ATI's R600 has 64 real pipes

[onethreehill]

"WHEN we first time heard that R600 was going to be a big chip we could figure out that ATI wanted to completely redesign the chip and fill it full of lot pipes."

http://www.theinquirer.net/default.aspx?article=33784

[NightCrawler™]

That's sick... if true....

[BlackX]

Thats a lot :slobber:

[grimREEFER]

im convinced the goal in these next gen cards is to be prodigously faster than quad sli in every possible situation...and have dx10 support of course lol.

[party animal]

That is so many pipes. :slobber:

[metro.cl]

seems almost imposible to go from 16 to 64 real pipes

[Starscream]

will be one big fat expensive chip.
meaning it more then likely will be expensive and ATI wont have much room with pricing or ATI must be ok with selling them under price.

Nvidias solution will be slower but will be on the market sooner and its likely that Nvidia will have room with the price.

[[TAG]Imp]

i call BS.

[BlackX]

will be one big fat expensive chip.
meaning it more then likely will be expensive and ATI wont have much room with pricing or ATI must be ok with selling them under price.

Nvidias solution will be slower but will be on the market sooner and its likely that Nvidia will have room with the price.

Same story like with the 7800GTX 512MB..
They've costed ~800$ each but second day they were out of stock :p:

[nn_step]

as opposed to what "fake pipes"

[Steensen]

Just the thought that this is true gives me goose bumbs.

What an awsome performance the R60 wil have, if it is true. X900XTX Cf @ 780/820 go home.

[Piotrsama]

64 real pipes?
Yes, and die size will be 500mm....:rolleyes:

[ahmad]

He is either confused with the naming or this is really going to be a monster.

Someone much more informed than those @ the inq said it could have 16 or 24 pixel pipelines and I think that sounds right.

[perkam]

16x4 still sounds far move believable. Plus, whenever you see the inq posting speculative stuff no one else is reporting...pls take it with some of this:

http://ww2.heartandstroke.ca/Images/English/no_salt.jpg

Perkam

[Gorgul]

The R580(1900XT) all ready has 48 of this so called real "pipes".

[Cobalt]

The article indicated that there will not only be the 64 shaders already predicted but will also have 64 ROPs

The X1900 has 48 shaders but only 16 ROPs. Basically the INQ is saying that ATI are moving away from the more shaders than pipes configuration.

[Nedjo]

And what could be the point of 64ROPs, when even with GDDR4 you cant have enough BW to truly transpire that resterization power!

It’s a waste of transistors, but then again if true, R600 must have some superior memory management technology!

[aMp]

That's gotta be physically impossible, unless by pipelines they mean something entirely different than we're used to. Even with a process shrink to 65nm the die would be completely unecomical to make.

[Starscream]

ATI can make such a GPU but it will cost alot and ATI cant afford it to sell this thing at a loss so they wont be able to drop the price for a while wich will give them the performence crown but i doubt it will get em cash.
this is ofc if its true as i doubt that its true.

[Mr. Popo]

What about AMD's tech? can't they use AMD's "Advance" tech to apply it?

[Hooker]

In that short amount of time?! I don't think so...

[Starscream]

What about AMD's tech? can't they use AMD's "Advance" tech to apply it?

to late.
when AMD bought ATi the R600 was already to far.
i dunno if AMD has to be involved from the ground up to put their tech to use so maybe R700.
I do hope ATi wont make the mistake to make R600 to powerfull.

[Mr. Popo]

to late.
when AMD bought ATi the R600 was already to far.
i dunno if AMD has to be involved from the ground up to put their tech to use so maybe R700.
I do hope ATi wont make the mistake to make R600 to powerfull.
too powerful? WTF?!

[NinjaWreck]

I do hope ATi wont make the mistake to make R600 to powerfull.

There is no such thing as too powerful.

[Mr. Popo]

It would be a mistake making the R600 too powerful, I wont buy a card that would get too high FPS! It's just wrong! (NOT) :slapass:

[milkcafe]

64x3 texture Units

[DilTech]

Physical pipes = FIXED PIPES.

Guys... If it's 64, it's 64 shaders, not pipes. With ATi bragging about their totally unified architecture, there's no room for 64 pipelines, or it'd be a hybrid design, not a unified architecture like they've been talking about.

This is just the inq trying to make a few filler stories. 64 pipes.... 64 Shaders, Yes. 64 pipelines? No.

[Starscream]

making it to powerfull meaning alot stronger then the G80.

Its almost a fact that the R600 will beat the G80 (unless Nvidia has a suprise) and seeing the rumors its also kind of hinted that the r600 will cost a good bit more to make.

so if the R600 really will be expensive to make it will mean that in retail it will cost a good buck and most of us wont be willing to pay that cash.
Also them rumored specs could mean that gettign a good yield could be hard wich is a thing non of us want and will again inflate price.

So id rather see it that these specs r fake and that ATi will be a bit more conservative with the power of the R600 making the card in the end easier to make and therefor more affordable for us.

Cause the rumored specs (pipes and speeds) seem kind of hard to pull of at acceptable yields and price.
So there are 3 options:
The specs r fake or ATI is actualy able to make sucha GPu at an acceptable retail price and good availability.
the third possibility being that wel get one kick ass monster card wich will rape all else but will cost a crapload and will be hard to get a hand on.

the last option is the one non of us want cause we already had enough cards around that no one could buy (price or availability being :banana::banana::banana::banana:ed)

[iddqd]

Empty speculations.

I'm not sure why INQ will even publish these, as this kills their credibility in everyone's eyes, but I guess they can do whatever they please.

[M.Beier]

If this is true, I'll just be goin' "DUUUUDE, WOOOW DUUUUDE" ;)

Seems like its worth skippin' X1950XTX and stickin' with X1900XTX unless X1950XTX will be as cheap as romours says :)

[fhpchris]

INQ printed same roumors about the X1900s if you newbies will remember that long ago :)

Grain of Salt here...

[M.Beier]

INQ printed same roumors about the X1900s if you newbies will remember that long ago :)

Grain of Salt here...

Please drop "that" stuff dude - and like I stated, if its true, its great, if its fake, well..

[nn_step]

to late.
when AMD bought ATi the R600 was already to far.
i dunno if AMD has to be involved from the ground up to put their tech to use so maybe R700.
I do hope ATi wont make the mistake to make R600 to powerfull.
There is no such thing as too Powerful.
However people please realize how we define Pipes Now has been changed with DX10. what used to mean Pipes Now doesn't

[richnewman]

Whens this thing coming out anyway?

I hope it's in the next so I don't have to buy the DS3.. One can dream, right? :P

[[TAG]Imp]

Imp']i call BS.
there's no way it has 64 physical pipelines...

[Kanavit]

I agree, i'm taking this with a big grain of salt. i don't believe that 64 rops is possible on one 90nm die. Unified shaders maybe. but not rops.

[fhpchris]

Please drop "that" stuff dude - and like I stated, if its true, its great, if its fake, well..

smile = joke, if that was confusing...

The x1900 was rumored to have like 64 physical pipes by the INQ too...

[BlaqMale]

would you bet your life that the inq is right? not saying it's not true but the inq is not credible, remember the 32 pipe 7900gtx?

[turtle]

Or better yet...the 24... no wait...32...no wait...24....wait...16...no wait...24 with some disabled...no wait...32...wait...16 pipe R520? That, for me, was the straw with The Inquirer. They get a lot of news (out to the masses) first, but when it comes to these filler, or as I call them "Fuad's bread and butter for hits" stories, I totally disregard them. After how badly we were led around with a carrot on a string about R520...never again.

All we know is 64 is a looking like a number that will be used most likely in coordination with shaders.

I agree with Perky, 16x4 looks possible if not probable. Like I've said before here and on my long-ass posts at R3D (http://www.rage3d.com/board/showthread.php?t=33864769&page=2), I also don't doubt the possibility of 32x3 either, but 32x2 active because of of one alu per pipe being used for the arbitor (load balancing), as it is on Xenos. Whatever the case, I seriously doubt 64 rops...It doesn't make sense for multiple reasons. That doesn't make it impossible, but seriously improbable. Whatever the case, you can bet ATi will be aching to shrink that sucker down to 65nm ASAP to make it profitable.

[STEvil]

They dropped a nice size bomb on us with R300. Be fun if it happened again :D

[NinjaWreck]

They dropped a nice size bomb on us with R300. Be fun if it happened again :D

Best GPU ever.

[SturmoV]

Best GPU ever.


Lasted me 4 years, so, AGREED! :cool:

[zakelwe]

Physical pipes = FIXED PIPES.

Guys... If it's 64, it's 64 shaders, not pipes. With ATi bragging about their totally unified architecture, there's no room for 64 pipelines, or it'd be a hybrid design, not a unified architecture like they've been talking about.

This is just the inq trying to make a few filler stories. 64 pipes.... 64 Shaders, Yes. 64 pipelines? No.

I think this is hitting the nail on the head and Fuad has got his terminology wrong. Current chip has 16 pipes with 3 shaders per pipe and so it is likely that the r600 has 16 pipes with 4 shaders per pipe doing all shading as they will be unified.

I think the same number of texture units will be present -16 - so the shader to texture ratio will go from 3:1 to 4:1.

Regards

Andy

[BlackX]

I think this is hitting the nail on the head and Fuad has got his terminology wrong. Current chip has 16 pipes with 3 shaders per pipe and so it is likely that the r600 has 16 pipes with 4 shaders per pipe doing all shading as they will be unified.

I think the same number of texture units will be present -16 - so the shader to texture ratio will go from 3:1 to 4:1.

Regards

Andy

Agree.
INQ did it again.. :slapass:

[phi|os]

Why is this news? Hasn't it been understood that the R600 would be a unified chip, so that pixel/shader/geometry use interchangeable pipes? 64 has been the number of this unified architecture for awhile.

And I doubt that ATI would just through a bunch of stuff into chip that doesn't have the bandwidth to fully utilize...that would be a waste of money.

Plus, this source is theINQ, which is starving to create headlines, so don't trip yet.

[EnergyStaR]

I think this is hitting the nail on the head and Fuad has got his terminology wrong. Current chip has 16 pipes with 3 shaders per pipe and so it is likely that the r600 has 16 pipes with 4 shaders per pipe doing all shading as they will be unified.

I think the same number of texture units will be present -16 - so the shader to texture ratio will go from 3:1 to 4:1.

Regards

Andy

Agree

if R600 has 64ROPs the die size will go up to the square centimeter class not the square millimeter class:D

[Rolle2k]

I was thinking that 32x2 would be possible, but 64x1 sounds as BS.

[Mr. Popo]

I think you guys misunderstood the whole thing.
R600 will come with a pack of 64 real stainless steel pipes for your own use! :D

http://img100.imageshack.us/img100/9310/pipesbb2.jpg

[alayashu]

I think you guys misunderstood the whole thing.
R600 will come with a pack of 64 real stainless steel pipes for your own use! :D

http://img100.imageshack.us/img100/9310/pipesbb2.jpg

oh yeah baby, this is so future!
i can now stay playing games and get food and pee & poo through those pipes!
there are so many, i can even invite friends at a "crossfire" =)

[aMp]

Assuming the 16x4 theory is correct (and it seems likely ATM) how much of a performance boost would we be talking about? How big would the die be @ 65nm?

We've heard plenty of rumors that R600 will have a huge die size, suck power like crazy, and be significantly faster than current-gen GPUs. While I take all those rumors with the requisite grain of salt, I have to wonder if something as relatively simple as adding another shader unit and unifying all shaders would really be as big a deal as R600 is supposed to be.

[Mr. Popo]

Assuming the 16x4 theory is correct (and it seems likely ATM) how much of a performance boost would we be talking about? How big would the die be @ 65nm?

We've heard plenty of rumors that R600 will have a huge die size, suck power like crazy, and be significantly faster than current-gen GPUs. While I take all those rumors with the requisite grain of salt, I have to wonder if something as relatively simple as adding another shader unit and unifying all shaders would really be as big a deal as R600 is supposed to be.
It's not 65nm, it's 65m. :D

[±Pink=FloYd±]

how much power would this monster consume?

[BlackX]

how much power would this monster consume?

Nobody knows exactly but probably a lot

[[XC] Lead Head]

Nobody knows exactly but probably a lot

150 watts or lower

[BlackX]

150 watts or lower

Because of PCI-E ?

[Mr. Popo]

I thought the R600 will be 200W, was I wrong?

[[XC] Lead Head]

The PCI-E max spec is 150 watts between the main connector and the 6 pin connector

[Vapor]

The 150W spec will be ignored (by both nV and ATi for this upcoming gen).

The GPU alone will likely consume >150W :( Then add the board and RAM's usage and you're pushing 200W. :(

How can it be ignored? Easy, put two 6-pin connectors on board, BAM! 225W spec ;)

As for the 64pipes, I'm withholding full judgement, but I'll explain some of what I'm thinking. The number 564Mil transistors is sticking in my head for some strange reason, which isn't quite double current gen (~340mil, no?). I made an analysis of R520 vs. R420/R300 awhile back (placeholder for link) and concluded that based off the die size/process, that 32 pipes WERE DEFINITELY possible with R420/R300 style pipes, but that all the new tech (SM3.0, etc.) that ATi was promising would limit it to 24.....but likely 16 new-pipes with extra mumbo-jumbo tech and stuff.

I'm gonna do a mini-version of that this time as well. So here goes:

Adding 32shaders from R520 to R580 only added ~25-30mil transistors, so going up to 64 unified pies with only 16 ROPs would be, well, not much considering you're throwing 8 vertex pipes out the window from the R5x0 design. That leaves >150Mil transistors to get filled. New tech? Surprisingly little outside of unifying pipes. Okay, so where do we turn?

Well, if there are 64 unified pipes, that limits ATi's options of pipes:ROP ratio to 1:1, 2:1, 8:3, and 4:1.

4:1, well, we're still 150+Mil transistors short, so that's a no-go. Plus, having the same 16ROPs would likely be more than a bottleneck, it'd be a hindrance.

1:1 is possible, but extremely unlikely due to ATi's stance of "3:1 is best", and would be tough to fit into the 564Mil transistors (not impossible, likely needing hardware features chucked)

8:3 (64 pipes, 24 ROPs) is very possible. It's a (logical) 50% boost in ROPs and still has the rumored 64 pipelines. It is very close to 3:1 and, with some added new hardware features, could likely fill the 564Mil.

2:1 (64 pipes, 32 ROPs) is my pick. It could provide the staggering performance increase we're expecting, it (likely) fits into 564Mil. It can alleviate bottlenecks to the unified pipes. It allows an upgrade cycle into ATI's favored 3:1 ratio (96:32).

Of course, this is all dependent on the 564Mil number (that's literally stuck in my head, probably from something completely different, lol) being correct/close :p:

[Pinnacle]

Pipes, I like pipes, fat ones too.

Wait and see, wait and see

[Mr. Popo]

Pipes, I like pipes, fat ones too.

Wait and see, wait and see
All sizes. :toast:


I think you guys misunderstood the whole thing.
R600 will come with a pack of 64 real stainless steel pipes for your own use! :D

http://img100.imageshack.us/img100/9310/pipesbb2.jpg

[nn_step]

I think you guys misunderstood the whole thing.
R600 will come with a pack of 64 real stainless steel pipes for your own use! :D

http://img100.imageshack.us/img100/9310/pipesbb2.jpg
Likes we might have to do some Plumbing work :rolleyes:

[Revv23]

Lasted me 4 years, so, AGREED! :cool:


My x800 died so im still using mine! :)

I had to stop using my DFI lanparty and go back to a 2.8c pentium, but this thing is stilla great card.

I dunno, they have been working on R600 design ever since they finished the r350 so there is huge potential here. I'll see it wehen i see it, and i may not be getting a x1950 after all.

[[XC] Lead Head]

No, The PCI-E MAX SPEC is 150 watts. Only 1 6-Pin connector is allowed. It would take quite alot to get the spec bumped. The only way to override the spec would be to cut the pins that supply power to the card from the main connector, and use an external PSU...

[[XC] hipno650]

so lets see 4 cores percard 64 pipes each feeding 256shaders each and 64rops each and 4gb of 3.2ghz ddr5 mem with built in phase cooling:slobber: :slobber: :slobber: and running quad crossfire:slobber: :slobber: :slobber: so lets see 64x4 is 256 pipes percard then 256x4 is 1024 shaders per card:slobber: :slobber: and that makes 1024x4 is 4096 shaders in total:slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: :slobber: THAT WOULD BE COOL!!!!!!! but mabye in about 20years:D i think it will be 64 shaders. but 64 real pipes would be cool to:D

[Staucher]

are waiting

[Revv23]

No, The PCI-E MAX SPEC is 150 watts. Only 1 6-Pin connector is allowed. It would take quite alot to get the spec bumped. The only way to override the spec would be to cut the pins that supply power to the card from the main connector, and use an external PSU...


i dont get it, whats wrong with using 2x6 pin connectors? Who is stopping ati/nvidia from doing that, sure the rules say one thing, but there is only one real rule in computer hardware - performance is king.

Or evenbetter ,like you said external psu, its been done before and i reckon we might see it again.

[[XC] hipno650]

what about a 6pin and a 4pin molex?

[aMp]

Specs are made to be broken. One of the major reasons for multi-rail PSUs was to keep the current on any given rail under 240VA, but we've all seen units with more than 20A on a single rail. If there's no technical reason two 6-pin connectors won't work, no manufacturer will stand on principle to keep it within spec.

[Revv23]

what about a 6pin and a 4pin molex?


An idea man i like it. :)

Power problem solved, box 'em and ship 'em, ill take two :D

[Vapor]

It's not that specs are made to be broken (though they will be), but ATi/nV *WILL* use whatever power is necessary for them to get the performance they think they'll need to compete.

nV will not draw more than 75W from PCIe port (at 100MHz ;)), ATi probably won't either, but I'm not sure.

There is no technical reason limiting any power supply to the cards....they can have their own independent PSUs (supplied with card), they can take any number/type of plugs. However, it's almost a known that ATi/nV will not require overloaded plugs (they're designed for 75W and, though they can certainly supply a lot more when necessary [like when we OC :p:], neither maker will require overdrawn connectors).

Having an array of plugs will actually ease card design as well....it can help simplify PCB design (as long as the second connector isn't an afterthought).

As for what PSUs will support the cards? 5.25" bay PSUs will, good PSUs will, and if you're running one (or two) of the super-draw cards, why would you even considering risking your hardware with a crappy PSU?

I must say, if the Inq is right about this...150W to the core might be slightly on the low side....current GPU cores are hot and consume a lot of energy, but due to diversity of functions, it's very hard to stress the entire core at once (impossible actually)....I wouldn't doubt unified pipes ability to move the actual power draw closer to theoretical draw :eek: 64 full-out pipes will cook for sure (back-up livestock brander, anybody? ;))....heck, even a 64:24 config will cook.

It'll be interesting, these upcoming months....

[STEvil]

just utilizing the whole 6-pin PCI-E plug would help, too.

There are 3 positive leads and 3 negatives on it. Each can supply in excess of 80w (3x80 = 240w+75w from PCI-E slot = 315w) assuming the PSU can cope with this and the cables and connectors can also handle these loads. Seeing that people have powered 86w+ peltier units with a single molex (including myself) and older 20 pin PSU's powered entire motherboards and PSU's using a limited number of positive leads i'd say we're probably ok. Some might burn (brown molex connectors, anyone remember those?) but ultimately if your PSU is high enough quality we'll be fine. Multiple 6-pin connectors on a single card would not be a problem either and may help to end the fad PSU manufacturers are going with for adding more and more different style connectors....

The PCI-E slot is rated at 75w but can be pushed past this limit (+10% current and other options on CFX3200-DR for example). In total I would not be surprised to see the ability to exceed 300w of power draw with the current available resoures.

[Revv23]

yeah my manta board can put over 100w in the pcie slots.

[MaxxxRacer]

wow, what a pipe dream....

(couldnt resist the pun)

[aMp]

I have a tough time believing they'd push individual parts (PCI-E slot, connectors) beyond spec because they can't guarantee the integrity of the other hardware.

Say they tried to pull 100W from the PCI-E slot. You'd see a lot of mobos blowing up, because there are always going to be a lot of people who don't do their homework or don't know any better, and will drop their new R600 into a two-year-old Asrock.

We've seen the same thing with people trying to power their shiny Crossfire rig with a 450W Rosewill PSU. The difference there is that it just doesn't work -- nothing catches on fire.

[Stuperman]

Agree

if R600 has 64ROPs the die size will go up to the square centimeter class not the square millimeter class:D

we are alread there, 1cm^2 = 1cm x 1cm = 10mm x 10mm = 100mm^2 = 1cm^2.

I tend to agree though, 64ROPs doesn't make sence, I mean as clock speeds go up wouldn't the need for ROPs go down, for example 16 ROPs @ 500mhz has the same fill rate as 12 @ 667mhz. Or do I not know what I am talking about?

[Revv23]

we are alread there, 1cm^2 = 1cm x 1cm = 10mm x 10mm = 100mm^2 = 1cm^2.

I tend to agree though, 64ROPs doesn't make sence, I mean as clock speeds go up wouldn't the need for ROPs go down, for example 16 ROPs @ 500mhz has the same fill rate as 12 @ 667mhz. Or do I not know what I am talking about?


whs to say they are raising clocks?

[Marlowe]

Guys, maby it will have more than one core.. :)

[kemist]

what he probably means is it'll be 4x16 array. (if its like R500, which is 3x16) I dont think he ever really showed that he knew what TMUS, ROPS etc are in any of his other articles, so i wouldnt take anything more than that out of it.

[MustardTheoRy]

Hey guys, I read that all 64 pipes would adjust in processing shaders, vertex and geomtry in realtime for whatever the frame required.

Also I believe AMD took over ATI to incorporate transcoding onto thier cpu's and also borrow some design to make thier chips faster.

Also AMD is designing a graphics chip built right onto the cpu

[KoHaN69]

Well, I sure hope it can run Counter-Strike!

/noobasm

[turtle]

Holy thread resurrection. :P

You know...the interesting thing pertaining to this article remains that in Xenos, Rops do not equal shader pipes which in turn do not equal shader alus. R600 could very well be similar.

Xenos has only 8 ROPs (because of the EDRAM), but a 16x3 shader array...All on a 186mm2 232m tranny 90nm package...Roughly half the size of what we think R600 will be on 80nm and with 40%-45% of the transistors. R600 obviously needs more ROPs per shader/tmus.

If R600 has 64 pipes...it probably has less ROPs and more shader ALUs.

Thinking about the possible combos still gives me a headache though...I still to this day like these (http://www.megagames.com/news/html/hardware/atir600specsrevealed.shtml) specs even though they're widely thought bs.

Even if you scrap the 65nm, lower the ram speed to a reasonable expectation (as I don't see Hynix announcing mass availability of that 3.6ghz GDDR4 they said they would...) and I think the rest still looks reasonable...and perhaps that spec as the 2007 refresh.

It does make sense it should have more than 64 shader alus, as ATi has said USA configured as DX9 separate pixel/vertex shaders in a 2ps:1vs gives a 20-25% performance loss which certainly doesn't sound like a performance-leading dx9 part. Take 64x2 though...Sounds a little more feasible...and doesn't contradict the 64 real pipe story. 24x4 or 32x3/4 also would make sense if shader arrays were equal to the amount ROPs...but would screw up the 64 being used somewhere.

It def would be nice if we had had either more information, or the same information passed through a source that actually speaks English and knows what they're talking about.

[revenant]

"oh my god, it's full of pipes"

the R600, a graphic pipeline odyssey ;)

[Lightman]

It looks like 64 shader pipes with 2 ALUs per pipe. And if this spec

65nm
64 Shader pipelines (Vec4+Scalar)
32 TMU's
32 ROPs
128 Shader Operations per Cycle
800MHz Core
102.4 billion shader ops/sec
512GFLOPs for the shaders
2 Billion triangles/sec
25.6 Gpixels/Gtexels/sec
256-bit 512MB 1.8GHz GDDR4 Memory
57.6 GB/sec Bandwidth (at 1.8GHz)
WGF2.0 Unified Shader
is close to real then R600 should be almost 5x faster in calculations than R580 (which have around 100GFLOPs)!

32ROPs are very probably for me because of 30'' cinema displays (2560x1600). For this kind of resolutions you need more than 16ROPs when core clock is under 1GHz. :cool:

[L&MFatty]

It looks like 64 shader pipes with 2 ALUs per pipe. And if this spec

is close to real then R600 should be almost 5x faster in calculations than R580 (which have around 100GFLOPs)!

32ROPs are very probably for me because of 30'' cinema displays (2560x1600). For this kind of resolutions you need more than 16ROPs when core clock is under 1GHz. :cool:
This would give conroe a run for it's money :fact:

[LowRun]

This would give conroe a run for it's money :fact:

:confused: :confused: :confused:

[Helmore]

No it might even give it more value for the money.

[nn_step]

This would give conroe a run for it's money :fact:
umm you can not compare GPUs and CPUs directly

[BlackX]

I think he meant that only conroe will be capable to show the real power of R600

[Lightman]

I think he meant that only conroe will be capable to show the real power of R600

And future CPU's (like K8L). Now I can think about good test for R600 math power --> BOINC (folding on GPU). 500GFLOPs is like hmmm 500.000.000.000 operations/second or ...... and in crossfire you will be very close to TOP500 supercomputer list :D

[Hicks]

I think he meant that only conroe will be capable to show the real power of R600

Hmmmmmmmm, at 1024, then maybe.

I just hope the R600 is good enough for me to run 4X AA and 16XAF, on all games. I hope it's a really good improvement from my X1800.

[gOJDO]

I will probably have one, sometimes in 2008.

[alexio]

I have a feeling it will be "just" 32 ROP's, resulting in a 32*2 configuration :)

[n-sanity]

And i have a feeling they said coming in late 2006. I guess i know what i want for christmass:D of course im not getting it tho :( Ima wait till dx10 tech matures, and get somethin then.

[BlackX]

And i have a feeling they said coming in late 2006. I guess i know what i want for christmass:D of course im not getting it tho :( Ima wait till dx10 tech matures, and get somethin then.

Probably a good decision

[DilTech]

And i have a feeling they said coming in late 2006. I guess i know what i want for christmass:D of course im not getting it tho :( Ima wait till dx10 tech matures, and get somethin then.

I agree with black, it probably is a good decision...

While the R600 will be beastly, word on the net is that these first gen DX 10 parts will be breaking all the rules about power draw, and in-result, heat output.

According to a few sources, this first round will be putting DX10 out there, and after that the focus will be more-so on getting power-draw down rather than performance from there. Atleast for a few refreshes that is.

[milkcafe]

comparing

7900GTX /24FP/8VP/24TMU/16ROPs
X1950XTX /48FP/8VP/16TMU/16ROPs

8800GTS /96FP/96VP/64TMU/20ROPs
8800GTS /128FP/128VP/64TMU/24ROPs

X2800XT? /196FP/128VP/32TMU/32ROPs

65nm
64 Shader pipelines (Vec4+Scalar)
32 TMU's
32 ROPs
128 Shader Operations per Cycle
800MHz Core
102.4 billion shader ops/sec
512GFLOPs for the shaders
2 Billion triangles/sec
25.6 Gpixels/Gtexels/sec
256-bit 512MB 1.8GHz GDDR4 Memory
57.6 GB/sec Bandwidth (at 1.8GHz)
WGF2.0 Unified Shader

7900GTX 256-bit 512MB 1.1ns GDDR3 @ 650MHz/1600MHz
Shader Operations: 15600 Operations/sec
Pixel Fill Rate: 10400 MPixels/sec
Texture Fill Rate: 15600 MTexels/sec
Vertex Operations: 1300 MVertices/sec
Memory Bandwidth: 51.2 GB/sec
Fragment Pipelines: 24
Vertex Pipelines: 8
Texture Units: 24
Raster Operators 16

X1950XTX 256-bit 512MB 0.85ns GDDR4 @ 650MHz/2000MHz
Shader Operations: 31200 Operations/sec
Pixel Fill Rate: 10400 MPixels/sec
Texture Fill Rate: 10400 MTexels/sec
Vertex Operations: 1300 MVertices/sec
Memory Bandwidth: 64 GB/sec
Fragment Pipelines: 48
Vertex Pipelines: 8
Texture Units: 16
Raster Operators 16

8800GTS 320-bit 640MB 1.0ns GDDR3 @ 500MHz/1600MHz
Core Clock: 500 MHz
Memory Clock: 800 MHz (1600 DDR)
Memory Bandwidth: 64 GB/sec
Shader Operations: 48000 Operations/sec
Pixel Fill Rate: 10000 MPixels/sec
Texture Fill Rate: 32000 MTexels/sec
Vertex Operations: 12000 MVertices/sec
Framebuffer: 640 MB
Memory Type: GDDR3
Memory Bus Type: 64x5 (320 bit)
Fragment Pipelines: 96
Vertex Pipelines: 96
Texture Units: 64
Raster Operators 20


8800GTX 384-bit 768MB 1.0ns GDDR3 @ 575MHz/1800MHz
Memory Bandwidth: 86.4 GB/sec
Shader Operations: 73600 Operations/sec
Pixel Fill Rate: 13800 MPixels/sec
Texture Fill Rate: 36800 MTexels/sec
Vertex Operations: 18400 MVertices/sec
Framebuffer: 768 MB
Memory Type: GDDR3
Memory Bus Type: 64x6 (384 bit)
Fragment Pipelines: 128
Vertex Pipelines: 128
Texture Units: 64
Raster Operators 24

[[XC] Lead Head]

Umm....bump anyone?

[XSAlliN]

Hmm, is it just me or the current CPU's from Dual Core Class might bottleneck that card? - for Quad Core might be just fine but where's the point in that?

[Revv23]

Hmm, is it just me or the current CPU's from Dual Core Class might bottleneck that card? - for Quad Core might be just fine but where's the point in that?


of course they will, looks at the 8800, its held back even with a 5ghz conroe.

Need better graphics to put the load back on GPU.. that or much faster cpu's. I bet graphics come first.

[nn_step]

of course they will, looks at the 8800, its held back even with a 5ghz conroe.

Need better graphics to put the load back on GPU.. that or much faster cpu's. I bet graphics come first.
The only way to do that is to take more features from the Standard CPU and drop them into the GPU. Which will make them far more CPU like. Which means ATi's R600 design would probably be a far better design than the G80

[perkam]

Too far away.

Perkam

[Turtle 1]

http://www.theinquirer.net/default.aspx?article=35614

[biohead]

ATI is almost five months late with its original plans.
Right :/

[Mr. Popo]

I think you guys misunderstood the whole thing.
R600 will come with a pack of 64 real stainless steel pipes for your own use! :D

http://img100.imageshack.us/img100/9310/pipesbb2.jpg
OMFG, that was funny (I'm a funny guy).

[Cooper]

shows the projected performance to help game developers to debug their games and get it ready for the new marchitecture

OMG
where does INQ take those authors ? Game developers emulate new games on huge farms and work with API not with DX10 hardware.

If I`d get a neckle for every BS from the INQ I`d be millionair by now :rofl:

[DilTech]

The only way to do that is to take more features from the Standard CPU and drop them into the GPU. Which will make them far more CPU like. Which means ATi's R600 design would probably be a far better design than the G80

Umm...read up on CUDA. NVidia's already in that game with the G80.

[generics_user]

Umm...read up on CUDA. NVidia's already in that game with the G80.
which is much more cpu dependant than any other card. i think nn meant that ati can take processes (like drivers) from the cpu onto the vga to make it less cu dependant and more scalable in CF

[Kunaak]

I really wish that ATI would stop making better cards then Nvidia, cause I love SLI and Hate Crossfire, hahaha...

[DilTech]

which is much more cpu dependant than any other card. i think nn meant that ati can take processes (like drivers) from the cpu onto the vga to make it less cu dependant and more scalable in CF

You'd have to have the instructions in place for the cpu to make the calls for the gpu in the first place to make those driver calls on the gpu. That'd be more work than needed, as you'd not only have the driver overhead on the cpu, but now also the driver overhead on the gpu taking away performance from rendering the image.

A lose-lose situation indeed.

[JoeBar]

Too far away.

Perkam
But not too late... ;)

[CedricFP]

Any news on the R600 release date? I am wondering if I should build my new rig now with a G80, or wait for the R600.

[DilTech]

Any news on the R600 release date? I am wondering if I should build my new rig now with a G80, or wait for the R600.

February...at the earliest. Possibly not until march though.


which is much more cpu dependant than any other card. i think nn meant that ati can take processes (like drivers) from the cpu onto the vga to make it less cu dependant and more scalable in CF

Read up on CUDA, the G80 is able to do work for the cpu.

http://enthusiast.hardocp.com/article.html?art=MTIxOCw4LCxoZW50aHVzaWFzdA==

Each one of those 128 "stream processors" can perform cpu tasks, and share data in the cache found on the G80.

[Hicks]

Jesus March? are you kidding me.

I really want a new video card. My X1800XT is getting a tad slow, but i god dammit don't want an Nvidia card. I was hoping R600 would be out by Janurary.

The 8800GTX does look great, but i's Nvidia, lol.:rolleyes:

[Metroid]

I put all my hope on this silicon.

I hope be right.:slobber:

[Turtle 1]

February...at the earliest. Possibly not until march though.



Read up on CUDA, the G80 is able to do work for the cpu.

http://enthusiast.hardocp.com/article.html?art=MTIxOCw4LCxoZW50aHVzaWFzdA==

Each one of those 128 "stream processors" can perform cpu tasks, and share data in the cache found on the G80.
Hows about a link showing a march release date. I think many would like to see it. I most certainly would. Sorry Charlie no links from Charlie he has never gotten anything right about ATI. Fact is guys from ATI love pulling his strings.

[vitaminc]

What about AMD's tech? can't they use AMD's "Advance" tech to apply it?

why use AMD tech when its lagging behind the foundries?

[Sparky]

I do like ATI over nvidia - always have for some reason - and I hope they pull it off for a powerful card and a good price. I guess I'll have to wait and see.

But I hope one thing they also do is fix the heat dump, considering my X1900XT gets up to 90 degrees C routinely under load I'd hate to see what heat output of the R600 if they don't do something about it.

"Central heating? Who needs that? I have my graphics card!"

[metro.cl]

I do like ATI over nvidia - always have for some reason - and I hope they pull it off for a powerful card and a good price. I guess I'll have to wait and see.

But I hope one thing they also do is fix the heat dump, considering my X1900XT gets up to 90 degrees C routinely under load I'd hate to see what heat output of the R600 if they don't do something about it.

"Central heating? Who needs that? I have my graphics card!"

They wont do that in the high end

[Turtle 1]

I do like ATI over nvidia - always have for some reason - and I hope they pull it off for a powerful card and a good price. I guess I'll have to wait and see.

But I hope one thing they also do is fix the heat dump, considering my X1900XT gets up to 90 degrees C routinely under load I'd hate to see what heat output of the R600 if they don't do something about it.

"Central heating? Who needs that? I have my graphics card!"

Ya the heat from ATI's latest and greatest is high. When wife is gaming and I am browsing it gets pretty hot in this large room and were all water cooled. Heat vents are all closed and window is opened a bit. Looks like G80 is putting out massive heat also. I told wife we should just put a pc in each room and shout heat off.

[Cooper]

Moved

[thunderstruck!]

Moved
:rolleyes: