different OSes.

As you can see, Bay Trail gets off to a very solid start, nearly cutting execution times in half compared to the prior-gen Clover Trail Atom Z2760. It outperforms all of the ARM-compatible SoCs we tested, including those based on alternative CPU architectures like Krait and Swift.

Also intriguing is the match-up between Bay Trail and the AMD “Kabini” A4-5000 SoC, which has four Jaguar cores clocked at 1.5GHz. The A4-5000 SoC alone has a TDP rating of 15W; we’d expect max power draw on an entire Bay Trail platform to be lower than that. Yet the Atom Z3770 prevails, with a slight performance edge over the Kabini-based laptop in both tests.

In these Windows-only applications, Bay Trail achieves something close to three times the CPU performance of Clover Trail overall. The new Atom SoC even challenges the low-end Ivy Bridge dual-core, the Core i3-3217U, in several cases. (It’s much faster in AES encryption because Intel has disabled the Core i3’s AES-NI support for product segmentation reasons.)

The new Atom’s performance continues to match or exceed the AMD A4-5000’s, as well.

In Cinebench, our most floating-point intensive test, Bay Trail more than doubles the single-threaded performance of its Clover Trail predecessor. The new Atom’s per thread performance is also about half that of the Core i3-3217U—and remember, the Core i3 is a 17W chip with even higher total platform power draw. The Silvermont microarchitecture is putting in a tremendous showing.

Here’s an Android-only graphics test that gives Bay Trail a chance to square off against Nvidia’s Tegra 3 and Qualcomm’s Snapdragon. What you need to know in order to understand these results is simple: both the Transformer Infinity and the Nexus 7 have 1920×1200 display resolutions, and they’re running the test at that native res. The Bay Trail tablet is pushing 60% more pixels, at 2560×1440, yet it’s still hitting the frame rate you see.

Now, a caveat: I’m pretty sure both the Nexus 7 and the Bay Trail system are hitting a vsync cap, limiting them to about 60 FPS, so we can’t say too much about how they truly compare. Ah, Android…

The first set of 3DMark Ice Storm scores is just chaotic. I didn’t lower the resolution on the Bay Trail tablet to match anything else, and I believe this test just runs at the native display resolution. Despite the handicap of having to push twice as many pixels as anything else, at least, the Bay Trail SoC scores reasonably well.

The “Unlimited” test is more what you want. It renders to a 720p off-screen buffer, providing a true cross-device comparison of performance. Unfortunately, this test is only built into the brand-new revision of 3DMark, which wasn’t yet out for Windows when we tested. In this much better comparison, the Atom Z3770 proves to be substantially faster than the Nexus 7 and the iPad 4.

Power consumption

We didn’t have nearly enough time to conduct battery life testing. However, we were able to measure the power consumption of a couple of tablets—one based on Clover Trail and the other on Bay Trail—at the battery connect point.

What we saw was very similar power consumption from one generation to the next. Both tablets tended to idle at about 2 W of power draw, and both used 3-4 W during video playback. Total system power draw is probably a bit higher during CPU- and GPU-intensive workloads, but we didn’t get any full-platform power use numbers for such scenarios.

We also got a look at the individual power consumption of our tablets’ graphics and CPU components in two scenarios: gaming and SunSpider testing.

While gaming, the Clover Trail system’s graphics drew about 650 mW, and the CPU drew 700 mW. The Bay Trail system’s total power use wasn’t far from Clover Trail’s, but the mix was very different, with 1.2W going to the IGP and 100-150 mW heading to the CPU. To be fair, though, the Bay Trail IGP was driving a much higher-resolution display.

In SunSpider, the CPU/GPU split on Clover Trail was 900/350 mW, while Bay Trail’s was 1000/475 mW—again, comparable total power use. Of course, Bay Trail finished the SunSpider test in half the time and then dropped back to idle, so it was easily the more power-efficient solution overall.

I think the big takeaway here is that Bay Trail’s power consumption habits should make it suitable for eight-hour-plus battery run times in tablets, much like Clover Trail before it. The big change is that you’ll be getting substantially higher performance at the same time.

Conclusions

The results we’ve shown you are admittedly early indicators, but they all seem to point to the same conclusion: that Intel has brought its considerable resources to bear on the tablet SoC market and delivered a best-in-class solution. The Silvermont CPU microarchitecture is a huge step up in per-thread performance, which is just what the Atom has needed for a while now. Bay Trail’s graphics are substantially better, too.

The question now is whether Bay Trail can find its way into the hands of consumers in really large numbers, which is the true measure of success for any consumer-focused SoC. For that to happen, Intel doesn’t just need to collect lots of design wins in various tablets and convertibles. Bay Trail has to ship inside of systems that are really good, the sort that can be truly popular with consumers. Whether that’s more likely to happen with devices based on Android or Windows 8.1 is tough to say. Windows-based convertibles still seem to have tremendous promise, and Bay Trail should allow for the creation of some compelling systems. But Microsoft has barely put a dent in the tablet market so far. Android may be the better opportunity.

At any rate, Intel certainly seems to be doing its part; its tablet SoC offering has made major strides from one generation to the next.

I often post stuff on Twitter using tablets and convertibles.

Comments closed
    • kamikaziechameleon
    • 6 years ago

    I would have liked a break down of how you perceive performance of bay trail in the conclusion. Not just the benchmarks but does it feel snappy when in regular use etc.

    The reason I ask for that is none of those numbers matter, sure its on par with tegra and snap dragon but its running a completely different OS with different demands. Benchmarks don’t matter if it produces an overly laggy experience in its native win 8 environment.

      • NeelyCam
      • 6 years ago

      You mean like this?:

      [url<]https://techreport.com/news/25375/hands-on-with-asus-349-bay-trail-based-transformer[/url<] [quote<]"Win8.1's touch-enabled Modern UI apps felt nearly ideal on the T100. App switches happened instantly, UI elements tracked effortlessly and scrolled silkily beneath my fingers, and web pages snapped open smartly. In fact, the T100 with Windows 8.1 felt smoother than the Haswell-based systems with Windows 8 that were on display across the room. The difference there may be due largely to software, but the T100 is also no slouch. The user experience was a far cry from the sadly compromised feeling of using a netbook. Seemed faster than a 4th-gen iPad, too."[/quote<]

    • FireGryphon
    • 6 years ago

    Ideal: A 64-bit Z3770 in a 13″ W8 tablet.

    Now we’re getting somewhere!

    • dashbarron
    • 6 years ago

    Will there be a general level of performance difference between Android apps run on this device vs. an ARM tablet?

    • jessterman21
    • 6 years ago

    I want this in an ultra-netbook, personally.

    10″ form factor, 12hrs battery life, good touchpad, good wireless, and aluminum body. For <$400.

    Edit – [url=https://techreport.com/news/25355/asus-transformer-book-t100-convertible-has-quad-core-bay-trail-soc-starts-at-349<]Oh.[/url<]

      • Waco
      • 6 years ago

      This. My EeePC 1025C is awesome at 11+ hours of battery life (with an SSD instead of the slow-as-molasses HDD it shipped with) but more performance would be well-appreciated.

      I’d buy an identical netbook with this chip without even thinking twice about it.

        • raddude9
        • 6 years ago

        [quote<]I'd buy an identical netbook with this chip without even thinking twice about it.[/quote<] For me they'd have to improve two more things, a screen with a higher resolution than 1024x600, and a higher than 2GB maximum memory capacity

          • Waco
          • 6 years ago

          Those would be nice, but similar battery life and more speed would be acceptable.

      • NeelyCam
      • 6 years ago

      [quote<]Edit - Oh.[/quote<] No sale. That Asus Transformer offers only 11h of battery life, falling crucially short of your requirements.

    • NeelyCam
    • 6 years ago

    (Even) Better than I expected. That performance is sick. Two things:

    1) This really puts the final nail in the WinRT coffin.
    2) This is going to eat into Haswell sales big-time.

    Because of 2), I’d really like to know what the BayTrail chip size is, so I can continue margin speculation.. I’m also wondering what this will do to Intel profits. This will definitely induce a refresh cycle with a temporary revenue boost, but long-term market revenue will likely come down even if volume grows (growth in BRIC) when high-price Core chip sales get replaced with low-price Atom chips. Even improved margins wouldn’t help when revenue goes down.

    I expect a two-year revenue and P/E bounce from all this, but maybe I should sell my shares late 2015 or early 2016..

      • chuckula
      • 6 years ago

      While these things aren’t going to obsolete Haswell notebooks, that 4.5 Watt Haswell part looks like more of an afterthought than a wildly popular product now.

      Intel has finally come around to the idea that it would be a whole lot better for Bay Trail to cut into big-x86 core sales than for ARM SoCs to be cutting into those sales.

        • NeelyCam
        • 6 years ago

        Intel will have to use some market segmentation to limit the damage to Haswell sales, and as Bay Trail just created a major performance gap to the nearest competitor, they can pretty easily do that.

        My guess is that Z3770 will be priced fairly high, while lower-end (“crippled”) versions will be priced closer to ARM and/or low-end AMD chips

          • dragontamer5788
          • 6 years ago

          [url<]https://techreport.com/news/25043/leaked-specifications-reveal-bay-trail-based-atom-celeron-and-pentium-processors?post=743851#743851[/url<] If the pricing rumor is right, then Z3770 will be over $130 per chip. Intel can fairly justify the higher prices due to the fact that Bay Trail is going to be 22nm, a more advanced process than pretty much everything else.

            • NeelyCam
            • 6 years ago

            Right, I forgot about that.. maybe the rumor is right. Price/performance ratios for various chips tend to be very similar, and if Z3770 has high performance, price is likely to be accordingly high

            • dragontamer5788
            • 6 years ago

            Indeed. The other consideration is that I’ve seen rumors that Snapdragon is between $20 and $30 per chip, which includes the LTE radio.

            If Intel can only achieve this performance at $100+ for the SOC (not including LTE radio), it might be a hard sell to the Android OEMs. Granted, Intel has now positioned this line of chips as the premium product in that power category.

            IIRC, margins for products want to be above 50%… as a general rule of thumb. So realistically, if a chip that costs $100 more than its competition will end up in a product that costs $200 more by the time it hits the end user. I don’t think Intel can sell the phone / tablet chips at $100 each and remain a competitor.

            • cal_guy
            • 6 years ago

            If $130 is correct then we’re not gonna see too many Z3770 tablets. Probably a better bet would be the Z3740 and Z3680.

            • TO11MTM
            • 6 years ago

            Now that the dust has settled, 130$ is shown to be an utterly absurd number:

            [url<]http://ark.intel.com/products/series/76761[/url<] We're looking at a 37$ for the top end according to ARK. This falls in line with Intel (thankfully?) understanding a 130$ chip would be DOA in this market. They learned from the original Atom.

            • chuckula
            • 6 years ago

            Thanks for that post! I’m sure we’ll hear all about how $37 is some obscene ripoff compared to [insert name of SoC here], but it’s nice to see some facts posted too.

            • NeelyCam
            • 6 years ago

            Don’t forget the 20MHz clock frequency crippling of the standard version compared to the D one that costs the same.

            I bet Intel does something evil like limit the max memory on the D-version just so you can never have the best of both worlds.

            EVIL MONOPOLY AND ITS ARTIFICIAL SEGMENTATION!!!

      • OneArmedScissor
      • 6 years ago

      Still obsessing over die sizes in a world of SoCs? It was inappropriate years ago when you insisted on comparing die sizes, when CPUs were already using multiple manufacturing processes with radically different costs.

      Now you’re just being silly, comparing devices which aren’t even put together the same way.

      Don’t you think the goal just [i<]might[/i<] be to reduce the total cost of the device itself, raising the sales potential? If the PC market maintains a 10-20% annual decline, cushy profit margins per sale don't mean squat.

      • Unknown-Error
      • 6 years ago

      Z3000 b!+ch goes past Ivy Bridge on a few occasions and with the exception of graphics, absolutely smashes A4-5000 especially when you look at the power consumption. Intel is really showing that they will be the one and only king from your smart watch to the supercomputer.

      • Klimax
      • 6 years ago

      Absence of some things like SATA or PCIe should fix that…

    • yuriylsh
    • 6 years ago

    I was definitely looking forward to CPU/GPU improvements (over what I see in my Z2760-based tablet which performance I can tolerate, but it’s far from ideal) and it seems Intel is delivering on this front. But I am very disappointed with the fact that they stick to eMMC, so it looks like I still have to pick something Haswell-based for my next tablet.

    Edit: on a second thought, I am [b<]definitely[/b<] skipping Bay Trail because of eMMC...

    • Sargent Duck
    • 6 years ago

    If the new Surface Pro 2 has this and is a decent price…win.

      • PixelArmy
      • 6 years ago

      This should be in a cheaper x86 Surface, but probably not the Pro. Should probably add a mid-level..

      RT = Tegra 4/Snapdragon 800
      Non-Pro = Atom
      Pro = Haswell

    • maroon1
    • 6 years ago

    Thats impressive. It even beats 15w kabini A4 5000 in CPU tests

      • NeelyCam
      • 6 years ago

      So, raddude, you predicted 2x efficiency advantage (BayTrail vs. Jaguar-based chips), I predicted 3x. Which one of us got closer…?

        • raddude9
        • 6 years ago

        Me of course.

        If you look at the power consumption of that Kabini system Anandtech benchmarked:
        [url<]http://www.anandtech.com/show/6981/the-kabini-deal-can-amd-improve-the-quality-of-mainstream-pcs-with-its-latest-apu/2[/url<] It could load a single core at just under 8 Watts. Which would be more than enough CPU power to play back video. And from this preview, Bay-trail plays back at video at 3-4W 4 X 2 = 8 QED 🙂 I was really talking about a full CPU load scenario, we'll see for sure when the full reviews come out.

          • NeelyCam
          • 6 years ago

          Hmmm… Let’s take another, closer look, and since you like Anandtech, let’s take Anand’s BayTrail numbers. We’ll use Cinebench since you wanted full CPU load:

          [url<]http://www.anandtech.com/show/7314/intel-baytrail-preview-intel-atom-z3770-tested/2[/url<] Baytrail is drawing 0.8-1.2W single-threaded, 2.5W multithreaded. From your link, Kabini draws 3.16W single-threaded (7.91W-4.75W), 6.75W multithreaded (11.5W-4.75W). [s<]Singlethreaded perf/watt: Baytrail: 1.48/1.2W=1.233/W Kabini: 1.5/3.16W=0.475/W [b<]Baytrail/Kabini: 2.60x[/b<] Multithreaded perf/watt: Baytrail: 1.655/2.5W=0.662/W Kabini: 1.323/6.75W=0.196/W [b<]Baytrail/Kabini: 3.4x[/b<][/s<] Huh - well, look at that. Seems that Baytrail is about 3x more efficient than Kabini. Sorry, I forgot - which one of us guessed 3x? Was it you? 😉 EDIT: I mess up the math again.. used the wrong benchmark numbers.. Here are the corrected numbers: [u<]Singlethreaded perf/watt:[/u<] Baytrail: 0.4/1.2W=0.16/W Kabini: 0.39/3.16W=0.058/W [b<]Baytrail/Kabini: 2.77x[/b<] [u<]Multithreaded perf/watt:[/u<] Baytrail: 1.48/2.5W=0.592/W Kabini: 1.5/6.75W=0.222/W [b<]Baytrail/Kabini: 2.66x[/b<]

            • raddude9
            • 6 years ago

            Now you’re looking too closely… at the wrong thing. I’m pretty sure I was talking about the whole system performance, not just the core performance.

            • NeelyCam
            • 6 years ago

            I’m pretty sure I was talking about the SoC performance, not the system performance. I vividly remember my argument being 22nm FinFET vs. 28nm planar, which obviously doesn’t impact the rest of the system one bit.

            • raddude9
            • 6 years ago

            Oh no, don’t try to back-track now that you’re losing. I started off the back-and forth with:

            [quote<]You can see that the entire platform only used 11.5W while running that multi-threaded benchmark.[/quote<] So clearly I'm talking about entire platform usage.... I'll take my apology in the form of an internet posting....;-)

            • dpaus
            • 6 years ago

            [AlecBaldwin] Just don’t suggest beer and wings – no, no, no!! [/AlecBaldwin]

            • chuckula
            • 6 years ago

            You know how in the Princess Bride you were never supposed to go against a Sicilian when death is on the line?

            Well, the TR corollary is: Always go against Neely when beer & wings are on the line!

            • dpaus
            • 6 years ago

            Even Neely acknowledges that so far, I’m winning that (non-) bet, so I think you mean “never go against dpaus when beer & wings are on the line”

            • NeelyCam
            • 6 years ago

            I went back and looked at the numbers; I used the wrong benchmarks for calculations. BayTrail SoC looks to be about 2.6-2.8x more efficient than Kabini.

            I really wasn’t talking about platform power, and even you started the discussion complaining about chip TDPs and that the whole Kabini system was consuming less in Cinebench than 15W. But since I can’t show that BayTrail platform wasn’t consuming more than 4.5W during Cinebench (as Anand only quoted the SoC power of 2.5W), I guess I am losing, and I have to now apologize to you in this internet posting.

            So, my apologies.

            • chuckula
            • 6 years ago

            NEVAR OPOLOJAIHZ!!

            [quote<]BayTrail SoC looks to be about 2.6-2.8x more efficient than Kabini.[/quote<] Well, using standard rounding techniques, 2.6 is closer to 3 than it is to 2... or did you overreach and try to claim a 4X advantage?

            • NeelyCam
            • 6 years ago

            That’s the SoC advantage. Raddude focuses on platform advantage.

            His argument is that AMD platforms are generally more efficient than Intel ones, so BayTrail’s near-3x SoC advantage over Kabini diminishes when you include Intel’s bad platform design. I can’t argue against that because neither Anand nor Scott listed platform power consumption numbers for BayTrail (BT) when running Cinebench multithreaded.

            I don’t quite agree with him that using some approximate power numbers for “video playback” is a good measure of platform efficiency (actually, he didn’t even have Kabini video playback numbers – he just assumed a single-threaded Cinebench was a good-enough proxy for Kabini’s video playback), and he used the worst-case BT number from the range Scott mentioned.. Seems more like a political argument than a scientific one..

            Anyway, that means that I ‘lost’ and needed to apologize.

            • chuckula
            • 6 years ago

            [quote<]platform advantage[/quote<] What exactly does that mean outside of the SoC in these types of devices?? There's no real Northbridge/southbridge/etc. using power. Do you mean things like RAM power usage or the back light on the screen? That's beyond the control of either AMD or Intel when a third party builds the final device.

            • raddude9
            • 6 years ago

            I’m talking about the amount of power the whole system uses. To me it doesn’t make much sense to test just the SOC itself, that’s beyond the abilities of most reviewers, but testing the power consumption of the whole platform is easily done by most people and that’s the amount of power they are going to be charged for at the end of the day.

            Sure, the screens in laptops and tablets can vary massively in the amount of power they consume, measuring just a motherboard would make more sense. Now we just have to wait for a nice mini-ITX Bay Trail motherboard (and a Kabini one as well I suppose) to test.

            • NeelyCam
            • 6 years ago

            It makes sense to compare the SoC itself, especially when the SoCs are very highly integrated. The differences are coming mostly from peripherals like display. To estimate how the two SoCs would perform in a system with the same display etc., looking at SoC alone gives useful information.

            • raddude9
            • 6 years ago

            I’m not saying it makes no sense to compare SOC’s themselves, but, if I stretch things to a car analogy (as is tradition). Looking at the power consumption of a SOC is like looking at the MPG of an engine, it’s an abstraction indication of performance, me, I’ll stick with looking at MPG values for entire cars.

            Sure, the screen is often the most variable component in terms of power consumption, that’s why I like my power-consumption numbers without screens as I said.

            • raddude9
            • 6 years ago

            [quote<]His argument is that AMD platforms are generally more efficient than Intel ones[/quote<] Not really, I know the first-gen atoms suffered from a particularly bad power hogging platform, but that's long since been rectified. The platforms these days seem to be quite close, but then there's only so much you can do without having to resort to low power RAM and such. My main point is that because the platform uses a fairly fixed amount of power (when you compare platforms with similar features), the law of diminishing returns kicks in, so going from a 15W CPU to a 3W CPU isn't going to give you the 5x power savings that it appears. Because, assuming the platform for both machines uses 3W, then you're going from a 18W system to an 6W system (for example), which means the real-world difference is just 3x And when you dig into the comparison even further, The Kabini chip uses nothing like it's full TDP when only the CPU cores are fully loaded, whereas the BayTrail comes much closer to using it's full TDP, likely because of the cores turbo'ing to their max. Thus the real-world difference in this particular scenario is closer to 2x. Yep, I don't like to use approximate numbers for video playback either, it's was just a rough guesstimate in lieu of some cold hard facts. So I wouldn't be sure that you lost just yet, I'll be waiting until I see some full reviews before I claim victory.

            • Klimax
            • 6 years ago

            Just small reminder that TDP is not max consumption, but max heat put out.

            • ronch
            • 6 years ago

            You both lose. I win!!

            Er, what do I get?

            • chuckula
            • 6 years ago

            What else ronch: BEER & WINGS!

        • shank15217
        • 6 years ago

        If you read the anand tech preview they actually have power numbers for kabini in different workloads and the power consumption would be around 6.5w for kabini on most workloads with about 11.5 watts in the highest end. Kabini also destroys these atoms in the gpu tests. Next version of kabini will probably have a turbo core feature that should make better use of the process tech. I consider this standard leap frogging, I dont think AMD is out of the running here.

          • NeelyCam
          • 6 years ago

          Actually, Kabini is 8W for most workloads:

          [url<]http://www.anandtech.com/show/6981/the-kabini-deal-can-amd-improve-the-quality-of-mainstream-pcs-with-its-latest-apu/2[/url<] Ivy Bridge 'destroys' kabini on most GPU tests, and Bay Trail 'destroys' Kabini on many CPU tests (even though it's running at a significant lower power). And AMD won't be leapfrogging anymore now that all the low-hanging fruit has been picked and process technology has become the main differentiator for performance and performance/watt.

            • dragontamer5788
            • 6 years ago

            There was an interesting point somewhere that Single-channel Ivy Bridge vs Single-Channel Kabini… Kabini seemed to win. It was a TechReport benchmark somewhere.

            I think Kabini is “losing” to Ivy Bridge, and similarly Atom is keeping up with Kabini, because of the Dual-Channel RAM. Doubling your bandwidth is of major importance to most games. GPUs seem to be mostly limited by RAM.

            AMD’s mistake seems to be making Kabini / Temash single-channel only. Hopefully they can correct that mistake soon.

    • chuckula
    • 6 years ago

    To TR: Thank you for this first look.

    As a quick question for your Android benchmarks, can you tell us anything about the Android release & Kernel version that was used? Was the software provided directly by Intel or through an OEM? I’m curious because there has been quite a bit of Bay Trail development both in power consumption and GPU drivers for the Linux kernel and it would be interesting to know how current your software stack is compared to the mainline Linux kernel.

    I know that Android is a problematic benchmarking platform, but thanks for these results!

      • Damage
      • 6 years ago

      The Bay Trail tablet was running Android 4.2.2. Intel had the OS pre-installed. I downloaded the apps from the Play store. No OEMs were involved in this testing of a non-OEM reference platform.

      The other tablets were our own. Running Android 4.3 on my Nexus 7.

        • chuckula
        • 6 years ago

        Thanks!

        • ermo
        • 6 years ago

        @Scott:

        Based on your current knowledge, how well do you expect hot-clocked (similar TDP/SDP) Qualcomm S800 and NVidia Tegra 4 parts to compete with the top-of-the-line Z3xxx parts? Do you expect the ARM parts to have better graphics or …?

          • NeelyCam
          • 6 years ago

          I’m not sure you need to “hot-clock” S800/T4 to reach the same TDP/SDP…

            • ermo
            • 6 years ago

            The idea is to compare performance between devices with a similar wattage during load, nothing more, nothing less.

            I’m on the cusp of buying a 2560×1600 dockable tablet thingy. Such a comparison would inform my purchase, which is what TR is about, right?

            • NeelyCam
            • 6 years ago

            [quote<] Such a comparison would inform my purchase, which is what TR is about, right?[/quote<] Sure. My point was that S800/T4 are probably already drawing a similar wattage already in stock (if not more) during load as BayTrail, them being 28nm planar instead of 22nm FinFET.. But I'm eagerly waiting for actual benchies, so I don't need to speculate so much..

    • Unknown-Error
    • 6 years ago

    By the early results, ….so much for “ARM is more efficient than x86”. CPU performance is astonishing. On rare occasions goes past Ivy-Bridge. Only thing using ARM ISA that might compete in that department is the A7 (Assuming the 2x performance of A6 is true). And, Intel has already shown 14nm prototypes. ARM, better get your act together or you’ll end-up like [b<]A[/b<]dvanced [b<]M[/b<]acro [b<]D[/b<]isasters.

      • My Johnson
      • 6 years ago

      So, the question is this: Does ISA matter regarding efficiency or performance and all evidence I’ve seen over the years is it does not. And I bet it’s because all these ISA’s are very mature. They are decades old.

        • mesyn191
        • 6 years ago

        ISA can still matter its just that ARM doesn’t really have any real inherent performance advantages over x86.

        A different example would be Intel’s IA-64 ISA, which is actually pretty good at FP operations…while also being mediocre to bad at scalar ops though.

      • TO11MTM
      • 6 years ago

      Arguing the merits of ISAs has been arguable since the Nx586. It all becomes Micro-ops anyway.

    • Ryhadar
    • 6 years ago

    I was pretty shocked how well this chip did compared to the A4-5000 in the CPU tests until I remembered the Z3770 has a pretty big boost range (1.46GHz – 2.39GHz).

    What was AMD thinking not giving Kabini boost?

    Anyway, I don’t mean to take away from the great engineering of the Z3000 series. The perf/watt numbers from these chips seem fantastic.

      • esterhasz
      • 6 years ago

      I am also very impressed. This is fast enough for fluid work in most everyday PC tasks if paired with an SSD and both price and battery life should be quite nice.

      • NeelyCam
      • 6 years ago

      [quote<]What was AMD thinking not giving Kabini boost?[/quote<] In order to be able to boost, the chip/circuitry will have to have clock frequency headroom. One of Jaguar's key points was that it's synthesizable, so it's cheap to design and port to different processes... but the downside to that is that the layout won't be as optimized for performance. That means less frequency headroom, and more difficulty boosting it. It's possible that high-clocked Kabinis just can't run any faster without crazy amount of extra voltage (and, consequently, power consumption). It's sort of like AMD FX-9000, that needs a really high supply voltage to run at 5GHz, and then consumes >200W. Or IvyBridge/Haswell chips overclocked to 5GHz. Every chip has a clock frequency point beyond which the power consumption goes through the roof (OneArmedScissor used to call this the "4GHz wall"). Process advantage (22nm FinFET vs. 28nm planar) helps Intel design in more clock frequency headroom, and they also have enough resources that they can use custom layout instead of synthesized layout.

        • mczak
        • 6 years ago

        That might be true for the fastest Kabini part (A6-5200). However, consider the A6-5200 can run all its cores at 2Ghz together with the GPU (which is faster clocked than the one in A6-5000) within a TDP limit of 25W. Whereas A6-5000 has a TDP of 15W. So based on these numbers it is very easy to see that A6-5000 should have absolutely no problem of running one core at 2Ghz within that 15W power envelope, and in fact I suspect it could run all 4 at 2Ghz (or at least very close to it) as long as the gpu is inactive and still not exceed 15W.

        • ronch
        • 6 years ago

        Yeah. And unlike AMD who has to be flexible when it comes to foundries, Intel can practically hand-draw every circuit and tailor it to their own process tech, which they also have complete control of.

      • cal_guy
      • 6 years ago

      I suspect that AMD needed to prioritize features that were useful to the consoles and a non-deterministic turbo isn’t going to be usable in a console. We will see a enhanced Jaguar early next year with connected standby and 2 Ghz+ operating speeds and probably a turbo that resembles that of Richland or Trinity.

      • raddude9
      • 6 years ago

      AMD did this exact thing before and it worked out well for them, so I imagine that’s why they’re planning to repeat the process. If you recall, the AMD C50 had a fixed clock at 1Ghz, but 8 months later they released the C60 which could turbo it’s CPU and GPU by about 33%. In lieu of moving to a new process node to improve performance, making a new version with good turbo functionality would give them a new and improved chip to sell to OEM’s early next year.

        • ronch
        • 6 years ago

        Yeah, Bobcat worked quite nicely, but it’s akin to the Athlon and Athlon 64 when Intel was flubbing with their Pentium 4 and Prescott chips and AMD had a window of opportunity. This time, Intel is taking their mobile lineup far more seriously, which means whatever worked for AMD before may not necessarily work for them again. This applies to the ARM bandwagon as well.

      • jihadjoe
      • 6 years ago

      I don’t think AMD really has any room to do boost, considering Kabini already uses 15W.

      What’s really impressive here is Intel is basically competitive with AMD’s 15W part using just 2-3W. That’s process advantage at work.

        • raddude9
        • 6 years ago

        Nope, Kabini has plenty of room to boost, here you go:

        [url<]http://www.anandtech.com/show/6981/the-kabini-deal-can-amd-improve-the-quality-of-mainstream-pcs-with-its-latest-apu/2[/url<] With the 4 CPU cores fully loaded (on the A4-5000), the entire Kabini platform uses only 11.5W. This is a clear indication that there is plenty of room for boosting the CPU cores, particularly when the GPU is mostly idle. In fact, most of the thermal headroom for that particular chip seems to be reserved for the GPU.

        • sschaem
        • 6 years ago

        what are you talking about. it require 2w just to playback video, and thats using dedicated decoding HW.
        the 4 x86 core loaded + GPU loaded + memory control + cache , do not max out at 2-3w when just video playback take 2w.

        and turbo, you re-allocate the TDP of all core to an isolated few core. so its not limited by TDP but clock rate.

        its funny how intel didn;t put out TDP numbers, only SDP.

        Wonder what you would see if you run prime95 + furmark. on a a5-3770 2w?

    • chuckula
    • 6 years ago

    [quote<]While gaming, the Clover Trail system's graphics drew about 650 mW, and the CPU drew 700 mW. The Bay Trail system's total power use wasn't far from Clover Trail's, but the mix was very different, with 1.2W going to the IGP and 100-150 mW heading to the CPU. To be fair, though, the Bay Trail IGP was driving a much higher-resolution display. In SunSpider, the CPU/GPU split on Clover Trail was 900/350 mW, while Bay Trail's was 1000/475 mW—again, comparable total power use. Of course, Bay Trail finished the SunSpider test in half the time and then dropped back to idle, so it was easily the more power-efficient solution overall.[/quote<] I know these numbers are not the full story but thank you TR for giving use at least a quick snapshot of real-world power usage. As has been shown in the past, the real-world power draw for these chips under practical workloads is quite competitive with anything else on the market, and Bay Trail is kicking up the performance while keeping the power consumption within reason.

    • bjm
    • 6 years ago

    Put this in a $349 or less Microsoft Surface and now we’re talking. Heck, put them in a Microsoft Surface Phone and I’ll buy them both.

    • ronch
    • 6 years ago

    Today’s die shots look like they’re just exported from the design tools used to draw up the chips themselves, with everything shown as a bunch of colorful squares put together. I wish they would bring back real die shots.

      • chuckula
      • 6 years ago

      Should I buy one and delid it to keep it real?

        • ronch
        • 6 years ago

        I thought you already did.

          • chuckula
          • 6 years ago

          No, that was Haswell 🙂

            • ronch
            • 6 years ago

            If you’re ready for another delidding (?) project, I don’t know if these Bay Trail chips are even an option. Do these even have lids/IHS?

            • chuckula
            • 6 years ago

            Yeah, I doubt they do. Taking the heat sink off would probably be how you’d get to see the chip directly and I’m sure someone will do that soon (we’ll also be able to find out the die size at that point).

            One thing about the chip when it’s on the package is that you don’t see all those cool etch patterns because the etched side of the chip is facing down into the package. You just see the substrate which is basically looks like a very dark reflective glass (it is taken from a silicon ingot after all).

            • ronch
            • 6 years ago

            Yeah, I know, kinda like the Athlon Thunderbird days. There’s a thin metal cover above the die itself to protect it from the heat sink, of course, and there’s probably an insulating layer between the cover and the die itself. You’d have a much easier time taking off a graphics card’s cooler than delidding a CPU if you wanna see that sort of thing.

      • WillBach
      • 6 years ago

      The “real” die shots are photoshopped to protect trade secrets IIRC 😉

        • ronch
        • 6 years ago

        Well, it’s not like engineers employed by a competitor can gain much by looking at die shots without the overlays / labeling.

    • ClickClick5
    • 6 years ago

    If Intel mows over the mobile market, ARM might need to start sweating.

      • ermo
      • 6 years ago

      With this chip, intel + Windows 8.x finally make sense in the 10″ dockable tablet + keyboard post-netbook niche.

      So, yeah, I see this as a better netbook. Too bad it looks like AMD is already out of the game after the success with Bobcat.

    • dpaus
    • 6 years ago

    Wasn’t I the first to hear that these would be in 30% of all cellphones by the end of 2013? [holdsBreath]

      • GeneS
      • 6 years ago

      I think Neely did (eventually?) make it clear that he meant the Merrifield chips, but the public semi-shaming is still good for a giggle (and it’s pretty obvious that Merrifield market share isn’t going to hit 30% in the smartphone space this year, or likely next year either)

      • NeelyCam
      • 6 years ago

      It was 10% in Dec. 2013, 30% in Dec. 2014.

      Nonetheless, I totally lost that first bet..

        • dpaus
        • 6 years ago

        Unlike raddude9, I [i<]will[/i<] accept beer-n-wings 🙂

          • NeelyCam
          • 6 years ago

          I don’t have time to travel to Canada right now.. will you accept some american cash (or maybe an Amazon gift card?) for your beer/wings night? Just let me know what the damage is

            • dpaus
            • 6 years ago

            [quote<]ust let me know what the damage is[/quote<] Well, his real name is Scott Wasson, and- Oh, wait. The 'real' damage is (and always was) that you'd have to take a night off from your Finnish heavy metal coffee bar and meet me next time I visit whatever U.S. city you're in. If you've come to accept your fate, let me know your locale and I'll check my itinerary.

            • GeneS
            • 6 years ago

            [Vader] Neely. It is your… destiny… [/Vader]

            • NeelyCam
            • 6 years ago

            [quote<]take a night off from your Finnish heavy metal coffee bar and meet me next time I visit whatever U.S. city you're in.[/quote<] WHERE DO I SIGN UP??!!

        • Beelzebubba9
        • 6 years ago

        You’ll probably lose the second one too unless Intel lands the SGS5.

        That said, by the end of 2015….

          • NeelyCam
          • 6 years ago

          [quote<]You'll probably lose the second one too unless Intel lands the SGS5.[/quote<] I haven't lost all hope of that actually happening.. and maybe Apple will also realize that Intel's chips are superior and fire all their chip designers. Hmm... if Apple fabs their chips at Intel, I wonder if that would count towards the bet...

            • chuckula
            • 6 years ago

            I’m more optimistic about Intel in tablets than I am in smartphones (for the next 2 years or so). 30% share in tablets by late 2014? That’s not guaranteed, but it’s within reason. 10% smartphone share by late 2014? Maybe.

            As has been pointed out before, it’s not the processor core that is the biggest stumbling block to Intel right now, but the RF and other bits & bobs. I definitely think Intel will take a chunk of smartphone marketshare, but it will be a gradual process.

Pin It on Pinterest

Share This