Four on the Floor

Just when we’re getting used to the idea of processors with two CPU cores, Intel ups the ante with the introduction of its first quad-core processors, the Xeon 5300 series processors (code named “Clovertown”) and the new Intel Core 2 Extreme QX6700 processors (code named “Kentsfield”). AMD is poised to follow suit later this year with the introduction of its own four-way processors, code named “Altair” and “Deerhound.”

 

The first quad-core-equipped workstation to arrive at Desktop Engineering is the HP xw8400. We previously reviewed this computer last summer (see “Xeon Takes Off” in the August 2006 issue of DE). That first system was equipped with a pair of Intel 5160 “Woodcrest” processors running at 3.0GHz, with 2GB of memory, a NVIDIA Quadro FX 3500 graphics accelerator, and two hard drives (including a 15,000rpm SAS drive).

The new quad-core version of the HP xw8400 workstation we received came with a pair of 2.66GHz Intel Xeon 5355 processors, 4GB of memory, and the same NVIDIA graphics board, although we opted this time for just a standard 160GB SATA hard drive, sacrificing a bit of I/O speed, and upgraded to a DVD +/-RW drive rather than a CD-RW/DVD-ROM combo drive.

Other than those differences, the two systems are nearly identical, housed in the same case, based on the same motherboard, and sporting the same ports on the front and rear panels. (Please refer to our earlier review for a complete description of the system details.)

Like Having Eight CPUs
Intel’s new Kentsfield and Clovertown processors are each really more like two dual-core Woodcrest processors combined into one chip, much like Intel’s original dual-core Pentium D “Smithfield,” which was two “Prescott” CPUs in a single package (see “The Dual-Core Promise” in the October 2005 issue of DE for our initial take on that first dual-core CPU). The Clovertown CPUs use Intel’s newer Core Micro-architecture, enabling Wide Dynamic Execution, Intelligent Power Capabilities, Advanced Smart Cache, and Advanced Digital Media Boost. All 5300 series CPUs are based on the same 65 nanometer (nm) fabrication process as the 5100 series. With two processors installed in our evaluation system, both Device Manager and Windows Task Manager indicated a total of eight CPUs.

 

The four cores in each 5300 series processor share an 8MB onboard L2 cache; up to 4MB of L2 cache can be allocated to one core. All run both 64-bit and 32-bit applications and operating systems, and use fully buffered DIMM modules enabling a maximum memory bandwidth of 21GB/s while using 667MHz modules. The downside is that each FB-DIMM module uses 8-10 watts — whereas DDR-2 modules use around 4 watts. Because the new memory modules run hotter than previous DIMMs, the xw8400 provides a second cooling fan directly over the memory slots. Our evaluation unit came with 4GB of memory, installed as eight 512MB DIMMs. The system also supports 1GB, 2GB, and 4GB DIMMs and will be able to accommodate up to 64GB of memory once the 8GB DIMMs become available.

 

   

With two quad-core processors installed, both Device Manager and Windows Task Manager indicated a total of eight CPUs. During our new multithreaded AutoCAD rendering benchmark, CPU usage reached 100% and all eight cores were active.

The reasoning behind the move to multiple cores is not hard to understand. For years, the way to get more performance out of a CPU was to increase its clock speed. But as the chips reached 3GHz and the number of transistors inside those chips continued to increase, the power requirements of the CPU and the heat generated made the chips less and less efficient. The way around this is to move to parallel processing.

By placing multiple processor cores inside a single chip, the computer can split its workload, getting more done at the same time without increasing the clock speed and power requirements, nor generating any more heat. At least, that’s the theory. In reality, multicore processors can’t deliver on their potential unless the applications being run are multithreaded — that is, unless those applications are designed to split their computational loads across multiple processor cores.

Some applications, such as FEA and CFD analysis, digital content creation, and some rendering applications, are multithreaded and show significant performance improvement when run on multicore processors. Unfortunately, most CAD applications are not significantly multithreaded. Neither are the benchmarks DE has used for years, which means they do not show any of the potential benefits of multicore processors. In fact, our usual benchmark results show the new quad-core-based xw8400 to be somewhat slower than the previous dual-core-based xw8400, due to the new system’s slower CPU speed (2.66GHz versus 3.0GHz). For example, the quad-core xw8400 took 221.57 seconds to complete the SolidWorks benchmark whereas its older dual-core sibling finished the test in 183.47 seconds, 17% faster.

 

For that reason, we had to look at some other measurements. Intel’s own tests show the 5300 series processors to be 50% faster than its own dual-core Xeon 5100 series and nearly twice as fast as AMD’s dual-core Opteron2220 CPU when rendering in 3ds max. Similar tests show the 5300 to be 175% faster running a Fluent CFD calculation and nearly 140% faster than the 5100.

To see for ourselves, we rendered a complex model in AutoCAD 2007, which uses mental ray, a multithreaded rendering engine. The same rendering that took 170 seconds to render on the dual-core xw8400 was completed in just 81 seconds on the quad-core xw8400, 70% faster. Interestingly, when we rendered the scene on the quad-core system, Windows Task Manager showed 100% CPU usage (indicating that the application is splitting its processes across all eight cores), whereas when we ran our SolidWorks Benchmark, CPU usage averaged around 14% and was confined almost entirely to one of the CPU cores (barely taxing the CPU and clearly not multithreaded). We also went back and ran this new test on several of the other systems we recently tested (the full set of benchmark results are provided here).

More Cores to Come
There are five members in the 5300 series, with CPU speeds ranging from 1.60GHz to 2.66GHz. All but the fastest processor consume 80 watts, with the 5355 using 120 watts. The 1.60GHz 5310 and 1.86GHz 5320 use a 1066MHz front-side bus, while the 2.0GHz 5335, 2.33GHz 5345, and 2.66GHz 5355 use the faster 1333MHz bus speed. The 5300 processors are available in the newest iterations of the dual-socketed HP xw6400 and xw8400 workstations.
The Kentsfield processor, now known as the Intel Core 2 Extreme QX6700, is available in the single CPU-based HP xw4400 workstation. But that 2.66GHz processor is priced the same as Intel’s earlier Core 2 Extreme chip, the 2.93GHz dual-core X6800.

In January, Intel launched several additional quad-core chips. The Core 2 Quad X6600 has a core frequency of 2.4GHz, placing it just slightly below the QX6700. The other two CPUs are the new quad-core Xeon X3200 series processors based on the same Kentsfield core: the X3210 at 2.13GHz and the X3220 at 2.4GHz. All three CPUs use a 1066MHz front side bus.

In the meantime, AMD currently offers a platform called Quad FX, but it actually uses two dual-core processors in separate sockets. Later this year, AMD will release its first true quad-core CPUs, Altair for desktops and Deerhound for workstations and servers. Both are based on AMD’s new 65nm “Barcelona” core architecture.

And don’t expect things to stop with quad-core. Intel will likely switch to a 45nm fabrication process when it introduces a new family of processors called “Penryn.” Within the Penryn family, the “Yorkfield” will be a 45nm quad-core CPU with two “Wolfdale” dual-core processors in a single package. This should be followed by “Tigerton,” expected to be a true quad-core CPU. Intel has already shown a prototype system with four Tigerton processors, for a total of 16 cores. And looking further forward, “Dunnington” is rumored to be either a four- or eight-core successor to Tigerton.

What Should You Buy?
That’s all fine, but if you’re in the market for a system right now, what should you buy? While prices for an xw8400 workstation start at $1,769 (including a single 1.60GHz dual-core Xeon 5110, 1GB of RAM, and a Quadro NVS 285 graphics board), a system with one 5160 dual-core CPU, 2GB of RAM, FX 3500 graphics board, and the DVD +/-RW optical drive would cost $4,217. Adding the second dual-core CPU and doubling the memory raises the price to $5,624. The same system with a single quad-core 5355 and 2GB of RAM would run $4,809, while a second quad-core 5355 processor and 4GB of RAM brings the total cost to $6,808.

If the applications you regularly run are already significantly multithreaded, more CPU cores will immediately equate to better performance. If, on the other hand, you’re running a current midrange CAD application, the faster dual-core CPU will probably perform better — for now. It appears clear that parallel processing is the future. It’s just not yet clear how fast we will all get there.

---

David Cohn is a computer consultant and technical writer based in Bellingham, WA, and has been benchmarking PCs since 1984. He’s a contributing editor to DE, an applications engineer with The PPI Group, the former editor-in-chief of Engineering Automation Report and CADCAMNet and the author of more than a dozen books. Please send comments about this article here. You can also contact David at [email protected] or visit his website at dscohn.com.