Doesn't work.
The Official RV630/RV610 Rumours & Speculation Thread
- Thread starter Arun
- Start date
- Status
Doesn't work.
What I get is: "Forbidden You don't have permission to access /short/images/pch-RV630XT.jpg on this server."Works just fine.
OK, that one works.
Hmm..It starts to look like that R600 has 16 TMU's so RV630 might be:
32-8-8. But the thing is it would be pretty huge chip as R600 is what - around 700M.
They must have crippled it further somehow.
Jawed
Legend
It seems very likely to me.
We already know it's a "huge" chip, it's ~160mm2 on 65nm. If it was on 80nm it would be well over 200mm2, far bigger than G84.But the thing is it would be pretty huge chip as R600 is what - around 700M.
They must have crippled it further somehow.
Similarly, RV610, which is seemingly exactly 50% the size of RV630 (judging from vaguely fuzzy pix), would appear to be destined to be 16-4-4. This is the only ATI GPU that's looking to be significantly smaller than its direct competitor.
Though you could argue that R600 is notably smaller than G80 and NVIO combined.
Jawed
caboosemoose
Regular
Well, if that's the case, at least ATI is back to 8 pixels and textures per clock for the midrange. Going down to four for the X1600 really was a bad idea.
ATI needs to get something right. If it can beat the 8600, then the disappointment of R600 will be a little easier to stomach.
ATI needs to get something right. If it can beat the 8600, then the disappointment of R600 will be a little easier to stomach.
Well it's fine with me. It does make sence as it would leave 48-12-12 for that Laka thing. Wonder how those 8 TMU's hold it back? The pixel pushing power should be pretty impressive, especially if it's clocked > 700mhz. The rumours have been going up to 800.
I agree with both of you to a T.
It leaves 12-12-48 for Laka (RV670?), and perhaps a 256-bit upgrade for RV630 (rv660?) down the road.
I still think RV530 is a terrible thing to hold against ATi. They were in a difficult situation switching architectures during one generation (1->3ps per rop), and I believe RV530 was a very unique situation; a failed experiment if you will to pour the half-step performance into a small die size. Since all of R600's gen (R650 etc) is likely to be based on the same architecture, I think we can expect similar derivatives from what we've seen in the past, ie X600->X800, 9500->9700,...6600gt->6800GT/U, etc...which have historically been 1/2 the high-end part with a 128-bit bus.
It leaves 12-12-48 for Laka (RV670?), and perhaps a 256-bit upgrade for RV630 (rv660?) down the road.
I still think RV530 is a terrible thing to hold against ATi. They were in a difficult situation switching architectures during one generation (1->3ps per rop), and I believe RV530 was a very unique situation; a failed experiment if you will to pour the half-step performance into a small die size. Since all of R600's gen (R650 etc) is likely to be based on the same architecture, I think we can expect similar derivatives from what we've seen in the past, ie X600->X800, 9500->9700,...6600gt->6800GT/U, etc...which have historically been 1/2 the high-end part with a 128-bit bus.
Last edited by a moderator:
caboosemoose
Regular
Yeah - if it's fully 1/2 R600 and clocked at 800MHz you'd have to think it comfortably has the measure of G84.
Well, of course it's seemingly still crippled by the 128-bit bus where 256 would seemingly make more sense...but yeah, I concur.
I also can't wait for Shamino to v-mod that sucker and throw his crazy HSF on it. If theorem, it should overclock well.
Last edited by a moderator:
caboosemoose
Regular
Well, the 8600 has the same 128-bit bus, so that is not an issue in terms of the head to head.
Still sucks though, don't it? Was expecting 256 on the midrange by now.
Still sucks though, don't it? Was expecting 256 on the midrange by now.
True enough, and yes, yes it does.
It also boggles my mind...I imagine adding a 256-bit bus to RV630 wouldn't increase the die-size or total cost past that of RV660, which they sold at a similar price...but it would be much more desirable. Sure it may require more power (which may be the reason as they don't want to add the extra plug/cooling) but unless it somehow would completely destroy yields, or somehow look overly attractive compared to R600 and/or upcoming higher-end parts, it seems like it would've been the right step to take for mid-range dominance.
caboosemoose
Regular
I think you'll find it is board complexity and manufacturing cost that has dictated the use of 128-bit buses on these cards.
Perhaps, but then that doesn't leave much margin for the impending RV670, which seems destined for the ~$250 market.
Obviously not very much if RV610 retains the 128-bit bus of RV630, and half the other specs, with ~54% the die size on the same process (82mm² vs 153mm²).
That's one of the reasons I believe at least part of that has to been done off the main ASIC of RV610 and possibley RV630.
Jawed
Legend
Ooh! Where did you get those die sizes from, that's very much more definite than we've seen before!
It seems likely. Though I was also vaguely under the impression that RV610's UVD wouldn't be capable of the highest bitrates at 1080p.
Jawed
Arun.
Yeah...I'm still confused about what RV610 will be capable of. It's possible it is on-die and doesn't have the ability, or perhaps it lacks that extra chip, although with the cheap RV550 using RV515 they were going to introduce (but alas was vaporware) with 1080p functionality begs the question why it would be left out. That seems the foggiest part of the architecture to me at this point.
Last edited by a moderator:
- Status
