|
 |  TG Daily -
Nov-04-2009AMD discovers several Fusion mantras, disses Larrabee(topic overview)
CONTENTS:
- First and foremost among the reasons cited for NVIDIA's alleged plans is the fact that we're about to make another turn on Sutherland's wheel of reincarnation, where graphics functionality will move back onto the CPU die. (More...)
- It's the NVIDIA rumor that won't die: no, not the one where the GPU maker buys tiny VIA, but the other one, where it jumps feet-first into competition with both Intel and AMD by producing an x86 processor of its own. (More...)
- At the tail end of last week, AMD's senior manager of developer relations, Richard Huddy, told HEXUS that he believes NVIDIA is "somewhat abandoning the gaming market." (More...)
- Will Nvidia enter the CPU market? At least one analyst thinks so. (More...)
- Codenamed Northern Island, HD 6000 will be a 32nm design. (More...)
- Probably the most exciting CPU of the bunch is AMD's upcoming hexa-core (6 core) CPU. Codenamed Thuban, it will be a 45nm part, guessed to be clocked at about 2.8GHz. (More...)
- I mean, seriously, HD 2000, 3000, and 4000 all only lasted one year. (More...)
SOURCES
FIND OUT MORE ON THIS SUBJECT
First and foremost among the reasons cited for NVIDIA's alleged plans is the fact that we're about to make another turn on Sutherland's wheel of reincarnation, where graphics functionality will move back onto the CPU die. When this happens, NVIDIA's lucrative integrated graphics processor business is, of course, toast, which has to be one reason why the company went ahead and euthanized it a bit early. This turn of the wheel is interesting because it's not just that the integrated graphics processors (IGP) will move from the chipset to the CPU, but retain their same basic degree of specialization. Both the integrated GPUs that go onto the CPU die and the more discrete models are essentially becoming many-core, general-purpose processors (albeit specialized for multithreaded throughput and floating-point). This makes the situation even more ominous for NVIDIA, because not only will Intel and AMD take their IGP market from them, but even in the discrete GPU market NVIDIA will also end up with a generalized processor that competes with x86. The ultimate point is that in both the discrete and integrated GPU markets, NVIDIA is already destined to compete directly with Intel and AMD, which means that an actual x86 product from NVIDIA means only that NVIDIA has decided to fight x86 with x86, as opposed to fighting it with some non-x86 architecture. The fact that NVIDIA has hired a bunch of ex-Transmeta guys doesn't necessarily mean anything at all for this supposed x86 processor. As the launch of Apple's Snow Leopard, with its pervasive use of LLVM, makes clear, just-in-time compilation (JIT) is the way that everyone is going to tackle the many-core problem, at least in the near-term. [1] "The question is not so much I think if; I think the question is when," Bit-Tech reported. There is an inevitable trend toward merging the CPU and GPU. Intel calls this its Larrabee project (although its first graphics chip appears to be a more traditional standalone GPU for workstations), while AMD refers to it as "Fusion," and will launch it in product form in the second half of 2010. Freedman said he believes that Nvidia would develop such a CPU in-house - naturally, since Intel and AMD own most of the available X86 patents (and share them through a cross-licensing agreement). GlobalFoundries, AMD's spinoff, presumably has access to that license, although Intel has already notified both companies that that's not the case.[2]
The story asserts that Nvidia has been quietly hiring former employees of Transmeta, a now-defunct producer of x86-based processors. The EETimes story is based on a report today by Broadpoint.Amtech analyst Doug Freedman, who asserts that the company "could become a supplier of x86 CPUs by necessity to preserve both GPU and chipset revenue." He says the company is more likely to develop the technology internally than to acquire a company with the capability, like VIA Technologies. "We believe Nvidia could (sooner rather than later) enter the x86 CPU business, with investor response largely a function of the market share opportunity given the CPU roadmap range," he writes.[3] Nvidia (Santa Clara, Calif.) has been quietly hiring former employees of Transmeta, a now-defunct, x86-based processor supplier. ''We believe internally developed x86 solutions are more likely than external acquisitions (i.e. Via Technologies),'' he said in a new report, referring to rumors that Nvidia would acquire Taiwan's Via. ''We believe that Nvidia has hired former Transmeta staff extensively, and that instruction code "morphing" requirements have declined as more x86 instructions have come off of patent coverage,'' he said. Nvidia is also set to report its results. ''For October, we expect revenue and GAAP EPS (for Nvidia) to beat our estimates of $846.4 million and $0.06, in-line with the Street at $835.2 million and $0.06,'' he said in the report. ''We believe revenue growth expectations may be as high as up 14 percent quarter-over-quarter, or close to $885 million given the strong sales reports from both Intel and AMD.''[4]
Broadpoint AmTech analyst Doug Freedman today revived talk of NVIDIA making its own x86 processors in a new research note. He claims that the graphics company has been recruiting "extensively" from employees of the one-time chip designer Transmeta and is using these to speed along one or more unknown designs. It may also not need to use Transmeta-like code translation as many of the original x86 architecture patents lose their effectiveness.[5] The idea has cropped up again in an analyst note from Doug Freedman of Broadpoint AmTech, in which Freedman claims that NVIDIA has been hiring former Transmeta engineers to work on a secret x86 processor that will appear sooner rather than later.[1]
It's the NVIDIA rumor that won't die: no, not the one where the GPU maker buys tiny VIA, but the other one, where it jumps feet-first into competition with both Intel and AMD by producing an x86 processor of its own. [1] Speaking on behalf of the GPU giants, AMD's senior manager of developer relations, Richard Huddy, and NVIDIA's senior technical marketing manager of EMEA, Lars Weinand, are at it again. This time around, AMD is once again accusing NVIDIA of preventing the use of anti-aliasing in Eidos' Batman: Arkham Asylum on an ATI graphics card, a subject first raised, to widespread furore, back in September, by AMD's senior manager of advanced marketing, Ian McNaughton. NVIDIA claims "AMD just did not do their work", adding that "it is up to Eidos to decide the fate of a feature that AMD refused to contribute to". Want to know what AMD's Huddy had to say in response? Check out the comments from both NVIDIA (in green) and AMD (in red), below.[6] I'm surprised and pleased by authorised NVIDIA spokesperson Lars Weinand's clarification that "Batman AA is not our property. It is owned by Eidos. It is up to Eidos to decide the fate of a feature that AMD refused to contribute too and QA for their customers, not NVIDIA." AMD received an email dated Sept 29th at 5:22pm from Mr. Lee Singleton General Manager at Eidos Game Studios who stated that Eidos' legal department is preventing Eidos from allowing ATI cards to run in-game antialiasing in Batman Arkham Asylum due to NVIDIA IP ownership issues over the antialiasing code, and that they are not permitted to remove the vendor ID filter.[6]
I would encourage Mr. Singleton at Eidos to move quickly and decisively to remove NVIDIA's vendor ID detection. It's also worth noting here that AMD have made efforts both pre-release and post-release to allow Eidos to enable the in-game antialiasing code - there was no refusal on AMD's part to enable in game AA IP in a timely manner.[6]
Is Fermi worth the wait. With your comment regarding locking DX11, do you try to indicate that AMD invented DX11 and could have been an AMD-only feature?? DirectX 11 is a new version of DirectX, that will be fully supported by Fermi, as we announced at GTC. It seems that AMD tries to create the perception that DX11 is a AMD only feature. It is not. Long story short and no argument coming from AMD will change this: NVIDIA loves games.[6] AMD is already working with games developers on over 20 forthcoming games which feature DX11 tech.[6]
At the tail end of last week, AMD's senior manager of developer relations, Richard Huddy, told HEXUS that he believes NVIDIA is "somewhat abandoning the gaming market." Such a sweeping statement isn't likely to go unnoticed by the green team over in Santa Clara, California, and NVIDIA's senior technical marketing manager of EMEA, Lars Weinand, has responded to AMD's remark in our very own HEXUS.community forums. [7] "In addition to the technical challenges of supporting an x86 product family, we believe Nvidia's fabless model is a headwind for manufacturing leadership in mainstream CPUs, namely 1) TSMC and other foundries may be unable to provide the most leading-edge geometries, 2) Foundries do not have processes optimized for high-yield, high-volume CPU manufacturing (though the Atom-license by TSMC is a step in the right direction, 3) Die size and performance may not be fully optimized by foundries as manufacturing investments address a broader range of semiconductor customers' needs, 4) Platform development costs are likely to be underestimated, and 5) Nvidia's brand recognition outside of the gaming community may be less than expected. Is this a possibility? Nvidia reports its quarterly earnings on Nov. 5, and it's likely Freedman will address the possibility in the subsequent question-and-answer session following the call. Personally, I'd expect Nvidia Jen-Hsun Huang to characterize the analysis as "interesting," followed by a short explanation of the company's decision to exit the chipset market and a hope that both Itel and Nvidia will resolve its differences amicably.[2] In 2004, Intel and Nvidia also signed a cross-licensing agreement, although the terms are considerably more vague. Nvidia could use something akin to the Transmeta code-morphing technology that would interpret X86 code, and transform it into its own native instruction set, or the company could buy the remaining X86 supplier, Via Technologies, a choice Freedman said was unlikely. "We believe internally developed x86 solutions are more likely than external acquisitions (i.e. VIA Technologies)," he wrote.[2]
An in-house x86 design may also be necessary for NVIDIA in the future if it hopes to continue developing and promoting integrated graphics. The company is embroiled in countering lawsuits as part of a dispute over NVIDIA's license to make chipsets for Intel processors and, without a settlement, can't make such hardware for Core i5, Core i7 and most other Intel processors with integrated memory controllers.[5] Rumors have resurfaced about Nvidia (NVDA) taking on Intel (INTC) and Advanced Micro Devices (AMD) in the x86-based microprocessor market, according to EETimes. The trade magazine's Web site says that that have been reports "for some time" that NVDA might move into the microprocessor sector to protect its core graphics chip business, which is now coming under threat from Intel and AMD, which are developing processors with more graphics capability.[3] Graphics chip vendor Nvidia has halted development of chipsets for next-generation Intel processors that feature the direct media interface (DMI) bus, pending the outcome of current litigation between the two companies.[4]
Fermi is an awesome graphics processor and we're confident that it will let us keep the performance crown. We will talk about this side of Fermi very soon. Fermi is an entirely new architecture with many new features specifically for the compute space, so it was important to us that we talk about them first.[6] AMD has a heap of new mantras but Fusion won'''t be released until 2011, when graphics processor and CPU end up on the same die and the interconnect is thus made far easier. He said that GPU scaling gives 2X performance every 18 months and AMD is not slowing down in CPU perfrmance either. The GPU gives the best performance per watt per dollar, he said.[8]
In terms of graphics, AMD's Evergreen series - already productised in part as the Radeon HD 5800 series - will remain prevalent throughout the year, and won't be superseded by a newer architecture until 2011. It isn't until 2011 that AMD's desktop roadmap begins to get interesting. That's the year in which we'll see Scorpius, a next-gen platform making use of the 32nm Zambezi processor - which, we can only hope, is based on the long-awaited Bulldozer core, successor to the K10 architecture.[9] The HD 5700 is simply the most sensible choice if you're looking for a future proof card but don't want to spend a fortune on 5800 cards or Fermi. In the U.S., it's just as hard to find any 5800s, and the same goes for HD 5700 series cards, so we won't even talk about North American pricing. AMD still has a strong lead on Nvidia, as there's little to no chance that the green team will manage to ship its Fermi based DirectX 11 products in volume by the end of the year. AMD doesn't seem to be exploiting its advantage as it just can't cope with demand. Although Q4 will be very bad for Nvidia, it won't be as good as it could have been for AMD's graphics division either as it is failing to capitalize its lead.[10]
NVIDIA is late to deliver DirectX 11 hardware to market. I agree with Mr. Huddy on this one. We launch later than AMD does in this case. Fermi is the world's first computational GPU architecture, with several world's firsts on the GPU. These take time to design and perfect.[6] Arguing, as NVIDIA did, that AMD working with Codemasters to add DX11 to DiRT2 is harming gamers is another. NVIDIA's behaviour in locking something as trivial as antialiasing to its own hardware (in Batman Arkham Asylum) shows that NVIDIA cares much more about money then gamers.[7] To address the real question, however: could Nvidia pull off an internally developed X86 CPU? I just can't see how, at least for the PC. Certainly, Nvidia could develop an X86 CPU, with some modicum of performance. In that case, both AMD and Intel would likely drown it out with their own integrated solutions.[2] As for whether NVIDIA is actually planning an x86 CPU, I have no idea. As I always say when this rumor crops up, the scuttlebutt from journalists who have sources is that NVIDIA is indeed going to bring such a beast to market. I don't have any sourcesjust a bit of senseand I think that anyone who jumps into the x86 market at this point is completely nuts, especially if they decide to roll their own instead of buying VIA.[1]
Will Nvidia enter the CPU market? At least one analyst thinks so. In a note sent to clients on Tuesday, Broadpoint AmTech analyst Doug Freedman aid that Nvidia could be forced into the CPU market by necessity, to preserve both GPU and chipset revenue. [2] According to Freedman, that chipset revenue represented about 30 percent of Nvidia's sales, revenue that the company would need to make up.[2]
"Nvidia will likely target mainstream and below performance, and NOT leading core-i7 like levels." In October, Nvidia confirmed that the company was halting development of its nForce chipset line, as an ongoing legal battle with Intel provided a roadblock to further design work.[2] I would predict that, within three weeks, Intel would discover some critical patent that Intel's lawyers would charge Nvidia with infringing. Is that a risk that Huang would accept? Actually, I believe that he would. He might just be betting his company while doing it. Editor's Note: This story was updated at 1:20 PM with comments from Nvidia and additional details from Bit-Tech. Mixx It Digg It More.[2]
An Nvidia representative declined to comment. Bit-Tech quoted Nvidia Michael Hara earlier this year, however, as saying that Nvidia could apply its Tegra integrated chip strategy to an X86 processor.[2] NVIDIA has hinted it would like to use x86 for future system-on-a-chip processors but hasn't set out any definitive schedules or product plans. Its Tegra chip used in the Zune HD is based on ARM and runs quickly but is incompatible with most variants of Linux and Windows.[5] The notion that NVIDIA might produce a non-x86 processor but use binary translation to sell it as an x86 competitor appears completely ridiculous. Now, this doesn't mean it won't happenit just means that if it does, it'll fail miserably and we'll all have a laugh.[1]
"We believe that Nvidia has hired former Transmeta staff extensively, and that instruction code "morphing" requirements have declined as more x86 instructions have come off of patent coverage." Freedman also lists "just a few challenges" to such a move.[2] Because the software side of the GPU is already headed towards a heavy reliance on binary translation techniques, it makes sense that NVIDIA is scooping up Transmeta guys regardless of any plans to produce a CPU. So, again, the Transmeta hires don't necessarily augur much of anything.[1] As the owner of a very expensive paperweight, formerly known as an HP laptop computer with Nvidia GPU and chipset, I can only hope that the Nvidia name will be prominently featured on any computers using their CPU.[3]
Our Devtech team is working closely with game develeopers all over the world, to ensure best support and compatibility of our GPUs for the latest games. If you're interrested to hear more on how NVIDIA works with game delvelopers, please feel free to watch the latest comments from Tony Tamasi on this topic.[7] As long as in game physics takes a step forward, we are happy, regardless of the path the developers chooses to get there. We support open standards, plus standards that allow NVIDIA to offer new innovations to customers well in advance of industry standards, such as CUDA C. Our goal is to lead the industry in new amazing directions and create value for our customers, which is exactly what PhysX has done.[6] I note from the most recent NPD sales figures that "Batman AA" figures at 96th place in the PC charts and yet that seems to be NVIDIA's ' showcase' for PhysX. I suspect gaming physics will be better adopted when as an industry we move away from the divisive proprietary standards that Lars advocates so heavily. Regarding your comment of "little traction", PhysX sure has managed to get the attention of AMD and their customers.[6] The debate is raging, and the beauty of the HEXUS.community forums is that the everyday gamer can take part and voice his/her opinion. Want to have your say or respond to the comments made by representatives of both AMD and NVIDIA? Head on over to the HEXUS.community forums.[7] Arguing, as NVIDIA did, that AMD working with Codemasters to add DX11 to DiRT2 is harming gamers is another.[6]
As Eidos confirmed (Not allowed to post links here, but check PCper for Eidos' statement) AMD refused the same opportunity to support gamers with AA on AMD GPUs.[6] I don't quite understand that proprietary argument. AMD was working with the Physics engine Havok, which would be a "proprietary" engine by that definition as well - since it is owned by Intel. What's the status there btw? Is GPU support for AMD GPUs coming in Havok in the near future? Or was it maybe only a single hardcoded demo without the involvement of Havok? Last but not least CAL also was proprietary as a language by that definition.[6]
In terms of desktop platforms, it looks as though AMD won't be able to deliver 32nm CPUs until 2011 - putting it roughly a year behind rival Intel, who plans to introduce 32nm parts in the first quarter of 2010.[9] The 32nm Llano Fusion APU that's destined to serve mainstream desktops will also power high-end notebooks, whilst Bobcat - a processor designed to compete in the sub-20W space - will power ultra portables as part of the Brazos platform. That's a whole lot of codenames to sink your teeth into, but it's a good look at what AMD will have to offer in the coming years.[9] Looking at the coming year, AMD's challenger in the performance stakes will be the upcoming hexa-core Thuban processor, apparently clocked at 2.8GHz and coupled with an RD890+SB8xx chipset.[9]
He'''s not even bothered about ARM, which has an office just down the road from the Lone Star AMD campus. He doubts that ARM will go down the X86 path. Said Bergman: '''We'''ll have a breadth of Fusion solutions and one of them will be a quad core.''' That'''s going to give good X86 performance, he reckons. In between now and 2011? AMD has several tricks up its sleeve on the big core little core front. In 2011, he said, Ontario will be a very compelling solution.[8] Even when you know at compile-time that your target architecture is x86, you probably don't know the number of x86 cores that users will have available, so you can delay some parts of the compilation process to runtime so that the output precisely fits the target machine's hardware.[1] If you're already using JIT by default to map code to hardware at runtime, then you can just target non-x86 cores as well, should any be present. This is how Snow Leopard is able to take advantage of GPU hardware alongside multicore x86 for general-purpose computing tasks.[1]
A new analyst note has revived a rumor that's a perennial hardware site favorite: the fabled NVIDIA x86 skunkworks project.[1] Before unpacking the rumor, let's lay out the full rationale for the "NVIDIA goes x86, competes with Intel head-on" idea.[1] In the note seen by the EE Times, Freedman emphasizes that NVIDIA not will take on Intel's Core i5/i7 lineup.[1] "Nvidia will likely target mainstream and below performance," and not more powerful chips. Freedman maintains his Neutral rating on the stock, but cuts his price target to $14, from $17.[3]
For some time, there have been reports that Nvidia would enter the x86-based fray to protect its bread-and-butter graphics chip business.[4] Pairing an inferior CPU with Nvidia's powerful graphics technology could create an unbalanced architecture.[2]
Physics in games stagnated when it fell to the CPU vendors like AMD, and has seen a resurgent when GPU vendors got involved.[6] Bullet is running on CUDA for over a year and is using many open source GPU algorithms from our CUDA and OpenCL SDKs. Technologies like 3D Vision and PhysX are unique features and technologies made by gamers for gamers and we're not intending to change this.[7] PhysX comes as a free feature for GeForce users - from GeForce 8 onwards. It does not cost a penny extra and you can turn it off if you do not like it for whatever reason.[6]
Anything that makes the PC gaming experience better is a great thing. This is also why we focus on adding things like PhysX and 3D Vision to PC games.[6] Our next generation solution will follow in that tradion: for games with full DirectX 11, 3DVision and PhysX Support as well as GPU computing with support for OpenCL, Direct Compute, Cuda C, Cuda C++ and more.[7]
Never missing a beat, AMD's Huddy fired back in a little over 24 hours, further suggesting that NVIDIA is "moving its focus away from gaming", before concluding that "GPU leadership is presently AMD's".[7] We did not lock anything out. AMD just did not do their work. This happened with previous UE3 engine titles before, where ATI owners had to rename the executable to make AA work on that title (Bioshock in example). It's not NVIDIA to blame here.[6] NVIDIA has done the right thing in bowing to public pressure to renounce anti-competitive sponsorship practices and given Eidos a clear mandate to remove the vendor ID detect code that is unfairly preventing many of Eidos' customers from using in-game AA, as per Mr. Weinand's comments.[6] The following comment was published by NVIDIA's Lars Weinand in the HEXUS.community forums on Tuesday November 3rd at 5:35am.[6] The following comment was published by AMD's Richard Huddy in the HEXUS.community forums on Tuesday November 3rd at 4:55pm.[6]
I'm sure Mr. Huddy knows how important QA is for game developers. I recommend AMD starts working with developers to make their HW work in a proper way. That's not our job.[6] NVIDIA is actively engaged with every major developer in the world plus we're also working with many smaller innovative game studios.[6]
While we don't expect Fusion to be able to cut through many DX11 games, we can guess that it will be a cost effective way of delivering integrated graphics for the needs of the mainstream.[11]
Codenamed Northern Island, HD 6000 will be a 32nm design. The boxes in orange on both roadmaps show AMD's new 'Fusion' designs. [11] There is no doubt that AMD's HD 5000 series Evergreens are the hottest cards available today, but this also seems to be their undoing to some extent. It's been just over a month since AMD introduced the first Evergreen cards, so we decided to take a look at availability and pricing in major European markets.[10]
Probably the most exciting CPU of the bunch is AMD's upcoming hexa-core (6 core) CPU. Codenamed Thuban, it will be a 45nm part, guessed to be clocked at about 2.8GHz. [11] Two detail-packed leak photos turned up in Japan recently. They are pictures taken of AMD's milestone roadmap, for desktop and notebook parts, stretching into 2011.[11] A couple of leaked AMD roadmaps have surfaced over in Japan, hinting at what the semiconductor giant will have to offer in the coming years.[9]
What is not to like about that? Besides, you should not cast a vote on PhysX based on one title, anymore than you should cast a vote for Dx 11 based on AMD showcase titles.[6]
I mean, seriously, HD 2000, 3000, and 4000 all only lasted one year. Aside from the GPU front, moving to 6-core processors will also be quite good for them. [11]
SOURCES
1. Are new rumors of an NVIDIA x86 CPU plausible? - Ars Technica
2. Analyst: Nvidia Will Enter the CPU Market - Reviews by PC Magazine
3. Will Nvidia Take On Intel In x86 Processors? - Tech Trader Daily - Barrons.com
4. EETimes.com - Is Nvidia devising x86 processor?
5. Rumors reappear of NVIDIA making x86 chips | Electronista
6. HEXUS.net - Beans :: AMD and NVIDIA in Batman Arkham Asylum AA fiasco - who's telling the truth? : Page - 1/2
7. HEXUS.net - News :: AMD and NVIDIA lock horns over graphics supremacy : Page - 1/1
8. AMD discovers several Fusion mantras, disses Larrabee
9. HEXUS.net - News :: Leaked AMD roadmaps plot route to 2011 Fusion : Page - 1/1
10. Fudzilla - HD 5800 series still facing availability issues
11. AMD roadmap shows plans going into 2011 - Neoseeker News Article
GENERATE A MULTI-SOURCE SUMMARY ON ANY SUBJECT
Enter your search query below. WAIT 10-20 sec for the new window to open.
Get more info on AMD discovers several Fusion mantras, disses Larrabee by using the iResearch Reporter tool from Power Text Solutions.
 |
|  |
|
