vertex_shader
Banned
Fruitzilla write the rv610/630 NDA expire only in july 1st, thats too long wait again, i except the same chaos than was in the last 10-12 days before r600 NDA expired, FUD/BS flood and with some reality.
The Official RV630/RV610 Rumours & Speculation Thread
- Thread starter Arun
- Start date
- Status
If that review is correct with its numbers, ATIs 2600 line should be competitive to NVs, but the 2400 will be a huge joke. Why Nv would release a 8400 card is beyond me unless they are trying to add insult to injury in the entry lvl segment.
HD 2400 is 64bit bus. It will cost a very litle price (50-70$)
Also in that review in company of heroes performance goes up with AA. So take de preview with lots of salt.
vertex_shader
Banned
This is what i write about in #661, BS started again with this lame/unprofessional site, more than 2 weeks left
leoneazzurro
Regular
HD2400 XT/Pro should compete with 8400/8300 cards.
HD 2600XT/Pro should compete with 8600 GTS/GT cards
Actually there is no direct competition for 8500 GT, but probably HD 2600 Pro will be positioned between 8600GT and 8500 GT. I think there are too much od SKU's for the low-end market.
There's a market where NV has still no competition here: the 8800 GTS 320 one.
The number of partners presenting a dual card (Sapphire, Gecube, and maybe others) make me wonder if AMD wants to use dual cards in that price range for now, and meanwhile developing an RV560/570 successor.
HD 2600XT/Pro should compete with 8600 GTS/GT cards
Actually there is no direct competition for 8500 GT, but probably HD 2600 Pro will be positioned between 8600GT and 8500 GT. I think there are too much od SKU's for the low-end market.
There's a market where NV has still no competition here: the 8800 GTS 320 one.
The number of partners presenting a dual card (Sapphire, Gecube, and maybe others) make me wonder if AMD wants to use dual cards in that price range for now, and meanwhile developing an RV560/570 successor.
vertex_shader
Banned
Here iy my quess:
dd2400pro/xt against 8300/8400
hd2600pro against 8500gt
hd2600xt gddr3 against 8600gt
hd2600xt gddr4 against 8600gts
Average i think all amd card will be slower (10-20%) than nv cards, but little cheaper, and i except very big frame drop with AA enabled even with the gddr4 version.
2.5 months later, under performing the competition cards, "good" job from amd again in "mainstream", i think the x1600 not was a huge fail like this cards will be.
I have a good name from the rv630 "almost 400 million transistor for nothing" :smile:
The only rumor from rv660/670 is, coming in q3,but its a 4 months old rumor.. and some site suggest that from the new amd roadmap, rv660/670 its coming only in 2008jan at 55nm.
The worst from user aspect is that amd has no card in 165$<->400$ range, and a cutdown card(s) (xl/gt/gto/gto2) with 80nm r600 won't help.
Dual rv630 won't help in this situation, its will be long, high, and has a bad performance/price ratio (gecube have only ddr2 memory), i'm sure no AIB can't sell the dual xt cards for 200$, so this cards are DOA.
AMD loose all the users respect when releasing lowend performance in masintream, NV do the same this time too, but NV is in the better position because they release first, and this time AMD have the chance to kick NV butt, but they miss.
AMD loosing fast they loyal customers because they are bored with the delays, and after the delays the cards can't show any exciting thing, so its a loose-loose situation for AMD, and very bad for the users.
I have no idea why Hector Ruiz still not make any personal consequences and kick some person out from the company.
If AMD want be a mostly OEM oriantated company (like ati was in some years ago), than say bye-bye for dekstop, and stop wasting the users time, thanks
leoneazzurro
Regular
I think you are too harsh, we have yet to see complete tests with release drivers, AA performance should and could indeed be limited by RBE, but this of course depends by the Software used. RV630 is not so texture limited, relatively speaking, as R600.
So, let's wait and see. 2.5 Months is a big delay, but anyway these cards seems to have good penetration with OEMs, seem to have lower power consumption than compettors (good point for HTPC) and if the price is right they could have their chance to be competitive in the mainstream.
vertex_shader
Banned
I'm not harsh just bored with AMD delays and strategy about mainstream cards, what nv started follow..
I wait for proper reviews, but i think everyone can figure out why the gddr3 xt will cost 100 euro..
And as i write already, i not care when its super cheap and have good price/performance when the 8600gts already a bad performer in mainstream.
Users want to buy mainstream cards for 180 euro/200$ what is not 128bit, and not cut-cut-cutdown version of the highend chip, and most of them not want follow nv in the 270-300$ segment where nv try to push the mainstream users.
Why you think AA is RBE limited with rv630, and not shader based like in r600? When AA done with RBE in rv630 than r600 shader based AA was because of silicon problems like some review suggest? when yes, than its a very bad design decision rv630 only have 4 RBE.
I not care about what OEM's do, they not care about performance, price counting only, and yes 65nm rv610/630 its cheap GPU's, power consuption very low and great for HTPC, but this is all pro.
I don't think so. In this case, RV630 is capable of 4x4x800 = 12,8G-Samples/s. Todays mainstream leader, X1950PRO (equipped with 256bit bus,) is capable of 12x2x575 = 13,8G-Samples/s. Thats almost the same number. RV630 with it's 128bit memory bus would be hardly limited by it's single RBE block...
leoneazzurro
Regular
When I say limited by RBE, I mean "in comparison with the competition". G8600 GTS has 8 ROPs@675 MHz and has an advantage in pure pixel fill rate AND sample rate, that in some situation (I never said in all situations) and with AA activated can give an edge over the 2600XT (and without AA, an RV630 can pull on the screen a maximum of 4 pixel per clock instead of 8 with G84 and 12 with RV570). I don't know yet if AA is shader based on the 2600XT, certainly it has more theoretical math power than G84, but in any case not all RBE tasks are handled by the shaders, so having only one RBE with 4 ROPs can be definitely be more limiting than having 8 ROPs, and having 4 Rops doing all the math work for box AA resolve is normally performance-wise better than having 4 Rops and having to do math work in the shaders.
PS: about "what users want": maybe buying 256 bit 180-200$ cards is what an experienced user want, but for many people even a 1950Pro is overkill, and one has to wonder how many 7600 GT nvidia sold per each 7900GS. And we do not have to forget about notebook market being really strong and that will absorb the most part of these 128 bit chips.
Finally, I have to agree, as I think I'm an experienced user with some passion with gaming, that current NV product line is much better as a whole, as it covers every price range from "sublow" to "very high end". And my wondering about if they think to compete with 8800GTS 320M with a dual chip card was not saying this could be a good idea.
Last edited by a moderator:
http://it-review.net/index.php?option=com_content&task=view&id=1419&Itemid=1&limit=1&limitstart=2
what is the main "limit pont" for RV630's performance ? 4 ROPs or 128bits ?
what is the main "limit pont" for RV630's performance ? 4 ROPs or 128bits ?
Which is the bottleneck, ROPs/RBEs or bandwidth? Fairly impossible to tell given
1) only three games benchmarked
2) with wildly different trends and
3) using old drivers on an unreleased card.
You could try extrapolating from R600 (BF2 LOL), which has RBEs and bandwidth in roughly the same proportions. But guessing the bottleneck involves enough variables that you don't need to add a hasty preview with early drivers to the mix. Best to wait for the launch reviews before you start spinning the hamster wheel. That's what I'll be doing.
1) only three games benchmarked
2) with wildly different trends and
3) using old drivers on an unreleased card.
You could try extrapolating from R600 (BF2 LOL), which has RBEs and bandwidth in roughly the same proportions. But guessing the bottleneck involves enough variables that you don't need to add a hasty preview with early drivers to the mix. Best to wait for the launch reviews before you start spinning the hamster wheel. That's what I'll be doing.
And it was a pretty good stab too. I think it would have gained more responses if you had spelled out clearly what the implications were for the 2600Pro/XT.
I might be looking at those numbers wrong, but said implications kind of seem to be pointing in the direction of 'boy, 2600XT better be dirt cheap'.
Some of the implications are as follows: If texturing is the primary bottleneck of R600 performance, 2900XT should give a pretty good account of itself VS G84. Conversely, you better hope that sample rate is not what is holding R600 back, because RV630 will be in for the world of hurt.
leoneazzurro
Regular
To find a limiting factor It's not so easy, as the bottleneck also can vary with resolutions, i.e.
It would be however interesting to have also other games thrown in the mix, maybe something more shader intensive as R6V, CoJ, and so on.
It would be however interesting to have also other games thrown in the mix, maybe something more shader intensive as R6V, CoJ, and so on.
Like all things in life, it depends. The most complex and accurate shader used to compute the lighting on a surface will do you jacksquat good if the art(i`m using the term loosely here) on that surface is ugly as hell, whilst right next to you there's some dude using lightmapping with some mouthwatering artwork and a few hacks and everyone goes:OMG, that`s so beautiful.
I think someone else stated this on these boards, but I find the following to be extremely accurate:math can get you quite far, but you still need high-quality textures/artwork/base data in order to achieve high-quality graphics, and this is not going to change...umm...ever.
I think someone else stated this on these boards, but I find the following to be extremely accurate:math can get you quite far, but you still need high-quality textures/artwork/base data in order to achieve high-quality graphics, and this is not going to change...umm...ever.
- Status
