Intel employs more than 100,000 people. A great many companies are larger. But still, you don’t change the direction of an organization doing more than $50 billion in revenue a year quickly, or without a specific endgame. And yet, Intel is intent on recalibrating its focus. In fact, a pre-IDF note I received was even more explicit than that.

Resetting the course of the company with a clear emphasis on mobile computing leadership. Wow. Take that Apple, Qualcomm, Nvidia, and Samsung. But while Intel’s home court rallying cry comes as a bit of a surprise, its playbook through the end of next year certainly isn’t.
Back in June, we spent time at Intel’s headquarters in Santa Clara to get briefed on the architecture that’d power its next generation of low-power desktop, notebook, tablet, appliance, smartphone, and server CPUs. The resulting story, Intel Silvermont Architecture: Does This Atom Change It All?, gave us a look at a design that promised either significantly more performance per watt or great power savings at a given performance level. We were naturally eager to get our hands on an example of Silvermont in action. More on that shortly.
Bay Trail, Powered By Silvermont, In Your Next Tablet?
Last week, Intel introduced the family of SoCs previously referred to as Bay Trail, which address the tablet, notebook, and desktop markets. Although you’re going to see Bay Trail-based processors branded as Atom, Celeron, and Pentium, they all employ the Silvermont core design, albeit within a range of power and performance levels.
In keeping with its mobile focus for IDF, though, Intel emphasized the tablet-specific quad-core Atom Z3700 and dual-core Atom Z3600 line-ups. They succeed the dual-core, quad-threaded Atom Z2760, which saw some play in a number of tablets running Windows 8. None of those tablets really caught on, though. Why not? Well...

...to begin, they were expensive. Samsung’s ATIV Smart PC 500T with 64 GB of storage sold for $750 with its optional keyboard. They weren’t very fast, either. Finally, we stumbled on some pretty severe quality issues with the Samsung and Acer units—two of the highest-profile Atom-based tablets. Though they were able to compete against Nvidia’s Tegra 3 and Qualcomm’s S4 Pro in measures of efficiency, after a while we found ourselves picking those Atom-based platforms up and setting them right back down.
And that’s actually part of what makes Bay Trail so exciting. Intel is expanding its horizons, maintaining its Windows focus, but branching out to enable x86 on Android (a process that began earlier this year with Clover Trail+). That should help widen Atom’s price band. As we know, Intel also makes some pretty broad claims about Bay Trail’s performance, which we'll put to the test shortly. We’re pretty certain that its improvements will translate to better responsiveness. Quality issues need to be ironed out by the manufacturers. But Intel has its engineers busy working with partners to improve their designs, and we can only hope many lessons were learned from Clover Trail.

Intel makes clear where you’ll find Bay Trail-based SoCs and Ivy Bridge/Haswell-based processors by charting price over a range between maximum mobility and peak performance. All the way to the left, tablets with Atom inside should stretch from $200 to about $550, while Core-based devices top out in the $850 range. Divisions are also made in the notebook space. Atom will not go into Ultrabooks-branded platforms, but is expected to show up in detachable and convertible systems, otherwise referred to as two-in-ones.
Alright, so we have an idea of where Intel is coming from and where it hopes to go with the Atom Z3000 series for tablets. Now let’s take a look at the six SKUs launched at IDF, along with their specifications:
| Atom | Z3770 | Z3770D | Z3740 | Z3740D | Z3680 | Z3680D |
|---|---|---|---|---|---|---|
| Process Tech. | 22 nm | 22 nm | 22 nm | 22 nm | 22 nm | 22 nm |
| Cores | 4 | 4 | 4 | 4 | 2 | 2 |
| L2 Cache | 2 MB | 2 MB | 2 MB | 2 MB | 1 MB | 1 MB |
| Core Clock Rate | Up to 2.4 GHz | Up to 2.4 GHz | Up to 1.8 GHz | Up to 1.8 GHz | Up to 2 GHz | Up to 2 GHz |
| Memory | 2-Channel LPDDR3 1066 MT/s | 1-Channel DDR3L 1333 MT/s | 2-Channel LPDDR3 1066 MT/s | 1-Channel DDR3L 1333MT/s | 1-Channel LPDDR3 1066 MT/s | 1-Channel DDR3L 1333 MT/s |
| Peak Memory B/W | 17.1 GB/s | 10.6 GB/s | 17.1 GB/s | 10.6 GB/s | 8.5 GB/s | 10.6 GB/s |
| Memory Capacity | Up to 4 GB | 2 GB | Up to 4 GB | 2 GB | 1 GB | 2 GB |
| Max. Res. | 2560x1600 | 1920x1200 | 2560x1600 | 1920x1200 | 1280x800 | 1920x1200 |
As we already know from our look at Silvermont, Intel is leveraging a version of its 22 nm process optimized for the SoC architecture. Four models sport four cores (or two modules—remember, this is a module-oriented design that couples two cores and 1 MB of shared L2 cache). Two are single-module designs, with two cores and 1 MB of L2.
Rather than simply referring to base clock rate and specifying a frequency ceiling, Intel decided it was a good idea to identify each SoC’s maximum Burst Technology speed. So, the quad-core Atom Z3000s are split between processors that top out at 1.8 GHz and those able to hit 2.4 GHz.
They’re available in a couple of different memory configurations, as well. A pair of the four-core SKUs support two channels (4 GB total) of LPDDR3-1066, pushing up to 17.1 GB/s of bandwidth. That’s key to enabling display resolutions of 2500x1600. The other quad-core SoCs and one of the dual-core models accommodate a single channel of DDR3L-1333. Two gigabytes of memory running at that speed does 10.6 GB/s, which is good enough for 1920x1200. The remaining Atom Z3680 takes one channel of LPDDR3-1066, is limited to 1 GB, and only offers enough bandwidth to support 1280x800.

All of this information is new. Back when Intel introduced us to Silvermont, it avoided product-specific questions. We knew about the module-based design of course, and we suspected that tablet-oriented SoCs would be dual- and quad-core configurations. But the engineers didn’t go into detail about the memory controller, for example—and for good reason. The specifics are implementation-dependent. Bay Trail exposes up to two 64-bit channels of LPDDR3 (last generation, you were stuck with two 32-bit channels of DDR2-800). But in Intel’s server-oriented Avoton, also built on a Silvermont foundation, faster ECC-capable DDR3 is supported.
We also now know that the Bay Trail SoC hosts an HD Graphics engine with four EUs. It runs at up to 667 MHz, though the base clock rate is 311 MHz. Intel says the architecture comes from Ivy Bridge, though it’s worth noting that four EUs is fewer than the GT1 implementation of Sandy Bridge, which featured six EUs. Even still, company representatives say to expect three times as much graphics performance as the PowerVR SGX545-equipped Atom Z2760. HD Graphics adds DirectX 11 and OpenGL ES 3.0 support, whereas Imagination Technologies’ solution was limited to DirectX 10.1 and OpenGL ES 2.0.
Heterogeneous compute via APIs like OpenCL is not supported on Bay Trail for tablets—Intel says its focus was to get the platform out with a focus on compute through the x86 cores. However, its system agent was designed to support a heterogeneous architecture in the future. In fact, the Celeron we're testing today does include the OpenCL runtime in its driver package, and we were able to get OpenCL working on the HD Graphics component, too.

As with its Core architecture, Intel’s Atom Z3000-series shares one power budget between on-die subsystems. So long as the SoC is operating under its power, current, and temperature limits, the x86 cores and other IP blocks are allowed to run at higher clock rates when more performance is needed. Although you won’t realize those peak Burst Technology figures in a thermally-restricted workload, you won’t slam into a ceiling and watch the chip throttle either, as we've seen from competing SoCs. There’s certainly something to be said for more elegant power management, made possible in large part to the central system agent.
The memory controller, cores, and graphics engine all tie together through this critical component, which maintains coherency between the shared caches. Image signal processing, fixed-function video decode, and display control attach to the system agent as well.
Beyond simply beefing up 3D potential, Bay Trail also gets the power and performance benefits of Quick Sync’s fixed-function and programmable media pipeline. H.264, VC-1, MPEG-2, and MVC are all decoded in hardware, while H.264 and MPEG-2 encoding can be accelerated.

Bay Trail supports HDMI 1.4, DisplayPort 1.2, eDP 1.3, and MIPI-DSI output across a pair of display pipelines. Given enough bandwidth, Intel says its DisplayPort connectors will do 2560x1440 at 60 Hz, while HDMI maxes out at 1920x1080. Both digital interfaces do support integrated audio.

An enhanced image signal processor is rated for up to 275 MP/s with support for auto-exposure, -focus, and –white balance, along with 1080p60 video capture, video stabilization, burst mode, continuous capture, and low-light noise reduction. The outward-facing switch supports sensors up to 13 MP, while the forward-facing switch supports up to 2 MP.
In a modern Intel desktop architecture, everything mentioned above is built onto the CPU die. The processor communicates over DMI to a Platform Controller Hub loaded up with storage, USB, audio, and networking—all of the I/O typically associated with a south bridge. Bay Trail takes that functionality and moves it up into the die through a switching fabric attached to the system agent. This fabric manages bandwidth and access priority for some of the new on-board subsystems. For instance, as you saw in the block diagram above, Bay Trail’s storage hub supports SDIO 3.0, SD 3.0, and eMMC 4.51. Native USB 3.0 is a big deal as well.
Again, Intel is most intently focused on the six tablet-oriented implementations of Bay Trail, and we share the company’s enthusiasm for portable devices with all-around better performance, longer battery life, and more advanced I/O capabilities. But because design wins are still forthcoming, there aren’t any tablets to test yet.
We did manage to get our hands on a Celeron J1750-based platform, though. Back on the first page, we mentioned that Bay Trail-based SoCs will also drive notebooks and desktops, albeit at different power targets. Intel is planning quad-core Pentium-branded processors with HD Graphics for both segments, along with quad- and dual-core Celeron SoCs with HD Graphics. For notebooks, expect N3000-series Pentiums and N2000-series Celerons. The desktop will initially see a Pentium J2850, a Celeron J1850, and a Celeron J1750. The latter model is a dual-core (single-module) chip; the other two are quad-core SKUs.
The Celeron-based mini-ITX platform in our possession gets away with passive cooling, since the dual-core Celeron only dissipates 10 W. The Silvermont cores run at 2.4 GHz, dropping as low as 500 MHz when the system is idle. Two SO-DIMM slots take 1.35 V DDR3L-1333 memory. Though storage connectivity is limited to SATA 3Gb/s, don’t think that your SSD is going to be bottlenecking this low-power SoC. Instead, we’re just happy that USB 3.0 is supported natively. Four lanes of second-gen PCI Express are also available.
| Clock Rate | L2 Cache | C/T | Mem. Data Rate | Max. Turbo Boost | Graphics | Dynamic Freq. | |
|---|---|---|---|---|---|---|---|
| BGA 65 W SKUs | |||||||
| Core i7-4770R | 3.2 GHz | 6 MB | 4/8 | 1600 MT/s | 3.9 GHz | Iris Pro 5200 | 1300 MHz |
| Core i5-4670R | 3 GHz | 4 MB | 4/4 | 1600 MT/s | 3.7 GHz | Iris Pro 5200 | 1300 MHz |
| Core i5-4570R | 2.7 GHz | 4 MB | 4/4 | 1600 MT/s | 3.2 GHz | Iris Pro 5200 | 1150 MHz |
| BGA 10 W SKUs | |||||||
| Pentium J2850 | 2.4 GHz | 2 MB | 4/4 | 1333 MT/s | N/A | HD Graphics | 688 MHz Base 792 MHz Max. |
| Celeron J1850 | 2 GHz | 2 MB | 4/4 | 1333 MT/s | N/A | HD Graphics | 688 MHz Base 792 MHz Max. |
| Celeron J1750 | 2.4 GHz | 1 MB | 2/2 | 1333 MT/s | N/A | HD Graphics | 688 MHz Base 750 MHz Max. |
Curious as to how Bay Trail-D fares against some of the most entry-level platforms, we rounded up Zotac’s D2700-ITX WiFi Supreme (sporting a 10 W Atom D2700 and GeForce GT 520 graphics), AMD’s 65 W A4-4000 based on the Richland architecture, and a 55 W Celeron G1610 based on Ivy Bridge. Naturally, the A4 and 55 W Celeron are in a completely different class. But they also represent two of the least-expensive drop-in upgrades you can buy, both under $50. Some of our charts also include Samsung's ATIV Smart PC 500T with Atom Z2760 as a point of reference.
If anything, we’re most interested to see how Celeron J1750 stacks up to Atom D2700 at 2.13 GHz. Sporting a technically inferior graphics engine, lower core clock rate, lower peak graphics frequency, and notably less memory bandwidth, it shouldn’t even be a contest. But the Cedarview-based processor, built using Saltwell cores, does enjoy the benefit of Hyper-Threading.
| Test Hardware | |
|---|---|
| Processors | Intel Celeron J1750 (Bay Trail-D) 2.4 GHz (29 * 83.3 MHz), BGA, 1 MB Shared L2, Dual-Core, Power-savings enabled |
| Intel Celeron G1610 (Ivy Bridge) 2.6 GHz (26 * 100 MHz), LGA 1155, 2 MB Shared L3, Dual-Core, Power-savings enabled | |
| Intel Atom D2700 (Cedarview) 2.13 GHz (16 * 133 MHz), BGA559, 2 x 512 KB L2 Cache, Hyper-Threading enabled, Power-savings enabled | |
| AMD A4-4000 (Richland) 3.0 GHz (30 * 100 MHz), Socket FM2, 1 MB L2, Turbo Core enabled, Power-savings enabled | |
| Motherboard | MSI Z77 Mpower (LGA 1155) Intel Z77 Express, BIOS 17.8 |
| Zotac D2700-ITX WiFi Supreme (BGA559) | |
| MSI FM2-A85XA-G65 (Socket FM2) AMD A85X, BIOS 2.0 | |
| Memory | Crucial 4 GB (2 x 2 GB) DDR3L-1333, CT25664BF1339.M8FKD at 1.35 V |
| G.Skill 4 GB (2 x 2 GB) DDR3-1066, F3-8500CL7D-4GBSQ at 1.5 V | |
| Patriot 4 GB (2 x 2 GB) DDR3-1600, PGS34G1600ELKA at 1.5 V | |
| Hard Drive | Samsung 840 Pro 256 GB, SATA 6 Gb/s |
| Graphics | Nvidia GeForce GTX Titan 6 GB |
| Power Supply | Corsair AX860i, 80 PLUS Platinum, 860 W |
| System Software And Drivers | |
| Operating System | Windows 8 Professional x64 |
| DirectX | DirectX 11 |
| Graphics Driver | Nvidia GeForce Release 320.18 AMD Catalyst 13.10 Beta Intel 15.31.9.64.3165 |
| Benchmark Configuration | |
|---|---|
| Adobe Creative Suite | |
| Adobe Photoshop CS6 | Version 13 x64: Filter 15.7 MB TIF Image: Radial Blur, Shape Blur, Median, Polar Coordinates |
| Audio/Video Encoding | |
| iTunes | Version 10.4.1.10 x64: Audio CD (Terminator II SE), 53 minutes, default AAC format |
| Lame MP3 | Version 3.98.3: Audio CD "Terminator II SE", 53 min, convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kb/s) |
| HandBrake CLI | Version: 0.98: Video from Canon Eos 7D (1920x1080, 25 FPS) 1 Minutes 22 Seconds Audio: PCM-S16, 48,000 Hz, Two-Channel, to Video: AVC1 Audio: AAC (High Profile) |
| TotalCode Studio 2.5 | Version: 2.5.0.10677: MPEG-2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG-2), Audio: MPEG-2 (44.1 kHz, 2 Channel, 16-Bit, 224 Kb/s), Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV |
| Productivity | |
| ABBYY FineReader | Version 10.0.102.95: Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| Adobe Acrobat X | Version 10.0.0.396: Print PDF from 115 Page PowerPoint, 128-bit RC4 Encryption |
| File Compression | |
| WinZip | Version 17.0 Pro: THG-Workload (1.3 GB) to ZIP, command line switches "-a -ez -p -r" |
| WinRAR | Version 4.2: THG-Workload (1.3 GB) to RAR, command line switches "winrar a -r -m3" |
| 7-Zip | Version 9.28: THG-Workload (1.3 GB) to .7z, command line switches "a -t7z -r -m0=LZMA2 -mx=5" |
| Synthetic Benchmarks and Settings | |
| 3DMark 11 | Version: 1.0.1.0, Benchmark Only |
| SiSoftware Sandra 2013 | Version 2013.01.19.11, CPU Test = CPU Arithmetic / Multimedia / Cryptography / Memory Bandwidth / Cache Bandwidth |
We're going to buck convention today and start with a look at power consumption over time. I can't help it; after slogging through several pages of unexciting benchmark results in Intel Core i7-4960X Review: Ivy Bridge-E, Benchmarked, only to discover that efficiency was ironically the most interesting characteristic of Intel's new flagship, I'm really curious as to how Bay Trail registers, even if it means spoiling some of the performance story.

The red line is Intel's 10 W Celeron J1750, and wow, check out that power consumption through our suite. We can tell by the length of the line that the new Celeron isn't the fastest option in our selection of processors. However, it does give us a taste of what's to come compared to Atom D2700 (with discrete graphics) and AMD's A4-4000.

Averaging power consumption across the entire run distills the line graph into something simpler. We get confirmation that the Celeron J1750 uses very little power. The entire platform, including storage and memory, averages less than 20 W. Zotac's Atom D2700-based platform is rated for the same 10 W, yet averages notably higher consumption. This is because it's complemented by a GeForce GT 520 discrete GPU, though. Can't blame this one on the host processor. And as a result, we can't generalize about Bay Trail's efficiency versus Cedarview.
The Ivy Bridge-based Celeron G1610 steps average draw up an additional 10 W and is followed up by the 32 nm A4-4000 almost 9 W higher than that.
What this figure fails to tell us is how efficiently each platform handles our benchmark suite. For that, we multiply the average power by the time it takes to finish the benchmarks, which are all the performance-sensitive, faster-is-better type.

Again, Intel is doing amazing things with 22 nm manufacturing and its latest architectures. Celeron J1750, based on Silvermont, is more efficient than even the Ivy Bridge-based Celeron G1610, which smoked through our benchmark suite, but used more power to do so.
The Atom D2700, dragged down on the power side by a GeForce GT 520 graphics chip that doesn't encounter much 3D work in the pared-back list of tests we're using, fares poorly. It still turns out .06 Wh more efficient than the A4-4000, however.
So, even before we break into the performance numbers, we have to be excited about Bay Trail's prospects in the tablet space, where we can be relatively sure that this is a faster, more efficient design than what came before. Remember also that we're looking at the Celeron J1750, the lowest-end desktop model. Intel has two other 10 W SKUs with four cores each that'll cut through threaded applications with more alacrity than the dual-core SoC on our test bench.
As mentioned, we're scaling back on the number of tests for today's preview. Much of our typical benchmark suite is workstation-oriented. Compiling lots of code, editing professional video, and rendering big 3D projects probably won't happen very often on a system with a Celeron inside. When it does, you're going to wish you had picked up something else.

Let's start with the basic synthetics, though. Sandra's Arithmetic module recognizes each processor's ISA extensions and capitalizes. Still though, the Ivy Bridge-based Celeron and Richland-based A4 are both quite a bit quicker.

The Atom D2700-based setup wouldn't complete the Multi-Media sub-test. Instead, we get a look at the advantage Ivy Bridge and Richland enjoy over the Bay Trail SoC.

The Silvermont processor architecture includes AES-NI support. Intel deliberately cuts that feature from the Celeron G1610 though, explaining why Bay Trail posts better numbers. Cedarview clearly lacks AES-NI, while Richland leverages it. Curiously, Bay Trail is outperformed in hashing by the Hyper-Threaded Atom D2700.

These platforms support different memory configurations. The Celeron G1610 supports two channels of DDR3-1333. AMD doesn't make clear what the A4 supports, but we ran it with DDR3-1600 modules that defaulted down to 1333 MT/s. Our Atom D2700 system took two 2 GB DDR3-1066 modules (albeit using one 64-bit channel). Meanwhile, our Celeron J1750 is rated for DDR3L-1333 at 1.35 V.

3DMark

Bay Trail might appear to falter, but remember that it's going up against two CPUs derived from more powerful desktop architectures (even if they're both $50 chips), and an Atom D2700 backed by Nvidia's GeForce GT 520.
Most important, perhaps, is that our dual-core Celeron J1750 manages to outscore the dual-core, Hyper-Threaded Atom D2700 in 3DMark's Physics test, which measures host processor performance.
As a point of comparison, you can see how badly the Atom Z2760 in Intel's tablet-oriented Cloverview process fares.

The situation similarly doesn't look much different for Bay Trail in the older 3DMark 11. What drags it down is the graphics score, though. You can clearly see that two Silvermont cores outperform a pair of Hyper-Threaded Saltwell cores. This test doesn't run on the ATIV Smart PC 500T, so we can't include that platform's numbers, which would have undoubtedly been ugly.

FineReader is a well-threaded optical character recognition app that consistently rewards platforms with the highest core counts. In the Atom D2700's case, it stays pretty close to the Celeron J1750 because of its Hyper-Threading feature. The dual-core Celeron is still faster thanks to a higher clock rate and superior IPC. However, we'd be even more interested in the performance of a quad-core model.

In contrast, printing a PowerPoint document into a PDF only uses one core. Celeron J1750 isn't quite twice as fast as the quickest desktop-oriented Atom, but it comes pretty close.
Of course, if you're building in a form factor able to accommodate a 55 W CPU, the Ivy Bridge-based Celeron is a very attractive option. AMD uses Turbo Core technology and a higher 3 GHz base clock to get the most from its A4-4000, approaching the G1610. But the Richland architecture just doesn't have Ivy Bridge's per-clock performance.

Typically I'd throw Photoshop in with Adobe's other CS6 apps. However, those are far too taxing, and anyone buying a desktop with a 10 W SoC in it will (or at least should) know better than try editing video in Premiere. Very regularly, though, I need to pop into Photoshop to knock out images for our stories. The hardware I'm using at the time simply has to cope with this.
The CPU-oriented test, which employs four heavily threaded filters, yields the hierarchy we were expecting. Intel's Celeron G1610 finishes the workload in less than half of the time as Celeron J1750. In turn, though, our dual-core Bay Trail SoC completes its task more than two minutes faster than the Atom D2700.
Although it's not part of the Bay Trail tablet story, OpenCL support is part of the Celeron's 64-bit driver package, helping explain the almost 2x speed-up compared to Atom D2700 and its GeForce GT 520.

The Celeron J1750 and Atom D2700 finish within 15 seconds of each other in our 7-Zip workload. Given the application's optimizations for threading, it's possible that Hyper-Threading is keeping the Atom well-utilized, while the Celeron's two cores are bottlenecked by something else. Either way, the task takes almost 10 minutes on both platforms.

We've shown WinRAR to be far less friendly to heavily parallel architectures in the past. Sure enough, Celeron J1750 shows up between AMD's A4 and the Atom D2700.

We run three different WinZip workloads. Like WinRAR, we don't get the best threading from 17.0 (this was improved in later versions). However, we can still see the CPU-oriented run, in red, favors the Celeron J1750 over Atom D2700. The EZ iteration, which goes for best compression (and consequently the longest run times), extends that lead over Cedarview.
Here's the thing, though. Intel rates the Celeron J1750 for up to 2.4 GHz using Boost technology. That means it enjoys the highest clock rates when it's operating under a thermal limit. But if we run our WinZip test as part of a scripted sequence, we get the numbers displayed above. If we run the workload on its own, without heating the SoC up first, we get significantly higher scores.
We wouldn't have figured this out had it not been for the curiously-lower results in our OpenCL-based benchmark. Re-running the test with Celeron J1750 completely cool, it's possible to shave a bunch of time from the outcome. The implication is that taxing the SoC depletes headroom for higher Burst frequencies. However, that's still preferable to a more dramatic throttle state in response to a thermal trigger.

Browsermark includes five test groups: CSS, DOM, a general group that measures resize and page load times, a graphics group that evaluates WebGL and Canvas performance, and Javascript performance.
Intel's Celeron G1610 dominates, followed by the A4. Celeron J1750 finishes in third place, with the Atom D2700 placing fourth ahead of the Z2760.

JSBench tests JavaScript performance using a series of real-world webpages that get recorded and replayed. The behavior of a human interaction is recorded and scored. What we see here, then, is the performance of the JavaScript engine, and not the Chrome browser we used to test.
Both Intel's Celeron G1610 and AMD's A4-4000 fare similarly, while the Celeron J1750 trails quite a ways behind. The thing is, it has little trouble besting Intel's Atom D2700, which in turn stomps the Z2760.

Also a JavaScript-based performance benchmark, Peacekeeper 2.0 reflects the same significant gains enjoyed by Bay Trail over last generation's platform, while still making it clear that the true desktop-oriented architectures are notably quicker.

The WebXPRT suite employs HTML5, but yields performance results similar to what we just saw from Peacekeeper. In both cases, the Celeron J1750 posts numbers in between the Atom D2700 and A4-4000. From one generation to the next, those are great gains, particularly when you consider that both low-power SoCs are 10 W models.

HandBrake is optimized for multiple threads. But despite the fact that our Celeron J1750 sample is only a dual-core CPU, it has little trouble trouncing the Hyper-Threading-enabled Atom D2700. Then again, both upgradeable desktop-oriented processors are notably quicker and quite affordable. If it's performance you're looking for, they're smarter buys. The 10 W Celeron is going to be best suited to applications sensitive to power and thermals.

Our iTunes workload, on the other hand, is decidedly single-threaded. We get another look at how effective the Silvermont architecture is compared to Saltwell. Though it's certainly true that the Celeron J1750 spins up to higher clock rates than the Atom D2700, even at the same frequency, the Bay Trail-based SoC would enjoy a commanding lead.

LAME tells us the same basic story. In short, the Celeron J1750 is quite a bit quicker than the fastest desktop-oriented Atom from last generation. The Ivy Bridge and Richland architectures have no trouble posting notably better numbers, but remember that they're 55 and 65 W CPUs, respectively. Celeron J1750 is rated for just 10 W, and it's cruising along.

Going back the other way, TotalCode Studio is well-threaded, and the addition of Hyper-Threading does good things for Atom D2700. Intel's Bay Trail-based SoC is still quite a bit quicker, despite addressing fewer threads. We'd want to see what four physical cores could do in these applications; good scaling could land a Celeron J1850 or Pentium J2850 within striking distance of the A4-4000.
When Intel introduced us to its Silvermont architecture, the company made grand claims of increased performance at a given thermal ceiling, or similar performance at reduced power. Naturally, we couldn't wait to get our hands on one of the first implementations, but it was understood that the Bay Trail-based SoCs featuring this new design wouldn't be ready until the second half of 2013.
Intel showed off the Atom Z3000 series for tablets at IDF last week, putting real benchmark data behind those earlier claims. And while we're positively inquisitive about how Bay Trail might deliver a better experience in more taxing applications and simultaneously stretch battery life out across a day, we were busy testing a desktop-oriented version of the SoC. Whereas the Atom Z3000s are only running in 32-bit environments, our testing took place under Windows 8 64-bit.
Indeed, there's certainly something to be said for using one form factor to compare multiple products against each other. To that end, our power and efficiency numbers are what impressed me most. It's clear from the logging power over time and charting performance that the dual-core Celeron J1750 isn't as fast as the least-expensive Ivy Bridge-based Celeron you can find on Newegg. And maybe Celeron branding isn't even appropriate for a 10 W SoC in an entirely different league.
But we still see that the entry-level J1750 finishes our suite of tests significantly faster than Atom D2700. Yes, the Atom-based platform has an on-board GeForce GT 520 GPU that we cannot factor out of our power equation, which affects our ability to compare their efficiency directly. The discrete graphics chip also conveys a performance advantage in 3D titles too, though. Just remember that the Atom D2700 is a 10 W part, similar to Celeron D1750, and that GeForce GT 520 uses up to 29 W. Remove the GPU and you'd likely be looking at very similar power from last generation's Atom. Based on our performance data, Bay Trail still comes away with an indisputable blue ribbon for efficiency.
And all of this comes from two Silvermont cores. The Pentium J2850 gives you four of them, the same 2.4 GHz peak core clock, higher graphics engine frequencies, and again, a 10 W thermal ceiling. Naturally, we're expecting an even more compelling performance story in our threaded tests once the quad-core models start surfacing.
Until then, we walk away from our first experience with Intel's Bay Trail SoC impressed. The company wasn't exaggerating when it suggested that the Silvermont architecture could as much as double performance at a given power limit. Snappy little passively-cooled platforms are almost certainly on their way toward the end of the year. Intel tells us that its partners are already working on fully integrated desktops and all-in-one designs, as well as channel-oriented motherboards with soldered-down CPUs.

