Which is why on 55nm it'd be about the same size as RV670... So RV670 doesn't exist, it's all just a dream, right?
AMD: R7xx Speculation
- Thread starter Unknown Soldier
- Start date
- Status
Which is why on 55nm it'd be about the same size as RV670... So RV670 doesn't exist, it's all just a dream, right?
It better be slightly bigger if you ask me, in order to hit higher frequencies than currently expected.
RV610->RV620 = +1 milion of transistors and 22% reduction of die-spaceWhich is why on 55nm it'd be about the same size as RV670... So RV670 doesn't exist, it's all just a dream, right?
using simple proportion, 55nm G94 would be 185mm2. It would be 9mm2 smaller than the smallest 256bit GPU, RV670. I think 55nm G94 is feasible, but die-space reduction wouldn't be as significant, as for ATi's 55nm transition.
See GT200 thread, although this isn't as much of a problem here as far as I can tell obviously:
Let's make this easy - no GPU-Z screenshots for either GT200 or R7xx unless accompanied with other plausible-sounding information, and no discussion that assumes the transistor count or the die size to be a perfectly reliable and obvious variable. A ridiculous amount of the noise in both threads is related to these factors so let's give that a try... please?
http://forums.vr-zone.com/showthread.php?t=269130
First set of claimed benches?
Fake probably, though, didn't even bother to check their original sources
First set of claimed benches?
Fake probably, though, didn't even bother to check their original sources
Freak'n Big Panda
Regular
Fake 4 sure
Why do you think that?? Maybe ATI has done great GPU and it`s the next r300?
We are talking about RV770, which is supposed be around ~250mm².
btw.
It the same site with early 29XT vs 88GTX@ Crysis benches:
http://www.bilgiustam.com/ati-hd2900xt-8800gtxa-karsi-crysis-demo-ve-3dmark06-testleri/
btw.
It the same site with early 29XT vs 88GTX@ Crysis benches:
http://www.bilgiustam.com/ati-hd2900xt-8800gtxa-karsi-crysis-demo-ve-3dmark06-testleri/
jimmyjames123
Regular
If the RV770 was truly a better performer most of the time compared to 9800GX2, then it would be priced much higher than ~$349.
ChronoReverse
Newcomer
Not only are those graphs showing it faster, it's a LOT faster. That's totally unpossible unless all the things speculated in this thread are completely off.
jimmyjames123
Regular
If RV770 can handle AA with significantly less performance hit than R6xx, then performance should be very good for the money.
Err, why?
If TMU's are the limiting factor on HD3870 as I believe well there are twice as many of them. Shouldn't it be twice as fast?
That is why I didn't get why early sources claimed RV770 might be 30-40% faster than RV670. It's doubling the most limiting stat (TMU's), 50% increase in shaders (more if clock domains are true), plus a core clock bump to top all that off.
Actually I just did the math and if the shaders are truly clocked at 1050mhz (which is basically just pure rumor) than you get 1.35 clockX1.5X SP's=>2.025X as much shader power as RV670 as well.
That said, those benches do seem too good to be true.
Last edited by a moderator:
As you could see below that under SM2 Quad SLI and Quad Crossfire are limited when CPU run only @ 3GHz.
http://www.hothardware.com/Articles/NVIDIA_QuadSLI_with_the_ASUS_EN9800GX2/?page=3
But under SM3 triple crossfire takes the lead which mean is NOT limited by CPU.
Last edited by a moderator:
Core clock over 1 GHz is very probable, I'd say certain. Just trust me on this
(btw, it's not just the shaders, but the whole chip. R6xx can't run out of sync like G8x/G9x does, there's an explanation from Jawed a few pages back, look it up if you want)....The catch is, however, whether the chip really has 32 TMUs. Texturing units take up a lot of transistors and if the chip only has over 800 million, they could stay at 16.
- Status
