Slashdot Mirror
Nvidia Talks About Next-Gen Geforce, Plus Pics
Per Hansson writes "Techspot was at Comdex in Sweden a few days ago; we have now posted a small interview with Nvidia along with some high-res pictures of the Geforce FX on this page in our new comments system." This is one of the strangest looking video cards I've ever seen (and it isn't cheap), though it may look different by the time you can buy it in a box. Which is not yet, despite all the hype.
I wonder if we're ever going to get to a point where "this is the hardware. You have 10 years to do something cool with it"
Only when, if ever, we can render something like the Final Fantasy movie in real-time. Something tells me Moore's "law" will have broken down before that though.
Kjella
Live today, because you never know what tomorrow brings
The fact that something may not be as fast (or as expensive) as the newest new computer hardware has nothing to do with obsolescence. It is only obsolete because they want you to believe it is obsolete.
There's very little reason someone with a video card made a year or two ago would need one of these. My Radeon 8000 works fine, thanks. $400 for a 10-frames-per-second improvement isn't what I call revolutionary progress.
~D:
At the moment it's easy for gamers and developers of software to imagine uses for next generation video cards. Hardware will continue to become obselete as long as humans want to do more with it. Something simple like a graphics card can be imagined to have limits though... for example: Imagine I have a video interface directly to my brain from my video card. Once that video card can render, in real-time, as much information as my brain can distinguish then there's no need for more powerful hardware...right?
Exactly. Everyone who complains of immediate obsolesence is either a naive fool or has too much money and nothing to spend it on.
A common sense view of the situation would be: yes, you have a Radeon 8000: you shouldn't even consider a GFFX. The GFFX SHOULD be marketed at people who have Nvidia TNT2's and 3DFX boards: people who are getting to the point where they want to upgrade have an extra option, people who don't need to upgrade shouldn't.
Common sense. It's pretty easy.
Mooniacs for iOS and Android
the most complex games, on the highest resolutions, get WAY too many frames
you're kidding me, right? Get UT2003 in 1600x1200 with everything maxed out and even on a 9700pro you won't see more than 30 or so fps on certain maps (alone, just looking around, in heavy firefights I'd suspect it'll drop to the teens: I don't have a system like that but I'm basing this on vidcaps I saw when UT came out).
Human eye is unable to perceive extra frames beyond a certain number
bs, it also really depends on what you're doing. If you're in a driving game going straight ahead and you get 30fps, you *might* not notice the difference between your 30 and 90fps. In a shooter or other game where the screen moves around quite a bit, I'm sorry but I can see the difference between 30fps and 70fps quite easily...
The moment somebody creates a card that is able to mantain refresh-rate-synced-updates (say 85fps) in any available game at any resolution regardless of what is going on, it's the moment a new game will be announced that will take a card 4x as powerful to do the same.
It really never ends... of course if all you'd like to do is play counterstrike you can get by quite well like myself with a really old p3-450 + geforce1.
-- the cake is a lie
The GeForce FX is in my opinion, not going to be what the world has expected from nVidia. It is simply too little, too late - 6 months too late. It may have the performance crown for a month, but it will be short-lived.
.13u process. In case some people haven't noticed, the leaked benchmarks of the GeForce FX show it to only be marginally faster than the Radeon 9700 Pro. Not to mention that it's 500MHz vs. 325MHz. It seems that ATI is faster in terms of IPC's.
ATI will simply respond with the R350, which is likely going to be an improved R300 core, as well as DDR2 and manufactured with the
It would be unfeasible for nVidia to respond until the summer with the NV31/34, at which time ATI will announce the R400.
I will have to give nVidia one thing though, their drivers are excellent. This is perhaps the only thing they have going for them at the moment. However, ATI is pumping out a new driver set almost every month, and at this rate, they will soon reach parity with nVidia.
why does the announcement of a $600 board from nVidia render a $400 from ATi obsolete? When Mercedes announces a new S-Class does it render the BMW 3 Series obsolete? WTF are you on about?
That was classic intercourse!
When I talked with the guys from BFG (who are already taking preorders) at the HardOCP Workshop, they had a FX card on hand that you could look at up close. I asked one of their guys about the huge ass coolers and they said that the manufacturers had the choice to put their own type of cooling on it if they wanted. So I'm sure there will be some 1 slot options out there if the customers demand it...
People call liquid cooling dangerious, unneccesary, and extravigant, and then buy video cards that have cooling such as this one, cpu coolers that are enormious, and put half a dozen case fans in their case to try to keep the temperature down.
I do security
Please, The market that this card is aimed at could care less that they lose a slot. Reason 1 being that the AGP adjacent slot normally shares and IRQ with the video card. Reason 2 is the target market is also obsessed about cooling, having a card in that slot reduces circulation. Look at the Abit GF4 Ti 4200 with the OTES, that's a production card. And the last reason being that the trend towards onboard peripherals has increased. Onboard audio has gotten better and LAN/USB 2.0/1394/RAID is onboard now on many high-end boards. Oh, and the people who buy those high-end boards will be the ones buying the GF FX. Hell, with all of that onboard and there being 5 or 6 pci slots, you really think that burning one slot is gonna keep someone from buying the GF FX? I don't have a PCI card in the AGP-adjacent slot, don't use the onboard sound and have a PCI nic. I still have 2 PCI slots left.
What I wonder is why just not put all the transistors, the chip and the ram...on the other side of the circuitboard! No-one looses a pci slot that way.
-- Waht? Tehr's a preveiw buottn?
bs, it also really depends on what you're doing. If you're in a driving game going straight ahead and you get 30fps, you *might* not notice the difference between your 30 and 90fps. In a shooter or other game where the screen moves around quite a bit, I'm sorry but I can see the difference between 30fps and 70fps quite easily...
It's interesting to note that the US military has done extensive testing in this area, specifically so that they can build simulators as absolutely 'real' as possible, and not produce any extra frames (and the increased cost involved in delivering them). According to a few engineers from Evans and Sutherland, who at least used to build the image generators for them, the vast majority of fighter pilots were unable to distinguish between framerates above 60fps.
Of course, then there's the whole 'aliasing' you get whenever you actually have a 'frame-based' video, compared with 'real life'. Case in point: Ever notice how helicopter blades, propellors, wheels, etc. seem to spin 'backwards' on TV? It's sample aliasing. Even your own eyes see this whenever your light source 'blinks', which is the case in nearly all artificial light. Take a bicycle tire, put it between your eyes and a flourescent light, and spin it; you'll see the aliasing artifacts with no problems. Take the same bicycle tire outside (in sunlight), and do the same thing-- no more aliasing!
To realistically remove all aliasing, we'd have to have much higher framerates than 60fps; however, it's generally considered a 'normal' thing, since we grew up seeing it, and nobody fusses about it.
-- Sometimes you have to turn the lights off in order to see.
C. It exhaust the heat onto those other cards.
I would rather it blows the hot air to the other PCI cards than to the CPU. Most modern CPUs are already hot enough by itself. So putting the GPU on the other side will essentially blow the hot air towards the CPU, which would make it hotter still.
I remember reading that John C is going to cap the FPS on Doom3 to like 30 or 40 FPS per seconds. I'm hoping so, I bet he is tired of people grading video cards by how many FPS they can get Quake 3 running.
The best thing about the FX isn't the overall FPS per second. It is the pixel shaders and such. The number of instructions it can excute per shader, and the rate at which it processes these is the real evolution of this card. The more complex the shader and the faster they run the more life like graphics will look.
We have been stuck in the same basic quake engine for a while now. Unreal II and Doom 3 ( doom3 more ) will be the first real change in graphics we've had. Now the GPU's can handle movie style rendering, without a ton of little tricks.
We really do need the horse power. The FX could probably render toy story in real time, that is pretty amazing. I can't wait till I can watch a movie and pause it and change the angle. The ability to have true 3-d movie projection is becoming more realistic with this type of hardware ( of course we need the 3d projector )
$400 dollars for this is nothing. You don't seem to realize that just 10 years ago a 486 DX system could cost over $4000 grand. With 16 megs of ram and 1/2 gig of harddrive. The price is rather low considering what it takes to create such wonders, stop bitchin.
Open source will help out in this arena as well. You got to think that the pros that did the work on Golem for LOTR are fans of open source, it won't be long until those kinds of shaders and techniques will be available for game programmers.
To me saying "why do we need all this power" is kind of sacreligous. Remember that increasing speed and creating a market for new hardware is what keeps most of us employeed. Never say more speed is a bad thing. And don't blame sluggish performance on the developers, as software becomes more complex you have to give up some performance for stability and expandability.
Of course, with this idea, you start running into other problems, namely reworking the bus to accomadate high speed tranfers from memory to the graphics processor. Current AGP spec is woefully underpowered, compared to the throughput the cards themselves manage. Actual numbers are in the range of 1GB/sec or so on AGP4x, and ~8.2GB/sec off local ram on the Radeon 8500LE card, ~19.2GB/sec on the Radeon 9700 Pro. Slightly less on the new GeForce FX card - Clearly not in the same ballpark. Conventional DDR memory cannot compete with the timings that video card DDR pushes - Buying ram sticks for the video subsystem will be rather expensive, unless you're willing to settle for vastly decreased performance. Consider - The average PC2100 DDR stick runs at 133MHz, doubled to 266. A PC2700 stick, so far as I know, runs at 166MHz. PC3200 is just beginning to flirt with graphics card speeds, at 200MHz, or 400MHz DDR. nVIDIA is mounting 500MHz chips on their GeForce FX card, and ATI packs ~350MHz chips on their 9700 part
(Actual clock speeds, not doubled DDR speeds). A drastic reworking of the motherboard layout, and a considerable increase in complexity, would be required to properly support this.
Then you get issues with the socketing standard - how long with ATI, nVIDIA, and everyone keep playing ball with each other? How long before nVIDIA leans on a motherboard manufacturer, using their nForce chipset, and creates a non-standard socket? Power requirements, as well - Will the motherboard be able to power the chip, or will we have to plug in a lead from the powersupply akin to these new powerhouse cards?
Interesting upsides to the situation would include the potential to use a G4 Mac style dual-, or perhaps quad-processor modules, for increased processing power - but that has the potential to easily saturate the bus, also bringing us back to the original concept of having everything mounted on an independent board module.
Problem.
Volume.
ATi doesn't ship lots of chips to be sold to OEMs on the cheap. nVidia does, and will still do. This was 3dfx's problem, and this will be what keeps nVidia alive. Whether or not it'll keep them competative or have them go the way of the Trident or not is another story.
Non impediti ratione cogitationus.
nvidia has lost this and probably the next generation of 3D to ATI. ATI's Radeon 9{5|7}00 is a very good card. Superior to the GeForce4. By the time the GeForceFX is released, ATI will have their next-generation chipset prepared. nvidia will be a generation behind. ATI cards are already close in price to their nvidia bretheren. nvidia needs a new product to get the performance crown back, or ATI will dominate.
Slashdot is a waste of time. I enjoy wasting time.
OTOH, a number of ATX motherboards I've worked on recently don't use the far-right (towards the CPU) slot, putting the AGP in the 2nd slot. In cases like this, one could not only have a 2-slot cooling system, but have a convenient exhaust vent attached to the card-cage, if they were to use the extra space behind the card.
Considering that even good motherboards barely break the $150 mark, while high-end GPUs can be $400+, it doesn't make much sense to make the GPU fit the mobo, when you can find a mobo to work with your GPU of choice.
my sig's at the bottom of the page.
Ever since AGP rolled out, most system builders have written off the first PCI slot.
The first reason is that the first PCI slot tends to conflict with the AGP slot in terms of resource managment. This may no longer be a problem, but old habits die hard.
The second reason is the damn heat-sink and fan is on the bottom of the card. I'll never figure this one out, but why did the hardware enginers do this? The heat from the heatsink rises back into the card and makes the ambient temp even hotter. Most people leave PCI 1 open to help dissapate this heat.
A third reason is that most people are not going to fill their slots anyway. Good mobos today have good sound, 10/100 NIC, and USB2 onboard. Add a good video card, and the rest of your slots are pretty much empty. Even if you add another card, just follow the urinal code. Never place 2 cards too close for comfort.
In short, the 2 card rule has been the de-facto standard for years now, why shouldn't nvidia embrace it for their own purpose?
I'd rather you do it wrong, than for me to have to do it at all.
There's a bit more to this as well:
You can't just take a current chip design, shrink it from the 180nm to the 130nm process, and expect it to run. If it does, it would be a miracle of a cosmic sort. As far as changing processes go, it's somewhat reminiscent of taking an SUV, and pulling out the engine, and putting in an electric motor-- and expecing everything to work fine, except 'faster' or 'better'. Ain't gonna happen
Most chips are written in a HDL (Hardware Descriptor Language); ATI and nVIDIA use, among others, VeriLog and VHDL. Both of these languages have their behavioral-level code, which is somewhat reminiscent of a traditional C program. (Make no mistake, HDL's are a totally different ballgame to a programming language). Then, after you have the behavioral code working (meets timings, etc.), you synthesize (compile) it.
Here's where it gets tricky:
Synthesis involves taking your process (fab size, power, material, and other characteristics), and create an optimized layout of gates to perform the tasks described by the behavioral code. The synthesized code almost definately does not behave exactly like the behavioral code-- but the synthesized code is close enough -- just barely, to meet the critical timings, and the whole thing works.
Quite often, the synthesized code will utterly fail, and the offending part will have to be identified, diagnosed, and fixed. But the fix will probably break something else. It's like putting carpet in your bedroom, and suddenly the ceiling caves in. Fix the ceiling, and the walls turn pink. Repaint the walls, and the bed becomes sentient.
The thing to remember is you get used to the 'personality' of a given fab process, and begin to pre-emptively put in fixes to avoid seeing them at all. But the instant you change fab processes, the entire 'personality' of the synthesis changes, and all bets are off. The entire design will have to be re-synthesized, re-simulated, and re-debugged. And that's before it hits silicon.
-- Sometimes you have to turn the lights off in order to see.
no it wouldn't, because the S-class and the 3 series operate in different market segments and command very different prices. Apples to oranges.
That was classic intercourse!
Speaking of motion blur brings to mind other practical uses for an accumulation buffer...
I wonder why today's uber-fast video cards don't offer temporal anti-aliasing to use with older, slower flat panels (that are in the 20 to 30 Hz max refresh region). Three to five consecutive frames blended together, then the result output at the slow refresh, and it wouldn't feel so slow at all. No jerkiness, no tails on screen, just steady going, smooth looking display.
I'd like to be offered that option. And plenty of those kind of panels around! (As to video cards, AFAIK, accumulation buffer has been part of the DX spec for some time. In OGL a very long time.)
This wouldn't solve a thing. In fact, it'd cause huge numbers of problems.
First off, the reason it eats 2 slots is because the 2nd slot is used for the blower. If you invert everything exactly where are you going to vent the blower? There's no standardized hole available for this kind of thing.
Second, it would render it incompatible with most motherboards. You'd hit either an I/O header, the CPU slot, or (most likely) support electronics like capacitors and the like. There is generally not a great deal of space between the AGP slot and anything above it because there are minimal (if any) specs requiring distance. A small number of MBs had problems with high end graphics cards right now because of heat sinks on the back of the cards -- they usually end up hitting caps, which is the last thing you want to do (ever short a cap? Not good)
But I wasn't talking about motion blur. I'm talking about TAA
I thought TAA and motion blur were the same thing, as this page seems to imply. Is there a difference?
I meant that it should be offered as a choice in the video card drivers. Much like Nvidia etc. offer enforced spatial anti-aliasing (FSAA)
The OpenGL rendering model introduces a big difference between spatial effects and temporal effects. The coordinates passed to OpenGL are three-dimensional (x, y, z), not four-dimensional (x, y, z, t)[1]. In OpenGL's coordinate space, coordinates of individual polygons within a rendered scene can be considered continuous. Only the rasterizer breaks this continuity, and a video card can choose traditional rasterization or FSAA rasterization because it has access to the (x, y) coordinates for interpolation within a pixel. On the other hand, OpenGL time is discrete, and each frame is considered a separate scene. There's no way to automatically map which polygons in one scene correspond to which polygons in another scene, so there's no way to interpolate an edge's coordinates along the fourth dimension.
[1] 't' (the time dimension) has nothing to do with 'w' (the homogenizing factor in homogeneous coordinates).
Will I retire or break 10K?