Slashdot Mirror
NVIDIA Responds To Intel Suit
MojoKid writes "NVIDIA and Intel have always had an interesting relationship, consisting of a dash of mutual respect and a whole lot of under-the-collar disdain. And with situations such as these, it's easy to understand why.
NVIDIA today has come forward with a response to a recent Intel court filing in which Intel alleges that the 'four-year-old chipset license agreement the companies signed does not extend to Intel's future generation CPUs with "integrated" memory controllers, such as Nehalem. NVIDIA CEO Jen-Hsun Huang, had this to say about the whole ordeal: 'We are confident that our license, as negotiated, applies. At the heart of this issue is that the CPU has run its course and the soul of the PC is shifting quickly to the GPU. This is clearly an attempt to stifle innovation to protect a decaying CPU business.'"
WTF? Does Intel sell more CPUs than NVIDIA sells GPUs?
What's next, Creative starts bitching too because their APUs (Audio Processing Unit) are being snuffed out by nVidia and Intel?
Hey you two, STFU. Your technologies are forever joined at the hip in modern computing. Stop the bitch slapping and grow up.
Life is not for the lazy.
Then why is NVIDIA trying to build a CPU themselves, if it's a decaying business?
"At the heart of this issue is that the CPU has run its course and the soul of the PC is shifting quickly to the GPU. This is clearly an attempt to stifle innovation to protect a decaying CPU business.""
Sounds like he reads slashdot.
Shai Schticks:"You don't make peace with friends, you make peace with enemies"
A GPU that replaces the CPU becomes the CPU. NVDIA created a GPU API that allows the GPU to be used for ANY computation and not be limited to graphics only. NVDIA is trying to become the CPU, or relegate the CPU to a very small role to grow NVDIA's business. They are trying to choke Intel by doing an end-run around the limits of their contract with Intel. As the GPU does more and more computation, it becomes indispensable.
Yeesh, does he ever say anything in public that doesn't sound like drug-addled desperate bluster...? He's like the Ken Kutaragi of the PC world...
We live, as we dream -- alone....
By locking all competitors out of the chipset business, a company can boost margins (and thus boost profit), as opposed to living with decaying margins and lower profitability due to commoditization.
As standalone CPUs get commoditized, the margins and profitability decay.
Also if you sell crappy integrated GPUs, you can protect the GPUs from competition and the CPUs from commoditization by bundling them and locking out competitors.
Intel didn't get to where they are today by not knowing how to play the game. They wouldn't be walking away from their standalone CPU business and move to integrated CPU/GPU if they didn't think their old standalone CPU business would suffer from decaying margins. As they move into this space, it also only makes sense to try to put up barriers to your competitors who might be trying to screw up your future business strategy. Remember how Intel made AMD go try and execute "SlotA" when before they made pin-compatible chips. This is seems like a very similar strategy to try to kick Nvidia out of the Intel eco-system.
What is at stake isnt just nVidia's chipset business, its their entire business.
They argue that the CPU is just the glue for highly parallel GPU operations, but Intel is planning to turn the CPU itself into a highly parallel monster needing no GPU at all... just some integrated display logic.
If nVidia loses its chipset market then they must drastically scale down their business.
I've said it before and I'll say it again... nVidia doesnt have anything to offer the other big players in the market in regards to licensing agreements. They have no essential IP to speak of. They are at a big disadvantage, even though they are currently king of the GPU's.
"His name was James Damore."
Jen-Hsun Huang has never been one to keep his trap shut when given the chance... even though Nvidia is in the red right now. Lesson one: When a CEO comes out and tries to use a legal dispute related to a contract as a pulpit to make a religious sermon, he knows he's wrong. See Darl McBride and Hector Ruiz as other examples of dumbass CEO's who love to see themselves in magazines but don't want to be bothered with pesky details like turning a profit or actually competing.
Intel is #1 in graphics when it comes to shipments... now I'm not saying I'd want to play 3D games on their chips, but guess what: despite what you see on Slashdot, very few users want to play these games. Further, I've got the crappy Intel integrated graphics on my laptop, and Kubuntu with KDE 4.2 is running quite well thanks to the 100% open source drivers that Intel has had it's own employees working on for several years. I'm not saying Intel graphics will play Crysis, but they do get the job done without binary blobs.
Turning the tables on Huang, the real "fear" here is of Larrabee... this bad-boy is not going to even require "drivers" in the conventional sense, it will be an open stripped-down x86 chip designed for massive SIMD and parallelism... imagine what the Linux developers will be able to do with that not only in graphics but for GPGPU using OpenCL. Will it necessarily be faster than the top-end Nvidia chips? Probably not... but it could mean the end of Nvidia's proprietary driver blobs for most Linux users who can get good performance AND an open architecture... THAT is what scares Nvidia.
AntiFA: An abbreviation for Anti First Amendment.
If the CPU has runs its course, why is Nvidia suing Intel to get a slice of a dying technology?
This is clearly an attempt to stifle innovation to protect a decaying CPU business.
So I can replace my CPU with a GPU the next time I buy a computer? Oh wait...
In five years nobody will need a powerfull GPU anymore. Run a real time ray trace benchmark on your PC and if you got a recent powerfull CPU (like me) you'll see why the scanline rendering is nearing it's end. At 1024x768 I get 60fps without shadows and 30fps with hard shadows. The benchmark wasn't even optimised for multithreading!
Here be signatures
There's got to be some information withheld here. Why would Intel and Nvidia put their relationship at risk? Intel needs them for high-end graphics, and Nvidia needs them for the CPU.
This dispute seems self destructive to both parties and is only of benefit to ATI/AMD.
Specifically why the "dying" x86 technology?
Shai Schticks:"You don't make peace with friends, you make peace with enemies"
Huang must have read this blog article Heralding the Impending Death of the CPU. Which is cool but Huang apparently declined to read this other article Parallel Computing: Both CPU and GPU Are Doomed, for obvious reasons.
"This, btw, is in my opinion the real reason AMD bought ATI. AMD wanted to work toward having a solution for that high volume market, and seemed to think they needed to own ATI to do it."
I think you're partially right. If they indeed wanted entry into the business graphics market. Matrox would have been a better purchase. But ATI makes better GPUs and they wanted entry there as well. It's easier to scale down a high-end GPU than it is to raise up a low-end GPU.
Shai Schticks:"You don't make peace with friends, you make peace with enemies"
An Ion is about as small a standard form factor as you can get (pico-itx).
Once you look at the Ion for an HDTV platform, I don't think you'd go back looking at Intel's offering...
Wake me when Microsoft gots somethin that runs on stuff somebody wants. Even I know Vista is the suckage.
No, don't. I really could pass a purple twinkie about what Microsoft thinks is good stuff even if they buy adds in my mags that say it's good enough. If you want to get on my stuff then wise up.
Thankfully, Intel is hearing me, yo. Otherwise I'd be waiting like until Jasmine textes me back, which is like for ev-er.
Help stamp out iliturcy.
You know real time raytracing isn't something that has to be done on the CPU. There's no reason you can't write a raytracer for a GPU.
In fact NVIDIA already has a fully interactive raytracer. They demoed it last summer at NVISION, and SIGGRAPH '08. I'm sure as they expand CUDA support you'll see more and faster raytracers.
Go check out http://developer.nvidia.com/object/nvision08-IRT.html
Dude. Even I know GPUs are optimised for compositing. Ray tracing is a way different thing. It has to have a way different system. Pretending it doesn't will not help you here.
Help stamp out iliturcy.
Waaaahhhhhhh.
I wish some company would create a new chip that runs at a THz or something.
Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
Intel, Nvidia and AMD helped Apple formulate the original proposal for OpenCL... Intel makes Apple's CPUs, Nvidia increasingly makes the GPUs (sometimes 2 in a single laptop). So there's bound to be some smack-talking about CPUs vs. GPUs and all that.
I think Apple will be the first to have OpenCL support in an OS, and as others follow suit and we see more CPUs and GPUs in machines, this little tiff might conceivably end up meaning... something.
Village idiot in some extremely smart villages.
Dude. Even I know GPUs are optimised for compositing. Ray tracing is a way different thing. It has to have a way different system. Pretending it doesn't will not help you here.
You didn't just write the above did you? You show your ignorance. A long time ago they did just compositing, but that was back in the VGA controller days.
Then they evolved to do fixed function rasterization, but those days are over (unless you're Intel doing integrated stuff).
GPUs are MUCH more programmable, and getting more so with each generation. You can do pretty much any floating point math function you want now. Go look up CUDA, and OpenCL they let you basically write C code for the GPU.
Sure the GPUs might not do so well when it comes to brancing, but you'll see that GPU's are being used to do more than just rasterization. Sure razterization would be an important target for NVIDIA/ATI but that doesn't mean it can only draw triangles.
If you look at the paper I linked (which you obviously didn't) it describes how they wrote a ray tracer using NVIDIA CUDA and EXISTING GPUs. If stuff gets more programmable as NVIDIA seems to be targeting, then it will only get easier to write ray tracers which run on the GPU.
If you want proof GPUs do more than rasterization go check out how NVIDIA's GPU tech is now in the Tsubame super computer.
Even Intel is getting into the GPU business with Larrabee, I bet they plan to write a ray tracer for that.
The computer hardware manufacturers are better off getting rid of all of this integrated stuff, and concentrating on the real hardware. I believe it would be a big boost to the industry as well. Plus a step up in the right direction.
You can plug an NVIDIA GPU card into an Intel motherboard (I did just that for the computer I am using).
I have no idea why Intel wouldn't want Nvidia to make chipsets for core i7. For some reason, even years after AMD bought ATI, the only Intel mainboards which support two linked graphics cards do so through Crossfire. So if Nvidia doesn't make chipsets to support core i7, Intel would be forcing the hardcore gamers to either (a) buy AMD's video chips to use Crossfire or (b) buy AMD's CPU's to use NVidia SLI.
Errrrrr, I think you'll find it's the other way aroud mate. That is, afterall, why you're maing comments like this?
I only need a very little 3d graphics so intel is actually very good... I had an intel chipset in my old laptop and I never had driver issues of any kind...
In my new laptop I've got an ATI card, which I regret... I have nothing but problems with it... And there is NO open drivers for it... Which is in fact the only reason I went ATI and not nVidia or Intel.
And unless ATI starts actually delivering drivers for their mobile hd series then I'll be looking for intel for my next laptop, that's for sure...
While Intel is trying to lock nVidia and ATI/AMD out of the chipset business by bundling the CPU and the chipset and bridging them with an interconnect - QuickPath - which they won't license to nVidia,
nVidia on their hand has tried to do exactly the same, locking Intel and ATI/AMD out of the chipset business by bundling them with the GPU and bridging them with a technology that they won't sub-license either : nVidia's SLI.
nVidia has tried to be the only chipset in town able to do SLI.
Intel is currently trying to be the only chipset in town usable with Core 7i.
Meanwhile, I'm quite happy with ATI/AMD which use an open standard* which doesn't require licensing between the CPU and the chipset (HyperTransport) and another industry standard for multiple GPU requiring no special licensing (plain PCIe).
Thus any component on a Athlon/Phenom + 7x0 chipset + Radeon HD stack could be replaced with any other compatible component (although currently there aren't that many HT-powered CPU to pick from).
*: The plain simple normal HypterTransport is open. AMD has made proprietary extension for cache coherency in multi-socketed servers. But regular CPUs should work with plain HyperTransport too.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
1. NVidia sells integrated GPUs too, and they too count crappy integrated GPUs as GPUs sold. And yes, even if you later go and buy an ATI 4870, Nvidia still counts it as a GPU sold.
So it seems to me like the GPs basic point still stands: Intel sells more GPUs than Nvidia. By a metric Nvidia too uses when they willy-wave about their market share being larger than ATI's.
2. You seem to assume that it's some inescapable misfortune for the users, or that that's somehow not included in the choice to buy this computer vs the other computer.
Newsflash: most people don't actually care about the GPU itself. They want a computer. And if they wanted a gaming rig that tops all benchmarks, there are enough companies selling them one. It's not like when they go to Dell's site there isn't a gaming computer category.
So, yes, the decision was made at some point to buy a computer which barely runs Aero well. Because they decided that they don't need more. And if an Intel integrated GPU was the cheapest there, I fail to see what's the problem.
Basically (for the mandatory bad car analogy) it's like when you buy a car, you don't actually give a flying fuck about the exact model of the gearbox under the hood. You might care about miles per gallon, price, whether your family fits in it, speed and acceleration maybe, insurance price, and/or the status-symbol value of that car brand. But if it's a Ford transmission or bought/licensed from Toyota, who cares? If they can save you some money by using transmission X instead of transmission Y, and the car still fits your criteria, why would you feel shafted? And if that saving came by getting it bundled with, say, the suspensions, again who cares?
Same here. If mom wanted a computer which runs Windows, does email and is good enough for her photoshopping photos taken with her digital camera, why would she care whether it's a discrete higjh-end GPU or an integrated solution from any of the manufacturers? The whole computer still does what she needs to do, and the latter costs less than the former. And if the one with Intel integrated chipset cost less than the one with Nvidia integrated chipset, so be it, that's the one she's going to buy.
A polar bear is a cartesian bear after a coordinate transform.
It is in fact NVidia that opens mouth where it should not. License valid or not, using every opportunity to do their usual "CPU is dead, long live (Nvidia's) GPU!" line won't fool anyone but the dumbest. Here they go again - "We are confident that our license, as negotiated, applies. At the heart of this issue is that the CPU has run its course and the soul of the PC is shifting quickly to the GPU. This is clearly an attempt to stifle innovation to protect a decaying CPU business." Of course you are confident, you have to defend yourself, what else would you be saying? But watch as the CEO hijacks the opportunity again and bash Intel and CPUs in general. Well guess what mr. CEO, you HAVE to have a central processor (oh-oh, that's what the CP in CPU stands for, did you know?), that's not just one doing 3D graphics, which is what you have been doing all these years, but actually centralizes resource management in the system, and even facilities that do not strictly need the CPU still use it for coordination - memory controller, FPU, DMA, interrupt controllers and more. Without the CPU, where do we send orders on bootstrapping the system to a usable and programmable and serviceable state? Tell you what, mr. CEO, - When you have a chip that not only does teraflops with SIMD but can actually CONTROL a motherboard, let us know. Until then be humble and let Intel and others provide. Nobody said you cannot compete, but a GPU alone does not make a system, even less so one that requires a special compiler to make it run generic code. NVidia should learn the meaning of humble, they think slapping together an SLI chip that consumes 200W and only works for two years is a good job done, they think wrong. I'll stick with a 25W x86 CPU for a while longer thank you. And no, NVidia "south"- and "north"-bridge chipsets are not good enough.
to slow or stifle competition. I saw it first more than 20 years ago.
-- Programming with boost is like building a house with lego. It's a cool but I wouldn't want to live in it
Well, the thing is that most people equate a good graphics card/chip with games, especially in terms of 3D. However, these days it's more than just games. Whether a good thing or not, modern OS's use 3d graphics.
With the advent of Vista, Intel's chipsets failed so miserably that they had to pressure Microsoft into changing what qualified as "Vista Capable" (and in doing so, MS pissed of a lot of other vendors by breaking their promise to stick by the spec).
Even in the world of Linux, KDE4 and recent versions of Gnome take advantage of 3D accelerated graphics.
So these days, people *should* care about 3D graphics to some extent, at least if they don't their modern GUI to run like crap.
The PC market may still be growing slowly, but I've noticed a lot of movement in terms of FOSS-OS based (Linux/BSD) proprietary systems.
We've got music players and phones based on BSD. Tivos and various other hardware based on Linux. Big companies such as Sony use a lot of FOSS in the back-end of such things as TV's, etc (Sony also promoted Linux compatibility for the PS3/PS2). Companies are out to save money, so perhaps in the future a new gen gaming console will run on a Linux or BSD base, and in that case it could be a fairly big contract for a graphics manufacturer to include their chipset and a compatible driver.
Why is this article under "Your Rights Online"? This is a contract dispute between companies, and at first glance there are no apparent copyright, DRM, or other YRO-esque issues involved. This belongs under "News" or maybe "Technology".
- T
It's exactly this kind of bullshit that has killed PC gaming, and driven most of the video game development towards consoles.
Absent the "Next-Gen" feel of a PC platform (3D performance has always felt one generation ahead of consoles) players have no incentive to upgrade their PCs every 18 months like they did at the turn of the century.
In five years, the only games you'll be playing on your PC will be independent titles that can't get licensing agreements on XBL or WiiWare, and flash/browser games.
Thank you, Intel and nVidia.
CAD needs a good video card also people who do dual display use needs a little better then on board gma. Low end amd and nvidia chips can do dual dvi out and the amd boards can have 64-128 side port ram. Do you really want to use 128-512 system ram on dual display with on board video and windows vista areo?
Intel should add 64-128 side port ram to intel gma.
If they finish working out the interconnect issues, an ideal future would be multiple processors in the system with the same instruction set but different blends of functional units and performance tradeoffs. So they can all execute the same general OS and application code, but some are much faster at vector-math and framebuffer-style memory ops, where others are faster at scalar code and complex branching. Similarly, a very low-power Atom-style CPU could keep running the OS and idle tasks while all the fancy processors get completely powered down, allowing much lower system power consumption in between the aggressive tasks (useful for all the embedded always-on application models).
The OS task scheduler can learn to place tasks on the right processors but still use virtual memory and timeslicing models to share the resources. Bad task placement may hurt absolute performance and efficiency, but the code would still run. And with some hardware performance registers, the OS could even monitor this efficiency and migrate tasks based on clear mismatch of code to functional units. But ideally, the OS policy would preemptively place tasks in the right place because of static knowledge or negotiation of which app tasks desire which kinds of performance environment.
CAD does need a decent card. But that's a very small market too. Most business users are running office apps and email, and that is it.
In a minute there is time For decisions and revisions which a minute will reverse. -T.S. Eliot
What I don't get is why doesn't Nvidia just buy out Via already? Since we have seen with Ion that an okay CPU+decent GPU equals a pretty nice platform for more folks needs, and with Intel trying to get into GPUs with Larrabee and AMD buying ATI it seems really foolish to me that Nvidia has left Via out there so long.
With the Nano they would have a good chip to combine with their GPUs in the Laptop/Netbook/ and desktop PCs, and then later if they wanted to integrate it with their GPUs they could. It just seems to me to be really stupid not to hedge your bets with the competition moving into their spaces. This way they could still sell GPUs and chipsets for both Intel and AMD motherboards and have a completely top to bottom Nvidia solution to offer OEMs as well. Frankly I am shocked that Intel hasn't tried to buy out Via just to cut off the chance of Nvidia snapping them up and to get all of those patents that Via holds. But from they way it looks like the market is going, especially in the mobile space(which Nvidia should have a nice jump on with Tegra) it seems crazy to me that Nvidia hasn't just bought Via by now.
ACs don't waste your time replying, your posts are never seen by me.
Many of us saw this coming over 5+ years ago. CPUs get more parallel due to technical and cost reasons while GPUs become more flexible to the point where they are programmable. The CELL processor design was the first big leap in this convergence that is now clearly happening (like anything that far ahead of the curve, its different and not anywhere near as successful as the designers hoped.)
Its totally expected that intel will have so many processor cores that they will start to tackle GPU sized problems; poorly, but intel has a history of using brute force over clever design (manufacturing talent aside.)
GPUs will get so flexible that they will almost resemble a well designed parallel cpu-- giving them an edge because they are not shoving an old idea into a new niche; they are evolving within their niche + someday adding on some cpu abilities (which could perform poorly, but those don't matter whole lot for typical usage.)
If you designed a general purpose parallel CPU, you'd work around the common use cases for that CPU-- and the most common operations that benefit greatly from parallelism are GPU friendly problems. Less common OR less speed intensive operations become less important.
Therefore, a flexible GPU could have CPU abilities added onto it and while the CPU part may run slower with it being an afterthought in the design-- it only needs to run about as well as a cheap CPU does today to handle 99.9% of the software used today. As apps get optimized they will try to exploit the programmable GPU just as they are trying to exploit multiple CPUs today (the line between the two will blur more over time.)
Democracy Now! - uncensored, anti-establishment news
Most CAD users aren't just going to be buying a discrete card, they'll be buying a Quattro or FireGL card for $500+, possibly $1000+.
Convert FLACs to a portable format with FlacSquisher
At the heart of this issue is that the CPU has run its course and the soul of the PC is shifting quickly to the GPU. This is clearly an attempt to stifle innovation to protect a decaying CPU business.
Now that is true talking out of your arse!
Yup it's looking like the GPU will be following in the footsteps of the FPU.
GPUs are MUCH more programmable, and getting more so with each generation. You can do pretty much any floating point math function you want now. Go look up CUDA, and OpenCL they let you basically write C code for the GPU.
I haven't actually written code, but from what I do know about CUDA/OpenCL this isn't quite true. See, the processors in GPUs aren't as full-functioned as one might like, and their capabilities have been changing dramatically from one generation to the next. Just for example, a generation or 2 ago, most of them didn't even support branch instructions, and could not run programs longer than a few thousand instructions.
So yes, CUDA and OpenCL attempt to put enough of an abstraction layer on this extremely raw programming environment to make coding for it simpler, but it's not always as simple as taking a C snippet, tweaking syntax a bit, and running it through a different compiler. Especially if you want to maximize performance. And the more they try to make the environment look like an ordinary CPU, the less likely it is you'll be able to approach the peak theoretical throughput of the hardware. I've seen commentary to the effect that OpenCL is cleaner and more abstract than CUDA, and as a result people don't expect to be able to get as much performance as CUDA. Despite this there's lots of interest in OpenCL simply because it is easier to program for and isn't tied to just NVidia hardware.
If you look at the paper I linked (which you obviously didn't) it describes how they wrote a ray tracer using NVIDIA CUDA and EXISTING GPUs. If stuff gets more programmable as NVIDIA seems to be targeting, then it will only get easier to write ray tracers which run on the GPU.
The thing is, it was a stunt. Yes, you can write a really simple raytracer which runs on existing GPUs, but it's not easy and it's not a natural fit for the hardware. Despite the fact that GPUs are getting more programmable, they are still designed around the needs of triangle rasterization and texturing.
Even Intel is getting into the GPU business with Larrabee, I bet they plan to write a ray tracer for that.
Plan? They already have. They've given plenty of demos of a software realtime raytracer running on 4- and 8-core x86 systems, and guess what Larrabee is? It isn't a traditional GPU, it's a multicore x86.
Intel's coming at this issue from the opposite direction: instead of slowly making a GPU more like a CPU, they're taking a real CPU, optimizing it for low power with high floating point throughput, and tiling a bunch of them onto a chip along with a few special function accelerators to support rasterization (texturing units). There is no need for something like CUDA or OpenCL on Larrabee; you can write real, standard C, run it through an ordinary x86 compiler, and it will work just fine on Larrabee. The only programming model difference from standard x86 is the addition of a new 512-bit vector instruction set to supplement SSE (which is only 128-bit).
While Quattro and FireGL cards would certainly be great for CAD applications, I'm going to need to give a big old [citation needed] to your claim that most CAD users have these high-end cards.
Unlike the situation with games, there's nothing stopping a CAD user from doing their work with lesser video cards. In fact, they could even do their work if you gave them a machine that was 10 years old; they would just have to hide more objects and spend more time in wire-frame mode.
The details are trivial and useless; The reasons, as always, purely human ones.
Very true; a few years ago I was playing around with Valve's "Hammer" map editor, which is essentially a CAD program. Even though I had a meager FX5200, I could still be productive, just not as productive as someone with a better card.
That's the thing with CAD users though -- if you're using CAD for more than a few hours per week, then the cost of a good graphics card pays for itself quickly in time that you don't have to sit there, waiting for the next frame to render.
I don't have a citation for CAD users vs. FireGL and Quadro (I must have said Quattro earlier because of Audi) buyers. However, I'd bet that very few people who use CAD as part of their daily job are running on integrated graphics from any company.
Convert FLACs to a portable format with FlacSquisher