Slashdot Mirror
AGP4X vs. AGP8X
An anonymous reader writes "With upcoming chipsets such as the SiS648 claiming support for the latest AGP8X standard, we asked ourselves if there were any performance benefits. We took the SiS648 and Xabre 400 reference boards, modified them and compared the results." I can't even get 4x stable under XP, so I figure 8x is
half as likely to let me play NWN :)
you building your own motherboards or something? What mainstream motherboard doesn't have AGP 4x?
2.18 gigabytes a second. Jesus - does anyone else see why this is wierd to me? I mean, I understand the need for faster hardware, but can't software producers just make their software more efficient? I mean, any game I'm playing that requires 2.18gb of data to be passed through my video card each second is going to require a better, faster computer than I've got now. I'm tired of always being forced into upgrading just to play the latest and greatest games - and then being told that I'm breaking the law when I want to play old ones that I can't buy anymore! It's absurd, and it makes perfect sense - too many software companies have a vested interest in hardware - the more advanced the game, the more hardware that sells. What we really need is another Mario Bros. or Tetris to come along to give us all a kick in the face - great games don't need outstanding graphics to be great fun.
dammit i was trying to post some replies on the forums when suddenly the server stopped responding
thanks for slashdotting the server. thanks a lot.
i was wondering why the sudden slowdown when its 4am here (singapore), and i launch a new browser window, and the first thing i see is the agp 4x vs 8x article, and the page linked is hardwarezone.
According to the article, so far there's only a 4.7% increase between the 4x and 8x cards. Personally, I'd say thats a pretty good start. Of course, I'm still using a GeForce2, TNT2, and Rage128...so as you can see, graphics cards aren't that big of a deal to me.
:(.
:(.
What I get worried about with these upgrades is that they're going to come out with games that actually require them! And them I'm screwed
Personally, I find it interesting that it continues to seem like every card *needs* more bandwidth, more power, etc (and yes, i know these cards operate at lower voltages, but still...). Someday I'm going to need that special SOUNDBLASTER QUASIEXTAGYWITHCHERRIESONTOP made for the SPECIAL SOUND CARD BUS WITH MORE BANDWIDTH. I dread the day when i need a special slot for every type of card i want
So gogo with the ultrauberbandwidth increases, but keep that backwards compatability! I like pci graphics cards sometimes!
Taco it amazes you can't get an operating system to work that was designed for the hopelessly clueless. Maybe my grandma could give you some tips.
Basically, SiS has come out of nowhere with motherboards that absolutely trash the competition in regards to performance and features. It started last year with the SiS 735, the best performing Athlon mobo of the year. Sadly, it was a poor overclocker, so it was shunned by AMD fans. But this year SiS has had a string of hits. It's the only 3rd party with a P4 license, which makes it the only choice for mobo manufactures in terms of 3rd party P4 mobos (obviosuly they're ansty about Intel frowning upon their Via-based P4 boards, seeing as Via doesn't have a valid P4 license). The SiS 645, 645DX, and now the 648 have consistently been of high quality with features no one else has. The 645 introduced MuTIOL which doubled the bandwidth between north and south bridges, to 533MB/s. The 645DX introduced unnofficial, rock solid DDR400 support. Now the 648 again doubles bandwidth between north and south bridges to 1 GB/s, it introduces AGP 8x, and it probably will officially support DDR400. SiS 648 boards also have Serial ATA support. This is a far cry from a decade ago, when everyone knew SiS=shit.
I have a Supermicro P6DGU, which is a great board (2mb mem, up to dual 1ghz, 6pci, scsi onboard, raidport option), but it only supports AGP 2x. I don't know what the real difference is between 2x and 4x, I can't really think that it's %200 faster.
Sometimes boards that have some of the features that you want, don't have all the features that you want, and when you spend alot on a good server/workstation board, you can't always jump to the newest standard on a whim.
Tibbon
tibbon.com
Maybe I missed something, but it really would have been nice if they explained how these tests stressed the AGP bus. You're not going to get much of a performance boost out of better AGP unless you're running tests with more textures than can fit in the on-board memory.
This review / test is bullshit. The only reason that they see an improvement in lower resolutions is that its the only resolution where the game / app is not limited by the video card.
/. editors will stop "jumping the gun" and wait until some real reviews come out. This is like testing a new high performance tires that can go upto 400/mph with a Yugo. Is anyone going to be surprised when the $25 tire performs just as good as a $400 tire? Sorry for the lame analogy, haven't had my morning Coke. :-p
I'd definitely take this with a grain of salt until someone can do a 4x/8x review with a NV30 or a ATI 9700.
What kind of hardware guy looks at this and doesnt say "WTF Xabre 4000?? What kind of video card is that to benchmark anything?"
Hopefully the
-- D3X
Most people will say agp8x is way too much and overkill and will introduce some bugs and firmware/hardware/signal issues with some lower quality cards, etc...
:)
Well, when AGP 1x was out, people didn't find it very useful because it wasn't fast enough
AGP2x was okay to offload the PCI bus and do some basic stuff, but not fast enough for high-speed games and transfering large chunks of information.
AGP4x seems to be okay for today's technology and all, and AGP8X seems to be way overkill, but I personnaly think that it's finally what it should have been since the start: a *VERY* fast graphics port on which the bandwidth bottlenect doesn't become an issue, * at any resolutions * , and that help cutting down the cost in other fields beside gaming. (one example: uncompressed video editing 1600x1200@24bits(or more for film and with newer card with better colorspace) @60FPS) Right now you require exotic hardware for this, especially for uncompressed playback. let's say you'd want to invest on a fast Ultra320 array (ok you'll say if you do so you can afford the exotic hardware as well, but the point here is actually CUTTING down the price, and this is one way), well now you could get way more drives for your system.
There are many more examples for this, but the main idea is there are new features that are going to come out for cards, bigger bitdepth, better this and that, that's going to choke the bandwidth and 256MB on a card won't be enough in a not so distant future, using system memory at almost local memory speed increases quality and possibilities tremendously, and while we don't see much use right now, I'm sure it won't take long after 8x is installed that we'll see a use for 12x or 16x
--- Metamoderating abusive downgraders since my 300th post.
"Don't hold your breath waiting for it. When we announced it we didn't realize that Linux owners are all cheap fucks who don't pay for games."
I have a ATI AIW 32MB DDR, and my tribes2 can get a lil laggy esp when their are some vehicles on my screen.
reading my X log I notices that DRI was using 1X mode for AGP. after some RTFM, I found the option to kick it into 4x.
Anyway the point being it didn't help speed up the game gfx (well i didn't notice much difference)
In case ur wondering for ATI cards the XF86Config option is:
Option "AGPMode" "4"
Also i noticed:
Option "AGPSize" "32"
But i cant tell if setting this bigger then the ram on the card helps or not (maybe that the buffer size opt?), was hoping to let the card borrow more of my sys ram (which is pc100, slow compared to gfx cards DDR, but better then hdd =)
anyone know any other good opts to help eek more speed?
I can't even get 4x stable under XP, so I figure 8x is half as likely to let me play NWN
It's always about you, isn't it?
"[...]there's actually quite a bit of advantage with AGP8X especially at lower resolutions."
What are these people smoking? The vast majority of the tests are all but identical. The VERY BEST performance difference is 3DMark2001SE Pro at 800x600x16, and it shows a whopping 4.7% improvement.
Clue: In the current 3D world, AGP4X IS NOT a constraint. Even AGP2X is fine. Hell, there was an early version of the (TNT2 or GeForce 1, I forget which) that was *PCI*, for chrissake, and it was only a whisker slower than the AGP cards at the time.
Geometry transfer, it would appear, just isn't very bandwidth intensive. The only time the AGP rate is going to matter much is when doing very heavy texturing from main memory, but that just isn't happening. Instead, manufacturers are putting more and more RAM on the video card instead, and all the games are oriented around pre-loading all necessary textures in that specialized, super-high-speed RAM.
At the present 1.06 MB/sec transfer rate of AGP 4X, that means that the entire video RAM of a 128MB card be filled in roughly 1/10th of a second. If you spend all the time, money, and effort to upgrade to AGP 8X, you can improve your load time by 1/20th of a second.
Just think...if you played 50 levels of some FPS a day, every day, you'd save over 15 minutes in your first year alone!
Obviously, this is a very important technology we should all rush out to buy. Thanks, hardwarezone.com! I'll trust you for all my technology reviews in future.
-----
AGP8X: Saving your time so efficiently, you won't even notice.
Well, when AGP 1x was out, people didn't find it very useful because it wasn't fast enough
AGP2x was okay to offload the PCI bus and do some basic stuff, but not fast enough for high-speed games and transfering large chunks of information.
Not fast enough to be useful? What reviews were you reading?
Back when AGP 1x and 2x were rolled out, they were found to be marginally useful because the graphics card was the bottleneck. This is true even today. Fill rate is still almost invariably the bottleneck for performance, and CPU power for geometry and physics is usually second.
The original intent of AGP was to transfer textures across the bus, with the card's texture memory just a cache of this data. But this is a _bad_ thing to do - bandwidth and especially latency of a card's on-board memory is likely to be much better than AGP transfer bandwidth and latency, so nobody in their right mind writes games that require streaming textures from system memory. This isn't going to change - the memory in your PC is optimized for being big. The memory in your graphics card is optimized for being fast. Even with a zero-latency, infinite-bandwidth AGP port, local memory is better.
All AGP is used for now is to transfer geometry data, and it's plenty fast for that (cards are still generally fill-rate limited). With on-board transformation and lighting, and further folding-in of the graphics pipeline on the way, the amount of data that needs to be transferred per frame is going to get _smaller_, not larger.
Very high AGP transfer rates are a marketing bullet-point, and not much else.
Oh, and if you're editing a 1600x1200 movie on a PC, you're limited by your disk transfer rate. No way are you storing *any* significant chunk of that in a PC's RAM.
from the bioware FAQ :
The Linux dedicated server will be distributed freely online, as close to the game being available in stores as possible. The Linux client will follow shortly thereafter. Linux users will need to own a Windows copy of Neverwinter Nights, as the Linux executables must import certain resources from those Windows CDs. All users will need to register their CD Keys (Linux users register the Windows CD Keys) at the Neverwinter Nights community site (www.neverwinternights.com). The Macintosh version will be available later in the fall (BioWare is completing the Macintosh Neverwinter Nights client and server programs, MacSoft is completing the Toolset).
-- It's always darker before it goes pitch black.
It seems like the editors have a particular list of text strings they grep all incoming submissions for, apparently among these is AGP8x. This comparison is ridiculous even for Rob to point heedlessly to. The wbesite itself is Yet Another Anandtech/Tom's Hardware ripoff design with an article that reads like a SiS fanboy on crack.
The whining and crying about AGP 8x is a bit premature and the AGP 3.0 standard has been pretty much supplanted in usefulness by graphics card manufacturers. Having a dedicated high speed port for graphics hooked up to the northbridge is a good design idea. It frees the traditionally low bandwidth nb-sb connection from needing to carry lots of graphics data. The memory sharing available in AGP has become increasingly useless as worthwhile graphics cards have scads of local memory now. About the only thing an AGP apeture is good for is an i845G chipset board or some other cheap piece of shit HPaq sticks in their computers.
The AGP 2.0 spec isn't much of a bottleneck either. Case in point, replacing the TNT2 based video card in my dual P3 500 with a GF2GTS more than doubled the 3DMark2001 SE score from 926 to 2068. The board is an IWill DBD-100 with a 2x AGP port on it. The fillrate or poly rendering ability was not adversely affected by the AGP 2x port, the only thing keeping the 3DMark score down is the relatively slow processors (as 3DMark is single threaded) and the low FSB bandwidth.
The fillrate of an ATi R300 or nVidia NV30 isn't going to affected much by an AGP bandwidth on ONLY 1GB/s. Most cards based on these chips will end up having >100MB of on board memory. It won't be too terribly long before the video card in the PC has more and faster memory than the system's main memory. Even Doom3's 80MB of textures isn't going to really stress a 4x AGP card, it would take all of a seventh of a second to transfer all 80MB of textures. Maybe AGP 8x will be on my upgrade path when the load time of a game's textures take a perceptible amount of time to load into the video card's local memory.
Rob it isn't Microsoft's fucking fault your AGP card doesn't work properly, you're probably stuck with some old VA Lin^H^H^HSoftware POS box. My system doesn't have any problems running reliably under Windows XP and I don't think too many other people running Windows 2000 or XP are having too many problems either. When do we get to mod the editors as -1 Troll?
I'm a loner Dottie, a Rebel.
Anybody else notice that their sponsors are SIS? :)
No wonder they're calling a "4.7% increase" worthwhile... jesus...
Well actually the driver he's using is a proprietary closed-source driver based on the same codebase as nVidia's windows driver. Linus and the rest of the kernel developers have nothing to do with them, in fact they basically hate them since they use their own kernel modules which add extra instability.
"(Man) tries to live his own life as if he were telling a story. But you have to choose: live or tell." --Sartre
SiS has been doing motherboard chipsets since at least the 486 era, and I/O cards before that. I've also had SiS-based motherboards, and they had bugs and instabilities I'd never even heard of before. I've long since come to associate SiS chipsets with the mobo mfgr cutting corners, and haven't seen anything yet to make me change my mind. (Tho the SiS-based I/O cards for 386/486 machines seemed to be pretty good in their day.)
... gee, I wonder if that's why SiS chipsets need heatsinks, even when no other chipset in the same class needs 'em.
As to another poster who says he likes SiS because of the "low heat/low power"
~REZ~ #43301. Who'd fake being me anyway?
This tomshardware article from a while back compares AGP 1x -> 4x... here were the results. You can see that even in the beginning of 2000, the benefits of higher AGP speed showed diminishing marginal returns.
My server
Where have you been? I have a ECS K7S5A with an SiS 735. The northbridge runs cool to the touch. Other chipsets require a heatsink fan. Shouldn't bother responding to trolls.
Just a Tuna in the Sea of Life
They need to run some tests where the memory to GPU bandwidth dominates the problem. For example, open up 3DS Max, Maya, or Softimage XSI with a complicated textured scene that can't redraw at full frame rate, and see if it helps.
The big win for more AGP bandwidth should be when the board's texture memory is full and the textures spill into main memory. Typically, game textures are tuned to avoid this, but you hit it all the time with authoring tools.
A bottleneck on geometry feed from the main CPU is unlikely, since it's hard for the CPU to generate a gigabyte/second of geometry.
right... so they DON'T DO THAT. Even 8X AGP is going to be very, very slow compared to the incredible speed of the RAM in most of the high-end video cards. There have been a few demos using AGP texturing, but all the real-life apps I'm aware of are carefully constructed to stay within that cache.
It may help doing background loads of 'seamless transition' games, but even so.... unless you're trying to stream all these textures out every frame, it's not likely to help much. AGP 4x can fill a 128MB card in 1/10th second; 8X can do it in 1/20th. Unless you get to the point of multiple updates per second, it's just not going to matter very much. Developers will use good caching algorithms and reasonably careful level design to work around AGP speed issues.
Streaming textures IS a pretty cool idea, and I would like to see games that use them. Maybe Doom 3 will, but it hasn't sounded like Carmack is trying to do anything like this yet.
The reason I was so acerbic in my original comment was that the website was talking like it mattered NOW, for the apps we have TODAY. (a whole 4.7% increase! in one benchmark! wow!).
In a nutshell: for everything out now and probably for another 18 months, AGP8X isn't going to matter a whit. Don't worry about it until 2004 sometime.
Faster memory isn't going to do much for that video card (the ATI AIW 32MB DDR). If you were using a GeForce 4, or the new Radeon card on an SDRAM system, then there would be a substantial bottleneck. However, on older, or more efficient 3D Renderers (more efficient, as in a tile-based rendering Kyro 2), SDRAM is perfectly fine, and DDR is pracially useless, as the card is not as bandwidth hungry.
His problem is A) Running Tribes 2 on an older Radeon. And B) The Tribes 2 Garage Games engine was horribly unoptimized in that game.
I agree with the observation that SiS has been an unexpected dark horse mobo/chipset candidate lately.
One correction though: Last I heard it was not a proven fact that Via doesn't have a "valid" P4 license. Via claims that the license is valid because they purchased S3, and S3 had a license. Intel claims S3's license was not transferrable. It seems the case is still up the in air, and the lawyers will have to sort it out. Via does seem to have a reasonable claim to the license, however.
Maybe that's so .. I haven't kept track of every chipset in the world, especially the newer ones. But SiS chipsets had heatsinks for about 3 years before anyone else found it necessary to do. That always made me wonder why they needed it when no one else did.
Sometimes a company sucks for years, but suddenly gets better. Maybe SiS has done so while I wasn't looking. But when I go to a computer show and examine dozens of motherboards, and the ones that have clearly cut corners are mostly SiS-based, it doesn't produce a sense of confidence in their product.
And I wasn't trolling (I *never* troll). I've been building computers for 9 years (and make part of my living that way) and what I posted are my consistent observations over that timespan.
~REZ~ #43301. Who'd fake being me anyway?