So, do we know anything about RV670 yet?

[nicolasb]

Fudzilla says that it will be 55nm and that AMD is claiming it will ship in Q4/07, but that it is lying and it will really ship in Q1/08.

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1881&Itemid=1

Does anyone have any information that might actually be accurate?

 

[vertex_shader]

Fudzilla says that it will be 55nm and that AMD is claiming it will ship in Q4/07, but that it is lying and it will really ship in Q1/08.

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1881&Itemid=1

Does anyone have any information that might actually be accurate?

Its coming again with lower process technology, so add 4 month what fruitzilla write and you got the launch date

I hope this information not real, 2008Q1 very late, all big games coming out late august to december, if they miss the target again with the performance part its will hurt AMD very bad.
Looks like NV 65nm performance part coming in november, the presure is very big on AMD, and the chance is there they miss the "in time" release again, maybe AMD surprise us? (i don't think so)

 

[chavvdarrr]

55nm ?!
Why ?! I don't believe they'll do that...

 

[Davros]

I hope this information not real, 2008Q1 very late, all big games coming out late august to december

Actually I think this is good, they will be able to play the lates batch of games before they release anything and we wont get a situation where they release cards that are incapable of playing them (2400 + 2600)

 

[Anon Lamer]

Actually I think this is good, they will be able to play the lates batch of games before they release anything and we wont get a situation where they release cards that are incapable of playing them (2400 + 2600)

I dont think its smart to waste engineering resources on making another turd that gets hammered by the nvidian counterpart. They should just make a low power 65 nm variant of R600 and make it their budget card like they did with the x800xl series.
All surviving engineering should be concentrated on R700.

 

[vertex_shader]

Actually I think this is good, they will be able to play the lates batch of games before they release anything and we wont get a situation where they release cards that are incapable of playing them (2400 + 2600)

Not good for AMD when NV release the 65nm performance part in november.(i don't think so NV miss the "in time" release, and the card fail performance wise.)
If AMD can't release real performance GPU this november they miss the big money sack(again), new games sell new hardware, and not new hardware sell new games, so users not wait for AMD performance card whats coming in 2008Q1 to play the good games what coming out this year.

Crysis will be the hardware seller game this year, this is what AMD can't miss without any performance part, users not want any cutdown highend cards like x800gto/x1800gto was, imagine a 250$ r600 80nm cutdown card with 170watt power consuption, most of the users in this segment already care about power consuption and not want buy new PSU, so they say no thanks. (me too)

One more problem can be with rv670, we see how hard rv630 fail with 390million tranistor on 65nm, so hopefully rv670 not based on r600, or they cutoff the HD features and stop wasting transistors for this.

 

[Sound_Card]

Stop wasting transistors for HD features?

IMHO, AMD needs to release a refresh part no later than september and a performance part no later than november or no later than G92.

My idea for the refresh part would be R600 on the 65nm(hopefully with drastically reduced power leakage) with increased texture units, a core clock of 900mhz and GDDR4.

My idea for the performance part would be a 65nm chip with a 256bit bus, and a 180 stream processors.

 

[vertex_shader]

Stop wasting transistors for HD features?

IMHO, AMD needs to release a refresh part no later than september and a performance part no later than november or no later than G92.

My idea for the refresh part would be R600 on the 65nm(hopefully with drastically reduced power leakage) with increased texture units, a core clock of 900mhz and GDDR4.

My idea for the performance part would be a 65nm chip with a 256bit bus, and a 180 stream processors.

Yes, like UVD and HD audio, when someone care can buy hd2400/hd2600 card, now a real gamer card needed without wasted transistors :smile:

Last generation rv570 was a 3/4 r580 so 180 looks very low, 240 sounds much better with 16TMU and 12ROP at 65nm without UVD and GDDR4 memory, 512mb GDDR3 memory at 1600-1800mhz more than enough for 256bit.

 

[Rys]

One more problem can be with rv670, we see how hard rv630 fail with 390million tranistor on 65nm, so hopefully rv670 not based on r600, or they cutoff the HD features and stop wasting transistors for this.

While RV630 isn't exactly a beacon of awesomeness in terms of perf/area/transistor, what the heck are you talking about when you say they should cut off the HD features? Do you mean UVD, or something related to 3D rendering?

 

[Rys]

OK, so it seems you mean UVD. Do you know how big that block is on the chip in terms of area? It seems not, because removing it would barely change what they did with the 3D side of things.

RV6xx's performance problems can't honestly be solved in the way you seem to want without a complete architecture change, and that's obviously not going to happen with any 6-family part.

 

[vertex_shader]

OK, so it seems you mean UVD. Do you know how big that block is on the chip in terms of area? It seems not, because removing it would barely change what they did with the 3D side of things.

RV6xx's performance problems can't honestly be solved in the way you seem to want without a complete architecture change, and that's obviously not going to happen with any 6-family part.

When not UVD and integrated audio wasting the the die space in rv630 than what?

When they keep the 16TMU with 12ROP i see no problem when rv670 based on r600, its can be a real killer card for ~200$ (when nv not release much better) without 150watt power consuption :smile:

 

[_xxx_]

I'd guess that UVD is not a huge chunk of the die, so I think that wouldn't change much in terms of transistor budget.

 

[vertex_shader]

I'd guess that UVD is not a huge chunk of the die, so I think that wouldn't change much in terms of transistor budget.

I think i confused myself with die space and transitors :smile:
I try again, rv630 55.57% of the R600 transitor wise, with 25% of the r600 ROP's, and 50% of R600 TMU's, where are the transistors gone? (i checked the GPU's diagrams but i can't find the answer).

 

[_xxx_]

Pretty much every part of the R600 is more complex than the according parts in the R580. DX10 functionality, the double-width ring-bus etc.

Another question, I always thought RV stands for the mid-low end, why is it supposed to be a performance part now?

 

[vertex_shader]

Pretty much every part of the R600 is more complex than the according parts in the R580. DX10 functionality, the double-width ring-bus etc.

Another question, I always thought RV stands for the mid-low end, why is it supposed to be a performance part now?

Rv570 was a performance part when its come out.

 

[Arnold Beckenbauer]

I don't know why, but I have the feel, that RV670 will be a shrinked R600 (+UVD?). This woundn't be new, compare R423/R481 <-> R430: The R430 is a full featured R423 at 110 nm.
But now there is a big hole between RV630 and R600, AMD could launch an official dual-RV630 to close it.

 

[CarstenS]

AMD could launch an official dual-RV630 to close it.

Inofficially, they encourage board partners to do just that and try to put this dual HD2600 XT against a single 8600 GTS (of course out-performing them by a wide margin).

 

[INKster]

Inofficially, they encourage board partners to do just that and try to put this dual HD2600 XT against a single 8600 GTS (of course out-performing them by a wide margin).

That is, if the game is fully supported under Crossfire multi-GPU setups...
And i don't see how any AIB can't do a dual-G84/8600 GTS card too. They did it with much more complex cards (G71, NV45, etc), at least in terms of power circuitry and thermal issues, so why not ?

 

[Tim Murray]

Does anyone have any information that might actually be accurate?

Somebody has to...

 

[cadaveca]

RV670 CANNOT be a R600 shrink. That would be some of the most stupid planning I have ever heard of.

Completely new toolset for a refresh? I thought ATI commited to no refresh by going 80nm in the first place?(90nm refresh should be 80nm, and 65nm refresh should be 55nm.)