Hardware Today: AMD Eats Into Intel's Server Chip Monopoly

by Drew Robb

In the two years since Opteron's launch, AMD has cut a large slice out of Intel's processor pie. What else is the chip vendor planning to serve?

What a difference two years can make. AMD chips were once firmly positioned as an inexpensive Intel desktop alternative and not ordinarily deployed in an enterprise environment. Sound R&D and excellent marketing is changing that perception.

As a result, Intel's server chip monopoly is looking a little fragile. In its nearly two years on the market, the AMD Opteron processor has gained more than 5 percent of the worldwide x86 server market according to Gartner Dataquest, and Opteron now claims more than 20 percent of the 4-way server market in North America.

"More than 40 percent of the top-100 global companies or their affiliates as ranked by the Forbes Global 2000 are now customers of AMD-processor-based systems," said Pat Patla, server/workstation marketing manager at AMD. "Customers running AMD systems include Deutsche Bank, ConocoPhillips, ChevronTexaco, Yahoo!, SingTel, Renault, and British Telecom."

AMD64 is the architecture on which AMD's eigth-generation processors (i.e., Opteron, Athlon 64, Athlon 64 FX, and Turion 64) are based. AMD64 was introduced when Opteron first launched in April 2003. Opteron can power x86 applications in both 32- and 64-bit environments. Its flavors span the 100 series (1-way), the 200 series (up to 2-way), and the 800 series (up to 8-way).

"Direct Connect gives AMD a substantial performance edge over Intel. Intel's very definitely late to this party and won't get there until around 2007." — Nathan Brookwood, Insight64.

The Opteron comes with AMD's Direct Connect Architecture, which consists of several facets that increase memory address space and eliminate the bottlenecks inherent in front-side-bus-based architectures. HyperTransport technology directly connects CPUs, making it possible to reach up to 24.0 GB per second peak bandwidth per processor.

In addition, integrating the memory controller onto the CPU changes the way the processor accesses the main memory. The result of this is increased bandwidth, reduced memory latencies, and increased processor performance. According to AMD, this enables Opteron to surpass most competitive processors that rely on faster speeds or bigger Level 2 and Level 3 caches.

"Direct Connect gives AMD a substantial performance edge over Intel," said Nathan Brookwood, an analyst with Insight64. "Intel's very definitely late to this party and won't get there until around 2007."

AMD did not invent the concept of a direct connect architecture, but it has effectively incorporated into its x64 designs what DEC, IBM, and Sun had done previously with their proprietary RISC processors.

Another feature of the Opteron is PowerNow, which manages power consumption.

AMD claims a huge cost savings for Opteron vs. Xeon in a data center. For example, it believes the Opteron processor would deliver a savings of $35,000 per year for a data center with 500 two-way systems, compared to a data center with 500 two-way systems based on Intel Xeon EM64T processors. This is based on energy costs — $250 to run and cool an Opteron for a year vs. $321 to run two Intel Xeon EM64T processors.

According to Brookwood, AMD can throttle an Opteron down to around 800 MHz, where it uses only about 20 watts. Intel's "Demand Based Switching," on the other hand, drops a 3-GHz-plus Xeon to 2 GHz and cuts its power consumption down to about 70 watts. Multiply that 50 watt savings by the number of CPUs in a rack, and it adds up quickly. The Opteron is also available in two low-power versions — the HE (highly efficient) 55-watt model and the EE (energy efficient) 30-watt model.

>> Up Against Intel

Intel vs. AMD

How does the Opteron compare to the Intel Xeon? It appears that, for now, the Opteron has the edge.

"Opteron is better in I/O intensive tasks and tasks that are sensitive to memory latency, and this advantage grows with the number of processors in the system," said Nathan Brookwood, an analyst with Insight64. "Xeon occasionally is better in tasks that benefit from very large caches."

But Intel continues to reign in other areas, namely the blade arena. AMD is making moves to close the gap, however. In the past two months, two major blade players have launched Opteron-based platforms. In February, Egenera announced new Opteron-based blades, and HP became the first tier-one OEM to offer Opteron blade servers (HP BL25 and HP BL35). Sun also announced its intention to offer Opteron blades in the coming months.

"Last month, we announced a new addition to our low-power AMD Opteron processor line-up that is ideally suited for the blade market, where lower power is needed to enable dense designs and meet end-user needs for lower power consumption," said Patla. "Each low-power AMD Opteron processor is designed to give the same performance as its full-power sibling with the same model number."

Dual Core

The biggest development on the horizon for AMD is the introduction of dual core processors by mid-2005. This is expected to significantly increase CPU performance — perhaps by a factor of 100 percent for some workloads — without the usual surge in power consumption.

CPU power consumption typically increases in direct relation to clock frequency. In practice, though, designers need extra power to speed up transistor performance to attain increased clock frequencies. Thus, a 20-percent boost in frequency might require as much as a 50-percent boost in power.

Dual-core designs reverse this, lowering the frequency of each core by 20 percent. The combined cores uses about the same amount of power as a single core at the higher frequency. This adds up to a performance boost of 1.7 over single-core designs using the same amount of power.

"AMD has a far more elegant approach to dual core than Intel's initial offerings in this area," said Brookwood. "It appears that Intel will win the dual-core desktop race, and AMD will win the dual-core server race."

The downside of dual core? Because dual-core chips run at lower clock frequencies than the fastest single core versions, clock-time performance will decline for single-threaded applications. Also, dual-core chips are double the size of their single-core equivalents, so they cost more to make. That said, rapid adoption of dual-core chips is expected

"Corporate IT systems currently optimized for symmetrical multiprocessing [SMP] multithreaded applications should see significant performance increases by using AMD multi-core processors, although performance will vary based on the application," said Patla. "Just as we brought 64-bit computing to the masses, we intend to the same with dual-core in 2005."

This article was originally published on Monday Apr 4th 2005
Mobile Site | Full Site