Chiphell RV770 rumour?

[Iconyu]

http://www.yougamers.com/forum/showthread.php?t=76368

Anyone got any insight about this one?

The previous numbers posted by couple sites said 480 SPs, 32 TMUs, over 1GHz clock and GDDR5 memories.

According to chiphell, however, RV770 in fact features 640 SPs and 32 TMUs, and clocked higher than RV670 but not over 1GHZ. They also report the memories being GDDR5.

Also, according to them A11 (first revision) of the chip was taped out recently, and it's apparently in production with fairly good yields, over 71%, and there's no reports of A12 revision needed.

Edit: If this doesn't class as news, due to it's questioning format, please feel free to move post.

[Craftyman.]

They also report the memories being GDDR5.

Yesssssssssss. :)

[xlink]

let see before:
50% increase in SPs
100% increase in TMUs

now:
100% increase in SPs
100% increase in TMUs


so they doubled the chip basically.

sounds like GeForce 6 to GeForce 7, frequency upscaling is close around 15-20%ish higher, basic doubling of chip components...

sounds like a nice leap.

[HKPolice]

:rofl: all lies.

Those fan boys really are fulla ****

[Nedjo]

part about high yield from initial silicon sounds promising

[Eastcoasthandle]

Wait a minute is this nothing more then another X2 or is this a single GPU solution???

[Craftyman.]

let see before:
50% increase in SPs
100% increase in TMUs

now:
100% increase in SPs
100% increase in TMUs


so they doubled the chip basically.

sounds like GeForce 6 to GeForce 7, frequency upscaling is close around 15-20%ish higher, basic doubling of chip components...

sounds like a nice leap.

Yep, I agree :yepp: . Needs an increase in ROPs though :shrug:

[xlink]

Yep, I agree :yepp: . Needs an increase in ROPs though :shrug:

they said nothing of ROPs...

lets assume that their engineers can figure it out on their own... I'd hope that they could

[Iconyu]

Wait a minute is this nothing more then another X2 or is this a single GPU solution???

RV770 is the single chip card, ATI are supposed to be using two of the for the R700 as far as I know. But as this chip was designed with that in mind, I guess there will be couple fewer issues with games only running on one of the two cores than we see presently.

[STaRGaZeR]

Yesssssssssss. :)

You should focus on SP numbers more than memory bandwidth. R600 architecture (and the rumours say RV770 is a R600 derivative after all) is not limited by bandwidth.

I hope these rumours are true, BTW even at 55nm that should be a huge chip with all the components doubled :eek:

[largon]

I'd be interested how they supposedly doubled the chip without going insane with the die size... :rolleyes:

Anyways, ATi R6 GPUs don't really need more render back-end (= ROP) muscle, only TMUs and SPs. And in that order.

[STaRGaZeR]

I've just seen it. 250mm^2 of RV770 vs 190mm^2 of RV670.

[[XC] gomeler]

MMMM Hot stuff. I want two of these, badly. My RV670's just don't scale high enough :p:

[Papu]

:rofl: all lies.

Those fan boys really are fulla ****

what an insightful and helpful comment.:down: :spam:

[Craftyman.]

You should focus on SP numbers more than memory bandwidth. R600 architecture (and the rumours say RV770 is a R600 derivative after all) is not limited by bandwidth.

I hope these rumours are true, BTW even at 55nm that should be a huge chip with all the components doubled :eek:

Hmm, interesting. regardless its nice to see technology move forward, right? :up:

[STaRGaZeR]

Of course, faster memory means we need smaller memory buses (and cheaper) to achieve the same final bandwidth. But the main thing here is the increase in TMUs and SP numbers, and the possible return of hardware AA. Can't wait to see the chip in action, honestly I'm done with all these rumours and FUD everywhere about everything.

[metro.cl]

On VR-Zone they dont say anything about more ROPs, TMUs, or else

I kinda trust VR-Zone more for the translation.

[Craftyman.]

Of course, faster memory means we need smaller memory buses (and cheaper) to achieve the same final bandwidth. But the main thing here is the increase in TMUs and SP numbers, and the possible return of hardware AA. Can't wait to see the chip in action, honestly I'm done with all these rumours and FUD everywhere about everything.

QFT, how's about they release this thing so I can start shelling out cash :rofl:

[LordEC911]

I hope these rumours are true, BTW even at 55nm that should be a huge chip with all the components doubled :eek:
Should be roughly the same size as G92, especially since they won't need two UVDs, single UVD is 3.3mm2, and they won't need to double the MC.

I'd be interested how they supposedly doubled the chip without going insane with the die size... :rolleyes:
Would be more interested in the trannie count than the die size.

Anyways, ATi R6 GPUs don't really need more render back-end (= ROP) muscle, only TMUs and SPs. And in that order.
Agreed, as long as the tweak the sampling.

I've just seen it. 250mm^2 of RV770 vs 190mm^2 of RV670.
Sarcasm?
I'll be guessing around 300-350mm2 if still on 55nm.

Of course, faster memory means we need smaller memory buses (and cheaper) to achieve the same final bandwidth. But the main thing here is the increase in TMUs and SP numbers, and the possible return of hardware AA. Can't wait to see the chip in action, honestly I'm done with all these rumours and FUD everywhere about everything.
Highly doubt AMD would switch back to hardware AA, especially since that isn't where the industry seems to be headed.

[STaRGaZeR]

Sarcasm?
I'll be guessing around 300-350mm2 if still on 55nm.

Have you actually read the linked article?

Final specification of RV770.
The core frequency of RV770 is a little higher than RV670, but not over 1GHZ. The ram used by RV770 is GDDR5, and the core size is 250 mm2. The stepping A11 GPU's yield is over 71% when recently taped out. And the stepping provided to retail will be the same A11 stepping. Whether stepping A12 will be provided is not known yet. The detail specification of RV770 is just more than two times SP number and TMU number of RV670. The release date of RV770 will be some day in May.

Highly doubt AMD would switch back to hardware AA, especially since that isn't where the industry seems to be headed.

Me too, I was just pointing out the possibility, as BrowncoatGR said here (http://www.xtremesystems.org/forums/showpost.php?p=2812541&postcount=98).

[LordEC911]

Have you actually read the linked article?
Yes but not before I posted.
Just the way you said, "I have just seen it."
I'm still trying to figure out why ChipHell is using that mockup of a RV670 card as if it were RV770...

Me too, I was just pointing out the possibility, as BrowncoatGR said here (http://www.xtremesystems.org/forums/showpost.php?p=2812541&postcount=98).
That's what I was gonna ask, if he had a link to that interview which I think I have already asked before.

[STaRGaZeR]

I don't now, but Chiphell seems to be a very reliable source for news like this one. If it's indeed true, 250mm^2 is pretty small with all these doubled stuff. But still, is an unconfirmed rumour until AMD release the final specs. We are always waiting for something :(

[JaD]

You should focus on SP numbers more than memory bandwidth. R600 architecture (and the rumours say RV770 is a R600 derivative after all) is not limited by bandwidth.

There's no way you can tell actually.

R600 being not bandwidth hungry doesn't mean it will be the same for a chip with twice the SPs/ROPs, even if it shares the same architecture.

In fact, it could easily be the opposite: you need memory bandwidth as long as the GPU is able to compute it. For R600, 70GB/s may be enough; by (theoretically) doubling processing power, you might need to feed way more data in order to take advantage of that power. :)

[OBR]

it is bull***, ATi cant to compete next Nvidias high-end and this is funeral march full of dreams ...

[xlink]

There's no way you can tell actually.

R600 being not bandwidth hungry doesn't mean it will be the same for a chip with twice the SPs/ROPs, even if it shares the same architecture.

In fact, it could easily be the opposite: you need memory bandwidth as long as the GPU is able to compute it. For R600, 70GB/s may be enough; by (theoretically) doubling processing power, you might need to feed way more data in order to take advantage of that power. :)

they said gDDR5...
think 160GB/s on a 256bit bus.
that's plenty.

[saaya]

they are talking about rd700 and not rv770 i think.
rv770 sounds like rv670 reloaded and this time with everything enabled and tweaked.

[zerazax]

Doubling things doesn't necessarily mean doubling area... after all who knows how much stuff was unlocked in the original design. After all, people thought R600 was too ambitious which is why a lot was cut down as yields were terrible and it was taking too long...

And a 1k+ MHz card with more power would be quite amazing to see...

[v_rr]

it is bull***, ATi cant to compete next Nvidias high-end and this is funeral march full of dreams ...

Your astrological power seem very powerful :eek:

Realy what a constructive way of posting. It´s the denial fase because nvidia seria 9xxx seem nvidia serie 8 with OC.

[Der_KHAN]

Highly doubt AMD would switch back to hardware AA, especially since that isn't where the industry seems to be headed.could you guys explain that for me? what does Hardware AA mean?

[clonez]

sounds promising, but wouldnt it be better yield wise to increase the shaders and tmus by 50% and clock speeds to 1-1.2 ghz, made possible by a switch to 45 nm?:D

[Syn.]

could you guys explain that for me? what does Hardware AA mean?

I will have bash at it.

In R6XX generation of ATi cards the AA is applied after the image is already rendered. This seems to be done more on the software level then all on hardware level. Which why sometimes you get blurred out text with AA applied on some games.

I for one cant notice any difference so far.

[LowRun]

it is bull***, ATi cant to compete next Nvidias high-end and this is funeral march full of dreams ...

Dude please stop embarrassing yourself with your crappy posts and speculations :rolleyes:

[biohead]

it is bull***, ATi cant to compete next Nvidias high-end and this is funeral march full of dreams ...
SNORE.

Let's wait and see.

[Nuker_]

it is bull***, ATi cant to compete next Nvidias high-end and this is funeral march full of dreams ...


Sounds like an nvidia-fanboy... your sig confirms that. That fact renders your post useless.

[thenrz]

it is bull***, ATi cant to compete next Nvidias high-end and this is funeral march full of dreams ...


Wowww, not even an attempt at quality conversation. OBR, show us some links from your godly realm of all knowledge.

[TheScavenger]

I am so excited! Please let this come out in June!

[Piotrsama]

In R6XX generation of ATi cards the AA is applied after the image is already rendered. This seems to be done more on the software level then all on hardware level. Which why sometimes you get blurred out text with AA applied on some games.

I for one cant notice any difference so far.

IMO:

IF the text gets blurred, then it has more to do with the algorithm used to AA the image than the fact that it's made at a software or hardware level.
If the algorithm is bad, it doesn't matter whether you AA by hard or soft, result will be the same. Speed may vary.

[NotFred]

My next build should be about due in june/july. Really hope this and Nvidias new offerings are surfacing by then... That should be a very interestign and fierce competition.

[trans am]

it is bull***, ATi cant to compete next Nvidias high-end and this is funeral march full of dreams ...

OBR, You are hilarious! Your garbage posts never cease to amaze!

[Bo_Fox]

The core in the picture does not have any heatspreaders on it or that guard thing around the chip that is supposed to keep the heatsink from chipping the er... chip.

So, given its impracticality, I will remain skeptical of this picture. Otherwise, the red color looks better than ever before and there's that tracing line on the PCB that follows the edges of where the heatsink is supposed to lay.

I'd guess that it's a single-core version of RV770, if it's made on the same 55nm process with bigger TMU units and more shaders. So, for dual-core RV770, it would be like 500mm^2, even bigger than an 8800GTX!! 45nm is needed badly to counter Nvidia's upcoming GT200, unless ATI is really really sure that the big dual RV770 will act like an 8800GTX did.

[whocaresbg]

RV770 will be bigger with only 71 mm2 from RV670 and the rumours "50-100%
more SPs,100% more TMUs" look unreal.

480 x 2 FLOPS x 1050 = 1008 GFLOPS but..
320 x 3 FLOPS x 1050 = 1008 GFLOPS

[LordEC911]

The core in the picture does not have any heatspreaders on it or that guard thing around the chip that is supposed to keep the heatsink from chipping the er... chip.

So, given its impracticality, I will remain skeptical of this picture. Otherwise, the red color looks better than ever before and there's that tracing line on the PCB that follows the edges of where the heatsink is supposed to lay.

I'd guess that it's a single-core version of RV770, if it's made on the same 55nm process with bigger TMU units and more shaders. So, for dual-core RV770, it would be like 500mm^2, even bigger than an 8800GTX!! 45nm is needed badly to counter Nvidia's upcoming GT200, unless ATI is really really sure that the big dual RV770 will act like an 8800GTX did.
There is no core in that picture, it is a mock-up...
The picture is of a current 3rd party non-reference design PCB.

RV770 IS a single die.... R700 is 2 dies.

[ownage]

Other sources say ATI also doubled the AA sampling rate which was responsible for the big performance drops when using AA. Nvidia uses 4x, ati cards 2x, and RV770 supposed to have an AA sampling rate of 4x.

I hope its true :)

[LordEC911]

Other sources say ATI also doubled the AA sampling rate which was responsible for the big performance drops when using AA. Nvidia uses 4x, ati cards 2x, and RV770 supposed to have an AA sampling rate of 4x.

I hope its true :)

Link?
I know certain users at B3d have stated that this would make up for sticking with 16ROPs but I haven't heard it actually rumored.
Also about the AA hit, it seems like it was mainly due to the lack of texture power rather than poor AA performance since all benchmarks seem to test AA with 16x AF...

[Jowy Atreides]

it is bull***, ATi cant to compete next Nvidias high-end and this is funeral march full of dreams ...

technically nVidias next high end is the 9800gtx,
or the gx2 if you count that

L o L

[safan80]

so should we expect a picture within the next 2 weeks?

[Swatrecon_]

so should we expect a picture within the next 2 weeks?

sounds reasonable. probably from Fud or Inq, though. Maybe Expreview?

[KoHaN69]

I still hope they recode AA on the current drivers. Using ROPs for AA is one of the biggest performance hits of the 3xx0 series.

I also hope the 4 series doubles the ROPs, and completely changes the way it handles AA. If that's the case, we can expect at least 1.5x the performance of 9800GTX (at least on stock)

[LordEC911]

I still hope they recode AA on the current drivers. Using ROPs for AA is one of the biggest performance hits of the 3xx0 series.

I also hope the 4 series doubles the ROPs, and completely changes the way it handles AA. If that's the case, we can expect at least 1.5x the performance of 9800GTX (at least on stock)

Ehhh....

Link?
I know certain users at B3d have stated that this would make up for sticking with 16ROPs but I haven't heard it actually rumored.
Also about the AA hit, it seems like it was mainly due to the lack of texture power rather than poor AA performance since all benchmarks seem to test AA with 16x AF...

[DeathReborn]

I still hope they recode AA on the current drivers. Using ROPs for AA is one of the biggest performance hits of the 3xx0 series.

I also hope the 4 series doubles the ROPs, and completely changes the way it handles AA. If that's the case, we can expect at least 1.5x the performance of 9800GTX (at least on stock)

The 3xxx series has broken ROP's as far as AA is concerned so recoding the drivers won't solve the problem.

[LordEC911]

The 3xxx series has broken ROP's as far as AA is concerned so recoding the drivers won't solve the problem.

Did you miss all my posts and some other peoples?

ROPs are NOT BROKEN, since AA is done by the shaders. On the same note, AA is NOT BROKEN.
The current theory on the poor performance is due to lack of texturing power.

[GAR]

i hope for the good of the industry we can have a good competition between nVidia and ATi.

[Natalia]

i hope for the good of the industry we can have a good competition between nVidia and ATi.

Agreed. Lack of competition produces things like the 9800's :(

[kemist]

So much FUD flying around here about AA in HD2x and HD3x series.

The reason they say its software is because resolve is now done in shaders instead of completely in the ROPs. Former designs did AA completely in ROPs, however going forward shader AA will be necessary to deal with games that use different shading techniques such as deferred rendering. People have calculated how much shader power the shader resolve uses and it isnt extremely high.

The hit that takes place is mainly due to the R600 design's ROPs, which are much weaker with Z only performance and also, the R600 is much much weaker with TMUs. This is why when you turn on AA and AF at the same time you see a relatively large hit. Though supposedly most of the hit is coming from the weak AF performance. By upping the ROP capability, which is what many are guessing is going to be done, you are essentially upping the ROP count so more ROPs wont be necessary.

Now a few details here or there may be wrong in my post but this is much more accurate than 90% of whats been posted so far.

[Eastcoasthandle]

So much FUD flying around here about AA in HD2x and HD3x series.

The reason they say its software is because resolve is now done in shaders instead of completely in the ROPs. Former designs did AA completely in ROPs, however going forward shader AA will be necessary to deal with games that use different shading techniques such as deferred rendering. People have calculated how much shader power the shader resolve uses and it isnt extremely high.

The hit that takes place is mainly due to the R600 design's ROPs, which are much weaker with Z only performance and also, the R600 is much much weaker with TMUs. This is why when you turn on AA and AF at the same time you see a relatively large hit. Though supposedly most of the hit is coming from the weak AF performance. By upping the ROP capability, which is what many are guessing is going to be done, you are essentially upping the ROP count so more ROPs wont be necessary.

Now a few details here or there may be wrong in my post but this is much more accurate than 90% of whats been posted so far.

Good Post,

I can sum this all up in one word MADDs. Once they return to distinct shaders instead of using MADD we should see a new light in how R700 series performs. We have to wait and see how this develops. Someone correct me if I'm wrong but I don't think ATI claimed using 320 sps, they claimed 320 madds.

[Monkeywoman]

It seems AMD will be going with a bit lower frequencies than previous stories have been speaking off. They will still be higher than with the RV670, but not in the gigahertz range. It's normal for a company to set the initial frequencies high, try to achieve them, but fail and lower them before the final design. Since this is a "new" architecture one shouldn't make too much out of this either. Back in February, figures said that final RV770 would be <50% faster than RV670. Right now, the reference design is somewhere in the range of 40% faster. We're three months from the first unveilment, but the figure is looking quite solid.

source;http://www.nordichardware.com/news,7470.html

[LordEC911]

It seems AMD will be going with a bit lower frequencies than previous stories have been speaking off. They will still be higher than with the RV670, but not in the gigahertz range. It's normal for a company to set the initial frequencies high, try to achieve them, but fail and lower them before the final design. Since this is a "new" architecture one shouldn't make too much out of this either. Back in February, figures said that final RV770 would be <50% faster than RV670. Right now, the reference design is somewhere in the range of 40% faster. We're three months from the first unveilment, but the figure is looking quite solid.

source;http://www.nordichardware.com/news,7470.html

The Q4 45nm GPUs should be fun.

[AliG]

Hey I think you guys are forgetting something. Remember back when the rv670 was launched some r700 info was leaked? Back then they said it would be manufactured on a 45nm process

I'll try to dig it up, because if it is manufactured on a 45nm process, then die size wouldn't be as much of an issue and I can see it being cut down a good 20% to 250mm

[AliG]

of course fuddo said it so take it with a bulldozer's worth of salt:p
http://www.fudzilla.com/index.php?option=com_content&task=view&id=4348&Itemid=1

and here's another interesting thing he said
http://www.fudzilla.com/index.php?option=com_content&task=view&id=4368&Itemid=1

2 tflops of processing power for the r700 (which is the dual rv770 card for those who don't know), if the rv770 die has 640 shaders and the 3870x2 has over 1 tflop of processing power with 640 shaders, then its very possible that this could be true

of course, once again, take with about this much salthttp://www.boatnerd.com/news/newpictures03b/large-salt-pile.jpg

[trinibwoy]

Good Post,

I can sum this all up in one word MADDs. Once they return to distinct shaders instead of using MADD we should see a new light in how R700 series performs. We have to wait and see how this develops. Someone correct me if I'm wrong but I don't think ATI claimed using 320 sps, they claimed 320 madds.

Correcting you since you're wrong :)

R6xx has 320SP's each capable of a single MADD per clock. So yes, it's both 320 SP's and 320 MADDs. What exactly do you mean by "distinct shaders" ? If you're referring to vector vs scalar ALU's then that's an entirely different topic.

[Swatrecon_]

of course, once again, take with about this much salt

that's a lot of salt...

[NH|Delph1]

of course fuddo said it so take it with a bulldozer's worth of salt:p
http://www.fudzilla.com/index.php?option=com_content&task=view&id=4348&Itemid=1

and here's another interesting thing he said
http://www.fudzilla.com/index.php?option=com_content&task=view&id=4368&Itemid=1

2 tflops of processing power for the r700 (which is the dual rv770 card for those who don't know), if the rv770 die has 640 shaders and the 3870x2 has over 1 tflop of processing power with 640 shaders, then its very possible that this could be true

of course, once again, take with about this much salthttp://www.boatnerd.com/news/newpictures03b/large-salt-pile.jpg

First FUD tells everyone the leaked Radeon HD 4000 specs are all wrong. Later he posts stuff confirming them ... :confused: :down:

//Andreas

[AliG]

don't forget that those links I brought up were put out back when the rv670 was debutting, so even if fud did get it directly from someone in ati, its quite possible things have changed since then, look at what happened to the r600

[NH|Delph1]

don't forget that those links I brought up were put out back when the rv670 was debutting, so even if fud did get it directly from someone in ati, its quite possible things have changed since then, look at what happened to the r600

The things is, I know Fuad has great connections at ATI, but that's not the problem. The problem is how he interprets the information he gets.

//Andreas

[BullGod]

The things is, I know Fuad has great connections at ATI, but that's not the problem. The problem is how he interprets the information he gets.

//Andreas

Nah the problem is that on that site there are cited more rumors than in a Starbucks during Monday morning lunch break...

[NH|Delph1]

Nah the problem is that on that site there are cited more rumors than in a Starbucks during Monday morning lunch break...

It wouldn't be so bad if he just published the exact quotes he got. I know for certain that he's got good sources, but as I said above, it's the interpretations that gets out of hand.

//Andreas

[AliG]

The things is, I know Fuad has great connections at ATI, but that's not the problem. The problem is how he interprets the information he gets.

//Andreas

I wouldn't be surprised if that's the case, reading his stuff for the past months has shown he has great connections as a lot of things he says tend to come true to some extent, but it is quite evident he new to the pc news world