AMD: R7xx Speculation

[turtle]

Who says each core only has 4 ROPs? I know 4x RV630 is a pretty popular theory around here, but is ATi really this stupid? To still be using a total of 16 ROPs after all these years?

Not me...I'm of the 160sp (with shader clocks ala g80) with 8rops/8tmus per core persuasion (as it's as close as I could see happening with Fudo's rumors of speed and die-size meshing.)

I think we should expect as much. With the speed per core suggested by Fudo to be ~500gflops, same as R600/RV670, but the die size of 72mm, assumingly on 45nm, I would expect at least 8 ROPs and 160 shaders.

If they keep 4 ROPs/TMUs per 120/160, and/or don't separate the core/shader clocks, I hope they plan on having 1.5-1.6ghz core speeds...Otherwise I would venture it to be another bottlenecked design; one that should have been sorted out a generation ago.

 

[ninelven]

Who says each core only has 4 ROPs? I know 4x RV630 is a pretty popular theory around here, but is ATi really this stupid? To still be using a total of 16 ROPs after all these years?

I didn't mention RV630... dunno where that came from.

As for 4ROPs per chip, I guess it is just a question of how much fillrate do you really need? I have a hard time believing even an 8 ROP chip would make much use of that much bandwidth (60GB/s should be more than enough and he is suggesting twice that or more).

 

[ShaidarHaran]

I didn't mention RV630... dunno where that came from.

It has been mentioned here before, by Geo, IIRC. Seems to be a relatively popular theory too.

As for 4ROPs per chip, I guess it is just a question of how much fillrate do you really need? I have a hard time believing even an 8 ROP chip would make much use of that much bandwidth (60GB/s should be more than enough and he is suggesting twice that or more).

Well, if they add hardware shader resolve back to the ROPs and up the sample rate to 8x per cycle, then I could see the need for a decent amount of bandwidth for even an 8 ROP chip. 4 ROPs don't need much bandwidth no matter how you slice it though.

 

[ninelven]

It has been mentioned here before, by Geo, IIRC. Seems to be a relatively popular theory too.

I was just saying that 4xRV630 is not what I was implying, nor is it what I believe to be the case.

 

[turtle]

I didn't mention RV630... dunno where that came from.

As for 4ROPs per chip, I guess it is just a question of how much fillrate do you really need? I have a hard time believing even an 8 ROP chip would make much use of that much bandwidth (60GB/s should be more than enough and he is suggesting twice that or more).

I have a hard time 8 ROPs would either, but this is ATi we're talking about here. Let's be realistic. I just can't see them making a 64 ROP (or 16x4) design, even if it needs it for balance.

It's possible 8ROPs/60+gbps could work though, assuming they get AA back on the memory and out of the shaders.

While it wouldn't make sense with R600's current design, it may make more sense when AA is tweaked.

Well, if they add hardware shader resolve back to the ROPs and up the sample rate to 8x per cycle, then I could see the need for a decent amount of bandwidth for even an 8 ROP chip. 4 ROPs don't need much bandwidth no matter how you slice it though.

Well said.

 

[Kaotik]

Kyle from HardOCP joins Fudzillas line - R700 coming in 2008 as planned:
http://www.hardocp.com/news.html?news=Mjk3MTcsLCxoZW50aHVzaWFzdCwsLDE=

Sometimes people are just full of BS. AMD is right on track with R700. What AMD is doing now that some “journalists” don’t understand is that AMD is holding their cards a bit closer to the vest than is normally the case. Don’t assume since AMD is not commenting on future products that they do not even exist...

...If you don’t see R700 this next year, I will paint the wheels on my truck pink, drive around Texas wearing a tutu, and take pictures for all of you to see.

 

[Arun]

Kyle in a pink truck wearing a tutu? Gosh, this would just be too ironic, now I'm going to have to hope for R700 to get significantly behind schedule in the next few months! j/k

 

[ShaidarHaran]

Kyle in a pink truck wearing a tutu? Gosh, this would just be too ironic, now I'm going to have to hope for R700 to get significantly behind schedule in the next few months! j/k

I don't know what I'd rather see - R700 shipping in volume next year and competitive, or that pompous assbag wearing a tutu.

 

[3dilettante]

I wouldn't trust my e-masculinity to AMD's product execution.

Let's not forget DAAMIT can be rescrambled to form ATAMID, or Abashed Tutu Attired Man In Drag.

 

[AlexV]

Kyle in a pink truck wearing a tutu? Gosh, this would just be too ironic, now I'm going to have to hope for R700 to get significantly behind schedule in the next few months! j/k

Kyle is simply pissed at nV for releasing Tri SLi(for whatever reason), so ge's getting back at em' in any way he can. Plus he kindof seems to like AMD(remember his review that said there was no reason to get a Core 2 Duo, and that Athlon X2s were just as good-the one that used high-res gaming to prove his point).

ATM it's unclear when the R700 is slated for release(yes, I suppose you know, but you're not goint to tell us, are you? )...on one hand you have Rivas' presentation(which could be wrong)+noises WRT shuffling of what the R700 actually is suggesting that it's a 2009 part, on the other you have the championship winning team of Fuad&Kyle, who use their "rep" to warrant that it's a 2008 part, and then you have the "No comment" attitude from AMD/ATi PR staff(which is logical).

As a mildly joking sidenote, wasn't Phenom also aimed for an early/summer 2007 release?In official statements, not "2 minute phonecall answers for guru-journalists"?That kindof turned out differently. Just like the RV670 coming back from the fab in such a great state, without needing respins. So how things actually turn out may be wholly different from what either common sense or industry noises suggest at such an early point, IMHO.

 

[turtle]

Kyle is simply pissed at nV for releasing Tri SLi(for whatever reason), so ge's getting back at em' in any way he can. Plus he kindof seems to like AMD(remember his review that said there was no reason to get a Core 2 Duo, and that Athlon X2s were just as good-the one that used high-res gaming to prove his point).

ATM it's unclear when the R700 is slated for release(yes, I suppose you know, but you're not goint to tell us, are you? )...on one hand you have Rivas' presentation(which could be wrong)+noises WRT shuffling of what the R700 actually is suggesting that it's a 2009 part, on the other you have the championship winning team of Fuad&Kyle, who use their "rep" to warrant that it's a 2008 part, and then you have the "No comment" attitude from AMD/ATi PR staff(which is logical).

As a mildly joking sidenote, wasn't Phenom also aimed for an early/summer 2007 release?In official statements, not "2 minute phonecall answers for guru-journalists"?That kindof turned out differently. Just like the RV670 coming back from the fab in such a great state, without needing respins. So how things actually turn out may be wholly different from what either common sense or industry noises suggest at such an early point, IMHO.

Point taken on sources, but 'planned' release seems to be 2008, after the 'planned' platform refresh and before the next one in 2009. The slide could be taken that way, or at the same time as the 2009 platform launch. Kyle cleared that up with the phone call. Last I heard, Kris over at DT was trying to clear things up as well, but didn't have an answer yet. I imagine AMD was trying to CTA by saying "by the 2009 platform launch" without releasing a target date/quarter/half...Which makes sense given the blunders with R600 and Phenom. I guess they told Kyle "sometime 2008" because they knew how (to be polite) outspoken and shrewd he is. If they hadn't commented you'd probably see that article but in a very different context...

Kyle also mentioned in the forums for that article that AA will be 'fixed in r700.' What that means, I don't know. I assumed he meant they are going back to the old way of AA; more bandwidth dependent and less shader dependent.

 

[Jawed]

Kyle also mentioned in the forums for that article that AA will be 'fixed in r700.' What that means, I don't know. I assumed he meant they are going back to the old way of AA; more bandwidth dependent and less shader dependent.

Nothing wrong with shader AA resolve.

Jawed

 

[AlexV]

Point taken on sources, but 'planned' release seems to be 2008, after the 'planned' platform refresh and before the next one in 2009. The slide could be taken that way, or at the same time as the 2009 platform launch. Kyle cleared that up with the phone call. Last I heard, Kris over at DT was trying to clear things up as well, but didn't have an answer yet. I imagine AMD was trying to CTA by saying "by the 2009 platform launch" without releasing a target date/quarter/half...Which makes sense given the blunders with R600 and Phenom. I guess they told Kyle "sometime 2008" because they knew how (to be polite) outspoken and shrewd he is. If they hadn't commented you'd probably see that article but in a very different context...

Kyle also mentioned in the forums for that article that AA will be 'fixed in r700.' What that means, I don't know. I assumed he meant they are going back to the old way of AA; more bandwidth dependent and less shader dependent.

As, I said, it's up in the air. They're probably aiming for 2008, but that produces about 0 certainty WRT what'll actually happen, based on the historical data we have available to us. It may come out early, late, precisely on time...we have no point of refrence as it's unclear what on time would mean and thus what would be earlier/later than that. Kyle's certainty is bogus...he could've called anyone from Hector to Eric Demers and back(BTW, Eric is about 10^10! more credible, don't get me wrong there) and still that wouldn't give certainty as they're uncertain when they'll actually release themselves(what, no vid-card has ever been late-anybody remember R600?X1800?GFFX?).

I guess I have some issues with his "saviour of the tech community" attitude and his implied ragging on TR and Xbit, based solely on his phone-call...which is presented in a light that makes Kyle Bennet look like the no.1 speed-dial setting for every CEO and CTO and Chief Scientist and Master Chief in Silicon Valley and beyond. Which I doubt he is. If he were to say:"there's strong indication that the slide is being read wrong, it being quite unclear to begin with, and AMD seems to be targetting a 2008 release, as they kindly relayed to me" it would've been OK.

 

[Rebel44]

I don't know what I'd rather see - R700 shipping in volume next year and competitive, or that pompous assbag wearing a tutu.

Will you bet against him ???

 

[vertex_shader]

and then you have the "No comment" attitude from AMD/ATi PR staff(which is logical).

I don't think so its logical, if customers think R700 delayed its not good for AMD, and they made this mistake at presentation where roadmaps showed to the journalist, whats the reason not show any GPU roadmap and clear the things about r700?

Silence never good in cases like this, but AMD (and ATi before) never was a great company in marketing side so its not a surprise they not do anything.
For r600 they can talk officialy before its released and turn out as total BS, if they learn about that than silence mean what is visible on the roadmap, no r700 in 2008, and no more officialy lie.

I'm sure R700 won't coming out when its needed for competation.

 

[AlexV]

I don't think so its logical, if customers think R700 delayed its not good for AMD, and they made this mistake at presentation where roadmaps showed to the journalist, whats the reason not show any GPU roadmap and clear the things about r700?

Silence never good in cases like this, but AMD (and ATi before) never was a great company in marketing side so its not a surprise they not do anything.
From r600 they can talk officialy before its released and turn out as total BS, if they learn about that than silence mean what is visible on the roadmap, no r700 in 2008.

There's a change WRT how much yapping goes on from both players, be it nV or DAAMIT. They're far tighter when it comes to spilling the beans. IMHO it's better than the incredible ammount of noise that was usually apparent before(the age of about a billion possible pipeline configurations for the 6800 one week before official release and so on).

What would you have them say at this point, beyond:IT'LL BE AWESOME?They're unlikely to have first silicon back and they're probably running on simulations. It's far better to have the public know that there's a R700, let it imagine stuff and have little dicked men like Kyle stir up shit in the puddle than being very verbose yourself at this point. Just like sometimes, for a woman, a little bit of mystery and incomplete exposure of skin might get far better final results than going topless from the first 3 seconds of a date. Especially if you don't really know how your tits will look after their recent surgery nor do you know what kind of tits your prospective partner has seen during his lifetime(translation:R700 is hardly a precisely known quantity back in ATi's labs, but is more like a very exact set of specifications and goals, and what nV has as an equivalent is even less of a known quantity ).

 

[ShaidarHaran]

Will you bet against him ???

You don't want to see my hairy ass in a tutu.

 

[Eternalightwithin]

what about ...

what do you think about the german site, pcgameshardware.de, R700 ?leak? speculation?

Here's is excerpt.

Das Spitzenmodell soll mit 32 ROPs, 128 Unified Shadern, 512 Bit Speicherbus und ein bis zwei GiByte GDDR4-Speicher ausgestattet sein. DX10.1-Untersützung und ein "digital power supply" (vermutlich eher eine digitale Spannungsregelung, wie AMD sie seit der X1950 Pro verwendet) sollen auch an Bord sein.

Besonders interessant: es wird auch ein R750 erwähnt und laut den dort gefundenen Angaben dürfte AMD Multi-GPU-Technik verstärkt einsetzen. Im Folgenden die im Forum genannte Auflistung:

- R750XT: 8 Kerne, 64 ROPs 256 Unified Shader, 1024 Bit, 20 MiB eRAM, GDDR5 0.5ns 1GB/2GB
- R750GT:6 Kerne, 48 ROPs 192 Unified Shader, 768 Bit, 20 MiB eRAM, GDDR5 0.5ns 768 MiB / 1,5 GiB
- R700XT: 4 Kerne,32 ROPs 128 Inified Shader, 512 Bit,10 MiB eRAM, GDDR4 0.7ns 512 MiB / 1 GiB
- R700GT:2 Kerne, 16 ROPs, 64Unified Shader, 256 Bit,10 MiB eRAM, GDDR4 0.7ns 256 MiB / 512 MiB
- R700LE:1 Kern, 8 ROPs 32 Unified Shader, 128 Bit , GDDR3 1.2ns 256 MiB / 512 MiB

Die GPUs sollen bei TSMC in 45 Nanometer gefertigt werden und HDMI 1.3, sowie UVD und Displayport unterstützen. Die Handelsbezeichnung wird mit HD4000-Serie angegeben.

here's the link to..sry couldn't find them in english...

http://www.pcgameshardware.de/?article_id=617895

 

[ninelven]

Nonsense...

 

[Shtal]

Nonsense...

If R600/RV670 has 320 Unified Shaders and R700 will have 128 Unified Shaders, it's looks like it doesn't follow R6xx road path, but it's more like Nvidia.