Printable View
That's multi Chip ;) not multi-card
And SimFusion have there own achitecture for deploying rendering farms
using multiple cards, any dual card system is not a ATI solution.
don't get fooled by the web babble ;)
And for info the X800 architechture can reportedly by ATI support 256 CPU's
and 1Gb of memory however they only have a 1 board multichip solution any
extension of that would be third party design.
being a fanboy is only gonna leave you settling for second best hardware because your pride will be hurt
but by the time you buy a second 6600 for less than you originally paid for your first 6600, the 6800 will be a lot cheaper as well. i think perkam is right there, you can probabaly sell your used 6600 and add just as much as you would pay for a second cheap 6600 and have enough to buy a 6800gt by that time, or even a newer and faster card!Quote:
Originally Posted by DudeMiester
so i think sli is really not improving the mainstream at all but creating an new "super ultra high" end market segment even above the 6800UE and x800xt-pe cards that already for the ultra high performence segment.
and PMM, they do have multicard designs, its not atis design but their custom pcb, but i dont think they wouldnt help or sell their design knowledge to ati if they asked.
http://xtremesystems.org/forums/atta...chmentid=13886
lmao, that looks rediculous. I think ATI is going to start falling behind again. I can picture them not doing anything to compete, or thinking it's not a big deal. Reminds me of intel.
My post does say that ;)Quote:
Originally Posted by saaya
Those quad 9700 pros are nice though :drool:
Edit... those quad 9700 pro's are Multi-board in the fact its 2 PCB they
don't require 2 AGP/PCI slots, i'm unsure without finding out the facts
but when they use more than 2 quad's there usings a multiple CPU system
that handles each... could be wrong though so don't quote me on it :)
Only problem with that is getting the data too the cards. With SLI you double the cards and the bandwidth. That's only thing a single card can't do, give you double the bandwidth. One thing I'm really interested in if anyone has benched 3DLabs new Realism cards. The 800 looks like it could seriously kick Geforce 6800 Ultra's ass. Since they also support Direct3D, you can use them in all games too. Why I bring them up is because in their multi GPU design they have one only chip that processes verticies, and all the GPUs just process pixels. All it really does is add more pixel pipes. This makes sense, because you can only send so much data through the AGP, which is going to limit the number of verticies you can process, so if a single vertex processor can saturate that amount, then there is no need for more until you get more bandwidth available (like PCIe).
DudeMiester, they dont both run with pciE 16x, one runs at 12x and the other at 4x ,or they both run at 8x from what ive heard.
agp bandwith is a joke when it comes to data, we dont even need agp2x speed for current games, agpx2 4x and 8x are only faster because they include extra pins and new commands, the bandwidth doesnt matter at all afaik.
the only thing we can max out the bandwidth would be by storing textures in the system memory... and thats kinda useless, i think only halo partially does that as its a bad port from a console to pc.
and PMM, i think they use one agp and one pci 64bit slot... but im not sure.
and their big servers are based on the same system, a dual cpu mobo with a 4way videocard, but they use several of those in a small cluster i think. im not sure, but i think i read that some months back when i found a pic of that card ans spend hours to find more ifnos about it :D
one of this card costs 1.2 million $ i think... something like that :lol:
maybe it was "only" 120.000$, but anyways, it was a huuuge amount of money.
yep bandwidth is overrated. AGP vs. PCI-E. There gonna come out about the same. Altho i'm not sure if we could live with AGP 2x...4x to 8x offered little to no improvment in preformance. Same with PCI-E. I never could understand this, why not make a board with 2 AGP slots. just make them both 4x. Why would that have been no hard?
Cool i'll have one :) ... checks credit card hmmmmm dratQuote:
Originally Posted by saaya
PMM goes off to rob a bank :D
edit... ponders if they consider lending oppainter one ;) to review & test.
jjcom, i once saw an article where a guy compared different cards running at agp1x (forced) agp2x agp4x and agp8x and there was a 0.1% difference betwen agp 8x and 4x and a 5% or less difference between 8x and 1x.
people always confuse agp rates with the agp standards.
there is agp 1.0, agp 2.0 and agp 3.0 each of them offers new functions and commands that slightly speed up the cards and are making up the biggest part of the increasements when going from one agp speed to another.
agp 3.3v signal voltage 1.0 1x/2x
agp 2.5v signal voltage 2.0 2x/4x
agp 1.8v signal voltage 3.0 4x/8x
and i think you can force agp2x or even agp1x on agp 2.0 and agp 3.0 slots by insulating some pins :) lostcircuits also wrote an article about it once iirc, i chatted with michael schuette who now works for ocz about this, and i think you can force and agp 3.0 slot to run 1x speed but with all the enhancements of agp 3.0. and the difference was very very small iirc.
pmm, those cards only work in server boards with an agp pro slot and pci 64bit slots :D
and the drivers only work under opengl and probably only support the flight simulations they are used for and wont let you run doom3 ... :D
ok I didn't know just how little AGP 1x to 8x was going to affect things. cool...then it would be intersting to see what the point of PCI-E 16x was....
First, I dont think it looks ridiculous...
http://xtremesystems.org/forums/atta...chmentid=13886
...I think it looks like Optimus Prime in Green :p:
Thats an excellent point saaya and I can see how you can get confused DudeMeister because even Nvidia acknowledges that its 2x PCIe 8x not really 2x PCIe 16x like they advertise on the dual PCIe boards including Nforce 4. The thing is though that, depending upon setup, you may be able to run one card with 16x and then when you upgrade, it divides the bandwidth to 8x each - which may or may not affect fps in future games.Quote:
DudeMiester, they dont both run with pciE 16x, one runs at 12x and the other at 4x ,or they both run at 8x from what ive heard.
Not quite so:Quote:
one of this card costs 1.2 million $ i think... something like that
Wildcat Realizm 800 640MB PCI-E 2xDVI-I Retail
Which translates roughly into US$2600 - about the cost of a well made Fx-53 rig with an X800 XT :cool:
Thing is though, since Nvidia has assumed that consumers are willing to pay $1000 for two 6800Us, I think an ATI solution (dual gpu, dual vpu, whatever) will be costing the same for, perhaps, one card. Cool thing would be though, that if someone tries to steal it, you could whack him unconscious with it. :p: :p:
i heared both ati and nvidia want to use the main system memory to store textures in it. this can give high end cards a boost of maybe 1-5% and it means that you can sell super cheap low end cards with almost no video memory (wich makes a big part of the costs of those cards) and then let them use the system memory... a bit like an add in "integrated graphics" card that uses the system memory as framebuffer like integrated graphics chipsets.
furthermore pciE lets you run multiple graphics cards, its bandwidth is bi-directional wich means the up and down bandwidth are now the same, i think with agp thats not the case.
and pciE means less room and pins/traces for the same speed, this means less traces/pins on the boards=cheaper to manufacture :)
and pciE slots can now provide up to 60W to videocards through the graphics slot compared to 45W(?) for agp slots wich means cards dont need an extra power plug. this means less traces on videocards=cheaper.
and about that agp 1x vs agp8x, im not 100% about it... hmmm i will look for the article on lostcircuits :D
Uh huh...Saaya: dont you have anything better to do than answer 5sec after a post :)
Ive waited days before getting answer on other forums. Sheesh
For the best though. ANyway, can you say without a doubt that future GPU configurations will be cheaper than today's even if they use system memory to store textures - I certainly wouldnt put any money on that.
here, i was wrong!
its 3.3v 1.5v and 0.8v and not 3.3v 12.5v and 1.8v :D
and you can actually run agp1x with agp 2.0 but not even agp2.0 with an agp3.0 slot/board/card
http://www.lostcircuits.com/video/asus_v9280s/2.shtml
theres a 1% difference between agp 3.0 agp8x and agp4x and another 1-2% between agp 3.0 and agp2.0
i think the "4200" is the 4200 agp 2.0 forced to agp1x speed wich shows an impact of 3-50% in performence!
i cant imagine that its the limit of the bandwidth for data though, i think ut2003 and specviewperf both store textures in the system memory and also use the new instructions in agp3.0 wich explains the big difference between agp2.0 and agp3.0.
in the end he says himselfso i think if we had the bandwidth of agp1x and the instructions of agp8x (=agp3.0) it would work with all current games just fine and would have no big performence impact :)Quote:
In most D3D applications, therefore, AGP2X still suffices
:stick: dude, wth is your problem? i wasnt sure about what i said so i posted again and DID SOME RESEARCH because i dont want to tell anybody some BS and give anybody wrong infos LIKE YOU DID ;)Quote:
Originally Posted by perkam
i never said they would be cheaper to buy, they will be cheaper for ati and nvidia=more money for them... :DQuote:
Originally Posted by perkam
Thats fine but my question is: What effect would running those games that use agp8x instructions have on a agp 2.0 slot ???Quote:
so i think if we had the bandwidth of agp1x and the instructions of agp8x (=agp3.0) it would work with all current games just fine and would have no big performence impact
And if there is a difference, then would there be a difference between running a PCIe card (ATI or Nvidia) at 8x or 16x irrespective of whether or not they are two slots or one?
Perkam
There will most likeley be no performance gain in running pci e over agp. There is currently no bandwidth limitations. So it's not like your going to lsoe performance by running to sli cards at 8x rather than 16x. And in perkams defense saaya I don't think he was trying to insult you, I think he was just joking around, you seemed a little po'd in ur response.
I'm pretty sure he was talking about the ones in the pic posted i.e. the quad 9700 pro card Evans and Sutherland do that our posts talked about ;)Quote:
Originally Posted by perkam
My 9700 is pretty cheap ot OC. :DQuote:
Originally Posted by blinky
I dont see what the point of SLI on these cards is when the 9500/9700 series cards play all current games at smooth rates.
we're getting an incredle gap between real performance and benchmark performance.
Thx, but that matter has been resolved.Quote:
And in perkams defense saaya I don't think he was trying to insult you, I think he was just joking around, you seemed a little po'd in ur response.
Oh, couldnt find a link on google for that - go figure.Quote:
I'm pretty sure he was talking about the ones in the pic posted i.e. the quad 9700 pro card Evans and Sutherland do that our posts talked about
Indeed the real gap lies between the performance we need (60 fps in most games) and the performance we want and I guess the difference is not an issue here at xs where, as I learned a few days ago, SPEED IS KING. :p:Quote:
we're getting an incredle gap between real performance and benchmark performance.
Its interesting to see that Nvidia did NOT release gaming benchmarks along with its 3dmark05 scores and I believe that might an issue. I dont think games can utilize SLI completely as of yet (1000 fps instead of 500 in Quake3 , etc) and so right now, SLI may not be worth the cost. And even if Nvidia gets partners for its SLI program (The way its meant to be doubled ?? :) ) I dont think Nvidia will be able to make the advances in market share that it has been losing to ATi over the past few years, in just a short period of time.
Nvidia did release gaming benchmarks did you not see the doom 3 and halo benches? I believe I also saw ones elseware for other games, tho I can't remember where.
doom 3 and halo are, again, TWIMTBP games and do not reflect REAL performance increases in games like hl2, ut2004, and more
In short, Nvidia can introduce patches for Doom 3 and Halo because of its partner relationship with those developers for SLI tweaks but cannnot do the same as easily with other games.
I completely agree with this....and I think ATI cards still the better for real game performance and overall use like 2d application...Quote:
Originally Posted by perkam
single card ATI solution to SLI's benchmarks
dual R520 cores on a single PCB, PCIx16 interface, + one dual DD cascade :slobber:
***slaps face*** wakes up from dream :::looks at 2yr old 9700pro and I sigh::::
Drooling Profusely - Getting Photoshop ready for dual core r520 mockup :)Quote:
dual R520 cores on a single PCB