Slashdot Mirror
Tackling AGP 8X
EconolineCrush writes "AGP 8X is popping up in new chipsets and motherboards, and graphics cards are also starting to support the standard, but is there a major performance advantage over the older AGP 4X spec? According to this review of NVIDIA's latest AGP 8X-enable graphics products, no. The review also covers some of AGP 8X's new functionality, which includes support for multiple AGP ports with multiple AGP devices per port. Whether future games and applications take advantage of AGP 8X's extra bandwidth remains to be seen, but more interesting should be what companies do with multiple AGP devices and ports."
In gaming, you could use multiple POV's in flight simulators (I think M$ flight sim supports three monitors, IIRC), or racing (Front, left, right). In desktop publishing it is usful for seeing two pages at once, or four pages at once depending on what resolution you are using.
At work I leave Outlook open on one all the time, have Visual Studio open on that one and an Internet Exploder screen open on the right screen. That way when I make changes in VS on the left I can instantly refresh the IE window on the right without doing all the toggling back crap.
I also used to do reports and presentations. Having dual monitors allowed me to have Excel/Access/whatever source program open on the left, and Powerpoint on the right. I could drag a chart from Excel full size and drop it into Powerpoint without having to do cut/alt-tab switch window/paste. Much easier, gives WYSIWIG some credence to its name.
I am running dual monitors on an NT4 box with 2 Matrox Millenium PCI's (have had dual monitors for 4 or five years now I think on that one). My other box has a Matrox G450 AGP and a Matrox PCI Millenium for dual capability on it (W2K).
IMHO, Matrox makes the best multi-display drivers/cards at a reasonable price and have had them for quite a long time compared to the others. They have a quad output card also but it is costs a bit more than the duals.
ngoy
--ngoy
three things:
1) Dual head AGP cards already exist, Matrox even has a triple head AGP card.
2) What's wrong with PCI cards? If you use it for work (like you said in the first part of your comment), I don't see what's wrong with it. I'm using 1 AGP and 1 PCI right now and I'm happy the way it is. usually I use my main monitor, which has a higher resolution, for coding and at the same time my second screen is cluttered with IRC, IM and online-documentation
3) I don't think dual AGP slotted mobo's will become standard real soon: people have lots of PCI slots and that din't encourage people to go dual/triple/... screen. I rather think that dual AGP will remain something for techies, geeks and professionals.
And remember kids: the more monitos you have, the larger your penis is!
"The majority is always sane, Louis." -- Nessus
http://slashdot.jp
The bus is very one-sided. It gives 2.1GB/s to the card, but nothing particularly special on the way back. After all, the intent was to allow the cpu to stream from main memory.
"We returned the General to El Salvador, or maybe Guatemala, it's difficult to tell from 10,000 feet"
In a way yes it is true. You can only shove so many textures into video memory until it fills up. One argument could be that extra textures can be cached in main
memory and then zipped to the video card when needed. The main thing is video card memory bandwidth. ATI's Radeon 9700 does something like almost 20
GBps. I remember a company called Bitboys that was working on embedding memory into the video processor silicon (as much as 24 MB) and using a 512 bit
wide bus running at core speed to provide 40+ GBps bandwidth. The chip also had an external memory bus to hold up to 256MB of additional video memory. Sony's PS2 uses a similar setup only using a 1024 bit bus and only 4 MB memory but it does over
50 GBps.
It would be cool to see a video chip that takes the 2D, video and AGP controller off the 3d chip and just have a 3d chip with 32 MB embedded ram and an external
memory controller. This would allow for more transistors to be used for rendering purposes. The AGP/2D chip acts as a frame buffer with its own memory to further
free up bandwidth inside the 3D chip. The two chips could then be linked together via a high speed low pin bus like Hyper Transport and possibly the 2D unit
could have 2 or 4 interfaces for adding more rendering units to double or quadruple rendering power. Of course this is a costly solution but the speed
gain would be incredible.
isn't the definition of graphics port that there is only one of them, that's why it's not a graphics bus?
I read that somewhere.
I see no reason for an AGP-based system.
The AGP's bandwidth is used mostly for uploading texture data. What else can the GPU not do besides quickly uploading texture data through the AGP interface? Well, some applications desire the use of Virtual Graphics Texture RAM, but that's too much of a hack and we should realy be interested in the advantage of a unified graphics/system RAM that may allow the GPU to allocate space upon demand. We are entering into an era of high-performance RAM and we are all left holding our dicks on whether we can actualy use multiple AGP ports on our motherboards? I don't find this a necessity. Re-design the PC, I don't like being stuck in the ATX form factor pushing a mere 6 PCI slots on me. I want more hardware expansion on my future computer, but cannot. I have invested in a 164UX Alpha 633MHz computer and it is solid as a whore's heart. I have also invested in another API Networks platform, the API Dual 833MHz Alpha CS20 Rackmount, and it came stock with only 2 64bit PCI slots. I coupled, with my CS20, a Rackmount MAGMA PCI bridge and now have 1 Ultra160 Adaptec SCSI RAID (60GB mirrored storage) localy on the CS20, and the MAGMA chassis has all my favorite SIIG, 3Dfx, ATI, 3DLabs, RME, Linksys, and DEC hardware. I have benchmarked until the cows came home and I have seen only 5% to 15% advantage in using AGP videocards over their PCI equivalents and only in the transfer of Texture. AGP is a verry sad happening in computer graphics portability. Manufacturers were tied into splitting production based on AGP and PCI, mass ammounts of advanced graphics clusters were being shadowed because of availability of certain hardware; yes, I have used homogenous GPU's to compute and render 3d images using the Chromium project. The Alpha architecture *should* not end with the 21364; it should be payed more attention now that it is a more house-hold name just from its broadcasted death. So, where will ATI and nVidia take us today...fucking USB videocards? Eat my ASssssP.
Sincerily,
The Alpha Troll (commonly seen posts on linuxgames.com)
Q: When will we see an 8X AGP card in a fucking Fag-intosh computer?
A: When they've dropped in price about 85% because something better has come along; and when Steve "I only stole *BSD twice" Jobs can figure out a way to make the Macintosh "do me next, Steve, please!" lemmings think they're hip and stylish for buying one on an overpriced Fag-intosh.
Hey, Taco and Hemos: FUCK YOU for buying a TiBook. Of course, with all your really valuable VA Software stock, I'm sure you bought the top-of-the-line models. It's only the pathetic college students who READ your fucking website who fall for the Steve Jobs reality distortion field and buy the iBook with the G3... which don't quite run as fast, does it? Not NEARLY as fast... But of course, since you're a Steve Jobs butt-buddy, you can't get your money back after you find this little fact out: Mac OS X only runs ACCEPTABLY on a G4, not a G3!
Yahhhhhhhhhhhhh! Fart!
Somewhere on nvnews.net (the official, unofficial support site for nvidia X drivers) I read that the X drivers support 16 cards running at once.
I can think of several applications for this, starting with the 3dfx approach to boosting 3d performance by having each card take turns drawing a scanline (sli)
There is also a possibility 3D displays on the horizon will require more information to draw the screen (Twice as much because the scene has to be drawn once for each eye)
Another possibility is for game house use. Standard counterstrike gamehouses charge about $3@hr to rent a machine to play CS. If a player could rent a machine with a wider FOV from multiple monitors the operator could charge more to cover the costs of the extra graphics cards. I would gladly pay $20@hr to be able to play doom3 in a psuedo holodeck enviroment.
Well thats my 2cents into the fray.
Games may not use 128MB of texture memory (or they may), but video memory is used by other stuff too.
There are colour, depth, stencil, alpha buffers (probably others.) Another big thing is vertex information. A static model can be loaded into video memory. Newer cards support programs (shaders) which are executed on the GPU. All this could add up.
RAM is cheap enough that adding more has a minimal effect on price, but could be useful for various things (even down the road.)
I vaguely recall some hack which let you use your video memory as a swap device. That's nice for a system which is only used as a gaming machine part of the time.
sh
"[A] high IQ is like a Jeep; you will still get stuck, just farther from help!" --Just d' FAQs, c.g.a
Two of my desktops have PCI-X (not available in normal desktop boards only workstation boards) and it is great. PCI-X Gigabit networking and fiberchannel. Very fast.
AGP wouldn't be as good as PCI-X is. It may have the data rate but the protocol is designed for a graphics card. You could put other cards on it but PCI-X/PCI is a much better choice! (To note: PCI66/64 will give you 0.515GB/s which is a really high data rate for a desktop system to sustain.)
WinFast GeForce2 MX DH Pro
I have this card, and it works rather well. I guess the Geforce2 is getting kind of dated, but it works well w/ my 1.2g Athlon. Couple this w/ a PCI card (dual-head or not) or two, and you should be able to have as many monitors as you want.
jred
I'm not a mechanic but I play one in my garage...
Remember VOODOO 2's SLI feature that we all so loved?
Yeah, those were the days. But SLI only increased fill-rate, and not triangles/second. Granted, it was one of the best features around. By one card, get kick ass graphics and speed. Then by the second card, hook up in SLI, and boom you've effectively doubled your fillrate.
The parent poster was incorrect when quoting the PCI Express bandwidth capabilities. The initial bandwidth will provide 2.5Gb/s in each direction (or 200MB/s when overhead is included). That's a single lane, i.e. 2 pins. Up to 32 lanes can be used to provide necessary bandwidth. So, if you'd like, you could set up a 32x2.5Gb/s connection, or 80Gb/s, in each direction. That's a little over 6GB/s.
:)
As the silicon technology improves, the maximum speed of the lane will increase to 10Gb/s, for a total of 320Gb/s in the widest implementation, or about 25GB/s.
Now, that's a lot of bandwidth.
Never underestimate the bandwidth of a 747 filled with CD-ROMs.
I can't imagine games using more than 128MB of texture RAM
Unreal Tournament 2003 has a textures directory of about 1,4 GB, and it's using relatively few stacked textures. Doom 3 might well use a dozen textures on one surface, so 128MB might not last that long into the future. Fortunately, texture compression helps a lot.
The upcoming specs for DX9 and OGL 2.0 have features (128-bit color, displacement mapping, much bigger shader program support) that can begin to render in real time stuff that used to only be possible on the massive render farms owned by folks like Pixar and SKG Dreamworks. However the fist chips that impliment these features, the Radeon 9700 and nVidia's NV30 likely don't have sufficient performance to be able to make heavy use of these features realistically using only one chip.
However, when using AGP 3.0 (AGP 8x) it is possible to put more than one AGP device on a port, and thus massive SLI configurations can be made realistically enabling the heavy use of the new DX9 and OGL2 features. ATI or nVidia may design boards with 4 or 8 chips per board, all running off of one AGP slot (would probably require and external power supply) that they couls sell for a few grand a peice to companies wishing to get into the realtime, high fidelity, near realistic 3d graphics buisness.
...and IN SOVIET RUSSIA, beowulf clusters imagine 1, 2, 3 profit!!!! jokes made out of YOU!!!
No.
AGP and PCI do not share bandwidth at all.
That is the whole point of AGP--to be a port in which the I/O of other devices would not effect the performance of the video I/O.
Additionally, if the AGP port shared bandwidth with anything, it wouldn't be a port, it would be a bus.
Computer Science is no more about computers than astronomy is about telescopes. --E. W. Dijkstra
actually, it is a bus. there are 32 signal lines for both addresses and data, and the graphics chip and the northbridge both send data to each other... so, when one is sending, the other has to tristate. hence, it's a bus...
I don't see why there is a need for them. Gigabit ethernet does a theoritical max of 125 MB/s, I think PCI 2.1's max is 133, I think. And we all know that these things don't always operate at their maxes.
11 was a racehorse
12 was 12
1111 Race
12112