Nanotech - The Circuits Blog

Correction, 1:30 p.m. PST: This story misspelled the last name of a Qualcomm director of product management. His name is Manjit Gill.

Qualcomm's four-year, $350 million effort to design a chip that goes into small notebooks and handhelds will come to fruition next year when device makers deliver products based on the Snapdragon processor.

I spent Friday morning at Qualcomm discussing the San Diego company's quest to build a processor for very small, very lightweight notebooks--what the Intel camp calls a Netbook.

Though Qualcomm's prototype looks like a Netbook on the outside, the Snapdragon technology inside is quite different from Intel's Atom silicon, which powers dozens of Netbooks on the market today.

And what's inside goes to the heart of how Qualcomm separates itself from Intel.

"Intel is a great company. I think they have great talent. But we believe there are limitations in the (Intel) architecture," said Manjit Gill, director of product management, Connected and Consumer Products Group, at Qualcomm. In short, he thinks Intel technology is wrong for this market, which values connectivity above processing power.

"Our vision is that (the device is) always connected. Even when you shut it down, it's still 'on.' (The laptop) goes to your Exchange server, gets your e-mail, puts it on the drive--solid-state or hard drive--and then when you're ready to do e-mail, you flip it open and it's right there. Instant on, always connected," Gill said.

Manjit Gill of Qualcomm demonstrates a small laptop prototype running Red Flag Linux. Next to it is a MacBook Air.

(Credit: Brooke Crothers)

"The question is, can you enable the same value proposition on an Intel platform?" Gill's answer, not surprisingly, is no. "Two or three hours later the battery's just completely drained out. You cannot rely on it to be there all day long in your bag and still getting all your e-mail."

And Snapdragon contains the technology that will enable Qualcomm to build an un-Intel, un-Netbook type of device, Gill claims.

Though becoming too un-Intel presents a challenge. Intel has consistently countered arguments like Qualcomm's by saying that consumers want to stay plugged into its x86 architecture that runs Windows and all the software and applications that go along with Windows. And this has been one of the reasons that Atom-based Netbooks have become so popular, so fast.

"Of course, Atom and Snapdragon are completely different beasts," said Tom R. Halfhill, a senior analyst at the Microprocessor Report. "Atom is an x86 processor with little integration. Snapdragon (has) more integrated features. In terms of real-world performance, Atom is much faster than Snapdragon."

But one of the big goals, according to Gill, was to build something that exceeds the current ARM processors found in hundreds of consumer devices today. (ARM is a design house that licenses its chip designs to companies such as Samsung, Nvidia, and Qualcomm.)

"About four years ago we had a bunch of guys join (who) used to work for IBM in the PowerPC group," Gill said. He said the engineers came from IBM's chip group that designs chips for low-power "embedded" devices.

The new team was tasked to turbocharge typical ARM designs that were "maxing out" at about 500MHz, which isn't enough speed to deliver the experience that Qualcomm is aiming for. (Intel's Atom for Netbooks, by comparison, now maxes out at 1.6GHz.)

"There was a need to go do something beyond this. So, we went and got the architecture license (from ARM) and we have this team of about 50 CPU designers and we put them to task. So, four years and $350 (million) to $400 million later, we have a CPU that actually works better than the (typical) ARM CPU."

The piece de resistance of this strategy is the Qualcomm QSD8672 dual-core Snapdragon that features two CPU computing cores capable of 1.5GHz performance, and a host of other features includes HSPA+, up to 28Mbps download speeds, 1080p high-definition video, Wi-Fi, mobile TV, and GPS. The graphics core is based on Advanced Micro Devices' ATI unit's technology.

Qualcomm is able to achieve this relatively high speed (1.5GHz) for a low-power processor because it did more than simply get a license from ARM. "We went and got an architecture license from ARM. The architecture license was for their new instruction set, the V7 instruction set. There's a difference between getting an architecture license and just getting a core license. A core license means ARM does the (chip) core and they give it to you. The architecture license is different: the actual implementation is your own," he said.

The 45-nanometer processor will be built by Taiwan Semiconductor Manufacturing Company.

Though Qualcomm won't necessarily build the laptops itself, the company's vision for how they will be used is clear. "We don't view these as laptop replacements. We don't think these are the devices people will use to do Photoshop editing or anything like that," he said. "Browse the Web, check your e-mail. At $399, or a carrier subsidizes it further to $199. (At this price), it becomes very, very compelling."

Companies including Acer, Asus, and Toshiba are planning devices based on Snapdragon, according to Qualcomm. Acer, for example, will initially introduce Qualcomm's Gobi 3G modem into its devices, then gravitate to products based on Snapdragon.

Apple's MacBook Air doesn't live up to its wireless promise.

To quote an Apple tagline, "without wires, you're free to go anywhere." But the wireless part of the "air" play on words fails to deliver. (The other half its light-as-air weight: here it does deliver.)

As I've written in the past, I like the Air. I got one in February as soon as it was available at retail and have been pleased with the performance, screen, keyboard, build, and, until recently, the battery life (which has dwindled to under an hour). Of course, the head-turning aluminum aesthetics is also a major appeal to many people.

That said, after a spurt of trips including a 10-day stay on the East Coast and a few treks to Los Angeles, the Air's wireless shortcomings have become painfully clear. In a word (or two), no 3G.

Now, before I get slammed, let me say that I fully realize that I'm not the first person to reach this conclusion so I'm not claiming any unique epiphany. There were a number of observers citing this paradox way back in January. Some frustrated users even attempted hacks to shoehorn a 3G modem into the Air. The point is: because of the price and the way it's marketed, 3G should be built in.

But the full brunt of not having 3G hit me on Friday when I made a trip to Qualcomm to get briefed on a new version of the Snapdragon applications processor (more on this in another post). Sitting there in the nerve center of one of greatest wireless companies in the world, I couldn't get a wireless connection. Everyone else in the room had 3G connections of one kind or another. The Air instantly became the proverbial doorstop (or paper weight--choose your simile, or maybe it's more apropos to say it was a dinosaur.)

It didn't take much prodding from me to get the Qualcomm product manager to point out this fatal flaw.

This came after weeks of not being able to use the Air in many situations when I desperately needed a wireless connection. For instance, not all LA airports have reliable Wi-Fi connections. The John Wayne Airport in Orange County being one example. And when I was on the East Coast, one sprawling place I stayed at for several nights had Wi-Fi only in one inconveniently situated area that was inaccessible at night.

In these situations, the Air is nothing more than a slab of beautifully sculpted aluminum.

Hewlett-Packard Compaq 2510p (L) and MacBook Air. The HP comes with a built-in WWAN option, the Air does not.

(Credit: Brooke Crothers)

Which brings me to the cult of Steve Jobs. Apple was brilliant enough to deliver a groundbreaking design like the Air but why wasn't it savvy enough to build in 3G?

Before I get slammed again, let me throw out some reasons (excuses) why Apple didn't build in 3G, based on reports I've read and my own observations. Apple didn't like the fact that 3G modems often made the user commit to one service provider, i.e., Verizon or Sprint or Vodafone. Or, it believed that if users wanted 3G, they could simply plug in a USB 3G modem.

I have serious problems with both of those reasons. Particularly when you're paying typically more than $1,800 (or $2,500, depending on the model) for a notebook billed as a wireless wonder (for Apple ad copy on why the Air is such a wonderful wireless laptop, just cruise over to Apple's MBA page.) And I have even more of a problem when ultrathin Netbooks are coming with 3G at one-third (and potentially a tiny fraction of) the cost of the Air.

For comparison, let's look at another Valley company, Hewlett-Packard. HP has been selling WWAN (Wireless Wide Area Network) modems in its business laptops for at least two years (and probably longer). The HP-Compaq NC6400 laptop introduced about two years ago was offered from day one with WWAN built in.

Granted, the choice of carriers at first was limited but look at HP's offerings today. Models (including those in the EliteBook line) are offered with Qualcomm's Gobi modem. Gobi obviates the need to have unique radios for each carrier. Gobi supports Verizon, Sprint, and others in one device.

Where was Gobi when Apple refreshed the MacBooks in October? I'm sure Apple has plenty of excuses (for example, not enough space in the Air's ultrathin design).

But Apple should have had 3G from the beginning and certainly in the October refresh. In today's 3G world, continuing to call it the MacBook Air brings another meaning to the play on words: lightness of weight with a touch of advertising hot air.

Additional comments::
The point is not that a user can potentially add an external WWAN modem (though even that's not necessarily easy to do), the point is that the Air should come with 3G capability built in considering how the computer is marketed.

The newest solid-state drives are just starting to hit retail. But would you buy one?

Intel X25-M solid-state drive has received glowing reviews for its performance

(Credit: Intel)

Solid-state drives are attractive because they're generally faster than hard-disk drives, particularly at reading data--generally something PC users spend most of their time doing.

But price is still an obstacle, especially to the frugal consumer.

Sunnyvale, Calif.-based OCZ Technology is now offering some of the most competitively priced solid-state drives based the high-speed Serial ATA (SATA) II interface.

OCZ Vertex SSD drives start at $129 for a 30GB SSD. Other capacities include a 120GB drive for $469 and a 250GB SSD for $869. Though $869 may seem pricey compared to a 7200RPM 250GB hard-disk drive that can retail for well under $100, it's relatively cheap for a large-capacity SSD. In the past, SanDisk had sold a 256GB drive through resellers that was priced, almost incredibly, at more than $15,000. Axiom had been selling 256GB solid-state drives priced above $6,000.

OCZ says the Vertex Series of SSDs have a 1.5 million-hour mean time before failure (MTBF), "ensuring peace of mind over the long term." Solid-state drives, since their inception, have been plagued by doubts about write durability. SSD manufacturers such as Intel, Micron Technology, and Samsung say long-term durability is no longer an issue.

The OCZ drives are backed by a two-year warranty.

Intel's X25-M SATA solid-state drive is now widely available at retail. An 80GB X25-M is priced at just more than $500 at most retailers. Reviews of the X25-M's performance have been very positive.

Unfortunately, the only way to get the newest SSDs in some cases is by ordering a laptop. Samsung is now offering its latest-and-greatest 128GB SATA drives in ultraportable laptops such as the MacBook Air and the Dell Latitude E4200 and E4300.

The Latitude E4300 can be configured with a 128GB SSD for $460 more than the standard 160GB 5400RPM hard disk drive.

Taiwan will likely rescue, in some form, failing memory chipmakers, as that country's industry falls behind the rest of the world.

Newest DDR3 DRAM memory module

(Credit: Micron Technology)

Recent reports depict an industry desperately seeking financial help. The Taiwanese government has responded by offering loans, according to a variety of reports.

(See also: Chip sales dip in October, flash sales dive.)

One report on Thursday said that Taiwan's economic affairs ministry has approved a rescue package. No specifics have been revealed, however.

At stake is Taiwan's dynamic random access memory (DRAM) industry. DRAM is the main memory used in personal computers.

Avi Cohen, managing partner at Avian Securities, which covers memory chip market movements, says "it's something (the government) will have to do if they want to keep all those people employed" at DRAM manufacturers in Taiwan.

And why is this happening? "The Taiwan DRAM industry is falling further and further behind in terms of cost and in terms of production volumes. They have the worst cost basis," Cohen said. "The guys that can produce DRAM on the newest equipment are the most competitive. And therefore can eke out a little profit. Everybody else loses money."

And the companies that lose money? Taiwan-based DRAM makers like Powerchip Semiconductor and ProMOS Technologies--which have been appealing to the government for immediate aid.

The winners (in relative terms since all companies are facing challenges) are companies such as South Korea-based Samsung and U.S.-based Micron Technology. Both Samsung and Micron are comparatively aggressive about moving to next-generation manufacturing process technologies to keep the cost per megabit of memory produced down.

But rescuing the companies will not necessarily be a good thing for the market. "The overproduction that exists in DRAM and all of memory for that matter will likely continue. If you want this market to get some semblance of sanity and reasonableness, you may want to see players go out of business."

Intel has completed the development phase of its next-generation manufacturing process that shrinks chip circuitry to 32 nanometers, the chipmaker said Tuesday night.

Intel 32-nanometer SRAM chip

(Credit: Intel)

Intel processors are currently made on a 45nm process. Generally, smaller geometries result in faster and more power-efficient processors.

"The company is on track for production readiness of this future generation (of transistors)...in the fourth quarter of 2009," the chipmaker said in a statement.

Intel said it will provide technical details about the 32nm process technology at the International Electron Devices Meeting (IEDM) next week in San Francisco.

Finishing the development phase for 32nm process technology keeps Intel on track with its "tick-tock" strategy. Tick-tock is intended to introduce either a new processor microarchitecture or cutting-edge manufacturing process about every 12 months.

"Producing 32nm chips next year would mark the fourth consecutive year that Intel has met its goal," the company said.

The 32nm paper and presentation "describe a logic technology that incorporates second-generation high-k + metal gate technology, 193nm immersion lithography for critical patterning layers, and enhanced transistor strain techniques," Intel said.

Other Intel IEDM papers will "describe a low power system on chip version of Intel's 45nm process, transistors based on compound semiconductors, substrate engineering to improve performance of 45nm transistors, integrating chemical mechanical polish for the 45nm node and beyond; and, integrating an array of silicon photonics modulators," according to the company's statement.

Intel will also participate in a short course on 22nm CMOS technology.

Bad bumps? A U.K. tech site is alleging that the latest Apple MacBook Pros contain Nvidia graphics chips with the same "bad bumps" problem that Nvidia addressed this summer and said was rectified.

Nvidia said in a phone interview on Tuesday that this is dead wrong.

First a little background. Nvidia issued a statement July 2 saying it would take a charge of up to $200 million to cover repairs due to a "weak die/packaging material set in certain versions of its previous generation GPU and MCP products used in notebook systems."

Both Hewlett-Packard and Dell have come out with statements addressing the issue in laptops. And both companies have programs that try to fix the issue.

U.K. tech site The Inquirer is saying that bad bumps--"tiny balls of solder that hold a chip to the green printed circuit board"--are still present in the GeForce 9600 graphics chips that ship in the newest MacBook Pros. An issue that The Inquirer claims is the root of the problem.

The Web site said it took a MacBook Pro off a store shelf, disassembled it, desoldered the chips, sawed them in half, encased them in Lucite, and ran them through a scanning electron microscope equipped with an X-ray microanalysis.

As a result, The Inquirer alleges that the MacBook Pros with the GeForce 9600 chips have the older, defective high-lead bumps, while the MacBook Air and MacBook have the newer eutectic solder (newer, low-lead bumps).

So, in essence, the MacBook and MacBook Air are fine, while the MacBook Pro is problematic.

Nvidia vehemently disagreed with the allegations, calling them completely untrue. The Inquirer's "initial analysis of problems with some of the older chips was already flawed," said Michael Hara, vice president, investor relations and communications at Nvidia.

The Inquirer reporter "believes high-lead bumps are bad. That's his underlying theory. It's not true," Hara said.

He continued: "When you build a device, it's the material properties and everything in combination that leads to the robustness of the design. What we call the 'material set.' It's a combination of the underfill (a kind of a glue that helps hold the chip down) and the bump together that creates that stability in that connection," he said.

Hara talked about how the original problem announced by Nvidia on July 2 was rectified. "A more robust underfill would have taken the stress off the bumps and kept that (original problem) from happening. What we did was, we just simply went to a more robust underfill. Stopped using that (previous) underfill, kept using high-lead bumps, but we changed the underfill. And now we don't see the problem."

"Intel has shipped hundreds of millions of chipsets that use the same material-set combo. We're using virtually the same materials that Intel uses in its chipsets," Hara said.

Hara also said Taiwan Semiconductor Manufacturing Company (TSMC) ships a "staggering" number of chips to many companies worldwide with high-lead bumps. TSMC is the world's largest contract chip manufacturer and makes chips for Nvidia, Advanced Micro Devices, and many other companies.

Nvidia also issued this written statement: "The GeForce 9600 GPU in the MacBook Pro does not have bad bumps. The material set (combination of underfill and bump) that is being used is similar to the material set that has been shipped in 100's of millions of chipsets by the world's largest semiconductor company (Intel)."

On Tuesday, an industry consortium ratified the OpenCL 1.0 specification, a standard that started as an Apple proposal but has gained many supporters, including graphics chip companies Nvidia and Advanced Micro Devices.

OpenCL, or Open Computing Language, is essentially an open industry standard for 3D graphics and computer audio and is meant to extend the capabilities of the graphics processing unit (GPU).

Not surprisingly, graphics chip companies have been quick to pick it up, including Nvidia and AMD's ATI graphics unit, which both made separate announcements Tuesday, along with the broader announcement from The Khronos Group consortium.

OpenCL has been developed on Nvidia GPUs and the company was one of the first to show working OpenCL. Nvidia is the graphics chip supplier for Apple's MacBooks.

Other computer graphics-related companies supporting OpenCL include Imagination Technologies and ARM. The complete list is long and includes 3DLABS, Broadcom, Electronic Arts, Freescale, IBM, Intel, Nokia, Samsung, and Texas Instruments.

Nvidia combines support for OpenCL with its CUDA language

(Credit: Nvidia)

The language itself--based on the C programming language--is designed for programming parallel computing across both GPUs and CPUs (central processing units). Apple proposed OpenCL in June for standardization work, targeting Mac OS 10.6 "Snow Leopard."

Apple's goal for Snow Leopard is to let any application tap into the prodigious parallel computing capabilities of GPUs, which integrate hundreds of computing cores.

"We are excited about the industry-wide support for OpenCL," said Bertrand Serlet, Apple's senior vice president of Software Engineering, in a statement. "Apple developed OpenCL so that any application in Snow Leopard, the next major version of Mac OS X, can harness an amazing amount of computing power previously available only to graphics applications."

AMD, which supplies both CPUs and GPUs, is well situated to exploit OpenCL. "The potential benefits of having applications run on both the CPU and GPU within a system are enormous," said Rick Bergman, senior vice president and general manager, Graphics Products Group, AMD, in a statement. "Unfortunately, up until now programmers could only choose proprietary programming languages that limited their ability to write vendor-neutral, cross-platform applications. With today's ratification of OpenCL 1.0, I'm happy to say those days are over."

Advanced Micro Devices will reduce its stake in the manufacturing operations it spun off in October, as it adjusts to repercussions of the financial crisis.

This follows a fourth-quarter warning earlier this month when the chipmaker revised its revenue estimate downward.

And like the shares of many companies, AMD's stock price has been in a free fall. Dropping from over $7 back in June to $2.10 on Monday.

Monday's action revolves around Abu Dhabi-based Mubadala Development and the Advanced Technology Investment Company (ATIC)--which Mubadala backs.

ATIC has equal voting rights with AMD in the newly formed manufacturing concern, the Foundry Company, and--per the October agreement--had owned 55.6 percent of the new entity.

However, as a result of amendments between AMD and ATIC, AMD will own approximately 34.2 percent and ATIC will own approximately 65.8 percent of the Foundry Company's fully converted common stock, AMD said Monday.

"Changing economic times" is the reason for a change in the terms, an AMD representative said.

Other changes include a restructured agreement that now says Mubadala will purchase 58 million shares of AMD's common stock "at a revised purchase price per share equal to the lower of (i) the average closing price per share of AMD's common stock on the NYSE during the 20 trading days immediately prior to and including December 12, 2008 or (ii) the average closing price per share of AMD's common stock on the NYSE during the 20 trading days immediately prior to the closing date of the transaction."

AMD will also issue to Mubadala an additional 5 million warrants to purchase AMD stock, for a total of 35 million warrants.

"All other material economic terms of the transaction agreements remain unchanged. ATIC will still invest $2.1 billion to purchase its stake in the Foundry Company, of which it will invest $1.4 billion directly in the new entity and will pay $700 million to AMD," the chipmaker said in a statement.

Nvidia's PhysX engine is intended to bring more realistic motion to games such as Backbreaker.

(Credit: GameSpot)

Electronic Arts and Take-Two Interactive Software are adopting Nvidia's PhysX technology, bringing more realistic gaming to the PC.

The largest graphics chip supplier is announcing on Monday that Electronic Arts and Take-Two have licensed its PhysX technology as a development platform.

"PhysX is a great physics solution for the most popular platforms, and we're happy to make it available for EA's development teams worldwide," Tim Wilson, chief technology officer of EA's Redwood Shores Studio, said in a statement.

"We are very impressed with the quality of the PhysX engine, and we licensed it so our studios can use this solution early in development," Jacob Hawley, technology director of 2K, a publishing label of Take-Two, also said in a statement.

Nvidia got its physics technology when it acquired Ageia in February. PhysX runs on the graphics processing unit, or GPU. Intel and Advanced Micro Devices, on the other hand, have been promoting technology that is executed on the central processing unit, or CPU. Intel's approach uses technology from Havok, a developer of a physics engine that Intel bought in September of 2007.

Adhering to the laws of physics
The goal of Nvidia's technology -- based on the laws of physics -- is to make game objects respond in a realistic way to physical events. More conventional technology uses a canned response, in which the same response is repeated over and over. For example, a window breaks, or a person falls the same way every time. In a PhysX-enabled football sports game, however, the angle and velocity of the impact is calculated by the GPU to generate a real-time response that is different every time.

The technology was meant to run on the GPU, according to Jon Peddie, whose firm tracks developments in the graphics chip industry. "It's a GPU thing, and the fact that EA and Take-Two are coming out (with support) gives you a clue why," Peddie said. "This really is a significant event," he said, "enabling the GPU to do physics."

Ujesh Desai, VP of product marketing at Nvidia, discusses physics strategy

(Credit: Brooke Crothers)

Ageia's secret sauce is its physics libraries, which are supported on Microsoft's Xbox, Sony's PlayStation 3, Nintendo's Wii, as well as on the CPU and Ageia's own PPU (physics processing unit), Ujesh Desai, vice president of product marketing at Nvidia, said in an interview last week. "It's a very open platform. Something game developers really liked, which is why a lot of game developers adopted it," he said.

The launch pad for Ageia on the PC is Nvidia's CUDA, or Compute Unified Device Architecture. CUDA already has a large installed base of GPUs that can run a C program, "which is what PhysX is," Desai said. "We bought Ageia, (and) they ported their PhysX API to our GPU, using our C compiler on top of CUDA. So now there are 100 million GeForce (chips) out there that can do PhysX processing."

And PhysX-enabled games will offer much greater realism. "Today, the way they do sports games is motion capture. They capture the different animation--running, falling," Desai said. "What you realize is that for the first 5 to 10 minutes of the game (or movie), it looks believable, but after you play for a while, you realize, wait a minute, every time he falls, he falls the same way. Every time I make that tackle, it looks the same."

The game Backbreaker uses PhysX. "They're calculating those tackles in real time, based on how the body interacts and the body mechanics interact. So no two tackles are the same," according to Desai. Another game, Mirror's Edge, is coming out in January from a company called Dice. The PC version will have PhysX in it, according to Desai.

"Ageia changed the rules on this," Peddie said. "It's much, much more realistic."

Ageia's physics was originally done on an Ageia Physics Processing Unit, Peddie said. "This was the only way to make it work. But now this capability (software) has been ported to Nvidia GPUs, and this can be done on Nvidia silicon," he said.

Physics can also be used to make things look more photo-realistic. "In today's games, cloth and hair look very fake because you don't have the right physical properties," Desai said. But with PhysX, "all these things can be physically simulated."

Unreal Tournament 3, a la PhysX.

(Credit: Nvidia)

Havok--the company Intel acquired--was the first to introduce physics into games and bring out a physics library. Havok's physics has been run on the CPU in a time-scheduled way, Peddie said. "Because of that, there weren't many CPU resources to really do a great job on the physics," he said. "Nothing would really happen. What happened, at most, is that you would hit this thing (a window or a wall, for example), and it would apply a decal to indicate that there was some change in it. It's not very realistic."

AMD, for its part, will pursue a balanced platform. "The GPU is a great place to do processing. We'll do the offloading (to the GPU), where it makes sense," said Korhan Erenben, product marketing manager at AMD Graphics Products Group. "(But) we are aligned with Havok, in terms of working on a future direction of physics. Right now, it is on the CPU, and we think that serves the broad installed base. Taking it to the next step would be to have a capability on the GPU--where and when it makes sense."

Physics is better on GPUs
Peddie explained why physics is more suited for the GPU than the CPU. GPUs today typically have hundreds of processors that are good at doing many things in parallel. "If you have threads or processes that can be run simultaneously, (and) if you have processors available to deal with each one of those threads, then you can get your results a lot sooner," he said.

He described a technique called Same Instruction Multiple Data (SIMD). "The same instruction is the physics equation. Things fall toward Earth all the time. And the multiple data will be what the things are. It might be a rock, might be a person, might be the wheel of a car. You have to be able to process this stuff and have it behave in a realistic fashion. To do that, you have to process it very quickly," Peddie said. "The advantage that GPUs bring is that they have this humongous number of processors. Certainly as good as the (Intel) 486 ever was. So they're really good processors, and you've got hundreds of them literally inside the GPU."

There will be challenges for users, however. "The tricky part is, why would I want to take one graphics card and spend $500 on it, and then not use it for graphics but rather use it for physics?" he said. "The answer is, of course, I wouldn't."

Peddie suggested that a gamer might use the really good card for physics and employ the old card "that you got last year" for graphics, assuming that there are enough slots in the PC.