dual 850xtpe

[saaya]

look what i picked up today :D

http://img118.exs.cx/img118/9224/74wu.jpg

http://img118.exs.cx/img118/4282/87oh.jpg

[perkam]

Saaya, did I ever mention how good of a mod u were :slobber: :slobber: omg :eek: :eek:

How long have u been secretly saving up for those ;)

Saaya: "oh no I only have an AXP" rig
XS: he he he he
Saaya: BAM ! TWO X850 XTs !!!!!
XS: *runs like a dog with tail between legs*

Nicely done saaya, I think i have an idea as to what there for :eek: Hope to see up there in the ranks when MVP comes out :D

EDIT !!!: THOSE ARE PEs !!!!

Perkam

[G H Z]

Nice :toast:

Are these yours saaya?

[saaya]

im a little confused though, where do i plug in the sli bridge? or do dual ati cards not need that bridge thing? but how do i enable the dual cards then? so far i get the same performence with two cards as i get with one card... :(

do i need a special driver? or does it have to be an ati chipset board? :confused:

http://img19.exs.cx/img19/5427/11ws.th.jpg (http://img19.exs.cx/my.php?loc=img19&image=11ws.jpg)

http://img19.exs.cx/img19/1182/28xj.th.jpg (http://img19.exs.cx/my.php?loc=img19&image=28xj.jpg)

http://img19.exs.cx/img19/807/35bx.th.jpg (http://img19.exs.cx/my.php?loc=img19&image=35bx.jpg)

http://img19.exs.cx/img19/8425/46fq.th.jpg (http://img19.exs.cx/my.php?loc=img19&image=46fq.jpg)

http://img19.exs.cx/img19/1337/58fp1.th.jpg (http://img19.exs.cx/my.php?loc=img19&image=58fp1.jpg)

http://img19.exs.cx/img19/8564/60hg.th.jpg (http://img19.exs.cx/my.php?loc=img19&image=60hg.jpg)

[saaya]

Saaya, did I ever mention how good of a mod u were :slobber: :slobber: omg :eek: :eek:

How long have u been secretly saving up for those ;)

Saaya: "oh no I only have an AXP" rig
XS: he he he he
Saaya: BAM ! TWO X850 XTs !!!!!
XS: *runs like a dog with tail between legs*

Nicely done saaya, I think i have an idea as to what there for :eek: Hope to see up there in the ranks when MVP comes out :D

EDIT !!!: THOSE ARE PEs !!!!

Perkam

Nice :toast:

Are these yours saaya?

havent been saving at all, one is for my buddy SAE and the other one is for me :D but only to do a review, after that imma sell it again :(

and yes, they are both x850xt platinum edition cards :D

[Ajin]

no sli for ati cards

[saaya]

no sli for ati cards

huh? whadda ya mean? :confused:

[metro.cl]

no sli for ati cards


Thats what i know

are you trying a new mobo with ati chipset??? is it arm ready???

if not you might need to sell one of those beautifull x850xt pe or wait until the bullhead is out

[d3birth]

I believe its only nVidia cards, 3dfx appearantly had something like SLI.

http://www.nzone.com/object/nzone_sli_faq.html

[saaya]

no, im using an dfi nf4 sli board... no sli for me? :(
are you guys sure? maybe its the yellow slot close to the power slot? or the white slot on top of the card? :confused:

[Ashe]

I never heard of any kind of SLi for ATI cards, sorry m8 :(

[Viss]

and were all stupid lol :) Or you gonna show us something weve been waiting for long for the first time or youll end up with 1 nice card in yr rig, whats it gonna be ? No way on earth you didnt know already only nvidia has SLI so far.

[saaya]

I never heard of any kind of SLi for ATI cards, sorry m8 :(

never heard of it? damn :(
how the heck did i think i could run two ati cards in sli then? im a stoopid :(

[FoxaS]

SLI for ATI will be AMR,

but there isn't any board that can make use of it... :rolleyes:

[saaya]

SLI for ATI will be AMR,

but there isn't any board that can make use of it... :rolleyes:

really? what board do i need then? any idear when it will come out?

[kryptobs2000]

no sli for ati cards, you need a mobo with a chipset that supports amr (ati xpress 200?)

[saaya]

hmmm so if i go get an ati xpress200 i will be able to run sli with my cards?

[metro.cl]

really? what board do i need then? any idear when it will come out?


ati bullhead mobo, release time in june maybe earlier but i dont think so.

also you might wanna sell the 2 x850xr pe and go for just a r520 fudo, it will perform like 2 x850xt pe.

also will be released on june

good luck mate, and arm only will need a driver not a bridge

[metro.cl]

hmmm so if i go get an ati xpress200 i will be able to run sli with my cards?


im not sure but ask grayskull he is fro ati. he might hel you with a beta board that would be sweet :)

[Agr3sive]

As far as I know there is no SLI for ATI at the moment, but there is sposed to be boards coming out in the near future (2months I think) which support AMR, which is ATI'S version of SLI.
You will not need a special bridge like the Nvidia counterpart as they will connect directly through the Pci-e bus. Thats gonna be a killa rig mate. :)

[saaya]

then what do i do with my two x850xtpe's now? :/

[Rippthrough]

Your kidding right?

[Holst]

LOL

Saaya makes himself look silly.

SLI is only for nvidia...

But you will still get great scores with just one X850.

[Noldor]

Enough with the games now sayaa, show us what you've been secretly hiding! :stick:

[saaya]

LOL

Saaya makes himself look silly.

SLI is only for nvidia...

But you will still get great scores with just one X850.

ahhh cmon holst, why did you have to tell them already? it was soo much fun :D :lol:

[Grov]

Oh dear, lol. :p:

[Noldor]

Spammer :D

[saaya]

Enough with the games now sayaa, show us what you've been secretly hiding! :stick:

unfortunately thats all ive got dude :D
no supa secret nda stuff for saaya :( :lol:

[Dynasty]

Sell and wait for r520 at end of May ;)

[Ferry82]

no, im using an dfi nf4 sli board... no sli for me? :(
are you guys sure? maybe its the yellow slot close to the power slot? or the white slot on top of the card? :confused:

Saaya you are not stuppid what going on??????? :stick:

[d3birth]

unfortunately thats all ive got dude :D
no supa secret nda stuff for saaya :( :lol:

How do you expect us to believe that?! :D

[scrible88]

lol, wow. Saaya, this is unbelievable.

That's why I don't believe it! At least right now. You must be fooling with us and have a new ATI board or something hiding up your sleeve!

[oublie]

Sayaa either this is a big fat hairy windup you playing on us or we should all go around to your house an poke you with sticks and laugh at you until you cry :stick:
Now come on spill the beans are you seriously trying to tell us the you thought you could run sli using the ati cards without some custom drivers & hardware or are you going to show us what your new toy really is?

... sharpens a big stick and puts some cat poo on the end of it in anticipation..... :)

[jkabaseball]

man, i was already for some massive scores with 2 x850's. Fugger said he has played with stuff that would knock our heads off. I say he has a board.

[perkam]

That's dissappointing saaya. You should have known sli is for nvidia only. I mean, common sense wise, why in the hell would an nvidia chipset invite ATi cards to work in SLI !!! dont you think that if that was possible ppl would have tried already ??? I mean ppl here have setups exceeding 5-6k and even then there isnt a single score at the ORB that shows two ATi cards working together lol.

If this mistake had been made dec04 when sli was still new, I would have understood but right now, its just plain silly :wave:

EDIT: Also if MVP (ATI's version of SLi (its not AMR anymore !!!)) was out, OPP/Macci would have results first.

At any rate, you can try the two cards and maybe find an sli mod that works with ATI and shut us all up...maybe....

Perkam

[saaya]

Saaya makes himself look silly. :D

nope, if i had a supa dupa secret board or anything id either have posted about it already or i wouldnt be allowed to talk about it hehe

and again, no, i dont have a supa dupa ultra top secret board or anything, just those two x850xtpe cards, one for SAE and one for me :)

[Magnj]

I don't think your telling the truth... Can't wait to see your results :-P

[S&M]

ROFLMAO !

I was reading this thread and was thinking - wtf is going on ? kept reading another post by saaya and again - WTF ?! another reply of him - and again - what the heck is going on ?

BUT ! when i hear him saying - i've got no other suprises ....
LMAO ! c'mon m8 ... what's that ? asus mobo ?

PPL ! don't forget where saaya is located ... and more important ...
where was CEBIT ....

[saaya]

ROFLMAO !

I was reading this thread and was thinking - wtf is going on ? kept reading another post by saaya and again - WTF ?! another reply of him - and again - what the heck is going on ?

BUT ! when i hear him saying - i've got no other suprises ....
LMAO ! c'mon m8 ... what's that ? asus mobo ?

PPL ! don't forget where saaya is located ... and more important ...
where was CEBIT ....

lol, dont wait for anything because theres nothing more to come, lol :D
i was just playing around and having a little fun :P

[C_X]

i have now a second sli bridge maybey you can push it very hard on the cards ;)

:D

[metro.cl]

man, i was already for some massive scores with 2 x850's. Fugger said he has played with stuff that would knock our heads off. I say he has a board.


i thin he is playing with a r520 or dual r520, thats gonna be awsome like 100k on 3dmark01 with one and 150k with 2 boards, the problem will be the cpu, amd should release an fx 70 for the cards maybe :)

[saaya]

i have now a second sli bridge maybey you can push it very hard on the cards ;)

:D

yhpm! :)





j/k :p: :D

[Ashe]

PPL ! don't forget where saaya is located ... and more important ...
where was CEBIT ....
And don't forget the rumors, saying a lot of stuff was stolen @CEBIT :stick:

Saaya, spucks aus :D

[saaya]

And don't forget the rumors, saying a lot of stuff was stolen @CEBIT :stick:

Saaya, spucks aus :D

those are NOT rumors! lol
they stole quite a lot of stuff, even 2 mxm laptops right from the nvidia booth LOL :lol:

and again, theres nothing else comming, i was just making fun :P :D

[Rippthrough]

Try it and suprise us :D

[XiN]

OMG, LOL

Hey saaya, you should contact ATI and show them this thread: I'm sure - after long laughs in ATI headquarters - they sent you a free Xpress 200 motherboard :D :D :D

[saaya]

OMG, LOL

Hey saaya, you should contact ATI and show them this thread: I'm sure - after long laughs in ATI headquarters - they sent you a free Xpress 200 motherboard :D :D :D

i think some of the ati people i met would indeed find this thread quite amusing :D
i doubt they would send me something though... but yeah, maybe its worth a good laugh for them :)

[cadaveca]

SLi is really all about the driver. Both cards get the duplicate info for frame buffers via the swapped SLI "bridge"(the one on the board, not the cards), and the primary GPU regulates which card draws which frame, according to the driver(hence the need for the card-bridge).

It's really alot more technical than that, but unless ATI is going to infringe on Nvidia's patent for the GPU bridge PCB, I doubt you'll see ATI cards on nVidia SLi. It's nVidia's way of trying to reclaim the lost marketshare from ATI, IMHO, and part of the reason that ATI cards will work without a bridge pcb.

I don't think the radeon Xpress 200 supports AMR, because of the wiring needed on the mobo pcb itself, but I am guessing either the xpress 250 or 300 will support AMR.(uless the 200 can "switch" lanes assignments and do the duplication itself...maybe smartgart has something to do with it?) :D
I also think that this is the major reason for Sapphire making the Grouper mobo's, and the pcb's for the x800 series of cards.

Saaya...look at the cards..see there a white 4-pin connector next to the pci-express power plug? (red circles)the one that looks like a fan plug, but isn't? Wonder what you do with that? :stick:
:eek: ;)
:toast:

[Grov]

Seeing as you have a spare one, wanna give it to me? :p:

[Bennah]

saaya ask FUGGER nicely to lend his bullhead board off him :D

Nice cards btw, you and SAE are going to enjoy them :toast:

[WiCKeD]

You can borrow my :am: sign for now, saaya. ;)

It's all good, I'm sure alot of people make the mistake, because it's not something either company widely advertises. (It's just immensely funnier when someone spends $900 on it) SLi board, two slots... makes sense you should be able to use any two cards, right? Except nVidia isn't going to just give away that catch-up-for-free-using-SLi "card" they bought with 3dfx. ;)

SLi is really all about the driver. Both cards get the duplicate info for frame buffers via the swapped SLI "bridge"(the one on the board, not the cards), and the primary GPU regulates which card draws which frame, according to the driver(hence the need for the card-bridge).

It's really alot more technical than that, but unless ATI is going to infringe on Nvidia's patent for the GPU bridge PCB, I doubt you'll see ATI cards on nVidia SLi. It's nVidia's way of trying to reclaim the lost marketshare from ATI, IMHO, and part of the reason that ATI cards will work without a bridge pcb.Basically. nVidia got SLi with 3dfx and it should have died with them. Alienware had a prototype going that used an add-on card to handle the balancing act of any two cards, but it died when they started using nVidia's SLi. nVidia uses the bridge, like 3dfx did, while ATi is forced to create their own design. It's supposed to purely use the mobo to split the load and I don't think anyone knows whether it will be cross-compatible. Probably not, but you never know. Current RX200 boards don't support it, though.

[cadaveca]

lol Wicked you edited your post right when i clicked quote. LoL you looked at my pic didn't you. LoL. It is NOT the MPC connector. Or are you gonna try to tell me that there are 2 on the pci-e aticards(pls note that only some of these cards have this connector..as in really...like getting a x800pro that will mod...luck of the draw getting the connector :D)

[WiCKeD]

lol Wicked you edited your post right when i clicked quote. LoL you looked at my pic didn't you. LoL.Yup. :D It wasn't coming up before, so I only guessed. ;) I always edit 1000 times anyway.

[cadaveca]

i still have my voodoo/ati card-combo...and have been following this SLi crap for a while. I am not inpressed with how nVidia is positioned on this.

[sobol]

Being serious now. I dunno if i remmember good. But doesent DFI NF4 SLI-D have semi SLI support. And tell me now , does it also have special bridge pin connector for 2 cards ?

[cadaveca]

LoL. 2x on second card, or is it 4x?

[enzoR]

did u buy those on the last day at CeBit?

[SLaY3r07]

2 R520's in AMR will be teh own4g3.

[bachus_anonym]

Saaya, you got me BIG TIME!!!
I really scratched my head thinking this just can't be true - Saaya trying to run 2 x X850XTPE in SLI ???????

:ROTF: :lol: :ROTF: :lol: :ROTF:

[SLaY3r07]

So, are you gonna sell an X850 XT PE?

[matt9669]

Saaya isn't stupid guys, he was just playing a joke on you all! :D :toast:

:YIPPIE:

As for AMR, it certainly isn't as simple as plugging two cards in, the main missing component is the driver.

[Bennah]

Bullhead = noes dual gfx. :(

http://images.anandtech.com/reviews/chipsets/ati/rs480/refboard_sm.jpg

:(...

could always ask him to borrow it anyways :p:

[naelq]

how about VIA's new chipset? :D :D

[isp]

This is a very confusing thread. However I hope you both enjoy the cards, Sayaa and SAE. :up:

[Nanobot]

Well I had a good laugh today,thanks Saaya :lol:

naelq: Via's new chips,the PT894 Pro, PT880 Pro(Intel ) and K8T890 (AMD 64) have DualGFX Express.
DualGFX simply enables two PCI-E video cards setups for multi monitor use the cards are unfortunately not bridged.

[dippyskoodlez]

does the "DXG" work?

[MaxxxRacer]

lol saaya.. you played this joke on me on aim. he was telling me how he was gonna get dual x850's and run them in sli... i kept telling him it wanst possible for like 10 minutes and then he busted up laughing as was like i know iw as just f'ing with u.. but i never thought he would go out and buy them... it begs the question why??

oh and according to reports (not sure how accurate they are) amr will be bridged through the CHIPSET. so that means no little bridge card like nvidia uses.

[cadaveca]

it begs the question why?? one for hime, one for SAE. maybe? LoL.


oh and according to reports (not sure how accurate they are) amr will be bridged through the CHIPSET. so that means no little bridge card like nvidia uses.

the bridge card is needed in order for one of the gpu's to take command, basically, and for shared vertex info between frames. The chipset in nVidia's SLi does nothing but duplicate the vertex info and shader info sent to the frame buffer of each card. ATI, on the otherhand, have had the same sort of functionality built into all of thier cpu's, since the r300, and was even shown in the rage fury MAXX.

The problem that is holding ATI back right now is patents..they have to approach the technology in a completely different way, as nVidia has a ton of patents on this technology, and even has fees for systems builders who use the SLi logos. The best i can see ATI doing is using hypermemory, so that each card can pull data from the same framebuffer, which would completely eliminate the need to cross-GPU communication. The driver would mearly need to verify that each frame was rendered before assigning the next frame to the second GPU (made really easy since they would be sharing the framebuffer. what holds the nVidia solution back from having full performance benfits is that hte cards cannot share the framebuffer, so have no idea what the other cards is really doing. the driver takes control of this.)

Of course, the hypermemory solution is not really feasible when the CPU/s chipsets cannot address large amts of memory that would be needed for such a system to top the charts, as this would take away from system memory, however this may change with the impending ram price fallout, and the introduction of ATI's chipset. The brink of multi-processing is getting nearer...once it topples over, we will see AMR.

[wfarid]

hmm did any of you guys see the asus amr board @ cebit... i think it was using the xpress 400 chipset or something along those lines... So from that im assuming that the AMR solution will be the 400 series chipset (some1 correct me if i'm wrong)... And sayaa you dont know how excited you gove me when i saw the title, i almost spit out my dr. peeper all over my moniter...

[drcrawn]

What's really funny here, is the Saaya vs. Sayaa spelling battle that's going on...lmao

[perkam]

wfarid, I have yet to see anyone talking about the radeon xpress 400 chipset, but you might have the right intentions there. the Intel ATI chipset is called the RD400 afaik where as the AMD one with MVP Enabled (STOP USING the word AMR !!!) is the RX482 chipset, which isnt out yet.

However at Cebit I read somewhere that they versions of the board working fine so they're probably less than 6 months from release. Expect it to be released the same time around the R520 will be released. :eek:

Perkam

[HiJon89]

I'm glad you were joking Saaya, when I first saw this thread I wasn't sure if my XP-120 was in good hands :p:

[R.Rabbit]

ya i saw the xpress400 board too(i doubt its an actual engineering sample, just a show board to get us excited)i also read somewhere that ati was gunna use a flexible circute(spell?) to bridge the two cards
hahahaha! i was freaking out when i read this, i thought Saaya must've had an xpress400 board! then he went on about SLI and such... then i started to question wheather sli can run on ati cards!!! good stuff! :banana: :banana: :banana:

[VSpecII]

Saw the first post and my jaw hit the floor - I thought it was going to be a first ever AMR run. Then I saw the motherboard. I wont deny it, I laughed.

[trakslacker]

Saaya's just trying to raise his post count. ;)

[IvanAndreevich]

Pretty funny. I believed he was sincere for like 2 minutes :banana4:

[perkam]

ya i saw the xpress400 board too(i doubt its an actual engineering sample, just a show board to get us excited)i also read somewhere that ati was gunna use a flexible circute(spell?) to bridge the two cards
hahahaha! i was freaking out when i read this, i thought Saaya must've had an xpress400 board! then he went on about SLI and such... then i started to question wheather sli can run on ati cards!!! good stuff! :banana: :banana: :banana:

There is no such thing as the Radeon Xpress 400 :rolleyes: The name "radeon xpress 200" will remain the same for both Intel and AMD chipsets, just that their codes will be different.

Like the first 939 pci-e cards were the RX480 chipset and the RS480 (Onboard Vid) Chipset. The next chipsets with MVP (two gpus) will be called the RX482/RS482.

As for the intel ones, heck here's a photo and a link to a more indepth analysis of how MVP (NOT AMR !!!) will work:

http://img.hexus.net/v2/internationalevents/cebit_hannover_2005/store/mvp_board_big.jpg


MVP supports split frame rendering using supertiling, where the screen is split up into tiled areas with each tile processed on a GPU, using any GPUs that support supertiling. That's anything from R300 up, but it's likely to be limited to R4xx GPUs. You can use X700 and X800, X800 XL and X850 XT PE, or any other mix that you can think of. There's the potential to increase anti-aliasing IQ using supertiling (multipassing the tiles through a GPU) and MVP.

Perkam

[metro.cl]

@ perkam, amr is so :banana: :banana: :banana:

edited thanx mate

[SLaY3r07]

@ perkam, arm is so :banana: :banana: :banana:

You mean "AMR" ;)

[ocmyface]

nono... he means arm. he has VERY nice arms :D

[IvanAndreevich]

Does anybody have a good link to well-informed AMR speculation? I was talking to a programmer friend of mine and we sort of couldn't figure out how it could be implemented with only PCI-E bandwidth available..

[aoc007]

havent been saving at all, one is for my buddy SAE and the other one is for me :D but only to do a review, after that imma sell it again :(

and yes, they are both x850xt platinum edition cards :D

How much are you planning to sell it for? Just wondering.... :)

[ocmyface]

ONLY pci-x bandwidth?

[IvanAndreevich]

PCI-X bandwidth is :banana::banana::banana::banana: since it's not PCI-E. But I meant PCI-E and it's not nearly enough. I suppose there needs to be bandwidth that's comparable to the memory bandwidth of one of the cards, unless there is some sort of clever compression.

There is no bridge possible on ATI cards simply because there is nowhere to plug it in.

[perkam]

Ivan perhaps these can help:

http://www.elitebastards.com/page.php?pageid=9524&head=1&comments=1
http://www.hexus.net/content/reviews/review.php?dXJsX3Jldmlld19JRD0xMDQx
http://www.techpowerup.com/reviews/Cebit2005/Day4/5

They are all on ATI's Cebit showing of MVP/AMR (i give up :rolleyes: ) so it should be helpful. :)

Perkam

[R.Rabbit]

haha, does it really matter? they'll change the name again next week anyhow just to get it in the news, cus they've got nothing else to put in the news(about mvp/amr/ati's sli/bullhead/xpress 400/RD-400/R2D2)

[mcnbns]

I don't even know why I thought Saaya was serious for 30 seconds. :wierd:

With regard to AMR/MVP/POS/BBQ/SLI/LOL/BYOB and other acronyms, I'll go with the one that gives me more 3DMarks.

[Cpt Twitchy]

I saw the two cards and was like boy would it be nice to have one of those. Then I saw his next post and was like why would he put them both on 1 mobo. I had no clue what he was trying to do and never thought SlI/AMR/etc. Then I was just ammused at all the people he fooled. Or am I the one he fooled :confused: :banana4:

[IvanAndreevich]

Thanks for the links. Unfortunately, they don't help me imagine the viability of "AMR" purely through PCI-E interface. If they have some compression tech, it should work on ALL SLI-capable boards (nVidia/VIA/Intel..)

[perkam]

If they have some compression tech, it should work on ALL SLI-capable boards (nVidia/VIA/Intel..)

That means we can only know once they come out so that we can compare the two...however when they come out we wont need to know...catch 22 :D

Perkam

[saaya]

im glad you all had some fun reading this thread :D


Can you say 250% of the peromance we can get today? ;)

400% ;)

perkam, at cebit all i heard was RD480...

[wfarid]

haha, does it really matter? they'll change the name again next week anyhow just to get it in the news, cus they've got nothing else to put in the news(about mvp/amr/ati's sli/bullhead/xpress 400/RD-400/R2D2)


you def have a point dude... But honestly (its been said many a times but i just cant help saying it again) i think ati's policy is pretty cheap although its smart... Wait for the competition to come out with something really kick ass, find out way to do it better and then release that product, leaving the competition saying, "What the f :banana: :banana: :banana:, i came up with that first!!!"


at first i didnt think the r520 was going to be anything anywhere near the rumored specs posted on anandtech, i thought it was just going to be near the improvement gained from the 9800 series to the x800 series... But if you look at the improvement for 9800 series to the x800 series, the rumored specs are somewhat right.

[cadaveca]

Does anybody have a good link to well-informed AMR speculation? I was talking to a programmer friend of mine and we sort of couldn't figure out how it could be implemented with only PCI-E bandwidth available..


it's not the bandwidth that might limit multi-processing over the bus, but the number of lanes assigned to a device, and the number of lanes the chipset supports. This is why the "PRO" versions of SLi use 2 chipsets...the bandwidth suddenly has become available.

This functionality is what makes PCI-E so great, and why it is now a standard. Once Hypertransport goes that route(i hear rumours of intel going that route now, too), you can truly have a customizable pc...multiple processors and chipsets, displays and gpu's, soundchannels and threads, OS'es and users...multi is a good word in IT halls right now. You want graphics and physics..buy the specific processors, slap them in a slot, and away you go. You need more cpu's and harddrives...buy the cards, slap 'em in a slot. Everything is about 3-4 years from being "slapped in a slot".

When that happens, then i can truly get excited, until then, wow..nice..but not that great. :stick:

[IvanAndreevich]

>>it's not the bandwidth that might limit multi-processing over the bus, but the number of lanes assigned to a device

More lanes = more bandwidth, so this is sort of a self-contradicting statement.

>>the bandwidth suddenly has become available
4GB is nothing for the task at hand if the images being passed are uncompressed.

[wfarid]

speaking of crazy tech, if you've been keeping up with some internet 2 rumors and what not, in the future our internet connection might be more important to getting more power in our computers than you think (all because of grid computing) so you could literally buy more processing power from your isp... wierd isn't it...

[cadaveca]

More lanes = more bandwidth, so this is sort of a self-contradicting statement.

how does that contradict? You say there is not enough bandwidth in PCI-E..bandwidth is dependant on the number of lanes only, and because PCI-E offers customizable lanes; if you need more, it can be assigned. :stick:


4GB is nothing for the task at hand if the images being passed are uncompressed.


i said it was possible...but not something you can go buy today. LoL. just because there are pictures of stuff you cannot buy yet, do not think that tech has not already surpassed what you are seeing.

:toast:

[saaya]

please keep it civil and friendly guys :D :toast:

[cadaveca]

LoL. not much of a debater? I'm sure it will always be civil. At least on my end.

[perkam]

perkam, at cebit all i heard was RD480...

Then that is certainly news to me. However, thats prolly for the 939 chipsets I believe ? and the ones withOUT MVP/AMR/WTH (I created my own Acronym :D Whatever The Hell) cos I really do doubt they're using the xx480 code for intel chipsets.

Perkam

[IvanAndreevich]

It's contradicting because adding lanes = adding bandwidth. You say lanes are important and bandwidth is not.

it's not the bandwidth that might limit multi-processing over the bus, but the number of lanes


if you need more, it can be assigned.
No you can't if the videocards don't support it. Somehow you have to find a solution which would mean CURRENT cards can support it, which are limited to 16x PCI-E. Besides, what's required is maybe 64xPCI-E bandwidth. Impossible to implement and if you imagine how large that would be physically because of the topology of the bus, you'll KNOW that this is obviously not the way to go.


i said it was possible...but not something you can go buy today
Nah, that's a stupid solution that will never be implemented.

In other words, if you don't know how AMR will work, don't imagine that you do ;)

[cadaveca]

um, two lanes are enough for today's cards where does the inportance of 16 play?


Seriously tho, i get what you are saying, but it's only the professional market, IE the guy's buying opterons, and 4mb cache Xeons, that need the bandwidth, not benchers of today or next year. Within the next 18 months, adding a second chipset is more than enough for the markets demands. just like having a dualcpu board...not very complicated or space taking.

Now, if we get a card that needs 64 lanes...well, it could happen. remember than technology is shrinking at the same time is is becoming more powerful, so to say that we will not be able to do it is not practical.

Now, you show me a vidcard on the market that can actually process 4gb of data in a clock cycle, and i'm there with you bud, but it is not a reality right now. PCI-e has it covered.

It's just like hypertransport...currently we are at what, 8gb of bandwidth? The spec allows for 22.8gb. Now here's where pci-e comes to play...the two techs are very similar, although entirely different. but to have a spec that can support such mass amounts of data makes sense only if there are other components that can use it now, or will in the next 2-3 years. we might see this multicore cpu's, but we are talking about 22.8gb per lane...that's 3x for the new opterons.

100mb/second/per pin(there are 40) is far greater than what is used today. it's just a matter of it being needed and implemented. So there is your 4gb/second. the only thing that may actually pass that sort of data would be a raid controller, which, unfortunately, would be limited more than i myself like, but is way faster than the 300mb/s of SATA...which bring me to my close...if a harddrive can only send 300mb/sec, and pci-e can handle 4 gb's, which is also the most any chipset can handle right now on the desktop market in system ram, why are you concerned over this? what system is sending 4gb/sec over PCI-E? Today?

4gb/sec is more than you can even source in shader information right now! so, how would AMR suffer?

http://www.pcisig.com/specifications/pciexpress/technical_library/pciexpress_whitepaper.pdf

[perkam]

Hmmm...would you guys know how the no. of pipelines makes a bigger difference when you use two vid cards instead of one.

As in, the ol X800 pro was almost at pace with the 6800GT with only 12 pipes, but my question is that would the 4 pipe diff between the pro and the gt be bigger when benching two pros vs two gts ??

Perkam

[IvanAndreevich]

perkam
Should be relatively the same IMO.

cadaveca
Monitor connects to only one card, right? If one card RENDERS the images on one end and then just passes those huge uncompressed 32-bit bitmaps in rapid succession -> calculate what bandwidth is required for 1600x1200x32bpp@150FPS, for example.

I am not really thinking on what could be done and what couldn't be, I am just trying to figure out HOW the AMR technology WORKS today [behind closed doors] with existing hardware

[cadaveca]

cadaveca
Monitor connects to only one card, right? If one card RENDERS the images on one end and then just passes those huge uncompressed 32-bit bitmaps in rapid succession -> calculate what bandwidth is required for 1600x1200x32bpp@150FPS, for example.

I am not really thinking on what could be done and what couldn't be, I am just trying to figure out HOW the AMR technology WORKS today [behind closed doors] with existing hardware


LoL. Oh. Well, Smartgart. You don't need to send uncompressed maps when you can easily compress them and send the already rendered frame, which is what happens. The info sent to the monitor gets translated at the ramdac, and is nothing more than really, a 2-d image. still just a flat monitor. The rendered frame merely needs to be the data for each individual pixel on the screen, and nothing more. the monitor does not care if the image is 3d or not, only our eyes do.

SO, you don't see big uncompressed textures, even with SLi. The cards only communicate vertex information shared between rendered frames(think wireframes), and the pre-rendered frame from the second card. This is what makes for the only slight 10% overhead in SLi, although this number has proved theoretical, as no real game out there has been optimized for SLi, as it does require some code in the game itself, and not just a driver. In order for there to be large texture swaps(which does not happen in SLi), the frame buffer for each card would have to be shared, or the memory on the card would have to be shared, and we all know that this is not the case; two cards with 256mb of frame buffer only ends up as one card with 256mb of framebuffer, and 2 gpu's. all info on the frame buffer, minus the 8-12 mb of data for the 3 pre-rendered frames, is duplicated, as each card need the same shader and texture data to draw from.

So, with only 8-12 mb need for the pre-rendered frames, and knowing that it's only this plus a small amt of vertex data that passes between the cards, realizing the the 100mb/sec of one PCI-E lanes should almost be sufficient, and that you would ideally want 1 for each way of communication. We already know that by 2x or 4x pci-e width is enough for two cards to render(kt890pro used to do it, but cannot becasue of patents of ATI's, IMHO, but i have yet to confirm this), and that the primary gets 8, that leaves 4-6 lanes left open for unidirectional communication between the cards, and how ATI will be implementing "AMR".

Class dismissed.. :banana4:

lol

[GregP24]

never heard of it? damn :(
how the heck did i think i could run two ati cards in sli then? im a stoopid :(

LOL. That's just funny. :p:

[IvanAndreevich]

So you agree with me then basically the compression is required ;) Hence, with proper driver support, this AMR or whatever you want to call it, should work on any current SLI-capable chipset.

[saaya]

It's contradicting because adding lanes = adding bandwidth. You say lanes are important and bandwidth is not.



No you can't if the videocards don't support it. Somehow you have to find a solution which would mean CURRENT cards can support it, which are limited to 16x PCI-E. Besides, what's required is maybe 64xPCI-E bandwidth. Impossible to implement and if you imagine how large that would be physically because of the topology of the bus, you'll KNOW that this is obviously not the way to go.


Nah, that's a stupid solution that will never be implemented.

In other words, if you don't know how AMR will work, don't imagine that you do ;)


32+ express lanes is possible however :D
uli had a southbridge at cebit that has 16 or 20 pciE lanes!
in theory you can use the southbridge on any chipset aht uses the hyper transport protocoll to connect aouth and northbridge and get 20 lanes from the north and 20 lanes from the southbridge :)


um, two lanes are enough for today's cards where does the inportance of 16 play?


Seriously tho, i get what you are saying, but it's only the professional market, IE the guy's buying opterons, and 4mb cache Xeons, that need the bandwidth, not benchers of today or next year. Within the next 18 months, adding a second chipset is more than enough for the markets demands. just like having a dualcpu board...not very complicated or space taking.

Now, if we get a card that needs 64 lanes...well, it could happen. remember than technology is shrinking at the same time is is becoming more powerful, so to say that we will not be able to do it is not practical.

Now, you show me a vidcard on the market that can actually process 4gb of data in a clock cycle, and i'm there with you bud, but it is not a reality right now. PCI-e has it covered.

It's just like hypertransport...currently we are at what, 8gb of bandwidth? The spec allows for 22.8gb. Now here's where pci-e comes to play...the two techs are very similar, although entirely different. but to have a spec that can support such mass amounts of data makes sense only if there are other components that can use it now, or will in the next 2-3 years. we might see this multicore cpu's, but we are talking about 22.8gb per lane...that's 3x for the new opterons.

100mb/second/per pin(there are 40) is far greater than what is used today. it's just a matter of it being needed and implemented. So there is your 4gb/second. the only thing that may actually pass that sort of data would be a raid controller, which, unfortunately, would be limited more than i myself like, but is way faster than the 300mb/s of SATA...which bring me to my close...if a harddrive can only send 300mb/sec, and pci-e can handle 4 gb's, which is also the most any chipset can handle right now on the desktop market in system ram, why are you concerned over this? what system is sending 4gb/sec over PCI-E? Today?

4gb/sec is more than you can even source in shader information right now! so, how would AMR suffer?

http://www.pcisig.com/specifications/pciexpress/technical_library/pciexpress_whitepaper.pdf
2 lanes are enough? define enough then... between 4 and 8 lanes there is still a small but notable difference and less than 4 lanes really has an impact on performence!
i dont know what kinda maths you are doing but tests have shown that less than 4 lanes nmeans quite a performence hit of 20% iirc.

ok, if this is ture then what i heard doesnt make any sence...

an ati guy told me in an interview at cebit to expect 32 pciE lanes really soon
and when i asked an nvidia guy in an interview whether we will see 32 pciE lanes for their boards/cards anytime soon the guy put a huge grin on his face and said "maybe" and said i can quote him, so here i go :D

if the pciE lanes are not needed and 4 lanes are really enough as your saying, then how come both nvidia and ati are going to double the lanes? for fun? for marketing? if theres absolutely no difference between 16 and 32 lanes then this doesnt make any sence...

maybe its faster to send small chunks on several lanes than sending big chunks at a few lanes? im sure you CAN send the data on just a few lanes, but splitting it up into small chunks probably gives a boost, probably lower latency...

[cadaveca]

yes smaller chunks are faster, and are easier set into memory...try setting your agp aperature to 64mb, you'll get a small boost. Smaller chunks of data, quite obviously, will get processed faster, and fill the bus less, so other things can happen.

But you have to remember..the cards need only send about 150-200mb/sec to each other...that is convered by 2 lanes. one card using 2 lanes by itself is slower, but only becasue of the data being sent along the bus to the one card from the cpu is limited. BUt if 2 are adequate for intercard communication, than the remaining 14 can be split between the 2 cards.

An x700 or x600 or x300 would not be hurt in performance by going to 4-6 lanes. remember that you can assign 6 per card if you want, and have 4 for inter-card connections. there are alot of ways to go about it, depending on whether the cards are going to share framebuffers or not.

But here's what is important..pci-e allows 40 pins(16x), each that can allow for 100mb/s. that 4gb/second, slplit 3 ways, if you want, is still more than can be source from a drive...which is only 300mb/sec.(sataII only, mind you, and who has a sataII drive)

BUt, once we get into the future generations, yes, 16 lanes may not be enough. in order to have a shared framebuffer, 6 lanes is not adequate. But for today's uses they are.

Ivan is looking too far ahead into the future...as are you saaya. It will be a long trime before they can make use of 32 lanes of pci-e, if they are not sharing the frame buffer. are you telling me that right now 16 lanes is not enough...no. But in the future, yes.

you are also thinking about multirendering..2 lanes DOES hamper performance a bit in multi-rendering. but going to 2, like you said, is only about a 20% hit, instead of the 10% of SLi. that's the important part. so, 80% faster than one card...and only 10% slower than nVidia's 8 by 8 lanes solution. then the jump to 6x6x4 makes more sense, does it not?...double the 2 lanes to four, and you should be right back at SLi performance, nevermind 6, which could theoretically surpass nVidia's SLi.

Maybe you are right saaya, and 16 lanes are not enough. Nvidia's solution in the professional market makes use of 2 chipsets, with one card on each chipset. But then i think about SM3.0 and 4.0...both are compression techniques that will allow for less data being sent over the bus. you play a game, the vidcard's ram get the shader and texture data, and once your playing, it's only small amounts of shader data that should be sourced from the harddrive, and then only vertex information that the cpu sends to the cards. SM3.0 and 4.0 both make the shader data smaller...so faster on the bus. they might not need go to 32 lanes if the compression works well. Of course, eventually they will need 32 lanes, but not too soon, i think.


needing only a driver to get 2 ati cards working...might be possible. in fact, becasue of smartgart, i know it is. but it being as efficient as SLi is, might not be possible, due to how the nforce4 is set-up. but we all know that they already have AMR up and running...whether it be only on thier chipsets(most efficient, imho), or on the competitions, still requires a driver.

[saaya]

hmmmm shared framebuffer sounds nice... and makes sence... for ati! but nvidia?
why would nvidia go for 32 lanes? shared framebuffer? it would mean they have to overhault their sli concept... hmmmmm

maybe nv48 is actually a dual core vpu and having two cards like that in sli would then def require the extra bandwidth they are adding. well not require it, but it would give them a boost, but nowadays its about those few percent of extra performence so it would make sence...

[cadaveca]

it would not require them to change much, imho. the chipset is what is in chagre of what happens with the lanes, along with the chipset driver. possibly a driver revicion may do it, but it must take some wiring as well.

wait until the second revison of SLi. it will happen, maybe, but maybe not, now that duality mania has started.

[SAE]

saaya, that's just too amusing... you should have asked ATI about it at Cebit :lol: You made me laugh reading through your first posts... :D


nono... he means arm. he has VERY nice arms :D

And you made me laugh hard second time... :up: :rotf:

[madman2233]

this is a great thread.

what i dont get is the part about AMR being avaliable on nvidia chipsets. it doesnt make sence, why would nvidia promote ati's technology?

So does this mean that if you want a nvidia motherboard you wont be able to have AMR? and the opposite of course.

thats what i think.

[VSpecII]

this is a great thread.

what i dont get is the part about AMR being avaliable on nvidia chipsets. it doesnt make sence, why would nvidia promote ati's technology?

So does this mean that if you want a nvidia motherboard you wont be able to have AMR? and the opposite of course.

thats what i think.
nVIDIA already made nForce4. It's not like they can change much about it. If ATi wants to re-write their own drivers for it, for AMR support, all the better to them. They saves themselfs millions in R&D, plus the market of both nVIDIA and ATi motherboards. Same on the flipside.

[skate2snow]

AFAIK, MVP is a tech to ASUS and not ATI....

[cadaveca]

AFAIK, MVP is a tech to ASUS and not ATI....

You'd be correct.
And for your prize, you get a big...... :cord: lol

[cadaveca]

So you agree with me then basically the compression is required ;) Hence, with proper driver support, this AMR or whatever you want to call it, should work on any current SLI-capable chipset.
if you call z-culling compression then sure.

You realize that the card only has to send the image that we see, and not the entire 3d-scene, right? stuff out of POV gets culled. that's alot of info dropped. It's not as full rendered scene. THis is what happens in the z-buffer.

[saaya]

it would not require them to change much, imho. the chipset is what is in chagre of what happens with the lanes, along with the chipset driver. possibly a driver revicion may do it, but it must take some wiring as well.

wait until the second revison of SLi. it will happen, maybe, but maybe not, now that duality mania has started.

well yeah, but then i dont get why nvidia told me there wont be a new sli. they said there wont be sli2 and they wont make changes to sli because it works just perfectly for them the way its now. thats what they told me, and if it was a lie it was a good lie as he told me all this totally seriously while he was smiling and giving me hints about other stuff...


saaya, that's just too amusing... you should have asked ATI about it at Cebit :lol: You made me laugh reading through your first posts... :D

And you made me laugh hard second time... :up: :rotf:

yeah, the arm thing made me lough out loud as well :D

if i would have told the ati guys i interviewed this they wouldnt have even smiled i think. rene was his name i think, he laughed a lot but mostly about his own jokes :P :D


this is a great thread.

what i dont get is the part about AMR being avaliable on nvidia chipsets. it doesnt make sence, why would nvidia promote ati's technology?

So does this mean that if you want a nvidia motherboard you wont be able to have AMR? and the opposite of course.

thats what i think.

ati told me they want amr to be way more flexible than nvidias sli. they want it to work on as many chipsets as possible and as many different cards as possible while for sli you are limited to have the same exact cards from the same vendor, ati wants to make their technique way more flexible than this so you can run different cards together, probably even a faster and a slower card, though i heard that in this card the faster card clocks itself down to the same speed as the slower card automatically.

and why not have amr on nforce4 sli boards? think about all the boards that have ben sold already, its just ati beeing realistically when they figure that all those people wont buy a new board just to be able to run amr.

and im pretty sure that amr will be faster on ati chipset boards, but the big bonus is that it also works on other chipsets unlike sli wich works on nvidia boards only.


nVIDIA already made nForce4. It's not like they can change much about it. If ATi wants to re-write their own drivers for it, for AMR support, all the better to them. They saves themselfs millions in R&D, plus the market of both nVIDIA and ATi motherboards. Same on the flipside.

the way i understand it there doesnt have to be a new chipset driver, all amr needs to work is to have two 16 slots with enough pciE lanes that can be configured to 8/8 or 16/16 (in future). thats all...
about the bridge thing, ati told me they are still not sure whether they will use a bridge solution or not! it might be that they will offer a bridge solution, but i think they are doing their best to make it work without a bridge.

ati told me that nvidia can easily prevent them from using amr on their boards with a bios/driver update, but i hope nvidia wont be that ....unfriendly... lol :D


AFAIK, MVP is a tech to ASUS and not ATI....
ahhh, that makes sence... so mvp is dual vpu on a card, right?
this is the tech that asus developed together with gigabyte then...
asus and gigabyte (and leadtek soon as well) use the same technique to make two vpus run in sli on a card i heard.

[cadaveca]

some cards in AMR will clock higher..most notably in the x800 series(just think about what was announced about overdrive).

[rhino56]

saaya,all i can say is you rock. please inform ati to catch up with you.

[saaya]

some cards in AMR will clock higher..most notably in the x800 series(just think about what was announced about overdrive).
like how? you think they will change the temp profiles so the cards clock themselves higher at the same temps?

well the overdrive thing is that it depends on the cards and the case temp... wich varies a lot... im not a big fan of overdrive at all tbh... it symbolizes how manufacturers try to make ocing go "mainstream" :D


saaya,all i can say is you rock. please inform ati to catch up with you.
lol, thx :D
catch up with me how?

[Stuperman]

my head hurts and I am very confused after reading all this about arms and most valuable players. Why are ATI and nVidia concerned with Barry Bonds?

j/k

note to self:sell kidney to afford 2xR520+RS482+FX-57

[Cpt Twitchy]

my head hurts and I am very confused after reading all this about arms and most valuable players. Why are ATI and nVidia concerned with Barry Bonds?

I think Stuperman is onto something. Don't you :confused: :p:

[cheece2001]

i just hope that the amr thing will work on the nvidia chipset cause i am gettin a nvidia board. Wouldn't it be just a matter of making very little small changes for it to work on nforce4 boards. Like why not make the little change that helps a TON and also helps out the companies.

[saaya]

stuperman RD480 ;)

well im sure that ati doesnt want to test its drivers with nvidia boards all the time, especially since nvidia might change their drivers/bios files to make it harder/impossible for them to get amr working on nvida boards...

but if they can get it to work fine on vnidia boards then they will def do it, im sure of that :)

[Stuperman]

for the longest time getting things to play nice with each other was really hard, the came DX and plug and Pray, and everyone was happy. but if what saaya says happens (nvidia makes AMR/MVP disfunctional with it's boards) that will be a giant step backwards in the PC world.

[saaya]

it wouldnt be the first time... amd and intel cpus used to be pin compatible, then intel sued amd and forced them to have their own sockets and boards, wich also slowed down the pc market quite a bit afaik.

imagine we would only need one nforce 4 mainboard and could switch from a prescott to an a64 just like that... :D
im sure it would speed up the pc industry...

[gundamit]

Saaya, you've started a hilarious thread here. I don't know what was funnier, the guys who got sucked in and thought you were a retard on SLI, or the guys who thought you had a top secret prototype motherboard sitting in your secret lab next to the laptops you swiped off the Nivida table at Cebit. No matter how many sticks with cat poo they threaten you with, never give up the location of your underground lab.

[aoc007]

stuperman RD480 ;)

well im sure that ati doesnt want to test its drivers with nvidia boards all the time, especially since nvidia might change their drivers/bios files to make it harder/impossible for them to get amr working on nvida boards...

but if they can get it to work fine on vnidia boards then they will def do it, im sure of that :)

I hope r520 amr works with nvidia sli, I have a feeling that the ati chipsets wont be as good as the nvidias, but I doubt it, amr probably has some proprietary technology in it. Besides amr is going to be only intel at first right?

[dyzophoria]

you must be kiddin us? right?, there is no way sli will work on those cards

[saaya]

Saaya, you've started a hilarious thread here. I don't know what was funnier, the guys who got sucked in and thought you were a retard on SLI, or the guys who thought you had a top secret prototype motherboard sitting in your secret lab next to the laptops you swiped off the Nivida table at Cebit. No matter how many sticks with cat poo they threaten you with, never give up the location of your underground lab.

*salutes* YES, Sir :D


I hope r520 amr works with nvidia sli, I have a feeling that the ati chipsets wont be as good as the nvidias, but I doubt it, amr probably has some proprietary technology in it. Besides amr is going to be only intel at first right? what makes you think so? read this thread, it looks like it basically doesnt depend on the chipset, so it should work on any ati chipset at least, intel or amd, if not also on via uli sis and nvidia and maybe even intel chipsets (though intel told me they dont plan to have a board with two 16x slots that can be configured to 8/8 anytime soon)


you must be kiddin us? right?, there is no way sli will work on those cards
welcome to XtremeSystems :D :toast:

[cadaveca]

what makes you think so? read this thread, it looks like it basically doesnt depend on the chipset, so it should work on any ati chipset at least, intel or amd, if not also on via uli sis and nvidia and maybe even intel chipsets (though intel told me they dont plan to have a board with two 16x slots that can be configured to 8/8 anytime soon)

I'd have to agree with you, saaya. It is probably just a driver needed, and maybe a crapload of ram! If i was a programmer, i'd be working at it right now....

[saaya]

too bad we both arent! :D

[jkabaseball]

does this mean that basicly if you switch from ATI to Nvidia or vise versa, you will need a new mobo with a new chipset?

[saaya]

does this mean that basicly if you switch from ATI to Nvidia or vise versa, you will need a new mobo with a new chipset?

? did you read any posts of this thread at all? :P

[cadaveca]

Just so everyone understands, a single rendered frame, @
1280x1024x32bpp =

(32 bits per pixel = )4 bytes x(1280x1024 pixels = )1310720 pixels

=5242880 bits per frame

/

1024(one k=1024bytes)

=5120k/(1024k=1mb) =5mb/ frame

each card will only buffer 3 frames, so even if the second card is sending 60 frames/second, that's only 300mb/second, which can be done with 3 pci-e lanes, without compression, assuming that there is no shared vertex data.

[saaya]

if theres a lot of vertex processing then the first card usually renders almost the entire frame, what does this mean? is the vertex info shared between the two cards or not? what do you think?

[cadaveca]

only does the data get shared when the vertexdata is shared between the frame..like doom of old and such.(the next frame needs to know how to draw from the previous frame-- the data gets sent to the second card after the first card renders the frame,and is checked while the frame is buffered, i suppose. or vice versa)


I konw for a fact that games with large amts of shared vertex data simply do not perform well on SLi.

Here's a tidbit for you, saaya...

SLI software only...no more mobo bridge(gpu bridge nessecary for the vertex/frame swaps in all situations, for nvidia)

http://www.tomshardware.com/motherboard/20050323/index.html


Go to the nVidia site, and downlaod the gpu programming guide for alot of the info that i have about nVidia's SLi, and even thier architechture in general.

i've attached a pic of the board.


SO you can truly say now, it's all about the driver. LoL. maybe this new asus board will do ATI AMR?