Slashdot Mirror
NVIDIA Gives Details On New GeForce 6
An anonymous reader writes "According to Firingsquad, NVIDIA will be announcing a new GeForce 6 card for the mainstream market at Quakecon this week. Like GeForce 6800, this new card will support shader model 3.0 and SLI (on PCI Express cards), so you can connect two $199 cards together for double the performance. NVIDIA will also be producing AGP versions of this card as well."
If there are as many people out there with fresh copies of Doom 3 in their hands or winging their way to them as I suspect, then this will be slashdotted veerrryy soon.
i cker=NVDA&script=2100 ), NVIDIA CEO Jen Hsun Huang confirmed reports that NVIDIA would be launching a new shader model 3.0 mainstream card shortly: "In a few days we're going to turn up the heat another notch. At Quakecon in Texas, a mecca for gamers and truly a phenomenon to witness, we will officially unveil our newest mainstream member of the GeForce 6 family".
So here's the content:
In last week's conference call ( http://www.corporate-ir.net/ireye/ir_site.zhtml?t
Jen Hsun went on to say:
This mainstream GeForce 6 will be the only shader model 3.0 GPU in its class and deliver performance well beyond that of the competition. PCI Express support is native and AGP support will be provided through HSI, once again showing the versatility of the HSI strategy...sampling started in June, production is in full steam on TSMC's 110 nanometer process, with shipments to OEMs soon.
Price points and product names weren't discussed, but Jen Hsun also confirmed SLI support for this upcoming card, and also mentioned by the end of the year NVIDIA will have a top-to-bottom family of shader model 3.0 cards. In fact, he mentions "we're ramping 110 on two GeForce 6 families right now at TSMC, and very shortly we'll start a third...and this quarter we'll have five GeForce 6 GPUs in production, and that ought to cover us from top to bottom."
PCIe is a switched network on your motherboard. If you're technically inclined, read this article for further details.
tasks(723) drafts(105) languages(484) examples(29106)
See this article.
tasks(723) drafts(105) languages(484) examples(29106)
well yes it can be split e.g. odd numbered lines and even numbered lines.
depending on the scene it won't always be a perfect split of the workload, but it should be pretty damn close.
Graphics is easily parallelizable, and SLI is actually almost perfect: have one card draw half of the scanlines, and the other card the rest. True, T&L and other stuff must be replicated, but that's a negligible part of the work nowadays.
Don't they interlace the lines, with each card doing an alternate line?
I think that this is actually a rare case where you can actually get close to 200% performance. For one thing, the job that is being done is very well understood and the cards need zero flexibility - hence they can write very specialised software that does one thing and does it very efficiently.
For another thing, many of the common problems of parallel computing are caused by communications, and in the case of SLI the two 'nodes' do not need to communicate - the mothership (i.e. the CPU via the PCIx bus) does all the organisation and communicating, and even that is basically one-way, so there is very little in the way of latency related issues. From a software point of view, the only real task is to shovel half the data one way, and half the other way - significantly easier than, say, a system where you have to constantly send and receive data to a range of nodes operating at different speeds.
I seem to recall that the Voodoo II (bless its zombie bones) was able to get near 2x performance in parallel.
Read Pynchon.
I don't think there is an "AGP camp."
PCI Express is a replacement for PCI and AGP on desktop class motherboards (I guess PCI-X might be better for servers, but I don't know).
Its advantages are that it has switched uplinks, so, if I understand correctly, each device can have its maximum bandwidth between any other component. PCI shares its bandwidth between all devices.
PCI Express 16x replaces AGP, and roughly doubles the bandwidth, I think. Then there's 8x, 4x, 2x and 1x for devices with lower bandwidth requirements. And you could probably expand to 32x if you really need more bandwidth than 16x. It's all about the number of "lanes" you devote to a card.
Someone here has a link to an article on this stuff, in case you want a description from someone who actually knows what they're talking about.
I've come for the woman, and your head.
This new SLI is not the same as old skool SLI.
The new one divides the screen up into two sections, I assume that if both cards are equally powerful then it will be 50:50 or thereabouts. I assume a little bit of overlap so that anti-aliasing and whatnot works correctly on the seam.
Then one card sends its generated half of the scene to the other, and they are merged and output to the display.
Am I right in thinking that most of the current crop of video cards don't really push AGP 8x at this stage? I seem to remember seeing some benchmarks where high end Radeons were not really that much faster on 8x vs 4x.
At least it will give 'gamers' a chance to brag about how fat their bandwidth is, I suppose.
Read Pynchon.
That's why it's called the "enthusiast market". It's for people who only care about top notch performance, not price. Your Geforce FX 5600 is fine for Doom 3, I'm running it on a Geforce 4 Ti-4200. You will NOT, however, be able to turn it all the way up. If you paid any attention at all 4 years ago when they first announced Doom 3, you would've known that you would need a top notch card at the time to play the game turned all the way up. Quit complaining.
Listen to my experimental-industrial-techno!
I have a 2.1ghz AMD with 1gb DDR and a 256mb GeForce FX 5900 Ultra that I bought a year ago for about $500 and I play Doom III just perfectly on my 23" widescreen LCD in 1024x768 resolution in ULTRA mode and it's as smooth as could be.
I'd crank the resolution up even higher if it weren't for the problem Doom III seems to have in displaying itself properly in some dimensions on this monitor. For some reason, unlike every other game, Doom III will do 1024x768, but 1280x1024 and 1600x1200 cause it to only take up a portion of the full screen (useless).
the SLI implementation allows sharing of texture memory. thats why there is a bridge chip connecting the two cards. it effectively makes 2 cards 1 super card.
Actually, some 3DLabs workstation cards let you do this. They have an external connector so that you can join a load of them (in different machines) together to make a rendering cluster. Of course, if you want to use commodity hardware (and don't mind a 2 frame latency) you could always use Chromium.
I am TheRaven on Soylent News
DirectX is Microsofts' standart to abstract software from the hardware (not only graphics, it also covers audio, controllers, networking, and so) as much as possible. Direct3D is DirectXs` 3D rendering part.
The thing with DX is that it's aimed mostly to games, and, while full-featured, it's incompatible with everything else. OpenGL, much like D3D, is dedicated exclusively to graphics but can be ported much more easily, and it's (IMHO) overall a cleaner implementation. Both can coexist in a single machine (if you have a modern videocard, that's most likely the case), but are independent, requiering separate drivers and so.
I don't care to google for the reference (I actually think I read it on dead tree), but I believe that what Nvidia is doing is to divide the screen in an upper and a lower part, with a boundry that is moved according to load. Then the two cards each render one part, communicating over the SLI link (and possibly the PCIe bus), and one of the cards output the finished frame (so one card would have nothing connected to its outputs).
OpenGL is doing just fine. I had lot's of worries about it, back in the DirectX 7 era, when Microsoft was rushing ahead, and the ARB was dragging it's ass with the standard, but those fears have since faded. OpenGL 1.3, 1.4, and 1.5 came out in quick succession, with each release maintaining feature-parity with DirectX. Vendor support, from NVIDIA anyway, has been excellent, with new driver releases supporting new features being released within months of each updated standard.
OpenGL is about to get a big overhaul for 2.0 (due out this year at SIGGRAPH, I think), and should compete well with the DirectX updates in Longhorn.
A deep unwavering belief is a sure sign you're missing something...
From what I read a while back, the screen isnt divided up into two equal sections, but rather in a proportion that would allow each card to do the same amount of work approximately, i.e. if they were working on a scene where the sky took up the top 2/3ds of the screen, while the bottom 1/3 was complex geometry, then 1 card would work on say the top 80% or so, while the second card would work on the bottom 20%, in such a way that neither card is ever doing significantly less work than the other.
Don't they interlace the lines, with each card doing an alternate line?
That's the way the old Voodoo cards did it, but that's not how it works with the new nVidia cards; they just split the screen into 2 halves (I believe the actual size of each portion is dynamic, to allow for a more even work load between the cards when one portion of the screen is recieving more action than the other) and each card renders its own half.
There is no mod option "-1: Disagree" for a reason. "Overrated" is not an acceptable substitute. Post something instead.
Reality was predicted to be 80 million triangles/second (with 25 pixel per triangle?). Just about every console system and graphics card now exceeds this.
The human retina consists of 120 million rods (wavelength insensitive) for peripheral vision and 6 million cones (wavelength sensitive for red,green and blue) for central vision. To match the full capability of human vision, you'd need a 12000x12000 monochrome framebuffer covering a field of view 170+ degrees, with a central region 2000x2000 with floating-point RGB colour, and it would have to update around 70 times/second.
Graphics cards and virtual reality headsets are slowly edging up to the resolution for central vision, since there isn't much demand to support peripheral vision resolutions.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
This is like the Betamax / VHS argument all over again. Betamax was the better technology but because of clever marketing only, VHS succeeded.
Not really. PC game developers want new features supported by the API they use and the graphics cards said API will run on. OpenGL has extensions, which make it easy to add features, but not all cards will support the same extensions. Therefore, support for any given feature is quite varied for a range of cards that support a particular OpenGL spec. Beyond that, how long did it take OpenGL to go from 1.0 to 1.2?
On the flip side, Microsoft adds features to Direct3D based on requests from video card manufacturers and game developers, usually releasing a new version (with backwards compatability intact in most cases) every year. This has allowed Direct3D to go from nearly 0 acceptance from developers to almost total domination of the games market in a relatively short time.
The quality of the APIs is somewhat relative (in this case), as many developers will simply stick with what they're used to. The features supported from any card supporting DirectX 7,8, or 9, on the other hand, are quite uniform where the API is specified by Microsoft. This makes the API more attractive to developers. The quality of a given implementation of either API is often left to the hardware manufacturers (and their driver developers), and over time the game developers have managed to push those manufacturers that wish to stay in the business to improve their implementations (in the case of both DirectX and OpenGL).
-PainKilleR-[CE]
Probably half the speed as a 6800 Ultra. Its called Geforce 6600 and only have 8 pipelines (16 pipelines on 6800) http://www.theinquirer.net/?article=17706
Open GL doesn't support many of the newer features of GPU's. By newer I mean anything in pretty much the past 2-3 years isn't really supported by OpenGL. As an example there isn't yet a HLSL or even a platform independant shader language at all.
That's not true. This *was* true 2-3 years ago, but in that space, the OpenGL ARB has been very quick to keep OpenGL competitive with D3D. 1.3, 1.4, and 1.5 all came out about a year apart, and 2.0 should be coming wout this year. 1.5, which came out last year, supports pretty much everything out right now, including a full high-level shading language (GLSL).
A deep unwavering belief is a sure sign you're missing something...
Okay, here's a guide for all you folks who are getting so unbelievably excited over something as mundane as pairing two video cards together.
FIRST OF ALL: THIS IS NOT "SLI".
Nvidia is simply leveraging the term to sell their version of the concept.
SECOND OF ALL: THIS IS NOT NEW.
In fact, every single consumer card that has attempted this in the past has been a failure.
** 3DFX Voodoo 2
The performance of a single Voodoo 2 was so good that people waited for prices to fall before buying a second Voodoo 2. Sales of the Voodoo3 also suffered heavily because, under many conditions, the Voodoo 2 SLI performed similarly. Thus, the long-term failure.
** Metabyte "SLI"
Shortly after 3DFX made "SLI" a household name, Metabyte developed a PCI-bridge technology that would split the framebuffer between ANY two cards and have them render in parallel.
Sound familiar? It should. There was one major drawback: both cards would have to operate in PCI mode, negating some of the advantages the newer AGP cards enjoyed. Metabyte tried to license the technology to TNT2 manufacturers, but none were interested...mainly because the upcoming GeForce 256 would make ir obsolete overnight.
** ATI Rage MAXX
This card featured two chips rendering a piece of the framebuffer, much like MEtabyte's technology. This was simply an attempt by ATI to get some experience designing a parallel-processor architecture, and to take some wind out of Nvidia's GeForce 256 sails. Because the parallelization was on-card, it could function as a normal AGP card. Bad drivers and lack of Win2k / XP support killed this card.
** 3dfx VSA 100 (Voodoo 5 5500)
The VSA 100 was designed to be used in parallel in a fashion similar to the Rage MAXX. Although this card boasted many fancy features, it could not keep up in the performance race. 3dfx also found out how hard it is to make money when the chipsets on your cards cost roughly twice that of your competitors.
** Alienware "SLI"
Yes, this is basically Metabyte's concept, but the appearance of PCIe has made it a reality for high-performance cards. PCIe also makes it possible for this to be developed entirely in software (Metabyte's vision required an on-card bridge), so why the hell wouldn't they market it?
** Nvidia Geforce 6 with SLI
Two things are readily apparent about this latest attempt:
1. The card is not a flagship, high-margin card. It is simply designed to lock-in users to a cheap Nvidia card now, and an upgrade in the future.
2. Even in SLI mode, this combo won't exceed the performance of their top-end card, meaning Nvidia won't cannibalize upgrades for their next card like the Voodoo 2 SLI did.
So sure, Alienware and Nvidia look like they've got a winner on their hands...except that there aren't many PCIe motherboards with dual 16x slots. Oh well, yet another niche-market-product-turned-failure waiting to happen.
Man is the animal that laughs.
And occasionally whores for Karma.
Since when is VW Japanese?
Boys from the City. Not yet caught by the Whirlwind of Progress. Feed soda pop to the thirsty pigs.
5200 FX doesn't require extra power connector and it has passive cooling. It costs about sixty euros and all the newish games run just fine with it. So that's very cheap, has all the features (TV-out, dvi&crt(dual monitor) output) and every feature is well supported on linux.
"so you can connect two $199 cards together for double the performance."
This stuff crops up all the time, I figured I was safe from it on a technical oriented website. A 64 bit processor is not twice as fast as a 32. Two processors are not twice as fast as one, or four twice as fast as two. Running two video cards in SLI will not "double the performance".
There are many factors that would prevent this from happening, CPU speed, bandwidth, and communication overhead to name three. A "signifigant increase in performance" would be an accurate discription.
This kind of broken logic may work for Apple PR firms, however it dosent play in the real world.
I believe the official reason was that Integras were immediately associated, in the US, with "most stolen car". It was a marketing move: Joe Consumer thinks they phased out the Integra and came out with a new model, when in actuality, all they did was re-brand it.
DirectX 8.x and 9 offer both vertex and pixel shaders. A vertex shader takes 3D coordinates (and constants) as input and gives screen coordinates and other vars as output. Although usually it transforms 3D to 2D with the standard multiplications with the world/view/projection matrices, you can easily use some constants to do vertex manipulation.
In fact, skeletal animation is very simple with vertex shaders. You just need one model and the vertex shader does all the animation.
Why wait for OpenGL when DX8 gave it to you in 2001 ?