AMD: R7xx Speculation

[hoom]

Appreciate being defined out of the hated 'enthusiast' category with my mere X1900GT with 24" 1920*1200 + 42" 1080P LCDTV waiting at least another week for my X3870 that can talk to the latter at higher than 1280*1024

Or on topic: Inq says R700 is based on R600/RV670, could it be RV670 minus a bunch of extraneous bits that you only need one of per SKU? Enabling say 3* R700 + 1*RV670 for '1280 SPs'?
Or if thats too much, surely 4* 1/2 RV670 is not enough (effectively same as 2*RV670) so maybe 4*3/4 RV670?

 

[ShaidarHaran]

hoom: strange, my X1650 XT does 1920x1080P via DVI->HDMI adapter to my 42" 1080P LCD HDTV.

 

[XMAN26]

Bar charts are irrelevant when when fps are greater than 60-70 fps because the human eye cannot tell the difference.

Now this is the biggest buch of bull I've read in this whole thread.

 

[XMAN26]

Errm, didn't there used to be a Radeon R7 generation thread around here somewhere?

Anyrate, I see that Inq is claiming it taped out recently, and is boldly predicting an easy time of it and a 2Q release. I think that's a little too bold for my stomach at this point, tho I think it likely that R7xx is not nearly so ambitious a change in arch as, say, R5 --> R6. If it really has taped out (and that's the kind of basic fact that Inq is usually more likely to be right about) then certainly a summer release is not an unreasonable target.

Just look at who wrote the article. Charlie, the same guy, who if it was Nvidia and G100 would be saying it'll be 6 months with +4 months per respin.

 

[Geeforcer]

Q3 Mercury numbers were 36% DX10 marketshare. (And NVIDIA's Q2 CC revealed they had 75% in Q2, where we had just begun the ramp of mainstream parts).

Which is fine - as I said, the timeframe was missing. It seems that some people interpreted CJs comment as AMD now having 30% share of the entire DX10 GPU installment base.

 

[itaru]

R700 has taped out
http://www.theinquirer.net/gb/inquirer/news/2007/12/31/r700-taped

 

[3dilettante]

Errm, didn't there used to be a Radeon R7 generation thread around here somewhere?

Anyrate, I see that Inq is claiming it taped out recently, and is boldly predicting an easy time of it and a 2Q release. I think that's a little too bold for my stomach at this point, tho I think it likely that R7xx is not nearly so ambitious a change in arch as, say, R5 --> R6. If it really has taped out (and that's the kind of basic fact that Inq is usually more likely to be right about) then certainly a summer release is not an unreasonable target.

There are possible implications of R680 that impact an early R700 rollout.

How much benefit would AMD get from slapping together a multichip SKU that would be good for 1-2 quarters, depending on where in the Q2 range R700 is ready?

If R700 is really as scalable as some make it out to be, its lower-range incarnations would match R680, and do so without the expense of a bridge chip. (If it's really scalable, what does it mean for the other RV*** cores if it launches that early?)

If AMD had confidence of such an R700 launch that soon, why go through all that effort?

Even if R700 is ready early, R680's existence becomes an obstacle that might prompt AMD to hold off on a release until R680 has had some time to ship and then clear from inventory.

There are of course scenarios where this might not be the case:
1) R680 winds up being a very limited release.
2) R700 isn't a significant jump, such that R680 can still supply the same enthusiast space it targets now (perhaps slightly lower-range enthusiast) without being cannibalized by R700. This implies that R700 is not compelling enough to really drown out R680 (I think this is unlikely).
3) R700 is much better, but AMD can decide to push it to a higher echelon of pricing and limited quantity and keep R680 around longer.

edit:
My speculative list here isn't meant to be an end-all be-all listing.

 

[Jawed]

I don't see how R700 can be much better than R680. The latter is 32 TUs, 32 RBEs, 640 SPs and 512 bits/140GB/s+ to memory in 1.3B transistors. 4x RV630 is 32 TUs, 16 RBEs, 480 SPs and 512 bits to memory in 1.6B transistors.

4x Z per clock in the RBEs and +15% core clock seems to be about the limit of the improvements we're likely to see in R700. There needs to be a hell of a lot of fat that can be trimmed off each chip, if starting with R6xx, in order to make R700 considerably faster than R680. I don't see that happening.

The only thing I can see is that R700 would recover R6xx's dreadful AA performance and otherwise be a near match for R680 performance, while being cheaper for AMD to make.

Jawed

 

[BRiT]

Nexus_alpha, what part of the world do you live in? I ask because I seriously wonder exactly what parts you're looking at to arrive at a $700 GPU and a $1500 CPU. From what I've been reading by others on this site, the following parts are described as borderline enthusiast and come up no where near the $2200 you posted. The Nvidia 8800 GT GPU even when price-gouged is $270. The Intel Q6600 CPU has been hovering around $280 for months now. That's roughly $550.

Even when I bought my GTX brand new 6 months ago, it was $520 and the Q6600 CPU was $380. Thats still $1300 less than the figures you threw out there.

To be on topic...

If the R700 did indeed tape-out, I suspect it's less of a change from the RV670 than what AMD truly needs to claim the performance crown.

 

[nexus_alpha]

Nexus_alpha, what part of the world do you live in? I ask because I seriously wonder exactly what parts you're looking at to arrive at a $700 GPU and a $1500 CPU. From what I've been reading by others on this site, the following parts are described as borderline enthusiast and come up no where near the $2200 you posted. The Nvidia 8800 GT GPU even when price-gouged is $270. The Intel Q6600 CPU has been hovering around $280 for months now. That's roughly $550.

Even when I bought my GTX brand new 6 months ago, it was $520 and the Q6600 CPU was $380. Thats still $1300 less than the figures you threw out there.

To be on topic...

If the R700 did indeed tape-out, I suspect it's less of a change from the RV670 than what AMD truly needs to claim the performance crown.

Sorry was using the very top-of-the-line figure i.e. gtx ultra anyways forget my rants Happy new year to all



Looking forward to the r680 in my machine shortly

 

[nexus_alpha]

There are of course scenarios where this might not be the case:
1) R680 winds up being a very limited release.
2) R700 isn't a significant jump, such that R680 can still supply the same enthusiast space it targets now (perhaps slightly lower-range enthusiast) without being cannibalized by R700. This implies that R700 is not compelling enough to really drown out R680 (I think this is unlikely).
3) R700 is much better, but AMD can decide to push it to a higher echelon of pricing and limited quantity and keep R680 around longer.
.

If you look at crossfire and sli it does not work in all games now if ATI was found some secret sauce for inter-gpu communication that lets it scale for all games then it would not matter what the performance of the r680 is. ..... my 2 cents

 

[Sc4freak]

R700 has taped out
http://www.theinquirer.net/gb/inquirer/news/2007/12/31/r700-taped

Isn't the time between tape-out and release usually a bit more than a couple of months?

 

[Unknown Soldier]

Now this is the biggest buch of bull I've read in this whole thread.

Yep, I agree with you 100%

Hint .. nexus_alpha
http://www.100fps.com/how_many_frames_can_humans_see.htm

US

 

[Unknown Soldier]

Isn't the time between tape-out and release usually a bit more than a couple of months?

Mentioned many times before ... about 3 months(maybe 2) for each spin .. so hopefully looking at only 2 spins .. 6 months(at the most) .. else 9 months(at the most) and hopefully not 12 months(gawd .. hope not). Then there's the time it takes to actually go out to the IHV's(2 months?). Then .. hopefully AMD doesn't decide to hold back it's 'enthusiastic' card to sell it as a 'mainstream' card. We don't need another debacle again.

Something like that(am subject to being incorrect).

US

 

[AlexV]

If you look at crossfire and sli it does not work in all games now if ATI was found some secret sauce for inter-gpu communication that lets it scale for all games then it would not matter what the performance of the r680 is. ..... my 2 cents

They're still pushing AFR through dev-rel and in most of their papers so no new rendering scheme for the R680 as far as I can tell. The R700 wil hopefully be another story.

 

[nexus_alpha]

Yep, I agree with you 100%

Hint .. nexus_alpha
http://www.100fps.com/how_many_frames_can_humans_see.htm

US

Read the article makes some good point.

 

[Geo]

Mentioned many times before ... about 3 months(maybe 2) for each spin .. so hopefully looking at only 2 spins .. 6 months(at the most) .. else 9 months(at the most) and hopefully not 12 months(gawd .. hope not). Then there's the time it takes to actually go out to the IHV's(2 months?). Then .. hopefully AMD doesn't decide to hold back it's 'enthusiastic' card to sell it as a 'mainstream' card. We don't need another debacle again.

Something like that(am subject to being incorrect).

US

I'd say 6-8 weeks for a spin, but it can be longer. Depends on how much needs changing. Inq was pointing at May, which feels a little optimistic to me. But then that also depends on just how ambitious the change is from RV670 to R700. . . "This summer" certainly feels likely if they don't hit some weird R520-like problem.

 

[turtle]

I'd say 6-8 weeks for a spin, but it can be longer. Depends on how much needs changing. Inq was pointing at May, which feels a little optimistic to me. But then that also depends on just how ambitious the change is from RV670 to R700. . . "This summer" certainly feels likely if they don't hit some weird R520-like problem.

It sounds to me as if they are predicting 2 spins, with the revision being quickly produced. Isn't that what most chips are? Rv670 was a miracle with 1, and the disasters 3, correct? Considering this is said to be on the same process (55nm) and perhaps more evolutionary, I think that's a decent guess. It's possible, but I think AMD is covering their ass in case there is 3, or a problem that causes a delay before the 2nd fabbing, which is also realistic.

If they really are planning on showing it at CeBIT (per the R800 blog) then it sounds like they are hoping to have working/production silicon by March, which would have to be the first or perhaps the second if something could be fixed quickly...but that would be cutting it awful close...especially if they only have a couple months to ramp from that point to release. Figuring it taped out early December and release in May; 2 spins plus ramp seems feasible, although slightly optimistic. June/July/Aug does seem to have a better feel to it, and if you look at the roadmap declaring it wouldn't be part of the mid-2008 platform, it looks like AMD is looking ahead and not making any promises it will make that cut either...but it seems like that's when they were aiming for (late Q2 May/June) but may slip to Q3 July/August...depending how much work is needed when the first spin comes back.

I smell mixed signals, uncertainty, and FUD.

I think it's def too early to predict a month until we hear how this first spin turns out. It could be May, June, July, or August per my thinking...Unless, as stated, there is some unforeseen tragedy...but we've had this discussion before back in 2004-2005. It's like R520 all over again...Who knows which way it'll go.

Barring someone forgetting to connect a ground wire (or whatever the laughable problem was with the early R600 spin), I agree a summer release sounds right...as of today.

 

[Geo]

Disaster is a bit strong for 3 on a new gen. And I don't see RV670 as a miracle. Good job by the team, but not a miracle. Getting R600 on the first spin would have been a miracle. Taking 3 for RV670 would have been a definite owie, but I have high standards for "disaster" I guess.

 

[Unknown Soldier]

We can only hope that there won't be a disaster and that they actually do release a proper Enthusiastic card and hoping the mainstream cards do pretty well too(lower prices ).

US