Cpu Reviews

Intel’s next-generation CPU arrives, ringing in the era of the integrated graphics core

In the Intel galaxy, the CPU is an inexorable black hole. A gravity well so strong that nothing can escape it as it consumes every function of the PC.

Don’t believe us? Witness add-in MPEG-2 decoders, hardware modems, hardware-accelerated soundcards, and Ethernet controllers, all of which have been swallowed by the all-powerful CPU. With Intel’s last CPU, the Lynnfield LGA1156 processor, the memory controller and even PCI-E functions were eaten by the CPU, too.

Now with Intel’s new Clarkdale (and its mobile equivalent, Arrandale) the company is taking the first step in trying to eat a gas-giant of functionality by moving a GPU core directly inside of the CPU.

But not only is Clarkdale the first Intel chip with graphics, it’s also our first glimpse at a CPU using Intel’s new, smaller-process technology. Current Core i7 and Core i5 CPUs are based on the original 45nm Nehalem design that Intel introduced more than a year ago. Clarkdale uses a newer 32nm process that is part of the Westmere family. For the most part, Westmere is an evolutionary step forward and a simple die-shrink of Nehalem, but Intel did add some interesting performance enhancements.

Read on for details about what makes Clarkdale unique.

*Graphics core runs at 900MHz

Hey, You Got Graphics in My Processor 

Until now, PC graphics have either resided in the PCI-E slot or in the motherboard’s core-logic chipset. With Clarkdale, Intel moves the GPU core directly into the CPU socket. It does this by packaging a new 45nm GPU core alongside the 32nm compute core, connecting the two via a high-speed QPI. It’s a method reminiscent of the company’s first quad-core proc, the Core 2 Extreme QX6700. Back then, Intel took a shortcut to quad-core land by combining two 65nm dual-core Core 2 Duo dies to make a “quad-core.” While chip purists scoffed that the multichip package was an inelegant hack, and AMD fanboys called it cheating, the move gave Intel a year-and-a-half lead over AMD to store shelves. (Interestingly, a parallel scenario exists today: AMD is working on its own integration of GPU and CPU, dubbed Fusion. As before, AMD’s plan is far more ambitious and elegant in its integration of GPU and CPU functionality. That product won’t see the light of day until 2011. See more on AMD’s Fusion efforts below.)

Clarkdale’s setup puts most of the logic in the GPU, which has a built-in single x16 PCI-E 2.0 controller, as well as the memory controller for both the graphics and compute core. Why use a multichip package instead of building a 32nm chip with graphics in it? It’s likely a matter of cost, technology, and timing. This move, again, gets Intel a CPU with graphics capability more than a year before AMD will deliver its version.

Got Speed?

You probably only care about one thing: How fast is the GPU inside the chip? By rough estimates, it’s about 1.5x times faster than the graphics in a current Intel G45 chipset found in most laptops and mainstream motherboards. If that sounds great, remember that Intel’s integrated graphics history hasn’t been stellar. Put plainly, Intel’s integrated graphics have stunk up the joint for years and it’s probably an insult to graphics cards to actually call Intel’s integrated parts graphics accelerators. A 3-year-old $65 discrete graphics card with a hairball jammed in the fan is slightly faster than what you get from the G45 chipset. In fact, we’ve long blamed Intel’s subpar integrated graphics for helping to push mainstream gamers to console gaming.

Intel’s reasoning is that if people are buying systems with integrated graphics, they probably don’t care about graphics. Sadly, that’s probably true. Mainstream consumers browse the web, use Microsoft Works, and don’t play anything more graphically intensive than Yahoo Bingo before heading down to the social hall for a game of bridge with Madge, Maude, and Betty.

Intel bluntly says the graphics core in Clarkdale is definitely not meant for hardcore gamers. We wholeheartedly agree. We first tested the Clarkdale using 3DMark Vantage on default and after getting a score of 0, abandoned all hope of it being capable of serious gaming.

To see if it was even capable of playing more moderate games, we fired up Left 4 Dead 2 and found the frame rate almost playable at 800×600 with the graphics set on maximum ugly. Borderlands at 1280×1024 was also over the Clarkdale’s head, but almost playable at a mobile phone resolution of 800×600. We did actually see 60fps in Counter Strike: Source at 1280×1024, with somewhat compromised graphics. Still, that’s better than nothing. As easy as it is to make fun of integrated graphics, it’s a moot point for someone who doesn’t play games.

Makes Good with Media

While Clarkdale may not shine in gaming, it certainly holds its own at media acceleration. Intel paid attention to the shortcomings of the Core 2’s G45 chipset. When released, the G45 accelerated Blu-ray content but it was short on features. Clarkdale adds a new sharpness filter, 24Hz refresh rates, HDMI 1.3a with Deep Color support, lossless Dolby TrueHD and DTS HD audio, and even dual HDMI output. With a 3.33GHz Core i5-661 Clarkdale, we were able to watch a Blu-ray disc with the processor running in its SpeedStep low-power mode.
 
By moving the GPU into the CPU, beefier centralized cooling can be used to keep both parts cool. (Unified cooling has even greater ramifications for notebooks with Arrandale). The relocation of graphics and PCI-E in the core also reduces the core-logic chipset from a north- and south-bridge design to a single chip. This lets board vendors design more compact boards with greater capability than traditional chipset-based graphics. If you don’t care about games, Clarkdale could let you build an extremely small yet Blu-ray-capable HTPC that can outperform many budget quad-cores.

One thing home builders will need to remember, though: Clarkdale is a unique part that could make shopping for a motherboard confusing. Today, pretty much any motherboard with integrated graphics will work with any CPU that fits into the LGA775 or AM3 socket. With Clarkdale, not only will you have to pair it with an LGA1156 board, but you’ll also need to make sure it supports the graphics capabilities of the CPU. And if you decide to, say, replace the Clarkdale with a quad-core Core i7 in a year, you will also have to install a graphics card to the system because you’ve just yanked out the GPU with the old CPU. You should be able to run Clarkdale in motherboards that don’t have graphics ports on back but the GPU will be disabled.

Clarkdale as a CPU

But enough about the GPU. The real gem here is the 32nm Westmere core in Clarkdale. Why? It basically gives us a preview of the performance we’ll see next year when Intel releases quad-core and hexa-core CPUs based on the Westmere core. So far, we like what we see. For starters, the CPU offers six new instructions to accelerate Advanced Encryption Standard encryption and decryption. In addition to that, Westmere retains Intel’s auto-overclocking mode, now called Turbo Boost. Originally dubbed Turbo Mode when Core i7 was first released, the feature has been fine-tuned to its current Turbo Boost iteration (yes, we know, Knight Rider’s KITT had a Turbo Boost, too).

One thing we’re not too impressed by is the CPU’s memory performance. In Clarkdale, the dual-channel DDR3 memory controller resides in the GPU side, and it’s apparently not up to the snuff of the memory controllers in 45nm Core i7 and Core i5 parts. With our Clarkdale sample CPU, memory bandwidth was about 25 to 30 percent lower than with a 45nm part. Latency was also significantly worse at 82ns in the 32nm Clarkdale versus 53ns in a 45nm Core i7-870. Why? We suspect it’s the result of having one memory controller manage both graphics and compute cores, but it’s not really clear to us yet.

Still, it’s easy to imagine what you can get if you put six of the Westmere cores along with a better triple-channel DDR3 memory controller and a high-thermal budget in the Core i9 next year.

Under the Clarkdale Heat Spreader

The transistor count and die size of a CPU have long been a fascination of chip addicts trying to glean insights about a processor’s capabilities. With Clarkdale, the simple question of how many transistors it packs gets quite complicated since there is now a GPU under the heat spreader, too. We dug into the spec sheets of the new processor and found out that the new 32nm core measures a very diminutive 81mm2 but packs 383 million transistors. The 45nm GPU is 114mm2 yet has just 177 million transistors in it. So the short answer is 560 million transistors.

 

AMD’s Take on the Hybrid Processor

AMD hasn’t been in the CPU hunt for several years, but that could change in 2011, when the company’s combination CPU and GPU is released. AMD calls its hybrid part an APU, or accelerated processing unit, and it looks to be a far more elegant approach than Intel’s method of jamming a graphics core and a compute core into the same CPU package. AMD’s Fusion platform is a true integration of the functions of a CPU and GPU. And we don’t mean simply because both are built on the same contiguous die. AMD’s vision is to closely enmesh the strengths of the GPU at running parallel code with the strengths of the CPU for general-purpose code. The first part is code-named Llano and will feature 1 billion transistors—roughly twice the number of transistors of Intel’s Clarkdale—and support for DirectCompute and OpenCL, which will let applications leverage the parallel portions of the APU for such tasks as encoding. Compared to Clarkdale’s GPU-in-a-CPU trick, Fusion appears to be far more forward-thinking. However, recall that AMD’s Phenom also seemed elegant and advanced when compared to Intel’s clunky Core 2 Quad. Though cruder, the Core 2 Quad was still faster, which is all that will matter in 2011 when Fusion hits the shelves.

Clarkdale in Action

After running the benchmarks, we declare it the fastest dual-core ever!

For our testing, we used an Intel DH55TC motherboard, 4GB of Corsair DDR3/1333, a Core i5-661, a Western Digital Raptor 150, and 64-bit Windows Vista Home Premium. Our benchmarks consisted of a standard suite of 3D rendering, encoding, photo editing, gaming, and memory-bandwidth and -latency benchmarks.

To be frank, we didn’t expect much from the Core i5-661. After all, a dual-core CPU in a quad-core world is asking for a beat-down, right? We’ve seen overclocked Core 2 Duo’s get spanked or barely break even with far lower-clocked quad-cores, so we didn’t think this was much of a match. Well, as Gomer says, “Surprise, surprise, surprise!”

The 3.33GHz Core i5-661 is actually faster than AMD’s budget quad-core, the $99 2.6GHz Athlon II X4 620, as well as its own sibling, the 2.33GHz Core 2 Quad Q8200. Against both chips, the Core i5-661 plowed ahead in the multithreaded tests thanks to its Hyper-Threading, and was significantly faster in gaming thanks to its Turbo Boost. So mark it an eight for dual-cores, dude.

The battle wasn’t so easy once we compared Clarkdale with CPUs in its own price range. At roughly $200, the Core i5-661’s real competition is against the $266 2.83GHz Core 2 Quad Q9550, the $200 2.66GHz Core i5-750, and the $200 3.2GHz Phenom II X4 965 Black Edition. Surprisingly, the 2.83GHz Core 2 Quad didn’t surpass Clarkdale in everything. While the Core 2 Quad was faster in multithreaded tasks, the higher clocks of the Core i5-661 gives it the edge in gaming. So if you’re still not convinced that Core 2 is dead, this should give you another sign.

Against the Core i5-750 and the Phenom II X4 965 BE, the dual-core Core i5-661 is clearly outclassed. Since all three CPUs are $200, we had to wonder if Intel didn’t fire a blank when pricing the Core i5-661. Sure, it’s a good chip for folks looking to build an ultra-quiet and ultra-small home theater PC, but it simply can’t run with the other $200 chips. The chip should really be priced about $30 to $40 cheaper.

We were far more interested in seeing how the lower-clocked Core i5 and Turbo Boost–denied Core i3 Clarkdales would do, but that wasn’t possible. Our Intel board didn’t allow us to underclock our sample processor and none of our P55 boards had BIOSes that support the new chip yet.

The upshot is that Clarkdale is the fastest dual-core today—and competitive with quad-cores, too. That’s damned impressive, and a testament to the power of the new Westermere core. Based on this glimpse of next year’s six-core Core i9, we can tell that it’s going to be a monster.

The problem is the pricing on the Core i5 dual-cores. With more competitive quad-cores priced the same, the Core i5-661’s only advantage is in HTPC or small formfactor designs.

Best scores are bolded.