Business Tech

Nvidia is readying silicon that would work with Intel's newest processor design, according to a report. Intel claims Nvidia does not have the legal rights to make companion chips for its newest processors.

In February, Intel alleged in a lawsuit that the 4-year-old chipset license agreement with Nvidia does not extend to Intel's future-generation processors with "integrated" memory controllers, such as its "Nehalem" Core i series of processors.

A chipset is companion silicon to the main processor. Integrated memory controllers are built into the processor itself to increase performance between the processor and memory.

According to a report on Chinese-language Technology Web site HKEPC, Nvidia is planning to bring out a MCP99 chipset that supports Nehalem processors and Intel's Direct Media Interface, or DMI. Nvidia cited DMI back in February as a technology that Intel was trying to prevent it from using.

Despite the report's claims, it is not clear yet whether Nvidia would in fact bring out a chip before the legal matter is settled with Intel.

Nvidia had no comment on the report.

Nvidia's current 9400M Intel-compatible chipset, which is used with Core 2 architecture-based processors, has been successful. It is used in Apple's MacBook and Toshiba's Qosmio lines, for example, and in Netbooks that use the Atom processor.

During Nvidia's July 26 earnings conference call, in response to an analyst's question about building chipsets for Intel's Nehalem processors, Nvidia CEO Jen-Hsun Huang said: "We're not necessarily building chipsets for future Intel buses. We've not commented anything on that and so you are just going to have to wait to see what we come up with," he said. "Our company is...pretty darn clever. There is a lot of ways to skin the cat," he said.

Intel said the matter is being left to the courts. "We tried many times to resolve the conflict but we couldn't resolve it. So we asked the courts to," said Intel spokesman Chuck Mulloy. "By the time something comes to market, hopefully we'll have some resolution," Mulloy said.

I've been getting a fair number of questions about multi-threading the past couple of weeks. The reason is that Intel has been previewing its "Nehalem EX" Xeon processor in advance of Advanced Micro Device's six-core "Istanbul" CPU launch. Intel's Nehalem generation has simultaneous multi-threading (SMT)--which Intel calls Hyper-Threading (HT)--while Istanbul does not.

I wrote about this topic in depth a couple of years back in "Gradations of Threading," but it's worth reviewing in the context of these new server processors.

First, a little terminology.

A thread is a sequence of instructions that can execute in parallel with other threads. The details of what exactly constitutes a thread and the relationship between threads and other structures such as processes vary by operating system. However, for our purposes here, think of a thread as an independent task.

Simultaneous multi-threading.

(Credit: Illuminata)

A core is, in most respects, a complete processor that includes all the hardware such as execution units, registers, and so forth required to execute a sequence of instructions. Although multiple cores on a single die or in a single package (i.e. a chip or socket) may share certain resources such as cache memories, logically each core is a full central processing unit (CPU). That multiple cores are packaged together today is essentially an implementation detail that relates to getting the best performance out of the most economically sized silicon die.

Absent multithreading, each core can execute one thread at a time, running that thread until it has completed or until the operating system scheduler swaps it out for another thread.

SMT changes that 1:1 relationship. On a processor with SMT, more than one thread can execute on a single core at the same time--in the case of HT, it's two threads per core.

SMT potentially allows a processor to be more efficiently utilized. The reason is that modern microprocessors have multiple execution units within each core. For example, they have separate logic to handle integer operations and floating-point operations. Thus, in principle, if a thread with mostly integer operations runs concurrently with a thread that mostly crunches floating-point numbers, we could keep the processor busier by running both threads at the same time than we could running them sequentially.

The other main benefit is to hide memory latency. CPUs have to operate on data and that data has to ultimately come from memory or disk. Computer designs incorporate all sorts of techniques--such as caches and prefetching--to keep data close to processors in time and space. Nonetheless, processors still spend a lot of time waiting for data to arrive from relatively pokey memory. SMT lets a CPU quickly switch away from a thread that's sitting idle waiting for associated data to arrive.

SMT is therefore essentially a technique to use a processor more efficiently. It does not itself add execution resources to a core. And, in fact, the duplicated hardware and other logic that SMT requires to function (such as registers) takes space away from implementing other features (such as larger caches) that could themselves provide alternative ways to boost chip performance.

Intel's HT implementation--a fairly "lightweight" approach relative to IBM's on its Power processor--uses on the order of 5 percent of the total chip area to deliver typical performance gains of between 10 and 20 percent. (Optimized applications can see bigger gains. On the other hand, applications that are already efficiently using the CPU's execution units--or that are bottlenecked in ways that SMT can't assist with--may see no gain at all.)

Ultimately SMT is just one performance feature among many that may or may not be a match for a given processor's design. In Intel's case, it's been in some x86 designs but not others since it debuted on the Pentium 4; Itanium uses a simpler Temporal multi-threading approach.

SMT's in the plus column of the features checklist. But what really matters is overall processor performance on relevant workloads and platform capabilities. SMT is one tool to get there.

Intel on Tuesday said it will ship a server chip that contains up to eight processing cores later this year, while IBM showed off a high-end server in the works that uses eight such chips, yielding 64 cores.

Intel's Nehalem-EX architecture supports up to eight processors and each processor can integrate up to eight cores.

(Credit: Intel)

Intel's Nehalem-EX processor, in production later this year and expected to be shipping in high-end server systems by early 2010, will feature up to eight cores inside a single chip that supports 16 threads, according to Boyd Davis, Intel's general manager of the Server Platforms Marketing Group, speaking at a teleconference on Tuesday.

Using threads, Intel essentially doubles the amount of work that can be done on each processing core.

IBM, which participated in the conference, discussed a server currently under development that uses 64 Nehalem-EX cores (eight processors) and can handle 128 threads, according to Alex Yost, vice president IBM BladeCenter. "We're very excited today to be the first to demonstrate Nehalem-EX," Yost said.

Nehalem-EX will also double the memory capacity with up to 16 memory slots per processor socket, and offer four high-bandwidth "QuickPath" Interconnect links.

Intel also said the currently-shipping Nehalem server chip is making market gains. Intel's currently-available Xeon 5500, the first server processor based on Intel's Nehalem architecture, will be "greater than half of shipments" for Intel's high-volume two-processor (aka, "two-socket") server shipments by August, according to Davis.

"Customer acceptance has been quite strong," Boyd said. "From an introduction at the very end of March to representing the majority of our shipments in the market for two-processor servers by the August time frame," he said.

Intel showed off a prototype server that can accommodate four eight-core Nehalem-EX processors.

(Credit: Stephen Shankland/CNET)

Intel's prototype Nehalem-EX server accommodates eight of these memory cards. They'll use relatively conventional DDR3 memory rather than the FB-DIMM technology Intel's current Xeon 7300 systems.

(Credit: Stephen Shankland/CNET)

Updated at 10:00 a.m. PDT with correction about launch of Nehalem-EX processor.

Intel is expected to announce details of an 8-core processor for the high-end server market next week.

The chip itself will not actually ship in systems until late 2009 or early 2010.

The 8-core "Nehalem-EX" Xeon processor is designed for servers that can use more than two processors (referred to as "sockets" in server argot). Currently, Intel is shipping Nehalem Xeon processors for servers with two sockets.

Nehalem is the same architecture used in Intel's Core i7 desktop processor line.

The Nehalem-EX is expected to become Intel's top-line Xeon processor. Currently, the six-core "Dunnington" processor, based on Intel's older Penryn architecture, is Intel's highest-performance chip for multi-processor servers.

Nehalem-EX packs 2.3 billion transistors and its eight cores are capable of executing 16 threads (or tasks) at the same time. The chip also has "integrated power gates" for lowering power-consumption.

The announcement of details is slated for May 26. Boyd Davis, Intel's general manager of Server Platforms Marketing Group, will host the roll-out event and "discuss how this new server product raises the standard in cost-effective RISC replacement solutions," according to an Intel note about the event.

Updated at 9:25 p.m. PDT: correcting for expected Clarksfield and Lynnfield availability.

Intel has been talking a lot about Westmere chips lately. So, here's a quick look at Intel's first chips based on 32-nanometer technology.

Chief Executive Paul Otellini addressed Westmere during the company's first-quarter earnings conference call this week, saying the Westmere chip design will ship later this year, earlier than expected. "We have shipped thousands of Westmere samples to over 30 customers already," Otellini said in the conference call.

Intel's current lineup is made up of processors based on 45-nanometer technology. Generally, the smaller the geometries, the faster and more power efficient the chip. The move to 32-nanometer will put Intel ahead of rival Advanced Micro Devices, which isn't expected to transition to 32-nanometer chips until late in 2010.

The first installment of the Westmere family, the Clarkdale and Arrandale processors, is expected later this year, according to published Intel documentation. Clarkdale is a 32-nanometer desktop processor with built-in graphics--what Intel describes as a "multi-chip package with graphics integrated in (the) processor." Arrandale is a version--also with integrated graphics--for the mobile market, due later this year.

Intel Nehalem/Westmere chip lineup

(Credit: Intel)

In 2010, a processor code-named Gulftown (see graphic) is slated to appear and will be Intel's first six-core desktop processor. The Westmere chip will plug into existing Intel motherboard designs.

Westmere will support Intel's Hyperthreading technology, which doubles the number of tasks that can be handled simultaneously.

In related news, Intel's Nehalem mobile "Clarksfield" (don't confuse with Clarkdale) and Nehalem desktop "Lynnfield" processors will come in the second half of the year, according to Intel.

Chinese-language technology Web site HKEPC says Clarksfield quad-core processors will have speeds of 1.6GHz, 1.73GHz, and 2.0GHz and range in price from $364 to $1,054.

Both Annandale (32-nanometer) and Clarksfield (45-nanometer) chips are targeted at the "thin-and-light" laptop market, according to Intel documentation.

New servers and workstation from Dell.

(Credit: Dell)

Dell is attempting to gain some ground in the server market, which is dominated by Hewlett-Packard and IBM. In order to grab a larger piece of the data center pie, the company has shoveled everything together into one announcement. And it's a big one.

There are 14 new products altogether: new Dell M-series blade servers, 11th-generation PowerEdge servers, Precision workstations, EqualLogic PS6000 storage arrays, and a host of revamped services.

The fact that they have joined competitors in taking on the needs of enterprise IT departments holistically--instead of piecemeal--marks good progress, according to Frank Gillett, principal analyst at Forrester Research.

"They're getting much more competent at engineering things together; they're not thinking about storage and servers by themselves anymore," he said.

It's all part of the "new Dell," according to Brad Anderson, Dell's senior vice president in charge of enterprise products. Five years ago, Dell was primarily concerned with cost-efficient supply chains. Now, he says, that same company sees "cost isn't just hardware and services, but the personnel around it."

To that end, Dell is stressing the services part of its new suite of products that are designed to save time and money. There's the decision to extend its ImageDirect offering to servers. A server can come with the image preconfigured, nullifying the need for a company's in-house IT people to install each image manually.

It's a service that Dell has previously offered for desktop and notebook PCs, but it's "an interesting twist" to offer it on servers, said Charles King, principal analyst at Pund-IT. It's something that will likely resonate with very large businesses.

"If you're buying a few dozen servers, it's not a big deal. If you're buying hundreds or thousands, that's where something like this can really make sense," said King.

Playing catchup
But the real question is whether any of this will help Dell move ahead of competition. There's a lot of ground to make up. According to IDC's most recent tally of the x86 server landscape, HP has just over 58 percent of the blade market, compared to IBM's 22 percent and Dell's 9.8 percent. Dell is slightly closer in its race in rack-mounted servers, with 27.1 percent of the market, just behind the leader, HP, who has 39.6 percent.

And just last week, networking giant Cisco Systems launched its own server push.

Anderson says there's still time to catch up. While Dell is starting to gain some ground in Europe and Asia, he said, there's much more to do. "It's a multiyear endeavor to catch up to HP in servers in Europe. It won't happen overnight."

In the meantime, Gillett of Forrester says that Dell is at least keeping pace with its competitors. "They're keeping up, but it's not clear that they're pulling ahead," he said. To be fair, however, for this new generation of servers, Dell is the first to lay its cards on the table. Not until IBM, HP, and Sun show their hands, likely after Intel unveils the Nehalem-based Xeon processors for servers, can a full assessment be made.

The timing of the product introduction does appear to be less than ideal. IT spending is expected to grow just 2.6 percent this year, according to IDC. But Dell is counting on customers that don't have a choice but to increase their data center capacity or replace broken and outdated equipment. As Anderson noted, employees need to be able to send e-mail, and a company's compliance obligations can't be neglected just because the stock market is down.

But is there ever a right or wrong time to introduce updated servers? asks Gillett of Forrester.

"One of our own clients said every year is a recession in IT. They said, 'This (economy) is nothing different for us. We always have to do a lot more with less.'"

The storage arrays and workstations are available starting Wednesday, while the servers won't be until next week.

Ponder this: Is an Intel product launch still a launch, if the product debuts very publicly in an Apple computer?

I won't presume to answer that question. But the fact is that Intel will launch Nehalem-EP server processors later this month, despite their manifestation Tuesday in the new Mac Pro under their official model names: the Xeon 3500 and 5500.

The chips--in their desktop variant known as the Core i7--are being offered in eight-core or four-core configurations and, like all Nehalam-architecture processors, come with an integrated memory controller for (theoretically) better performance. (Intel's Core architecture does not integrate the memory controller.)

Other Nehalem-architecture features include: Hyper-Threading for, according to Apple, "up to 16 virtual cores" (which improves multitasking), and Turbo Boost Technology, which dynamically increases the processor's frequency, as needed.

The Mac Pro also offers high-end Nvidia and ATI graphics. Systems can be configured with either Nvidia GeForce GT 120 or ATI Radeon HD 4870 graphics chips.

SAN FRANCISCO--Intel plans to bring its first dual-core Atom to market next month, it was revealed here Monday during the Intel Developer Forum. The chipmaker also disclosed more details of the Nehalem processor.

The power-efficient processor will be targeted at Atom-based desktops called nettops. Currently, Intel offers the Atom N230 processor for nettops. This chip has a slightly higher power envelope than the Atom processors built exclusively for mobile devices.

That news was revealed to this reporter by an Intel employee as senior vice president Pat Gelsinger was delivering his IDF keynote, which included more specifics about Nehalem, the family of chips the company plans to begin rolling out in the fourth quarter. Gelsinger, the general manager of Intel's Digital Enterprise Group, showed the first wafer holding individual eight-core processors, detailed the power-saving features of the Nehalem processors, and confirmed future mobile Nehalem processors.

Intel Nehalem processor lineup as shown at IDF 2008

(Credit: Intel)

Also due in September is the six-core Dunnington server processor, the final member of Intel's 45-nanometer "Penryn" family, which will ship to customers next month, Gelsinger said.

Most of his keynote centered on Nehalem, and one of the features Intel was pushing hard at IDF was a technology called Turbo mode.

Turbo mode is essentially a switch that turns off unused processor cores and then uses the remaining active cores more efficiently. This kind of sophisticated power-management technology will be used in both Nehalem-based laptops and servers, according to Gelsinger, and will become increasingly necessary as Intel brings out chips with more cores like the eight-core Nehalem processor due next year.

In short, in multi-core processors, cores not doing much can still use power. So, it's better to use, for example, a couple of cores more efficiently than four cores inefficiently.

The power saving technology is enabled by "an integrated microcontroller which only works on power management," said Rajesh Kumar, an Intel Fellow, who spoke during Gelsinger's keynote. There are about 1 million transistors dedicated solely to power management, Kumar said.

The feature "requires no operating system intervention. It is fully detected and managed by the hardware. If it has detected an idle core, it is able to reallocate that power budget to the other cores," Gelsinger said in an interview after his keynote.

On another front, Intel showed the first eight-core Nehalem chip. "This is the first showing of the eight-core Nehalem-EX," Gelsinger said in his keynote. He said the chip is a monolithic design, meaning that all eight cores are on one piece of silicon.

Nehalem-EP, or Nehalem Efficient Performance, will be a quad-core chip for mainstream servers and workstations. What Intel traditionally calls two-socket servers, Gelsinger said.

The mainstream desktop will be the Core i7. "With the i7 we have high-end desktop and extreme," Gelsinger said. The extreme edition is for overclockers, he said. Enthusiast gamers often overclock processors (ratcheting up clock speed beyond the rated speed) to gain extra performance.

"Turbo Mode" is a linchpin Nehalem technology

(Credit: Intel)

But there will be more pedestrian dual-core versions of Nehalem too. "There will be versions for the desktop that will be dual-core as well," Gelsinger said.

Gelsinger also talked about Intel's plans to put graphics directly onto the same piece of silicon as the processor. This will be a first for Intel.

He described why Intel is putting graphics right next to the processor. "There's a big sucking sound near the CPU. It keeps pulling things closer to it. This is uniquely enabled by Moore's Law...and as things get closer together I'm able to drive down thermal envelopes (i.e., heat) and decrease physical form factors (i.e., enable smaller computer designs)," Gelsinger said.

Gelsinger broke down the future processor lineup--with graphics on the processor die and without--as follows. "Lynnfield and Clarksfield are the versions without graphics. Havendale and Auburndale are the versions with integrated graphics." (Even Intel executives occasionally get confused by all the code names and it took two tries for Gelsinger to get this right.)

Intel has scores of futuristic, potentially game-changing research projects but Nehalem is bet-the-farm reality. The first Nehalem chips--and the first drops of what should become a giant revenue stream--will arrive later this year. So, it is not surprising that real silicon and real systems are starting to appear.

What is Nehalem exactly? The architecture will scale from two to eight processor cores, have faster chip-to-chip communication (Intel calls this technology QuickPath), do a better job of adjusting performance levels to suit power needs, and have a higher level of integration (more logic will be built directly onto the processor die).

Other salient features include more scalable memory (each processor will have its own dedicated memory), the ability to do more stuff simultaneously (up to 16 threads with simultaneous multi-threading), and new instructions to increase efficiency (called SSE4.2 instructions). Here's how Intel describes Nehalem.

A four-socket Nehalem system with integrated memory controllers

(Credit: Intel)

Maximum PC appears to be the first to build a Nehalem "Bloomfield" desktop system for everyone to see.

The system uses a 2.93GHz Bloomfield processor and an Intel motherboard with an X58 chipset (which had been codenamed "Tylersburg"), which will is also due to ship in the fourth quarter.

The Bloomfield chip is larger than current Intel quad-core processors (e.g., the Q6700), according to Maximum PC. This means more fans, bigger heat sinks, and more heat to dissipate overall.

Nehalem will support faster DDR3 memory. And this points to one of Nehalem's major departures from past Intel processors: the memory controller--which talks to the DDR3 chips--is now on the processor die. Previously, this was off-chip. In short, higher levels of integration generally means higher performance.

New "overclocking features" are also offered in Nehalem, according to Maximum PC. Overclocking--running the chip faster than its rated speed--is an absolute prerequisite for gamers. Which means, of course, that initially one of Nehalem's biggest draws will be gamers.

In the more distant Nehalem future, the mobile platform has gotten a name. At this point, Intel will confirm the code name only: Calpella. But otherwise "won't comment on speculation."

Other information posted on various tech Web sites about the 2009 Nehalem mobile platform in the past few days has been in the public domain for almost a year. That is, it will have an on-die memory controller and one version of the chip will have an integrated graphics processor--which will be a first for Intel. One new twist is the timing: it may launch in the third quarter of 2009.