It’s usually at about this time in the console life cycles (4-5 years in) that the PC begins to pull ahead with raw power, and we start seeing noticeably superior games (graphically). Moore’s Law is still in effect, but being able to see large graphical improvements in games is getting harder to discern. There is usually always one developer who steps up with an amazing looking game on the PC, even if it’s only for bragging rights, but at this stage the two year old Crysis still holds the crown and there is not much else in the pipeline except maybe Rage, Project Offset, and S.T.A.L.K.E.R.: Call of Pripyat, although from the trailers I’ve seen none of these look better than Crysis.

This raises some questions. Have we reached a plateau with graphics? Is what we have “good enough”? If the PC is a barometer of where graphics are heading, as it has been in the past, then things are definitely slowing down. There is a case of diminishing returns as you get closer to photo realism and you don’t need any more power to make the side of a building or a race car track look more realistic than they do now on current technology. But surely there is still a long way to go with animations and rendering people? and physics could still be greatly improved. Maybe it’s just another indicator that the PC is a shadow of it’s former self as a major games platform?

Intel are hoping that their new Larrabee graphics cards will revitalise the market but unless they are blindingly fast and developers get on board with them they will only add to the problem because they have a different architecture to current GPUs which will mean more development costs when making games for them.

I hope we haven’t already had all the great leaps forward in visual fidelity and further improvements prove to be only incremental. I still want to be blown away. I guess we’ll have to wait for the next-gen of consoles to find out how significant the improvements can be.

After the Nvidia GPU conference I’ve been reading up on the Fermi architecture. Now for gaming, the card is cool, but you quickly realize that gaming is not its targeted audience. It just happens to be that the work I do has a lot to do with images and doing stuff in parallel. The thing with images are that they can usually be very of very high resolution and the output might consist of voxels which also might be of high resolution. So far bandwidth has been a big bottleneck. Even though the PCI-express is incredibly fast, the calculations might actually be faster than the transfer. Therefore if one applies SIMD instructions and take advantage of the considerably larger memories on 64-bit machines thinks are done faster since you don’t need to shuffle stuff around.

This has been, in my own humble opinion, a big issue for the professional market. Of course at the GPU conference many companies shared their thoughts on the subject and showed that although this is a limitation, it is possible to work around. The bad news of course is that it requires one to have good knowledge of how CUDA utilies the raw power of graphics card. Such knowledge takes around a year or two and hasn’t been much of an option for companies without proper R&D.

Thankfully Nvidia acknowledges the issue and therefore they implemented 64-bit memory addressing capabilities on the Fermi card (actually the first gen is going to use like 40 bits) which in turn allows for those 8 GB graphics card. For the professional market this is the right way to go, and programmers will always be happy with more memory (if a bit more lazy).

The next thing that blew me out of the water is the raw computation power of the Fermi architecture. Now, for single precision calculations graphic cards have been viable for quite some time. But when doing stuff in a more scientific way you always want double precision. The Cell processor has so far been a real killer at this, and if the memory wasn’t so lousy on the Playstation 3 could possibly become the weapon of choice (4 SIMD registers per SPU if I recall correctly, proved 10 times faster number crunching than Intel processor). Now the Fermi has 512 cores, which I think would translate into registers? Anyway, with 16 Streaming Multiprocessors each with 32 single precision calculations per cycle the Fermi card shows incredible potential. Especially as doing double precision is not suppose to be slower (giving 16 calculations per cycle). Now this I won’t know before testing, if given the opportunity of course. Still the sum becomes 256 double precision calculations per cycle compared to 2 per core on a normal processor. Even with new quad cores with utilizing hyper threading and having 2-4 times the clock rate the Fermi should still outperform considerably.

I can only assume the Larrabee will blow Fermi out the water (because it has to), and there fore the parallel computing power on the horizon must be crazy. Exciting times for the performance geek. Also, Nexus allows for proper debugging of CUDA (FINALLY!) which will allow companies to capitalize on this sick performance more quickly.

For about two decades, when Stamford police officers needed to track a fleeing suspect or scour a house for drugs, they turned to canine units from neighboring departments.

That meant officers often had to wait for an hour or so as departments from as far away as Milford sent canine teams to Stamford. On Tuesday, a State Police dog was sent to downtown Stamford to help track a bank robber, who eventually got away with an undisclosed amount of cash.

Last month, however, the department got its newest members — two German shepherds bred and trained for police work. The revived canine unit should be on city streets in December, but for now the dogs are living with Stamford police officers, who will become their full-time handlers.

The dogs should help police track scents of suspects and missing persons, sniff out drug stashes and subdue violent criminals. Lt. Sean Cooney, a police department spokesman, said the two dogs cost $12,000 each, including their training. The money comes from the Federal Asset Forfeiture program, which allows police to seize property and cash from suspected drug dealers.

Over the years, the canine unit became a point of contention during contract talks between the city and police union, said Sgt. Joseph Kennedy, the union president. Chief Brent Larrabee approached the police union a few months ago and suggested setting up the canine unit outside the contract because of its necessity, Kennedy said.

“It just dragged on and on and on and went from contract negotiation to contract negotiation,” Kennedy said, adding that waiting for another department’s patrol dog “causes a lot of investigations to go south.”Seth O’Brien, one of the Stamford police officers in the revived canine unit, got Stoki last month. O’Brien is a fan of the breed and already has a 5-year-old German shepherd. He said they’re smart dogs, loyal and willing to work, so when the police chief sent out letters gauging interest in housing police dogs for a start-up canine unit, O’Brien jumped at the opportunity.

After stints in a narcotics detail and patrolling the West Side, O’Brien said he always wanted to work with dogs.

Stoki lives with O’Brien in Trumbull with the officer’s wife, two children and Ty, his 5-year-old German Shepherd. A 3-year-old, Stoki spends most of the time in a crate to limit his time with people other than his trainer and handler. Stoki is from Eastern Europe and was bred from a long line of working German shepherds. O’Brien isn’t quite sure what language the name Stoki comes from, let alone what it means.

“I Googled the heck out of it and couldn’t find any meaning for it whatsoever,” O’Brien said.

Officer Dave Dogali has a 19-month-old German shepherd named Bobi. The dog lives with his family in Oxford. On Friday, Dogali and O’Brien took their new partners to Stamford for trips to the veterinarian and some basic drug-sniffing exercises.

Next to a grassy hill outside Stamford High School on Friday, the officers doused wrapped-up towels in synthetic powder that produced scents similar to cocaine and heroin. Stoki barked immediately when O’Brien produced the towel.

Dogali threw the towel into a patch, and Stoki led O’Brien to the area, sniffing around for a few moments before grabbing the towel.

“That was all nose,” O’Brien said. “That was nice.”

William Scribner, a sergeant at the New Milford Police Department, trains police dogs through his private company, Renbar Kennels. He said 18 different police agencies in western Connecticut use canine units for patrol, drug searches and locating explosive materials.

Scribner helped Stamford police shop for the police dogs. He found Stoki and Bobi in the Czech Republic. Before coming to the United States, each was trained in the sport of Schutzhund, which teaches tracking skills and how to protect handlers, among other skills.

He said German shepherds are the breed of choice among police departments and military organizations because they are well-rounded dogs who can excel in several skills at once. He called them “jacks-of-all-trades.”

At the end of the October, Stoki and Bobi will enroll in Scribner’s eight-week training program. The graduation date for both dogs is Dec. 18.

By Jeff Morganteen

LINK/PICS

Information to protect the public against scams by Robert Firebaugh.

A stabilizing economy, companies with larger cash piles, a general need for newer technologies (and products) and most importantly, revenue growth are among the reasons that we expect to see strong technology M&A activity in the coming months, as we noted two weeks ago. Since then, three major deals totaling more than $13 billion have been announced: Dell (s DELL) is buying Perot Systems for $3.9 billion, Xerox will acquire ACS for $6.4 billion and Cisco (s CSCO) agreed to snap up Tandberg for $3 billion. So what about chip companies? Will they go shopping, too?

Today, EETimes.com came out with its list of the top 10 cash-rich chip makers, with Intel (s INTC), Qualcomm (s QCOM) and TSMC leading the charge. And revenue is on the rise again for chip firms, according to the Semiconductor Industry Association, which said today that total sales reached $19.06 billion in August. That’s up 5 percent from the month before, though still down by 16 percent from the same month last year.

So they have the cash, they have revenue growth and some of the top 10 clearly need to add new products to their existing arsenals. With that in mind, we think Intel (with $10.45 billion in cash) and Qualcomm (with $9.89 billion) will be the two that actually go shopping, though their picks might surprise us. Intel will likely keep up its software splurges rather than buy any hardware companies. In a bid to make multicore programming easier, the chip giant bought bought two compiler makers this summer. It also bought Wind River, a software maker for the embedded chip market, in an effort to get its chips inside our set-top boxes, televisions and phones.

Another hot area is that of 3-D, making companies such as Organic Motion, which makes software that tracks a person’s movement and turns it into a 3-D representation, potential targets. Those that focus on 3-D using x86-based chips could prove to be likelier buys, however, especially as Intel gears up to release Larrabee, its x86-based graphics chip. (Related Research from GigaOM Pro, subscription required: How 3-D TV Will Go From Hasseloff to Must-Have)

Less likely buys for Intel would be semiconductor equipment makers. The equipment industry is struggling, and Intel has already invested in keeping one key player alive, but it may take a step further into outright ownership, if it feels forced to do so. Intel (along with Samsung and TSMC) is also trying to get the industry to move to making chips on larger wafers, despite protest from the equipment makers that such a shift is unnecessary and would be too costly for them. Owning a company that can help jump-start the process may convince the rest of the equipment companies to follow, or it may push Intel back into becoming a vertically integrated chip shop.

Meanwhile, Qualcomm could diversify away from chips and start bulking up on infrastructure services, such as when it bought Firethorn to increase its presence in mobile banking. Or we could see it go after players such as Jasper Wireless or Kore Telematics to grow its M2M connectivity business. If it wants to pursue a mobile CDN play for its MediaFLO network, it will likely have to buy some technology to bring a CDN service together.

Readers, who do you think Intel, Qualcomm or other chip vendors should buy?

En esta miniserie trataré de resumir a grandes rasgos el transfondo cultural, económico y tecnológico que ha dado como fruto a las nuevas tecnologías y productos que comenzamos a ver en el mercado de cómputo como son: netbooks, las familias de procesadores Sun Microsystems UltraSPARC Tx, Fujitsu SPARC64 Venus, IBM Power7, Cell, Intel Atom, Larrabee y ARM Cortex-A9, GPGPU’s de ATI e Nvidia, y OpenCL.

Durante muchos años la industria de microprocesadores nos quiso vender la idea de que lo único que importaba al comprar una nueva computadora eran los Ghz que ofrecía, y desde hace unos 4 años nos vienen diciendo que lo importante ahora es el número de núcleos. Sin embargo las condiciones del mercado han traído a una nueva generación de usuarios, más preocupados por el costo, el gasto energético, la duración de la batería en equipos móviles y la velocidad con la que sus aplicaciones responden a sus comandos. Provocando una gran innovación, la creación de nuevos mercados y de nuevos productos radicales.

“El número de transistores que puede ser colocado en un chip se duplicará cada 2 años”

Gordon E. Moore

Esto es lo que nos dice la ya famosa ley de Moore, no obstante, aún no sabemos que hacer con todos estos transistores, ¿Cuál es la forma más eficiente de usar estos transistores para procesar X tipo de información? aún nadie lo sabe.

Los Ghz y el procesamiento secuencial
La máquina de Turing nos muestra una forma sencilla de escribir cualquier algoritmo, una operación después de la otra, llevándonos a lo que se conoce como programación secuencial; el paradigma de desarrollo de software más exitoso, adoptado por la industria del software a nivel mundial y donde se han invertido hasta hoy en día billones de dolares en la creación de software bajo este principio elemental.

Esto impulso a los fabricantes de chips a crear procesadores cuyo objetivo principal era ejecutar lo más rápidamente posible este tipo de programas. Todo el diseño de estos chips se basaba en su capacidad para ejecutar una o más instrucciones cada vez que un reloj central hacía un tick-tack, y buscando que cada vez, todo el sistema se moviera al ritmo de relojes más y más rápidos, llevándonos a una vertiginosa carrera por frecuencias de reloj más altas. Más Ghz.

Ejecutar este tipo de programas no resultó fácil, ni tampoco eficiente. Diseñadores de procesadores tuvieron que recurrir a toda una serie de ingeniosos métodos para ejecutar instrucciones más rápidamente: pipelines, out of order execution, caches, branch prediction, speculative execution,… que se enfocaban más en “administrar” el código que en realizar trabajo útil para el programador. Al mismo tiempo, conforme aumentaban las frecuencias de operación, así también aumentaban la temperaturas de los chips, esto provocado por los constantes cambios de estado de los componentes, al ritmo del reloj. Requiriendo continuamente soluciones de enfriamiento más eficientes.

Finalmente las altas temperaturas desaceleraron el ritmo con el que avanzaban los Ghz. Debido no solamente a las altas frecuencias, sino también a las crecientes corrientes de fuga provocadas por las escalas de fabricación tan increíblemente pequeñas de los nuevos chips, a lo que se conoció como la barrera térmica.

Esta nueva limitación provocó grandes fracasos en el mercado de microprocesadores como se vio con el Intel Pentium 4/ Tejas y forzó grandes cambios en la industria. Como podemos apreciar en la última conferencia de Hot Chips 2009, el procesamiento secuencial a muerto, la última linea de procesadores enfocándose en este tipo de arquitecturas, IBM Power, ha preferido dar un cambio de rumbo e ir en la dirección del Throughput-Computing.

Durante o Intel Developer Forum, foi finalmente revelado um pouquinho do potencial na prática do futuro processador gráfico da Intel: o Larrabee. Sean Maloney, vice-presidente de arquitetura da empresa, foi quem apresentou uma demonstração gráfica utilizando-se do Larrabee, que você pode ver no decorrer do vídeo abaixo:

Trata-se de uma bonita cena em 3D, não? A renderização foi feita através de ray-tracing, em modo nativo. Bom, agora resta aguardar mais informações, pois nada mais foi falado sobre o Larrabee, muito menos data de lançamento, até então.

Segundo a Intel, alguns desenvolvedores já receberam a tecnologia para testes iniciais.

Fonte: Adrenaline

Hard-disk drives will hold back the performance benefits of Intel's upcoming high-end graphics chip Larrabee

With the first generation of Intel’s high-end graphics chip – code named Larrabee – only a few months away from shipping, Intel is trying to set some expectations in regards of its performance.

Speaking at the Flash Memory Summit in Santa Clara, Calif., Intel’s desktop “performance guy” François Piednoël pointed that the performance and the user experience of a Larrabee-based PC will be “less interesting” with a hard-disk drive installed, highly recommending a solid-state drive (SDD) instead.

Hard-disk drives are what is holding PC performance back

Luckily, it so happened that the Santa Clara, Calif., chipmaker had recently shipped its next-generation 80GB SSD for about $220. At this price, Piednoël argues, consumers will begin to pay attention on this new category of devices.

“On a lot of benchmarks you will get more performance benefits to put an SSD into a laptop than having a discrete graphics card… This accelerate dramatically the user experience.”

A statement that I largely agree with having experienced a jump of performance after I installed a Kingston’s SSD on a MacBook used for video editing. The jump in performance was immediately noticeable and would have only been possible if I had upgraded the processor and/or the Nvidia discrete graphics.

With that in mind, Intel’s heavy investment in SSD technologies makes a lot more sense. “The CPU and the SSD are intimately linked. The faster the SSD is, the more you need processing power, and vice-versa.” CQFD.

Follows 2 video excerpts of Piednoël’s presentation at the Flash Memory Summit:

And why Intel is so interested in SSDs:

Can AMD change the game in graphics with its upcoming Evergreen GPUs? We'll know on September 10th!

[Update 1] We’ve got confirmation from AMD that the Evergreen cards are being shown this weekend, at the QuakeCon video game convention in Dallas, Texas.

[Update 2] AMD will host its Evergreen’s official launch on aircraft carrier U.S.S. Hornet moored in Alameda, Calif.

The word is finally out. AMD will launch it’s much anticipated next generation graphics processors code name “Evergreen” on September 10th; ahead of Windows 7’s launch in late October..

For AMD, these 40-nm, Microsoft DX11-compliant GPUs will fundamentally change the graphics industry and give it a clear advantage over Nvidia, again!

The prior generation of ATI cards was such high performance and so cheap that they forced Nvidia to hastily put together competitive video cards.

Sadly, AMD’s Santa Clara, Calif., rival hasn’t shown much of its DX-11 chips yet. However, Nvidia might choose to show off its wares at its own GPU Technology conference at the end of September in San Jose.

The GPU market is finally kicking some tires, just in time for the holiday season!

Early Fort Humboldt, established to address the "Indian problem"

Recently blog  reader Ernie B    observed  HERE  that at least some of the perpetrators involved in the massacres may have been coerced into doing so. 

 “… As you continue your research you will find that rule by intimidation was very prevalent. Many thugs had henchmen to do their dirty work. People that crossed the thugs were shot, and witnesses would swear that it was self defense. Many small ranchers were poisoned with strychnine…

It is not a stretch to think that the people that joined Larrabee in the Indian Island massacre were there through intimidation, or the desire to prove themselves to be one of the gang. Very much similar to some motorcycle gangs of today…

Most likely the people that were helping Larrabee were intimidated into being there, and that is why we don’t know who they were. They didn’t want anybody to know. Just a educated guess. “ ~Ernie B, 8 Aug 2009

Note  to readers unfamiliar with our history:   Hank Larrabee was often named in military and other documents as a perpetrator of the Indian Island massacre and other atrocities against the natives.

Ernie may be  on to something here… one of the most striking things about the massacre was the lack of justice demanded by the public on behalf of those murdered.   The perpetrators, though apparently known by at least a few, were never called to answer for their crimes.  It would be easy to assume that the  residents around the Bay didn’t care enough to bother…. but perhaps there was another reason.  

After  the massacre, the following letter was sent to the editor of the San Francisco Bulletin:

”Having lately arrived here from Humboldt Bay, I take the opportunity to inform the public…  of a few of the recent instances of shameful and horrible crime committed upon the Indians  in Humboldt county by white men (here the author describes the February massacres and then)…  Some time about the 18th March last, three desperate ruffians, armed with hatchets, entered the hotel at Hydesville, and demanded of the proprietor by what authority he had written a letter to Liuet. Hardcastel, of the U.S.A. at Fort Humboldt,  and if he had not convinced said ruffians that the letter was strictly private, and had no allusions to Indian affairs, and no communications for the Bulletin, he would have been assassinated on the spot. The names of these ruffians I shall withhold for the present.   Society is completely demoralized on Eel River; and the Thugs are largely in the majority, led on by Wiley of the Humboldt Times, and by Van Nest the Sheriff. Young men talk and think of nothing else but hanging and killing young Diggers and their mothers.  The pulpit is silent, and the preachers say not a word.  In fact, they dare not…  Men who detest and abhor the thugging system, from circumstances which surround them, are silent.  Two or three men who were on the last Grand Jury which sat at Eureka were thugs. … I append my name, privately , to this record of some of the atrocious deeds that have recently been perpetrated in Humboldt county. I have left that quarter for good; but as I have a few friends in the place, I do not wish that they should be molested for any doings of mine, and you had better, therefore, not communicate my name, except under such circumstances as you may consider necessary or proper for the public good. [Daily Evening Bulletin-: San Francisco, June 1, 1860]

It is unfortunate for us that the Bulletin editor chose not to make the writer’s name public and that the writer declined to name the thugs, though both decisions were probably fortunate for the author and his friends.  This was an ugly time….  and though I would love to judge those who refused to advocate on behalf of the Natives,  I do know it wouldn’t be fair.

My family recently watched Pale Rider, with Clint Eastwood, which was a typical western story, gold rush, good guys, bad guys who have the local law in their back pocket, and a hero that comes in and saves the town.

Kinda similar here.  Except it looks like perhaps the bad guys were the law… and no good guys came in to save the Indians.

nVidia (NVDA) is reporting earnings tomorrow.  Some analysts expect them to beat estimates and give pretty good guidance.  I have no idea about how they are likely to do over the short-term, about 6-9 months, but over the long-term, I think they will do quite poorly.  I first discuss the short-term, and then the long-term.

Short-term.  nVidia competes with AMD in the discrete PC graphics business.  The current offerings are competitive, and over the last 6+ months ASPs have been declining.  In addition, according to Mercury Research, AMD gained a bit in market share at nVidia’s expense.  nVidia’s quarter ended near the end of July (vs. end of June for AMD).  Although the ASP of discrete graphics products continued to decline in July, it was a better month than previous ones.

nVidia also makes an integrated graphics chipset (sometimes called ION and sometimes 9400M) that competes with Intel chipsets in both consumer PCs and Netbooks.  This part has gained design wins, and should be gaining in quarterly volume/revenue.  This growth is likely to continue for the next 6 months.

Looking at the rest of this year, there are 2 factors that will affect nVidia – one positive and one negative.  The negative one is that AMD is set to announce a slew of 40nm discrete graphics parts on September 10th that will support DX11, the new Windows 7 graphics interface.  AMD seems to be ahead of nVidia in the 40nm ramp and supporting DX11.  nVidia’s new high-end graphics part, which some label GT300 and supports DX11 is unlikely to be available this year.

The positive factor for nVidia’s (and AMD’s) discrete graphics business comes in September with Intel’s launch of Lynnfield (for desktop) and Clarksfield (for mobile).  These are 4-core Nehalem CPUs, but will have a more cost-efficient platform architecture (with the P55 chipset) than the current Nehalem one.  This platform does not come with an integrated graphics solution from Intel.  So all P55 chipset PCs this year will need to come with either an nVidia or AMD graphics processor.

So, if we assume a decent back-to-school and Xmas selling season, nVidia should do reasonably well with its ION/9400M integrated graphics (for Core 2 and Atom platforms), and with its discrete graphics chips over the next 6 months.

Long-term.  In early 2010, we have the beginning of a MAJOR change in the competitive environment.  In late 2009, Intel starts its ramp of 32nm with 2 key processors, Arrandale (mobile) and Clarksdale (desktop).  They are 2-core versions of the Nehalem microarchitecture.  Both come with integrated graphics.  To be precise, if you looked inside one of the component packages for these, you would actually find 2 die: one the processor, and the other a 45nm graphics processor.  Of course, if one wants higher graphics performance than that offered with these, it is still possible to attach a discrete graphics processor via the PCIe bus.  Intel has not disclosed the performance of its integrated graphics processor, but it is reasonable to assume that it is at least 2X that of previous versions.

The major significance of the above for nVidia is twofold: (1) as the Nehalem generation replaces the Core 2 generation, the market for an nVidia integrated graphics chipset (like Ion/9400M) diminishes; and, (2) assuming that the performance of the Intel integrated graphics is 2+X better than previous generations, the TAM (total available market) for  discrete solutions decreases.  The latter will also affect AMD’s discrete graphics business.

In Netbooks, we have a similar development.  In late 2009, Intel starts shipping its 2^nd generation of Atom, referred to as Moorestown or Pine Trail or Pineview.  A good non-Intel overview can be found here (AnandTech: Intel Unveils Next-Generation Atom Details).  This 2^nd generation Atom integrates graphics and memory controller onto the processor chip, Pineview.  The 2^nd generation Atom platform reduces idle power by 50X, and active power by more than 50%.  Also, since Pineview is still on 45nm, it can quickly ramp to replace the 1^st generation Atom.  So the modest success to date that the nVidia ION has had in Atom platforms should be quickly reversed by mid-2010.

In ~1Q2010 (+ or – 1 quarter), Intel’s Larrabee is scheduled to ship.  nVidia’s high-end GPUs (graphics processor units) go beyond the normal role of a graphics processor.  They are programmable, and can be used to accelerate various other applications, e.g. video compression, 3D rendering in movie special effects production, seismic modeling, etc. As such, they are also referred to as GPGPUs (general-purpose computing on GPUs).  nVidia gangs up multiple of these GPGPUs to provide computing platforms for such high-compute intensive applications in their Tesla product line.  Like nVidia’s line, Larrabee will be targeting the same markets, high-end graphics for PCs and workstations, and high-performance specialized computing platforms.  Intel started talking about Larrabee almost 2 years ago, but more intensely about one year ago (e.g. check out Intel, Dreamworks Animation Form Strategic Alliance to Revolutionize 3-D Filmmaking Technology, and First Details on a Future Intel Design Codenamed ‘Larrabee’).

So, in the same timeframe that nVidia will be coming out with its next generation high-end GPGPU, the GT300, it will have a new competitor, Intel.  Will this be a game-changer?  It is too early to judge.  Also, since this is Intel’s 1^st effort in this kind of architecture, it would be unreasonable to expect a home run.  But I have little doubt about the eventual end-game, and it is not a good one for nVidia.

Now lets look at one final element.  Toward the end of 2010, Intel’s next generation x86 microarchitecture is due out, Sandy Bridge, on 32nm.  One persistent rumor on the net is that the 1^st implementation will incorporate graphics on the processor die.  Apparently, a French site (canardpc.com) came into possession of an annotated Sandy Bridge die photo.  I have no idea if this is real or a fake.  (Type in “Sandy bridge die photo” into google, without the quotes.)  But if true, the graphics die area is just a little less than that of 2 Sandy Bridge CPU cores, which implies another significant jump in graphics performance.

To summarize, in 2010, the available TAM for nVidia’s integrated graphics chipsets decreases over time.  By mid-2011, if not sooner, it will be ~0.  Also, the TAM for discrete graphics (starting in 2010) will decrease over time, but not as dramatically, due to significant improvement in Intel’s graphics integrated with the processor.  And then starting in ~1Q2010, AMD and nVidia will have a new competitor in Intel’s Larrabee. Over the next 2 to 3 years, Intel could become the market leader.

2 esoteric footnotes:

1. Intel and nVidia are in a legal dispute as to whether nVidia has a right to produce chipsets for Intel’s Nehalem platform, under their current license agreement.  I can’t comment on this because: (1) I have not seen the agreement; and, (2) I am not a lawyer.  But regardless of the legal issues, an integrated graphics chipset for a Nehalem platform makes no sense to me.  Conceivably, an nVidia chipset (without graphics) that competes with Intel’s P55 chipset would make technical sense, but the margin (without graphics) would be small (but some Intel chipset competition would be good).
2. nVidia is getting into the ultra-mobility segment (cell-phones, digital media players, MIDs, Smartbooks) with its Tegra product line.  This uses an ARM CPU architecture – not an x86 one.  Tegra is going to be used in  Microsoft’s HD Zune.  In June, nVidia claimed a total of 42 design wins.  So perhaps, Tegra is really the long-term future of nVidia.  However, I am skeptical.  There are many better ARM designs currently, eg, Qualcomm’s Snapdragon.  There are a lot of competitors in this space, and margins are likely to be thin.

Disclosure: I have a short position in nVidia.  But I may be as much as 6 months too early.

Remembering that the Intel Itanium was supposed to be a ground-breaking departure with the past, can Larrabee be all that and more for graphics? Itanium is still not what Intel had hoped. And poor early adopters are still buying new and vastly over-priced minor incremental revs of the same CPU architecture to this day. Given the delays (2011 is now the release date) and it’s size (650mm^2) how is Intel every going to make this project a success. It seems bound for the the Big Fail heap of the future as it bears uncanny resemblances to Itanium and the Intel i740 graphics architecture. The chips is far too big and the release date way to far into the future to keep up with developments at nVidia and AMD. They are not going to stand still waiting for the behemoth to release to manufacturing. I just don’t know how Larrabee is ever going to be successful. It took so long to release the i740, that the market for low end graphics GPUs had eroded to the point where Intel could only sell it for the measly price of $35 per card, and even then no one bought it.

According to current known information, our source indicated that Larrabee may end up being quite a big chip–literally. In fact,we were informed that Larrabee may be close to 650mm square die, and to be produced at 45nm. “If those measurements are normalized to match Nvidia’s GT200 core, then Larrabee would be roughly 971mm squared,” said our source–hefty indeed. This is of course, an assumption that Intel will be producing Larrabee on a 45nm core.

via Intel’s ‘Larrabee’ to Be “Huge” – Tom’s Hardware.

Livermore, ME, June 1, 2009 – The Coombs Motorsports Ford Fusion was back in action this Saturday at Oxford Plains Speedway. After overcoming a rough heat and slow cars in the feature Coombs piloted the 57 into a solid ninth place finish.

The day started off well for the Coombs Motorsports team running consistently in the top ten for two practice sessions. In the final practice, the team chose to scuff race tires and ended up 12’th on the speedcharts.

This was also the second week that Coombs Motorsports had to start at the rear of the field for not racing last season. This placed the team deep in the first heat with many new drivers. On the start, the car running in the third spot lost control and spun in the middle of turn two. Most of the drivers checked up to avoid the wreck, except the 48 driven by Kenny Harrison who made hard contact with rear end of the Coombs Motorsports Ford.

On lap three, the 57 made its move to the outside, but the position would be short lived as another car forced a three-wide situation in turn one sending Coombs off the track. Since the team would have to start from the rear of the feature anyway, Coombs elected to bring it in and not deal with the poor driving exhibited in the heat.

Coombs Motorsports started the feature in the 15’th position. Once the green flag fell, it was obvious that the 27 driven by Jason Larrabee was holding up the entire back half of the field. Larrabee also seemed out of control letting the car slide up into the third groove of the track on both the front and back straights of the speedway. After letting off once as Larrabee chopped down into turn one, Coombs decided he would make his move on the inside of the 27. Coombs was up to the door of the 27 when Larrabee decided he wanted the inside. Larrabee made contact with the front of the 57 and immediately lost control of his car relegating him to the back of the field.

With the slow cars out of the way, Coombs set his sights on the front runners. He reached Donnie Wentworth and Shawn Knight within a few laps and set up to pass them on the outside. Coombs made easy work of Wentworth and tried to overcome the 25 of Knight.

Coombs and Wentworth soon found themselves held up behind the 25. Wentworth would make the first move clearing Coombs, who slid up behind him to help overcome Knight. When the checkered flag flew, Wentworth was the only one who could overcome Knight placing Coombs in the ninth position for the second consecutive week.

Coombs Motorsports will be taking the weekend off from racing but will be back in action at Oxford Plains Speedway on June 13.

Larrabee saab olema 32-tuumaga protsessor, mis on vastuseks NVidia ja AMD lahendustele sel alal. Larrabee südameks on tavalistest keskprotsessoritest tuntud x86-käsustikuga tuumad ja iga tuum on paaris vektorprotsessoriga. Larrabee tuumad jagavad ühist suurt vahemälu ja protsessorikivi servadel on ühendused põhimäluga. Inteli sponsoreeritud instituuti Saksamaal Saarlandi üikoolis avades ütles Joseph Schultz, Inteli mikroprotsessorite tehnioloogia laboridirektor, et Larrabee’d võib oodata järgmise aasta esimeses pooles. Inteli teatel on Larrabee’d võimalik kergemini ümberprogrammeerida muud tegema kui tänaseid graafikaprotsessoreid. Firma toob 26. mail välja spetsiaalse tarkvaraarendusvahendi Parallel Studio, mis aitab graafikaprogramme kergemini paralleelseks muuta ja toimib koos Microsoft Visual Studio paketiga. Ilmselt hakkab Intel tootma Larrabeed varieeruva arvu tuumadega – võibolla 8, 16, 32, 48 ja 64 tuumaga. Esialgu kasutatakse praegust 45 nm tootmisprotsessi, hiljem võidakse ehk üle minna 32nm tootmisele, mis looks võimaluse ka veelgi suurendada protsessorituumade arvu.

Täpsemalt võib lugeda TheRegisterist.