• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

RV770 specifications revealed

B

benchmark_hunter

n00b
Joined
Mar 4, 2008
Messages
42
gpu.jpg


1.gif


2.gif


3.gif


55nm process by TSMC
480 SP (4D+1D) - 1D consists of 96 SP
32 TMU
800-900MHz core clock
PCB design similar to RV670
Replace current RV670 price point

From here: http://forums.vr-zone.com/showthread.php?t=255294
 
Intresting cards. HD4600 series doesn't look that good, but HD4870 is total yes. Well that means that these cards launch H2/2008..at least those using GDDR5
 
the 4870 looks beastly it has the processing power of a 3870x2 on 1 chip without xfire holding it back it should destroy the 3870X2
 
and the memory bandwith ?

oh looks awfully crappy with 256bit !!!
 
leSLIe said:
and the memory bandwith ?

oh looks awfully crappy with 256bit !!!
Click to expand...

Last I checked, the G92 and RV670 actually decreased their memory bandwidth to 256-bit from 384 (G80) and 512 (R600) respectively and it didn't make much of a dent in their performance. I'm pretty sure it's not going to be a major bottleneck. I'm just wondering whether or not these will support then new F@H GPU client out of the box or at least within a month or two.
 
Biskquik said:
Last I checked, the G92 and RV670 actually decreased their memory bandwidth to 256-bit from 384 (G80) and 512 (R600) respectively and it didn't make much of a dent in their performance. I'm pretty sure it's not going to be a major bottleneck. I'm just wondering whether or not these will support then new F@H GPU client out of the box or at least within a month or two.
Click to expand...

well, lets just hope so
my old Radeon 9800 PRO had a 256bit memory bus :eek:
 
Hm. Thought it was going to be something...better.. I dunno. I guess they are doing another refresh kind like the 3000 series and the G92. We'll wait and see.
 
Well, it's not 'just' a refresh. The difference between R600 and RV670 was a die shrink (and smaller bus). The difference between G80 and G92 was, again, a die shrink (and smaller bus). The difference between Geforce 9 and Geforce 8 was... nothing?

With all the recycling that's going on, going from 320 to 480 'shaders' is quite welcome. Even though it's not 'revolutionary' it's a lot more than we've seen for the past 16 months or so. And if the 4870X2 is two cores in a single package rather than two GPUs connected by a bridge chip, that might improve CF scaling. I'm pleased.
 
leSLIe said:
well, lets just hope so
my old Radeon 9800 PRO had a 256bit memory bus :eek:
Click to expand...


Keep in mind they are moving to GDDR5 here. Your 9800Pro would likely need to be using a 1024-bit bus to get the same amount of bandwidth as the GDDR5 card at 256-bits wide at the same clockspeeds. I liken it to F1 car engines, seemingly small displacement but high efficiency and high rpms (transfers/second) are what make them perform so well. A car with the same displacement at 1/4 the rpms will be eating dust.
 
Also don't forget that z culling has improved since the 9800, so they are more effectively discarding textures that don't need to drawn, thus reducing overdraw.
 
Silent.Sin said:
I liken it to F1 car engines, seemingly small displacement but high efficiency and high rpms (transfers/second) are what make them perform so well. A car with the same displacement at 1/4 the rpms will be eating dust.
Click to expand...

hmm.. interesting analogy
 
Thats a nice price for something 2x as fast as the 3870x2, imo the 3870x2 should really be retailing around $330 to be competitive with nvidias stuff.
 
Would someone please explain to me 4D + 1D.

I am not seeing where ATi is going here, more SPs always = better, But its all about IPC, how much betters are those pipes compared to RV670, thats what I want to know.
 
Simpson5774 said:
Would someone please explain to me 4D + 1D.

I am not seeing where ATi is going here, more SPs always = better, But its all about IPC, how much betters are those pipes compared to RV670, thats what I want to know.
Click to expand...

I actually think that part of the spec listing has a pretty big technical error (aside from the fact that it could be totally bogus). Take a look at the setup of an SP "cluster" as they stand in the R600 and R670 cores: http://www.beyond3d.com/content/reviews/16/8.

They are already in a "4D + 1D" configuration, but those numbers correlate to the kind of work the SPs are capable of, not groups of x number of SPs. I think the more likely scenario is that they've increased the number of these 80ALU clusters from 4 (320SPs) to 6 in the higher end core (480SPs), and they have half of that available in the mid spec core. Technically there would be 4 x 96 + 1 x 96 units in total as that spec sheet lists, but I don't think they're actually divided into sets of 96 (ie- "1D") like that on the core, they are grouped into clusters as stated in that B3D article. Does anyone have an actual die layout picture of an R670 core with the processing areas outlined to confirm that?

Doing things that way should ease driver development because the way these clusters are fed shader instructions would remain mostly unchanged from the current generation; they simply have more SP clusters at their disposal. We should continue to see improvements through better handling of this particular setup, so owners of R6xx based cards won't be left behind when it comes to performance tweaks in the drivers. When those groups of 4+1 SPs can be used effectively they can be quite powerful but it sometimes takes some massaging by the software to get the most from them.
 
Simpson5774 said:
Would someone please explain to me 4D + 1D.

I am not seeing where ATi is going here, more SPs always = better, But its all about IPC, how much betters are those pipes compared to RV670, thats what I want to know.
Click to expand...

Yeah I'm a bit curious if there are any changes that aren't showing up on this spec sheet. All those stream processors, running at a higher clock, have not done anything versus Nvidia's cards using less SPs at a lower clock.

Im hoping there are some other changes we're not seeing.
 
Oh god why 16 ROPS again?
Do they never learn?

Altrough they will have about 25% more power from core clocks.
 
Numbers don't matter...it's the size that counts! G92 has the same number of ROPs but per-clock has 4 times the Z-fillrate and 2 times the AA sample rate of RV670. Fatter ROPs ftw!
 
ooooh ho ho... 4870 gddr5 @ 1 TFLOPS

not just that $299 too ohhh la la hope this card delivers
 
Might be fake, but if it genuine, sounds quite interesting.

Idle power comsumption looks quite good (my PC is 24/7). gflops number is interesting, but only if it results in better graphic performance, and ATI seems to have trouble translating power efficiently into real performance.

Really tired of my rather old GPU from 2006 now (8800GTX) and am hoping someone makes something worth upgrading to, even if I have to go crossflex... There's got to be some motivation to open up and completely upgrade my PC....
 
When's the expected release date? I was about to pull the trigger on a new HD 3870 Toxic, but I can wait if something new is close (I work all summer at a camp, so if it's any time between now and then I can wait to save money now instead of trying to flip on eBay later).
 
RikkAndrsn said:
When's the expected release date? I was about to pull the trigger on a new HD 3870 Toxic, but I can wait if something new is close (I work all summer at a camp, so if it's any time between now and then I can wait to save money now instead of trying to flip on eBay later).
Click to expand...
June/July, but the RV770/R700 will only be 50% faster than current cards so it may not we worth the long wait.
 
E.R said:
June/July, but the RV770/R700 will only be 50% faster than current cards so it may not we worth the long wait.
Click to expand...

I can hold over by playing GTA IV on the 360 until then. I just need a better GPU for CoD4 and World in Conflict on the PC, all my other games don't really have any problems with frame rates.
 
Only interested in hearing if they un-broke their AA engine and if they finally figured out a good way to get use out of most of their shaders, it seems they are still using Vec design.
 
Trimlock said:
Only interested in hearing if they un-broke their AA engine and if they finally figured out a good way to get use out of most of their shaders, it seems they are still using Vec design.
Click to expand...

R700 is not a new architecture, so it's still ultimately based on R600. Hopefully that doesn't mean the AA logic stays the same, otherwise R700 will be awesome on paper and not so good playing games, just like R600 was.
 
Am I on drugs or do those power consumption numbers seem WAY low? ... Iam going to throw hoops if its accurate though :)
 
Michaelius said:
Oh god why 16 ROPS again?
Do they never learn?

Altrough they will have about 25% more power from core clocks.
Click to expand...


They finally increased the texturing capability, that's a big step in the right direction. I'm disappointed we're not hearing anything about different clock domains for the shaders. I think that's what they really need to do to beat or at least stay competitive with Nvidia. Faster speeds for the shaders, without increasing the pipeline stages. It's time they put that Fast14 tech they licensed eons ago to use! :D
 
Trimlock said:
...it seems they are still using Vec design.
Click to expand...

Why does this keep coming up from time to time? It is not vector. Going down to its simplest unit, it's a 5-way superscalar. There are five scalar units, each one can process a different instruction. Vector means the one same instruction for multiple data.
 
Mat3 said:
They finally increased the texturing capability, that's a big step in the right direction. I'm disappointed we're not hearing anything about different clock domains for the shaders. I think that's what they really need to do to beat or at least stay competitive with Nvidia. Faster speeds for the shaders, without increasing the pipeline stages. It's time they put that Fast14 tech they licensed eons ago to use! :D
Click to expand...

dude at 1:1 your still looking at upward of 900MHz...
 
Michaelius said:
Oh god why 16 ROPS again?
Do they never learn?

Altrough they will have about 25% more power from core clocks.
Click to expand...

ROPs don't seem to mean much these days

8800GTX = 24ROP? 9800GTX = 16ROP same shaders, high shader clock, less mem bandwidth and mem size, and somehow it seems to match the 8800GTX =p
 
Silus said:
R700 is not a new architecture, so it's still ultimately based on R600. Hopefully that doesn't mean the AA logic stays the same, otherwise R700 will be awesome on paper and not so good playing games, just like R600 was.
Click to expand...
FWIU, all AA related issues will be resolved with the R700. Hopefully it will be the case when it's released.
 
APOLLO said:
FWIU, all AA related issues will be resolved with the R700. Hopefully it will be the case when it's released.
Click to expand...

lets hope so.

Anyone remember 24X aa? Did anyone ever try that on the HD2900XT? Must have just murdered it. Wen't from a nice little extra to a cruel ironic joke.
 
I think 24AA only worked in CF mode, as it seemed it would be nice I don't think it held any of the extras that wide/narrow tent added it was just plain MSAA.

Maybe this time they won't sacrifice their AA performance with density problems.
 
MrWizard6600 said:
lets hope so.

Anyone remember 24X aa? Did anyone ever try that on the HD2900XT? Must have just murdered it. Wen't from a nice little extra to a cruel ironic joke.
Click to expand...

24x AA was just 8x with a edge detect filter, so its only a little bit slower than 8x:)
 
You must log in or register to reply here.
Back
Top