Posted by
Hemos
on from the look-at-the-purty-pictures dept.
~fly~ writes "Tom's has a detailed review with benchmarks of Nvidia's new GeForce256 'GPU'. " The synopsis: High expectations, but it appears to meet the demand.
True, anybody can put up a web site and review products, but I'll go to bat for Tom to say he is (or was about a year ago when I was frequenting the site) one of the best on the web when it came to technical information - especially on the whole dual Celeron overclocking thing.
Then again, I haven't really seen his site in about a year, since he became more of a video board review site.
"The number of suckers born each minute doubles every 18 months."
-- These are my friends, See how they glisten. See this one shine, how he smiles in the light.
Re:Comparison to Pro Graphics?
by
randolph
·
· Score: 2
It's there--competitive with at least the low-end SGI hardware. Basically, there is a hierarchy of computations in 3-D graphics. (Copied from the flightgear hardware requirements page.)
Stuff you do per-frame (like reading the mouse, doing flight dynamics)
Stuff you do per-object (like coarse culling, level-of-detail)
Stuff you do per-polygon or per-vertex (like rotate/translate/clip/illuminate)
Stuff you do per-pixel (shading, texturing, Z-buffering, alpha-blend)
At each level of the hierarhcy the amount of computation goes up an order of magnitude or so. The GeForce256 moves up the hierarchy to the per-polygon level, providing, (eventually, when the software properly supports it) an order-of-magnitude improvement in 3-D rendering, just like an SGI system does. There is apparently going to be Linux OpenGL support, too. Price, I believe, is in the $250 range.
Because the Geometry Processor Unit (what GPU stands for) will be optomized for processing geometry, which currently is a task of the CPU. With the GPU, all the processor will be responsible for will be feeding geometry data to the GPU (well, it's the only graphics function it'll be responsible for).
In the end, the GPU should be faster at geometry than the CPU, which is the goal.
My good old Viper550 (TNT) card seems to work with Mesa-3.0 and NVidia's "glx 1.0" driver, but I'm under the impression this is only a "partial" set of drivers.
It's a complete OpenGL driver AFAIK, but it doesn't do direct rendering (it goes over the X pipe), and it's not nearly as optimized as the Windows drivers yet.
(Myth II supposedly only works with 3dfx brand boards because they're the only ones that have a 'complete' set of OpenGL drivers - though I've been trying to find out if that's really true or not...)
It's not. Myth II uses Glide. 3Dfx is actually unique in being one of the last people to *not* have a complete set of OpenGL drivers; that's why they have their "MiniGL" to run Quake* games.
Has Nvidia said or done anything on the Linux front since the initial release of their drivers?
There was an interview where an Nvidia rep said they'd have GeForce Linux drivers (but X server? Mesa drivers? Who knows?) when the card shipped, but I haven't heard anything since.
but I would've like to see some more support for DVD playback. HDTV support is cool but
Proper HDTV support requires mostly a superset of what is required for DVD support; everything but the subpicture decode.
I think that a full MPEG 2 decoder, that would be a little excessive since that would mean adding an audio output to the card.
No, adding an MPEG 2 decoder to the card doesn't necessitate adding an audio output.
This can be done two ways:
Leave the demux of the MPEG transport or program stream to the main CPU, and only hand the video PES (or ES) to the card. Demux is fairly easy and won't suck up but a tiny fraction of the main CPU as compared to doing full decode.
Let the card demux the transport or program stream, and hand buffered audio PES (or ES) data back to the main CPU.
As for software decoders, the ones for Windows don't seem very optimized.
They're not very good, but that's not because they're not very optimized. Just compare any of the Windows players to the NIST code if you want to see the huge difference between majorly optimized and non-optimized decoders.
Why should he have to respond? Because some people question the validity of his OPINIONS?
Q: How do we evaluate someone elses OPINIONS?
A: The same way we evaluate anything else, if you want to, go for it, if you don't like him, don't read him.
Apparently, not reading his articles has expanded to the realm of trash-talking him from the comfort and safety of the AC post.
If you're going to lay into somebody, please, have the courage to accept personal responsibility, and link to the allegations instead of giving a vague, biased (but presented as unbiased) description of what these allegations were.
You're lucky I ran out of moderator points already.
-- by Mike Buddha -- Someday the mountain might get him, but the law never will.
Tom should have mentioned...
by
ParallaX-
·
· Score: 2
The reason why most of the benchmarks were so close is because none of these games (with the exception of parts of Quake3) are using the OpenGL T&L pipeline because at the time they were made there were no hardware T&L engines and so by 'rolling their own' T&L they could get significant speedups. The nVidia Tree demo should be evidence to anyone what a dramatic difference having hardware T&L can make. That tree demo has far more complexity than your average shoot-em-up game, and these are the kind of things we can expect when developers make games for hardware T&L (most new games will use the hardware). So the real problem with the benchmarks was running a bleeding edge graphics card on yesterday's software. It does well, even better than the competition, but don't expect a 3X increase...you can't get much faster than 100FPS no matter how you try. But the GeForce should be able to do 60FPS with 10X the polygon count of current cards (assuming the developer is handling T&L with OpenGL)
"Why not just slap another processor on the motherboard and call it a "GPU" instead?"
Theoretically, you could do that but you would need a mighty powerful CPU to achieve the level of performance of the GeForce since CPUs aren't optimized for graphics processing (note: GPU stands for Graphics Processing Unit not Geometry Processing Unit as someone earlier posted.) The GeForce is a much more cost effective solution for graphics processing than getting another CPU.
According to Nvidia web page about the GPU, their technical definition of a GPU is:
"a single-chip processor with integrated transform, lighting, triangle setup/clipping and rendering engines that is capable of processing a minimum of 10 million polygons per second."
The review of the GeForce 256 at Ace's Hardware has good info on comparing CPUs to GPUs. As another poster mentioned, graphics processing exists in a limited form in CPUs (3DNow!,etc.). Possibily in the future CPUs will integrate more advanced graphic processing functions. But, even if you had a CPU with complex graphic processing functions you still need some sort of display adapter. Personally, I think that it makes more sense to have the display adapter and graphics processing integrated on one unit.
Re:Tom's...and every other hardware site too
by
jkorty
·
· Score: 2
The previous links are defective. These should work:
Kinda off subject, but shame shame Tom for not using the new Matrox G400 drivers that were released on Oct 8th that includes the new Turbo GL (mini GL) drivers. Would have liked to see how the G400 Max performed with the newest drivers compared to the GeForce at the higher resolutions. From some of the benchmarking I have seen, it is giving the TNT2 Ultra a run for it's money on OpenGL games at higher resolutions.
GPU Acceleration
by
Anonymous Coward
·
· Score: 3
A lot of confusion seems to be going around about that whole GPU T&L thing when applied to Quake3, well Shugashack amazingly enough has the answer from one of the developers working with Quake3 technology. Here you go, right from the Shack. His benchmarks of the card are pretty good too.
Quake 3 does indeed use T&L and will take advantage of any hardware supporting it. It uses OpenGL's transformation pipeline for all rendering operations, which is exactly what T&L cards such as the GeForce accelerate.
Well what if Q3 used the other stuff besides the transform engine? The other three real features are the per-vertex lighting, the vertex blending, and the cube environment mapping. Since Quake 3 has static world lighting, one of the only places for the lighting to be useful would be for the character lighting, especially for dynamic lights. The current character lighting implementation is pretty quick though, I don't really see *too* much of an improvement there, though it is worth mentioning. The vertex blending may help skeletal animation, but since the current test has no skeletal animation, it would not help it at all in the current benchmarks. And the cube environment mapping won't help the game at all, since the game doesn't use cube environment mapping to begin with.
While I'm at it, the use of OpenGL doesn't necessarily mean that all games will be accelerated by the GeForce's T&L. Such examples are Unreal engine (including UT) based games. Its architecture is very different from QuakeX's and cannot benefit from T&L hardware without rearchitecting the renderer, as Tim Sweeney has said before.
Tomzilla
by
Anonymous Coward
·
· Score: 3
Tom Pabst (of Tom's Hardware) has gotten himself mixed up in a lot of tough questions about his journalistic integrity (or lack thereof). There have been many accusations that he was a little too generous with certain reviews in exchange for getting hardware to review before anyone else on the 'net, and there was a big stink about him rushing to publish Q3Test "benchmarks" without even looking into whether such a thing would have any basis in reality. Tom has responded to some of these allegations, and his responses have not been particularly professional.
Anyone on the net can put up a web site and review products. Don't take other people's reviews (which are really opinions) as truth. Question seriously those who are writing. Many online authors have not displayed much professionalism, and those types are probably best avoided.
Re:Comparison to Pro Graphics? Here:
by
TraumaHound
·
· Score: 3
Riva3D ran the GeForce256 through Sense8's Indy32 benchmark. The results are here.
Far as I can gather, looks pretty promising. (with the right CPU. They used an Athlon.)
Seems to me like the next logical step is to have a graphic card that can handle more of the game duties. If a box is built right, the CPU can be slow but everything flies because the work is handed out to chips specialized in different tasks (see: Amiga, mid 1980's, which is still a superior design than any current PC). This chip makes a good first step in that direction, taking over the lighting and such, eliminating the need for faster AGP transfers and such. Ideally, I would like to see a graphics board that actually takes over some of the program itself. Of course it would be even better to have a NUMA motherboard and have chips dedicated to I/O, another to graphics, another to sound (not through the ISA/PCI slot), thus the CPU itself wouldn't have to be the latest greatest to turn out incredible results. These guys are turning out a chip in the ballpark of $100/piece wholesale that runs circles around any CPU. The whole computer needs to get that way. The only time you should ever need a fast CPU is for science/math, not for a normal desktop machine. ***Of course Transmeta might change the whole scenario, because if their chip can be reprogrammed on the fly to do things like graphics then there's no need for so much hardware.
random, possibly baseless points and conjecture
by
LocalYokel
·
· Score: 3
At first, I thought the moderators were all smoking crack again, but I see that they probably ran out of moderation points... Why is it that the subject of 3D graphics cards seems to bring out such obnoxious folk?
Frankly, I'm just not interested in these new components. Is an extra $100 enough to justify a 5% increase in performance, and if so, how many generations should be skipped after that before upgrading? Nvidia is talking about a 6 month schedule (though nine months to a year seems more realistic).
At the rate things are going, graphics cards will soon be the most expensive component in every system, even with RAM at its current prices. I'm also willing to bet that NetBSD will be ported to exclusively use the GPU, bypassing most components altogether, before the product is even released...
For me at least, I can't justify the costs of upgrading my system every six months just so I can play the newest rehash of a ten year old game. It doesn't impress me that the *new* version gives you more control, gore, levels, and/or 3D graphics -- I liked the *old* game just fine.
The CPU or component speed haven't been the bottleneck in games for a long, long time. The imagination of game developers has been occupied with utilizing the hardware acceleration buzzword of the moment, not with developing new groundbreaking ideas...
Doing a search for geforce on www.linuxgames.com revealed this snippet from an irc log: ----------- ([Jar]2) (orlock) WIll they still be supporting Xfree86/Mesa3D/glx/linux/etc like they have in the past? (nvdaNick) Yes.
(MicroDooD) (LaRz17) Will drivers for multiple operating systems be released at the same time? (nvdaNick) As for driver releases, I think NVIDIA is planning to release all drivers at once.
([Jar]) (MfA) Will the non windows drivers be open source? (ie not run through the pre-processor) (nvdaNick) What would you want with open source drivers, by the way? (nvdaNick) I'm not sure what our plans will be regarding that. -----------------
\begin{speculation} Anyway, if this is correct and nVidia is going to be have official support for Linux they are probably going to use SGI sample implementation and thus cannot release their driver as open source. \end{speculation}
Boy, does he hate 3dfx o_O
by
Chris+Johnson
·
· Score: 4
What happened with that, did they make fun of him or not give him cards to test or something? Like anybody, I have pet vaporware that I'd like to see succeed and become real, and for me that's the next generation 3dfx stuff with the antialiasing and motion blurs (in which the former would work with old games too). It's OK with me if it doesn't fly, I'll still wait and see what happens with it, but it's pretty boggling to see this guy kicking at 3dfx so bad. He was coming up with these big benchmarks for a GeForce card that people can't even get yet, and making nasty remarks about how poorly the Voodoo3 measured up (when actually Glide ran competitively when available), and how old is the V3 by now? Compared with a GeForce that people can't even get ATM?
It's nice to see that they've apparently added some of the MPEG 2 motion compensation support that ATI has had for a while. But I really wish they would bite the bullet and add a full MPEG 2 decoder. It would only take about a half million transistors; no one would even notice the extra die area.
Software MPEG 2 decoders for Windows basically suck, and there aren't (yet) any real-time decoders for Linux anyhow. Hardware decode is the way to go.
I keep hoping that someone will ship an inexpensive VIP-compatible MPEG 2 decoder daughterboard that I could use with my Asus V3800 TNT2 card, and it hasn't happened yet, but simply building it into the next generation nVidia chip would be even better.
Eric
Tom's...and every other hardware site too
by
sugarman
·
· Score: 5
Slashdot Mirror
True, anybody can put up a web site and review products, but I'll go to bat for Tom to say he is (or was about a year ago when I was frequenting the site) one of the best on the web when it came to technical information - especially on the whole dual Celeron overclocking thing.
Then again, I haven't really seen his site in about a year, since he became more of a video board review site.
"The number of suckers born each minute doubles every 18 months."
These are my friends, See how they glisten. See this one shine, how he smiles in the light.
It's there--competitive with at least the low-end SGI hardware. Basically, there is a hierarchy of computations in 3-D graphics. (Copied from the flightgear hardware requirements page.)
At each level of the hierarhcy the amount of computation goes up an order of magnitude or so. The GeForce256 moves up the hierarchy to the per-polygon level, providing, (eventually, when the software properly supports it) an order-of-magnitude improvement in 3-D rendering, just like an SGI system does. There is apparently going to be Linux OpenGL support, too. Price, I believe, is in the $250 range.
Because the Geometry Processor Unit (what GPU stands for) will be optomized for processing geometry, which
currently is a task of the CPU. With the GPU, all the
processor will be responsible for will be feeding geometry data to the GPU (well, it's the only graphics
function it'll be responsible for).
In the end, the GPU should be faster at geometry than the CPU, which is the goal.
I was wondering the same thing.
Me too.
My good old Viper550 (TNT) card seems to work with Mesa-3.0 and NVidia's "glx 1.0" driver, but I'm under the impression this is only a "partial" set of drivers.
It's a complete OpenGL driver AFAIK, but it doesn't do direct rendering (it goes over the X pipe), and it's not nearly as optimized as the Windows drivers yet.
(Myth II supposedly only works with 3dfx brand boards because they're the only ones that have a 'complete' set of OpenGL drivers - though I've been trying to find out if that's really true or not...)
It's not. Myth II uses Glide. 3Dfx is actually unique in being one of the last people to *not* have a complete set of OpenGL drivers; that's why they have their "MiniGL" to run Quake* games.
Has Nvidia said or done anything on the Linux front since the initial release of their drivers?
There was an interview where an Nvidia rep said they'd have GeForce Linux drivers (but X server? Mesa drivers? Who knows?) when the card shipped, but I haven't heard anything since.
This can be done two ways:
- Leave the demux of the MPEG transport or program stream to the main CPU, and only hand the video PES (or ES) to the card. Demux is fairly easy and won't suck up but a tiny fraction of the main CPU as compared to doing full decode.
- Let the card demux the transport or program stream, and hand buffered audio PES (or ES) data back to the main CPU.
They're not very good, but that's not because they're not very optimized. Just compare any of the Windows players to the NIST code if you want to see the huge difference between majorly optimized and non-optimized decoders.Why should he have to respond? Because some people question the validity of his OPINIONS?
Q: How do we evaluate someone elses OPINIONS?
A: The same way we evaluate anything else, if you want to, go for it, if you don't like him, don't read him.
Apparently, not reading his articles has expanded to the realm of trash-talking him from the comfort and safety of the AC post.
If you're going to lay into somebody, please, have the courage to accept personal responsibility, and link to the allegations instead of giving a vague, biased (but presented as unbiased) description of what these allegations were.
You're lucky I ran out of moderator points already.
by Mike Buddha -- Someday the mountain might get him, but the law never will.
The reason why most of the benchmarks were so close is because none of these games (with the exception of parts of Quake3) are using the OpenGL T&L pipeline because at the time they were made there were no hardware T&L engines and so by 'rolling their own' T&L they could get significant speedups. The nVidia Tree demo should be evidence to anyone what a dramatic difference having hardware T&L can make. That tree demo has far more complexity than your average shoot-em-up game, and these are the kind of things we can expect when developers make games for hardware T&L (most new games will use the hardware). So the real problem with the benchmarks was running a bleeding edge graphics card on yesterday's software. It does well, even better than the competition, but don't expect a 3X increase...you can't get much faster than 100FPS no matter how you try. But the GeForce should be able to do 60FPS with 10X the polygon count of current cards (assuming the developer is handling T&L with OpenGL)
Where is the cheapest place to buy one of these things? (including shipping)
-- Virtual Windows Project
Thanks, I was serching for Ge Force and nothing was turing up. Here is a link to the complete listing at pricewatch.
Prices at Pricewatch
-- Virtual Windows Project
Why not just slap another processor on the motherboard and call it a "GPU" instead?
--Adam
Anandtech GeForce 256 Review
Ace's Hardware GEForce 256 Review
RivaExtreme GeForce 256 DDR Review
The FiringSquad GeForce 256 DDR Review
GA Source Guillemot 3D Prophet Review
3DGPU Geforce 256 DDR Review
Fast Graphics Guillemot 3D Prophet Review
CGO GeForce 256 Preview
Shugashack GeForce, V3 and TNT2 benchmark roundup
Riva3D Full GeForce 256 DDR Review
GeForce 256 DDR Review at Planet Riva
Kinda off subject, but shame shame Tom for not using the new Matrox G400 drivers that were released on Oct 8th that includes the new Turbo GL (mini GL) drivers. Would have liked to see how the G400 Max performed with the newest drivers compared to the GeForce at the higher resolutions. From some of the benchmarking I have seen, it is giving the TNT2 Ultra a run for it's money on OpenGL games at higher resolutions.
Anyone on the net can put up a web site and review products. Don't take other people's reviews (which are really opinions) as truth. Question seriously those who are writing. Many online authors have not displayed much professionalism, and those types are probably best avoided.
Riva3D ran the GeForce256 through Sense8's Indy32 benchmark. The results are here.
Far as I can gather, looks pretty promising. (with the right CPU. They used an Athlon.)
Seems to me like the next logical step is to have a graphic card that can handle more of the game duties. If a box is built right, the CPU can be slow but everything flies because the work is handed out to chips specialized in different tasks (see: Amiga, mid 1980's, which is still a superior design than any current PC). This chip makes a good first step in that direction, taking over the lighting and such, eliminating the need for faster AGP transfers and such.
Ideally, I would like to see a graphics board that actually takes over some of the program itself. Of course it would be even better to have a NUMA motherboard and have chips dedicated to I/O, another to graphics, another to sound (not through the ISA/PCI slot), thus the CPU itself wouldn't have to be the latest greatest to turn out incredible results. These guys are turning out a chip in the ballpark of $100/piece wholesale that runs circles around any CPU. The whole computer needs to get that way. The only time you should ever need a fast CPU is for science/math, not for a normal desktop machine.
***Of course Transmeta might change the whole scenario, because if their chip can be reprogrammed on the fly to do things like graphics then there's no need for so much hardware.
At first, I thought the moderators were all smoking crack again, but I see that they probably ran out of moderation points... Why is it that the subject of 3D graphics cards seems to bring out such obnoxious folk?
:)
Frankly, I'm just not interested in these new components. Is an extra $100 enough to justify a 5% increase in performance, and if so, how many generations should be skipped after that before upgrading? Nvidia is talking about a 6 month schedule (though nine months to a year seems more realistic).
At the rate things are going, graphics cards will soon be the most expensive component in every system, even with RAM at its current prices. I'm also willing to bet that NetBSD will be ported to exclusively use the GPU, bypassing most components altogether, before the product is even released...
For me at least, I can't justify the costs of upgrading my system every six months just so I can play the newest rehash of a ten year old game. It doesn't impress me that the *new* version gives you more control, gore, levels, and/or 3D graphics -- I liked the *old* game just fine.
The CPU or component speed haven't been the bottleneck in games for a long, long time. The imagination of game developers has been occupied with utilizing the hardware acceleration buzzword of the moment, not with developing new groundbreaking ideas...
My US$0.01 (lousy Canadian pennies
--
E2 IN2 IE?
Doing a search for geforce on www.linuxgames.com revealed this snippet from an irc log:
-----------
([Jar]2) (orlock) WIll they still be supporting Xfree86/Mesa3D/glx/linux/etc like they have in the past?
(nvdaNick) Yes.
(MicroDooD) (LaRz17) Will drivers for multiple operating systems be released at the same time?
(nvdaNick) As for driver releases, I think NVIDIA is planning to release all drivers at once.
([Jar]) (MfA) Will the non windows drivers be open source? (ie not run through the pre-processor)
(nvdaNick) What would you want with open source drivers, by the way?
(nvdaNick) I'm not sure what our plans will be regarding that.
-----------------
\begin{speculation}
Anyway, if this is correct and nVidia is going to be have official support for Linux they are probably going to use SGI sample implementation and thus cannot release their driver as open source.
\end{speculation}
What happened with that, did they make fun of him or not give him cards to test or something? Like anybody, I have pet vaporware that I'd like to see succeed and become real, and for me that's the next generation 3dfx stuff with the antialiasing and motion blurs (in which the former would work with old games too). It's OK with me if it doesn't fly, I'll still wait and see what happens with it, but it's pretty boggling to see this guy kicking at 3dfx so bad. He was coming up with these big benchmarks for a GeForce card that people can't even get yet, and making nasty remarks about how poorly the Voodoo3 measured up (when actually Glide ran competitively when available), and how old is the V3 by now? Compared with a GeForce that people can't even get ATM?
Software MPEG 2 decoders for Windows basically suck, and there aren't (yet) any real-time decoders for Linux anyhow. Hardware decode is the way to go.
I keep hoping that someone will ship an inexpensive VIP-compatible MPEG 2 decoder daughterboard that I could use with my Asus V3800 TNT2 card, and it hasn't happened yet, but simply building it into the next generation nVidia chip would be even better.
Eric
Did an NDA expire today or something?
Just a couple quick links:
Anandtech GeForce 256 Review
Ace's Hardware GEForce 256 Review
RivaExtreme GeForce 256 DDR Review
The FiringSquad GeForce 256 DDR Review
GA Source Guillemot 3D Prophet Review
3DGPU Geforce 256 DDR Review
Fast Graphics Guillemot 3D Prophet Review
CGO GeForce 256 Preview
Shugashack GeForce, V3 and TNT2 benchmark roundup
Riva3D Full GeForce 256 DDR Review
GeForce 256 DDR Review at Planet Riva
Any others?
--sugarman--