AMD: R7xx Speculation

Status
Not open for further replies.
1024bit memory interface with 20MBs of eDRAM which apparently does @%# all b/c it isn't tied to the cores or the 64 ROPs... I mean if you are going to make stuff up at least do it properly. I miss the days of well crafted b.s.
 
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.

I don't think so... According that rumors each core will have 32 unified shaders and 8 rops. Maybe each shader have 5 stream processor.... It's look like more reasonable to me. I think that 8 rops, 160sp (32*5) and 8 TMU could fit in a 72 mm^2 die.
Achieving 2 Tp with a similar architecture don't look too difficult: a 1,5 Ghz of shaders frequency give an amount of 1,92 Tp (160 sp*4 cores* 2 flops * 1,5).
 
Markolt, Eternalightwithin, to spell it out directly for you: Given current fabrication technology any GPU rumor which has "eDRAM" in it is complete nonsense with absolutely no basis in reality.
 
Markolt, Eternalightwithin, to spell it out directly for you: Given current fabrication technology any GPU rumor which has "eDRAM" in it is complete nonsense with absolutely no basis in reality.

Yes, offcourse, i know that :D
10 mb edram , like Xenos, required almost 100 milion of transistors and they could help GPU only at low-classic resolution with low livell of AA. For appling 8xAA on 1080p, without using tiling, would need more than 32mb. Too big and too expensive
But what the rumors said about shaders and rops, could be a good supposition.
 
Nothing wrong with shader AA resolve.

Jawed

Indeed, it makes you wonder why AMD did not implement it though and instead implemented something that really knocks AA down performance wise.

To be honest this entire thread has started way way too early, the only way out is for B3D to buy AMD around April when it is worth about $30m and then we can order them to do an entire write up on R700 and then the thread can be

CLOSED.


:D
 
interesting rumor right here

Last week I received word that ATI has taped out their R700 core... That means they are just a few weeks away from having running R700's in the lab. A little insider FYI for you as no one else reported this! ;)

So, I guess we can start calling these Radeon HD 4000s?

At the end of that thread:
This is very bad news!
Latest ATi product launches have taken place 9-12 months after core tape-out. If R700 taped out just recently then launch can be expected in 4Q 2008.
 
9-12 months is ridiculous; it's basically always going to be less than 9 months. Even R600 wasn't much more than that IIRC (although I don't remember the exact dates) and I think RV670 was 5-7 months.
 
9-12 months is ridiculous; it's basically always going to be less than 9 months. Even R600 wasn't much more than that IIRC (although I don't remember the exact dates) and I think RV670 was 5-7 months.

Depends if there is need of respins and the such, RV670 seems to have had a very successful first spin, while R600 had some problems, according to what the rumors said. What I heard is that R600 was taped out at the end of Summer/beginning of Fall 2006, but rumors said it took 2-3 respins before coming out.
 
Rumours? Why rumours? The number of spins is always right on the chip! Also, assuming problems are found and fixed instantly, each respin adds ~2 months. 9-12 months is just ridiculous... :)
 
Also, assuming problems are found and fixed instantly, each respin adds ~2 months. 9-12 months is just ridiculous... :)

For sure AMD not need 3-4 spin to tape out sucesfully a less than 80mm² GPU. (or they need than i'm speechless).

What about when not for respin's the 9-12 months needed, its needed to make acceptable (at highend not 10% faster than r680) multi core products and mature drivers?

BTW. don't forget this
 
Rumours? Why rumours? The number of spins is always right on the chip! Also, assuming problems are found and fixed instantly, each respin adds ~2 months. 9-12 months is just ridiculous... :)

If I recall correctly, the R600 launch revision was A13, that is, at least 2 respins needed. That's 4 months at the minimum. So we are talking about tape out-debugging-respin-debugging-respin-production launch, and that's likely to have taken 6-8 months: What I knew is that R600 was taped out in September at the earlier, so that's perfectly compatible with the launch dates. RV670 seemed to not require respins, so it was a pretty successful execution. So, the bottom line is that if R700 has non big problems, we could see even a 4 months between tape out and launch, but if it needs more respins...
Anyway, IF R700 is a multi-chip solution, and each chip has a smaller number of transistors than RV670, I don't expect to have such enormous problems in execution. And, of course, multi-chip drivers are needed...
 
leoneazzurro: R600 was ready at the end of January 07 - look at the date of R600 at B3D review...

But it was only released in what .. May. That was a bad decision at that time(something I said). Last I heard it was driver problems too.
icon_rolleyes.gif


http://www.tomshardware.com/2007/05/14/r600_finally_dx10_hardware_from_ati/

So what we looking at here .. R7xx tapes out now, 6-9(or if you believe 12 months) then AMD might still delay further for driver issues .. add another 3-5 months .. hmm yep .. 2009 release. 1Q09 .. and dare I say 1H09

US
 
leoneazzurro: R600 was ready at the end of January 07 - look at the date of R600 at B3D review...

August/September->January =5 months, January was probably first A13 revision, that is going from first spin (A11) to A13 took 5 months, then there have should been the "production and launch" part of the process (because you have to go in mass production in the final spin), which usually takes a couple of months or a little more.
But this needed 2 respins .

If R700 is more successful (à la RV670) then ATI could save 4-5 months respect to R600.
 
Why in the world would anyone think it likely that a new gen part would spin closer to a refresh rather than a new gen part? At least if it's a real new gen part and not another R420.
 
Why in the world would anyone think it likely that a new gen part would spin closer to a refresh rather than a new gen part? At least if it's a real new gen part and not another R420.
Because of the supposed size and trannie count.
I thought most of us agreed that this would be a update of the R600 architecture or are we back to thinking this is something completely new?
 
Status
Not open for further replies.
Back
Top