Nehalem Vs. Shanghai

Damon Poeter

Intel and Advanced Micro Devices are finally back to where it all started when the personal computer boom kicked off in the early 1980s – squaring off with chips that bear more similarities than differences.

That sets the stage for the kind of competition between the two main x86 microprocessor makers that has system builders such as Brian Corn energised for a lively 2009.

“We’re really going to be able to finally have an ‘apples-to-apples’ comparison on performance,” said Corn, vice president of marketing at Source Code, discussing how Intel’s new “Nehalem” processors will stack up against AMD’s recently launched “Shanghai” chips.

“The next statistic to look at is who has the best power consumption. If you can have both the performance and power consumption lead, that’ll be the crown jewel. AMD’s new 45-nanometer chips look great out of the gate, but Intel’s holding back on all its information about Nehalem on the server side, so they might have an ace up their sleeve.”

Corn first voiced his excitement about the upcoming battle between Intel and AMD more than a year ago, when he pinpointed the current convergence of the two chip maker’s product road maps as the beginning of an epic battle for CPU supremacy.

That’s because the two companies took divergent architectural paths over the past decade, but in many ways return to the same general thoroughfare with the coming ramp of Intel’s new Nehalem chips and AMD’s 45nm Shanghai transition.

Intel’s Nehalem microarchitecture, first represented in three Core i7 desktop processors released in mid-November, eliminates the Front Side Bus, integrates the memory controller on the die itself and independently powers each of the processor cores in these multicore chips for a “native” multicore design similar to the one that AMD embraced several years ago.

In transitioning to the 45nm fabrication process Intel pioneered a year ago, AMD will match its larger rival on that technology node until Intel makes its next major transition to 32nm in late 2009.

The first products to emerge in this renewed battle aren’t exactly alike.

Intel’s first publicly available Nehalem products are desktop chips, including the Core i7-965 Extreme Edition, a blazing-fast 3.2-GHz, quad-core processor with 8 MB of L3 cache that is priced at US$999, or about US$500 less than the top Intel quad-core built on the older Core 2 architecture.

AMD, on the other hand, introduced its 45nm processor a few weeks ago with nine new Opteron server processors that feature higher clock speeds than the previous Barcelona generation for equivalent prices.

But as Intel and AMD build out Nehalem and Shanghai across more product lines, that “apples-to-apples” scenario Corn anticipates will really start to materialise.

Intel already has Nehalem server chip samples in the channel, with parts for two-socket installations planned for release early 2009 and the full array due out over the course of 2009.

AMD recently confirmed that it will release 45nm quad-core desktop chips to be branded as Phenom II beginning in early January. The top Phenom II part will be listed at 3.0 GHz, but AMD overclockers have actually taken the chip north of 6.0 GHz using liquid nitrogen – similar scalability to what’s been shown in Intel’s Core i7.

Enter Nehalem
Intel officially lifted the curtains on the Core i7 on 17 November at three different global locations, kicking off with a “Midnight Madness”-style block party in Tokyo.

Events in San Francisco and New York rounded out the slate.

System builders, software developers and other guests were floored by the new processors’ raw power at the San Francisco launch party, held in that city’s new hipster mecca, Dogpatch Studios.

“It’s a lot faster than I’m used to,” said one 3-D animator, referring to a custom-built Core i7-965 Extreme Edition desktop built by Puget Systems.

The first Nehalem lineup also includes the Core i7-940 (2.93 GHz, 8-MB L3 Cache) and the Core i7-920 (2.66 GHz, 8-MB L3 Cache).

CRN’s Test Center, reviewing the Core i-965 Extreme Edition on Intel’s DX58SO Extreme Series “SmackOver” motherboard, described “nearly historic levels of improvement over previous generations of processors.”

Intel’s move to DDR3 memory on upcoming Nehalem server chips that will carry forward the chip giant’s successful Xeon brand is a big leap forward, Source Code’s Corn said.

But as of late October, he was still concerned about the availability of memory to meet the new processors’ demands.

“They’re integrating the memory controller and with the QuickPath interface they’ll compete right up against [AMD’s] HyperTransport. It’s triple-channel DDR3 memory, so you start running three DIMMs at a time and we should see some spectacular memory performance,” Corn said.

Others in the channel voiced concerns about the immediate availability of memory, graphics and other components to build around Nehalem, but Intel’s Steve Dallman quelled those fears about a week ahead of the Core i7 launch.

“I’m really glad you didn’t ask me six months ago,” said Dallman, Intel’s worldwide reseller channel chief, when queried directly in early November about component availability in the first weeks after the Core i7 launch.

“But today I’m feeling really good. Six months ago, we were really worried about chassis availability and who would get certification on the x58 chipset.

“And look, memory is out there and available,” he added.

“Luckily, we put DDR3 memory on our earlier Extreme products. Right now, we don’t have any major gaps in components.”

With regards to the new Core i7 desktop chips, system builders say that sounds about right because DDR3 on the client side is not new.

But questions still remain about memory availability on the server side, where Intel is moving to DDR3 while AMD has chosen to wait until about 2010 for that transition, according to the smaller chip maker.

It may also take some time for software developers to fully embrace Intel’s new triple-channel memory architecture, said Philip Pokorny, chief architect at Penguin Computing in San Francisco.

“Writing for three memory channels is tough for a software engineer,” Pokorny said.

“My guess is we’ll see customers and applications vendors needing to take time to optimise for a new platform and a memory allocation system on three channels.”

Destination: Shanghai
On 13 November AMD unveiled the first nine Shanghai processors for two-, four- and eight-socket x86 servers in its Opteron stable.

These include five Opteron 2000 series chips ranging in clock speed from 2.3 GHz to 2.7 GHz and four Opteron 8000 series chips ranging from 2.4 GHz to
2.7 GHz.

All nine quad-core chips are socket-compatible with the previous Barcelona generation and slot into the 75W thermal envelope.

AMD is also set to introduce 55W and 105W versions of 45nm Opteron processors in the first-quarter time frame.

Joe Toste, vice president of marketing at Equus Computing, feels Intel and AMD have directly opposing problems of perception going forward with their new chips.

Intel’s Nehalem drops into a new socket and requires a platform refresh, which is why questions have been raised about whether the rest of the hardware ecosystem is going to be ready out of the gate.

But there are no doubts, Toste said, that Intel will be able to deliver its own new products quickly and in volume.

AMD’s Shanghai, on the other hand, is socket-compatible with the previous Opteron generation code-named Barcelona.

That means the ecosystem is already in place for AMD’s delivery of its first 45nm chips.

But as Toste points out, the chip maker still suffers from the bad feelings left in the channel by the sluggish ramp of Barcelona and a glitch on that chip that caused volume distribution to be delayed for about five months after the release date in September 2007.

AMD will simply have to prove it is able to deliver its new parts, say market watchers. But the first reviews of Shanghai have been glowing.

CRN’s Test Center found that a pair of the new 2.7-GHz Opteron 2384 processors outperformed a pair of Intel’s Xeon E8450 chips while consuming less power.

And some system integrators, even though wary of AMD after a rough couple of years in its relationship with the channel, sound ready to give the chip maker
a second chance.

“I like what they are doing with Opteron in terms of power consumption and virtualisation. I have not been a big AMD fan, but I think I’ll have to look at them again,” said Mitch Miller, president of Dynamic Computer Solutions.

Some of AMD’s biggest advocates are builders of high-performance compute clusters, such as John Lee, vice president of Advanced Technology Solutions at Appro International.

Lee believes Shanghai solidifies the chip maker’s position in the four-socket server space.

“The advantages of four-socket, utilising the Direct Connect architecture, are huge. With HyperTransport, it makes migration very easy and affordable. Performance scales linearly,” Lee said.

“We’re very excited to launch general-purpose servers, commodity clusters and supercomputers on Shanghai.”

Rivalry and co-operation
Even as Intel and AMD go head-to-head, the graphics side of AMD’s house actually stands to gain from its longtime rival’s success with Nehalem.

AMD acquired graphics chip maker ATI more than two years ago, but it still takes some getting used to when an AMD spokesperson such as Rick Bergman openly praises Intel’s Core i7, as he did in a recent chat with CRN.

Bergman, general manager of AMD’s Graphics Product Group, has reason to hope for steady adoption of Intel’s new desktop chips, particularly the Core i7-965, which is already being called the best desktop processor on the market by CRN’s Test Center, among other reviewers.

That’s because AMD’s discrete graphics products stand to benefit from Core i7 sales right out of the gate for a couple of reasons.

For one, AMD appears to have wrested the GPU performance crown from market leader Nvidia, with its ATI Radeon HD 4000 series of graphics cards, particularly the Radeon HD 4870 X2.

Another is that AMD was quicker to validate its CrossFire technology for the new Core i7 platforms than Nvidia was to ready its SLI, as confirmed at the San Francisco Core i7 launch by Pat Gelsinger, general manager of Intel’s Digital Enterprise Group.

The top two parts in the initial Core i7 lineup are tailor-made for enthusiast rigs, and the “mainstream” 2.66-GHz chip is a monster in its own right, meaning discrete graphics are going to be added to most systems featuring Intel’s newest desktop processors.

And right now, anecdotal evidence suggests AMD is winning the battle to get its cards into Core i7 systems.

Don’t count Nvidia out for long, though.

CEO Jen-Hsun Huang, while admitting recently that the performance of the top ATI cards had taken Nvidia by surprise, promised his company would have an answer early next year.

And don’t count on some synergy between Intel and AMD over high-end graphics to be the dominant story for these two bitter rivals.

Especially worth watching will be the introduction of DDR3 on the upcoming Nehalem server chips, a crucial area where Nehalem and Shanghai will be quite distinct.

The upside for the channel is that in these tight economic times, a clear-cut winner on value between Intel and AMD could trump brand loyalty, according to system integrator Cameron Janzen.

“People are way more open-minded to other technologies when it means a better bottom line for them,” said Janzen, vice president of marketing at Pro-Data.

May the best chip win.