For budget and entry-level systems, the APU — a CPU with an onboard GPU  — is most certainly the future. While AMD’s ambitions in this field won’t turn up until 2011, Intel is priming a selection of Westmere APUs for the winter months. Engineering samples of Westmere APUs are already circulating in the Orient, and HKEPC has one of them on the bench for review.

The roundup pits a Westmere APU (Clarkdale) and a Lynnfield APU (Havendale, more on that in a minute) against a Lynnfield-based Core i3 540 and a Core 2 Duo E8400. In all, the 540 trounces the competitors, but the APUs turn in fairly respectable numbers in computationally-heavy tasks.

There are a few issues with the review, however:

• HKEPC claims it is testing a Havendale, but Havendale was canceled for the 32nm Clarkdale and Arrandale models. Clarkdale serves the dekstop, while Arrandale serves mobile products, respectively. Even Intel’s own roadmaps from April support Havendale’s cancellation.
• HKEPC claims it has the Clarkdale at 3.06GHz in its test platform table, but all subsequent tables indicate 2.4GHz. HKEPC claims that it downclocked the Clarkdale part to match the maximum Havendale frequency of 2.4GHz.
• HKEPC regularly provides the incorrect spelling of “Havendale” as “Havandale.”

Does HKEPC really have an engineering sample of the ill-fated 45nm Havendale? The world may never know, but it raises enough questions that we must take the benchmarks with a grain of salt.

That said, the APU Intel is definitely launching, the Clarkdale, is within a stone’s throw of the Lynnfield Core i3 with a 600MHz clock deficit. If the benchmarks are at all legit, it appears Westmere has some significant bounce in its step.

Despite ongoing legal proceedings designed to block NVIDIA from producing Nehalem-compatible chipsets, big green is allegedly working to do just that.

A report from the Chinese tech website HKEPC suggests that NVIDIA is preparing two IGP chipsets for Nehalem and Westmere CPUs. The first, codenamed MCP99, is said to be a single-chip solution employing DMI, which means it is being prepared for the upcoming crop of 45nm Lynnfield chips which we have previously discussed. MCP89, meanwhile, will offer a similar set of features in a lesser performance profile for the more budget-conscious amongst us.

The MCP85 is also mentioned as a SoC solution which condenses the GPU, northbridge and southbridge functionality down into a single chip to simplify boards and cut costs.

NVIDIA has predictably refused to comment on unconfirmed chipsets, but the wily NVIDIA chief exec Jen-Hsun Huang was cryptic and suggestive as usual.

“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 during a July 26 analyst call. “Our company is…pretty darn clever. There is a lot of ways to skin the cat.”

We’ve been discussing Intel’s upcoming Gulftown for quite a while now. Tapped to succeed the Bloomfield (Core i7) at the top of Intel’s performance heap, the new Core i9-branded part will launch with a raft of enhancements to make sure it fills those shoes.

Gulftown’s membership in the 32nm Westmere family of architectures has allowed Intel to drop temps, boost cache, raise clockspeeds and increase the cores– six, to be exact. With a dash of HyperThreading and X58 compatibility, enthusiast desktops will be primed to chew through twelve concurrent threads when the chip debuts in early 2010.

While the details surrounding the chip have been available for quite some time, hard performance has remained elusive until today. HKEPC has recently published a small review that pits the Core i9 against a battery of tests, all of which paint it as an improvement to the Bloomfield.

To whet your appetite, here’s a snippet we’ve polished after Google spit it out:

In power consumption and temperature testing, the 32nm manufacturing process used by the Intel Gulftown offered amazing performance. Although Gulftown has six cores to the Bloomfield Core i7’s four, power consumption and temperatures are lower than the quad core. This proves that the next-generation 32nm manufacturing process is very mature.

UPDATE: 27 July, 2009 @ 11:57 AM

Ex-Inq journo and long-time Valley rabble-rouser Charlie Demerjian has recently offered new insight on the wild world of ATI’s DirectX 11 codenames in a brutal piece on plagiarism in the world of tech journalism. Here’s the breakdown:

Hemlock: Highest-end desktop part, probably dual Cypress GPUs a la the rumored Radeon 5870 X2.
Cypress: High-end single-GPU part. Similar to the Radeon HD 4870 when it launched.
Juniper: Mid-range desktop part. Similar to the Radeon HD 4850 when it launched.
Cedar & Redwood: Low-end desktop parts. Similar to the Radeon HD 4650 and 4670.

The confusion in the codenames contained in the original story is most likely owed to a simple transcription error on behalf of The Inquirer. It happens, especially when (as Chuck D points out) you’re being blitzed with PR, buzzwords and codenames. Anyone who’s been to a trade show or a press event understands.

Original story follows:

Today’s semiconductor news from the likes of AMD and Intel aren’t quite headline material, but they’re better than the proverbial page bury. Times like these call for a roundup, but our taskmaster editor doesn’t like them. Good thing he’s sleeping off a lake party and isn’t around to stop us.

AMD’s DirectX 11 codenames

DirectX 11 is just around the corner, and NVIDIA’s struggles to get its shit together at 40nm has left AMD the sole heir to the estate this winter.

While considerable speculation remains over just how AMD plans to get into the spirit of the season, we do have codename confirmations to tide you over.

The family of 40nm DirectX 11 GPUs is collectively known as the Evergreen. The Evergreen architecture will launch with five parts that will span the price spectrum:

• Cypress (Desktop: For those with more money than sense)
• Redwood (Desktop: An enthusiast part like the Radeon 4890)
• Juniper & Cedar (Desktop: Mainstream parts like the Radeon 4670 and 4650)
• Hemlock (Desktop: Poisonous parts for the cheapskates amongst us)

Word has it that AMD has already received functioning Evergreen silicon back from the fab. If that’s true, we are indeed looking at a DX11 launch timed to coincide with Windows 7. It’s like the Radeon 9800 coup all over again!
(more…)

Update 31 July, 2009: In the time since Intel announced its new branding initiative, the firm has wizened and kicked the impending 32nm Gulftown under the “Core i9″ banner. While the rest of their branding scheme remains inane as ever, it’s nice to see some clarity bubbling to the surface.

Original story follows:

Yesterday Intel Corp. announced a sweeping change to their branding which is every bit as confusing for the consumer as the branding it replaced.

The world’s largest CPU company announced the changes as a major pillar in a sweeping strategy allegedly designed to simplify branding for customers. Intel spokesman Bill Calder wrote that the Core branding currently suffered from an identity crisis at multiple levels.

“Today the Intel Core brand has a mind boggling array of derivatives (such as Core^TM2 Duo and Core 2 Quad, etc),” he said. “Over time those will go away and in its place will be a simplified family of Core processors spanning multiple levels.”

Intel’s new strategy is hinged on permutations of the “Core i7″ branding unveiled with the introduction of the firm’s new Nehalem family of architecture. Here’s how it works: (more…)

Intel CEO Paul Otellini used the April 14 Q1 2009 earnings conference call to reveal that the 32nm Westmere family of Intel processors is already in the hands of OEMs for testing.

Westmere is considered the culmination of a broader initiative that will invest $7 billion over two years to create additional 32nm plants in the United States. The investment will focus on improving or converting existing 45nm sites in Oregon, Arizona and New Mexico. Intel states that their investment will help create more than 7,000 high tech/high-wage jobs in the United States alone.

New processors based on the 32nm technique are the smallest produced to date, and were slated for year’s end. Collectively known as the Westmere family of processors, parts based on the design will include the Gulftown (enthusiast desktop), Lynnfield (mainstream desktop), Clarkdale (value desktop), Clarksfield (enthusiast mobile) Arrandale (value mobile), and Beckton (server).

Given Intel’s considerable lead on 32nm fabrication over rival firm AMD, there has been mounting speculation that Intel would push the Westmere launch back to 1Q10 to give existing Nehalem parts more time as firm’s flagship in the channel. However, the revelation that Intel is already sampling Westmere parts to OEMs confirms a 2009 launch and proves the process is as healthy as claimed.

(more…)

We did some digging around on Twitter today, and we’re pleased to inform that the Gulftown chip from the 32nm Westmere family will be X58-compatible.

With an expected release date of 4Q09/1Q10, the Gulftown is the successor to today’s Bloomfield core that serves the Core i7 920, 940 and 965 CPUs. Gulftown will continue in the tradition of QPI, DDR3, and tri-channel memory, but it should be a drop-in upgrade on X58 motherboards. The perk? It has six cores, not four.

In other news, if you’ve been waiting to see how Intel’s P55 chipset pans out, you might want to abort those plans. Intel confirmed that the next five chipsets to emerge from Intel HQ will be based on the LGA1156 socket. This socket will serve the upcoming Core i5 series, most notably composed of the Lynnfield core, which drops QPI and a memory channel in favor of the old frontside bus.

Don’t let this scare you, though. The “next five” claim probably refers to the ecosystem of P55-derived parts, such as the G55 and Q55. The P35 and P45 received a similar treatment to serve all its various market segments.

If we had to hazard a guess, we would say that Intel has adopted this tack to clearly stratify its enthusiast/mainstream market segments. We can only presume that Intel was unhappy with P35 and P45 cannibalizing significant sales of the X38 and X48. We can also presume that Intel was rather unhappy with speedbinned Core 2s adopting a massive following that all but sabotaged the sales of higher end parts.

Bless Twitter’s little heart.

Yesterday we reported Intel would release 32nm processors by the end of the year, presumable in the fourth quarter, but today more information is coming out that 32nm processors could hit a full quarter early.

After yesterday’s demonstration of a working 32nm processor, new Intel road maps and information from company sources suggest the 32nm Westmere processor, a die-shrunk and improved Nehalem, is on the fast track to being released around September for desktops and notebooks. The race to 32nm is interesting, but perhaps Intel was even quicker out the gate than most thought.

Intel has freshly announced that processors based on their new 32nm fabrication will be available by the end of the year.

The move is one piece of a broader initiative that will invest $7 billion over two years to create additional 32nm fabs in the United States. The investment will focus on improving or converting existing 45nm sites in Oregon, Arizona and New Mexico. Intel states that their investment will help create more than 7,000 high tech/high-wage jobs in the United States alone.

Intel’s investment will be made at existing manufacturing sites in Oregon, Arizona and New Mexico and will support approximately 7,000 high-wage, high-skill jobs at those locations — part of a total Intel workforce of more than 45,000 in the U.S.

(more…)