Slashdot Mirror
NVidia Vs. Intel: Fight To Come?
Mostly Monkey was the first to write to us regarding
a new article on Tom's Hardware. The article is about the launch by Nvidia of "...nForce, its integrated graphics, audio, Northbridge, and Southbridge chipset, also referred to as Crush." The implication is that Nvidia is setting itself up to move past the graphics/audio market and get into competing with Intel in the full on chip market. What with AMD's recent success, that doesn't sound so unlikely.
This is pretty interesting in the sense that no company since the original Amiga team has tried to create a fully integrated system with multimedia capabilities from scratch. Instead of building an initial architecture and keep tacking on bus loads until performance starts to suffer, build *everything* into the architecture so that bus transfers from competing chips don't step on each other (well, as little as possible anyway).
With all the talk about the new Amiga (software only), I'm more excited by this development.
Sounds very nice, but it doesn't mean a thing if it doesn't push bits as fast as the old guys. Or if they can't get the thing to market.
Hmm... what part of dual-bank DDR with h/w prefetch didn't you understand? That's double the memory bandwidth (4.2 GB/s) of any existing mobo, with a possibility for greatly reduced latency due to prefetch. And while the CPU still communicates to the chipset over a mere 1.6GB/s bus, remember that this leaves another 2.8GB/s for prefetches and DMA from the graphics chip and other components. Also, EV6 still has room for further speedups, and this will give AMD a reason to do so. I don't find their claim of 20% improvement to be unbelievable (I'll believe it when I see it, but it isn't unbelievable).
I have no idea where the "cross-bar" part of the memory system comes into play... The idea doesn't even make sense. You have a chipset, and two banks of DDR. That's two point-point links, for which the term cross-bar seems ill-fitting. Anyway...
Also, they've got a demo board using only 4 layers! Number of layers has a high impact on board cost, and having only four is very good. Also, according to the article a number of major board manufacturers have made commitments. So it looks like this thing will see the light of day.
The enemies of Democracy are
Well, as per usual, it seems to prove that extra competition in an industry forces innovation. :)
With the Intel/VIA etc. wars, it just seemed to be a war to see who could produce the same solution at lower cost. Very little new actually happened.
When NVIDIA talked about their memory specs in the GeForce3, I was wondering how long it would take for the powers that be in the industry to have the same kind of foreward thinking on motherboards.
Now it seems that it took NVIDIA to jump in with both feet to wake everyone up.
Well, succeed or fail in this new venture, they've certainly given everyone a lot to think about, and I'll definately be putting money aside to buy one as soon as I hear they're settled in and stable.
It'll be fun to see where this race of new ideas ends.
Long live innovation, competition, and hats off to NVIDIA.
Malk
Well, as they said in the article, when queried about why they didn't have an Intel version:
"We don't have an Intel bus license"
From this, I assume that Intel are trying to do what IBM did with the MCA architecture over a decade ago.
They're forcing developers into restrictive licenses, and possibly expensive ones, to use an architecture.
This was a really bad idea, as IBM found when nobody used MCA, and instead went for the inferior EISA bus, which didn't have such heavy restrictions.
I can't blame NVIDIA for trying to get their dedicated graphics cards looking and performing to the best possible standard (as they already have optimisations for the PIII and AMD processors), and it's only logical for them to optimise for P4 if they can. And they have.
However, this new chipset is simply saying that AMD are more open, and easier to develop with, rather than having Intel breathing down their neck with licensing issues at every step of the way.
My view on it is that AMD just gave NVIDIA the license to innovate properly, and good things are now happening.
Malk
The Tom's article says that the chipset does speculative data prefetch to improve memory-CPU throughput by up to 20%...cool... but why are they bothering to do this when Athlon is about to be replaced by Palomino that has hardware prefetch built in?
The reason they are using a GF2 and not a GF3 is simple really - if you want a GF3 then you have to buy it from nVidia and they get to make more money. If you don't want a GF3 then they've sold you a GF2 already which will keep you from spending money at their competitor.
Remember the GF3 driver is likely to be included in the unified driver they are going to be shipping, but if you go with the competitor then you have to worry about driver issues. Looks like nVidia is set to become the Microsoft of the chipset/hardware industry!
Fear: When you see B8 00 4C CD 21 and know what it means
While this looks like an interesting development, it looks like nVidia is trying to play both sides of the battle. The top graphics article on Tom's Hardware, here at http://www4.tomshardware.com/graphic/01q2/010511/i ndex.html, discusses the next nVidia driver release, which focuses entirely on Pentium 4 optimizations to the nVidia drivers. I would decalre an all-out war yet, guys. ;-)
Interested in open source engine management for your Subaru?
- ability to guarantee bandwidth and latency for specific I/O requests (what they're calling "StreamThru"), a feature previously only available on mid-range (Octane) and high-end SGI systems with the XIO bus.
This can be quite handy for things like handling multiple video streams without dropping frames, or mixing video with other I/O requests. I wonder if there are other cool things you could do once you can allocate bandwidth for I/O and have said allocations enforced in hardware?I've never seen this capability on the Intel platform before. It was a cutting-edge SGI feature back in 1996 or so. Sun/HP/IBM/Intel were never concerned enough about video to implement this on their workstations, but I do recall some discussion about whether it'd be in Infiniband or not (don't remember the outcome).
--LP
It's all been caught on video, right here.
I see people making this mistake time and time again. Everyone seems to think that a graphics driver is just like any other driver. They think that the code for the driver is mostly useless for any card other than the one it is written for. With graphics drivers, that is not the case.
In order for your driver to be OpenGL certified, is has to implement *all* of OpenGL. That includes T&L, NURBS, and much much more. NVidia has an *excellent* software implementation of T&L, and I'm sure the people who made the Kyro and Kyro II (neither of which have hardware T&L) would *love* to get ahold of it. You expect NVidia to just give that to you for free? Um, I don't think so.
Motherboards are a different issue, and I don't really think NVidia would do something dumb like keep their motherboard specs closed. I suppose we'll find out, but their record with graphics cards says nothing about it.
------
Now, NVidia is just jumping right on in to the motherboard and sound processing markets. And from the looks of it, their very first product already blows the competition out of the water. How long has this competition been here? We're talking about *Intel* and *Creative*, for crying out loud! You don't just beat these guys overnight! What have they been doing all this time? Why is NVidia able to take them out so easily? Why can't ATI or Matrox touch the GeForce 3? Why is NVidia not afraid to spend Microsoft's money to develop Linux and Mac drivers? What is going on here? At this rate, NVidia will rule the world in five to ten years!
Obviously, NVidia's technology is being provided by aliens. There just is no other explanation. We have to stop this conspiracy before it's too late!
------
First, Microsoft was expected to lose about $200 per unit on early Xbox units. With this NVidia announcement, there's a path to an NVidia/AMD Xbox implementation at a lower price point. This gives Microsoft much more leverage with Intel in bargaining for discounts on parts for the Xbox. Microsoft probably won't lose $200 per unit. Intel will get squeezed hard.
Second, it makes it very clear that entertainment apps now drive the PC industry. More transistors are now devoted to 3D graphics and audio processing than to the main CPU. The PS2 had that balance too; all the innovation is in the graphics and media processing, while the main CPU was an old design. Now it's taking over. Business desktops will have full 3D graphics and 5-channel audio, even if they never use it.
Third, it's now clear that one more generation and one-chip PCs will go mainstream. There have been one-chip PCs, but they were low performance. As the chip count goes down, motherboards get smaller. With this level of integration, there's little need for slots for plug-in boards. AGP goes away, and probably so does PCI. Add-ons will go on FireWire, USB, or Ethernet.
Finally, this all leads to downsized packaging. The computer can fit in the base of a flatscreen, and will probably end up there.
It's not clear why this new chip is GeForce 2, rather than GeForce 3 architecture. The XBox uses GeForce 3. It's probably a production bottleneck; NVidia was having trouble getting GeForce 3 parts out the door in volume. But NVidia will fix that.
This is the one angle of this story that I feel has not been sufficiently explored. This new chipset is a potential godsend for IT managers and workers. Why? When you troubleshoot a PC, one of the first things you have to look at are Video Drivers, Sound Drivers, NIC Driver, IDE Driver, Chipset Driver, etc. NVidia is going to make sysadmin's lives much easier if this product is 1/3 as good as it looks. Much easier to properly maintain PCs this way. This gives AMD a much bigger wedge to use to pry into the corporate market. Saving a few dollars per unit when buying PCs is nice, but anything that drops TCO (Total Cost of Ownership) gets major attention.
Help save the critically endangered Blue Iguana
The reason their Video Drivers are so closed is first because of various NDAs that they (Nvidia) had to sign to get bits of tech from other people and also because one of their great strengths in the video market is the unified driver setup that they have. Whether you agree with then or not (I don't) it seems clear why they would think that keeping that info close to their chests would be a good thing. In the MoBo chipset market it is a diffrent story these things need to be open and everybody else has them open. Also the entire history of the Linux drivers seems to indicate that they really did want to open them and just found themselves unable to do so. Comparing that with this is apples and oranges and it seems obvious they will do the right thing here.
Cypherpunks: Civil Liberty Through Complex Mathematics. Those who live by the sword die by the arrow.
What, so now I have to sign an NDA to get chip programming manuals, if I can get them at all?
No thanks... nVidia is one of the least-open companies out there, repeatedly denying requests for programming information. They finally did produce Linux drivers for their cards, but they are binary-only. How would you like binary-only kernel support for your CPU?
Or maybe they would support Linux. It's a fairly lucrative market in the server area, at least. But based on their past attitude I wouldn't count on it.
-John
nVIDIA is trying to be the next MS? Isn't everyone trying to be the next MS? You speak as if all companies aren't trying to take control of their respective markets.
Uninnovate - Only the finest in engineering.
That's awesome. Now I get to have outdated and nonexisting drivers for my motherboard from nvidia instead of just this goofy video card!
(If you don't see the sarcasm in that statement I pitty you)
to address part of your question about directx, directly here is a link to the page about the audio processing unit.
As it appears that this chipset is an outgrowth of the Xbox development I can imagine that integration with directx* will be pretty much a given.
As far as earlier comments about drivers go, this second article mentions that "A motherboard with nForce chipset will be installed with ONE driver"
Obviously this type of chipset won't appeal to everyone, but hey for me it sounds great. No more Nic, audio card, graphics card etc to buy. The only thing they missed was integrating 128 MB or system RAM into the mix..