AMD: R7xx Speculation

[Sound_Card]

IMHO only;

the RV635 has 120mm² with 4 TMUs and 120 ALUs; the RV670 has 192mm² with 16 TMUs and 320 ALUs. If you make a simple excel chart then you can see that a RV6xx with no TMUs and no ALUs at all would have ~90mm². Therefore the RV635 needs only 30mm² for 4 TMUs and 120 ALUs. If we devide this then 4 TMUs need roughly 15 mm² and 120 ALUs again 15 mm².
So a hypothetical RV770 with 32 TMUs and 640 ALUs would need 32/4 x 15 mm² (=120 mm²) + 640/120 x 15 mm² (=80 mm²) die area for the functional units.
So 90 mm² + 200 mm² = 290 mm² for a RV770 with 32TMUs and 640 ALUs.

A RV670 would have ~190mm² using this math.

I know very well that this small calculation is bonkers. But it shows very well that the hypothetical specs are not out of range.

If the RV6xx with no functionality would have 95mm² instead of 90mm² then a RV770 with 32 TMUs and 640 ALUs would have a dia-area of ~266 mm² (but a RV670 would have only 180 mm² using this math).

RV670 I believe also has other logic over RV635/630(or am I thinking of RV620?) and you forgot to mention that RV670 has 4x the amount RBE's. So maybe perhaps their is some more space than I thought. Although I don't picture ATi spending all that space on just shaders and texture units.

 

[BrynS]

DigiTimes is reporting that RV770 will show up by the end of Q2 2008 and that AMD is expecting it's share of the discrete market to jump from 35% to 50% as a result. However, as TR notes, the recent JPR numbers show AMD's share at the end of Q4 2007 (calendar year) at 29% for discrete, although perhaps it has increased in the interim; saying that the 9600GT must be impacting 3800 sales, but when exactly will that impact be counted, i.e Q1/Q2 share?

 

[Razor1]

50% is a pretty unrealistic expectation if they don't have one of the higher selling model lines in Dell or HP.

 

[BrynS]

Yeah, it seems like a tough ask, even if RV770 becomes very compelling until GT100/200. In the high end it often appears as if ATI have struggled more to gain/reclaim share than Nvidia for a bunch of reasons, including NV's stronger marketing. IIRC, it took ages for ATI to gain dominant share in the R3xx days and at least anecdotally, NV seemed to win round more consumers to the 7900 series, for example, when R580 was clearly a much better enthusiast proposition.

 

[Kaotik]

Well, assuming that the rumors from chiphell saying 2x SPs, 2x TMUs, higher clocks, faster AA and what not compared to RV670 are true, they do have a monster of a chip in their hands while nVidia (it seems) has only G92-based products. This could lead to G80-like domination at least until the release of GT200 or whatever it's called, especially if they release HD4870X2 at the same time, which could raise their marketshare a LOT in quite short time, with all the back-to-school sales etc without nVidias answer.

 

[Razor1]

Verification for OEM's will happen May and June for back to school, the card has to be ready in some form now for that to happen. Thats not very likely.

 

[Kaotik]

Verification for OEM's will happen May and June for back to school, the card has to be ready in some form now for that to happen. Thats not very likely.

Well if they're out by the end of Q2 like Digitimes now said (too), there HAS TO be cards ready to show for OEMs in may/june for that.

 

[v_rr]

VR-Zone has learned that Nvidia is preparing to shrink G92 process technology from 65nm to 55nm to better compete against the upcoming RV770. DigiTimes today reported that RV770 will come along by end Q2 which is about the same timeframe as the 55nm G92. Most probably, you will see RV770 based mobile parts first as notebook makers have already received RV770 samples for testing. Right now, we are able to confirm some 55nm G92 mobile parts coming up to pit against the 55nm RV770. Most likely, there are 55nm G92 for desktop parts too as Nvidia in desperate needs to lower cost and improve yield in the next couple of months ahead to stay in competition. In any case the 55nm G92 even with higher clocks can't match up against RV770, Nvidia will have to push ahead the schedule of next generation GT200.

 

[BrynS]

Interesting, particularly about the mobile variants. Explains why there hasn't been a mobile RV670 yet and ties in with the recent AMD slide that mentioned a new high-end mobility part towards the end of Q2.

 

[Kaotik]

VR-Zone has learned that Nvidia is preparing to shrink G92 process technology from 65nm to 55nm to better compete against the upcoming RV770. DigiTimes today reported that RV770 will come along by end Q2 which is about the same timeframe as the 55nm G92. Most probably, you will see RV770 based mobile parts first as notebook makers have already received RV770 samples for testing. Right now, we are able to confirm some 55nm G92 mobile parts coming up to pit against the 55nm RV770. Most likely, there are 55nm G92 for desktop parts too as Nvidia in desperate needs to lower cost and improve yield in the next couple of months ahead to stay in competition. In any case the 55nm G92 even with higher clocks can't match up against RV770, Nvidia will have to push ahead the schedule of next generation GT200.

CJ pretty much said it here: http://forum.beyond3d.com/showpost.php?p=1133180&postcount=88

A nice little birdy told me that G92 is still gonna be around for a while longer... *cough* G92b *cough*.

/runs away and hides

That was nearly 2 weeks ago

 

[Arty]

VR-Zone has learned that Nvidia is preparing to shrink G92 process technology from 65nm to 55nm to better compete against the upcoming RV770. DigiTimes today reported that RV770 will come along by end Q2 which is about the same timeframe as the 55nm G92. Most probably, you will see RV770 based mobile parts first as notebook makers have already received RV770 samples for testing. Right now, we are able to confirm some 55nm G92 mobile parts coming up to pit against the 55nm RV770. Most likely, there are 55nm G92 for desktop parts too as Nvidia in desperate needs to lower cost and improve yield in the next couple of months ahead to stay in competition. In any case the 55nm G92 even with higher clocks can't match up against RV770, Nvidia will have to push ahead the schedule of next generation GT200.

55nm G92?

 

[Arun]

Remember that 50% presumably includes notebooks - and also likely assumes RV770 & R700 will both be taking back-to-school design wins. Either way, if AMD really are responsible for that claim, I'm going to offend a bunch of people here but they should just shut up and stop trying to predict the market. This is neither the time nor the place to make such ridiculous claims, especially when everything I'm seeing points at disappointing numbers for Q1.

AMD's credibility in terms of accurately predicting market dynamics right now is zero - and when I say zero, I really mean minus infinity. They got everything wrong in every single one of their businesses, both CPUs and ex-ATI, for the last two straight years. I'm really giving them the benefit of the doubt when it comes to discrete notebook GPU market share right now (with their claim of achieving ~50%+ unit share), as I see very good reason for that to be correct. If that doesn't materialize though, I'd be forced to no longer even take their credible predictions seriously again...

Either way, I'll STFU now (and encourage everyone else to stop discussing both this and the 55nm G92 here) because there's no point diminishing the importance of an hopefully very well engineered product (RV770/R700) because of AMD's management inability to grasp reality.

 

[INKster]

CJ pretty much said it here: http://forum.beyond3d.com/showpost.php?p=1133180&postcount=88

That was nearly 2 weeks ago

Well, he hinted at it, but i said it.

 

[pharma]

AMD's credibility in terms of accurately predicting market dynamics right now is zero - and when I say zero, I really mean minus infinity. They got everything wrong in every single one of their businesses, both CPUs and ex-ATI, for the last two straight years.

I don't think you've offended many here. In fact it's refreshing to see someone blurb the truth concerning the past & current state of AMD/ATI affairs . I'm beginning to wonder whether things will ever change in their favor again ...

Marcate

 

[Rangers]

So am I to understand that RV770 isn't a true next gen chip ala GT200, but judging by it's die size is more like a "fixed" R600? Since according to VR Zone it is still 17% smaller than G92 and only 29% larger than RV670?

If true and the fixes involve mostly more TMU's (along with debugged rops) that is some validation for the R600= TMU limited crowd (me:smile

Normally if you see a new high end part you're looking at near double size.

Looks like AMD may soon be on top again in performance.

 

[INKster]

VR-Zone is reporting that RV770 has a total of 800 (!) stream processors in 5 "arrays", with 160 stream processors in each of them.

 

[Shtal]

VR-Zone is reporting that RV770 has a total of 800 (!) stream processors in 5 "arrays", with 160 stream processors in each of them.

Who cares about 800 stream processors, it's worthless; I only look at actual real 160 streams physical pipeline.

ATI 160 streams vs. Nvidia 128 streams
ATI 16 ROP's vs. Nvidia 24 ROP's
ATI 32 TMU's vs. Nvidia 64 TMU's

I wonder if 16 ROP's and 32 TMU's are not going to be bottle-neck for RV770, but then again if you look at Nvidia G92 with less ROP's and TMU's compare to G80 are just fine.

 

[INKster]

Who cares about 800 stream processors, it's worthless; I only look at actual real 160 streams physical pipeline.

ATI 160 streams vs. Nvidia 128 streams
ATI 16 ROP's vs. Nvidia 24 ROP's
ATI 32 TMU's vs. Nvidia 64 TMU's

I wonder if 16 ROP's and 32 TMU's are not going to be bottle-neck for RV770, but then again if you look at Nvidia G92 with less ROP's and TMU's compare to G80 are just fine.

I don't think you've understood it correctly.

That's (4+1) * 32 * 5 arrays = 800 stream processors.
HD3870/HD3850 is (4+1) * 64 = 320 stream processors.

 

[turtle]

I don't think you've understood it correctly.

That's (4+1) * 32 * 5 arrays = 800 stream processors.
HD3870/HD3850 is (4+1) * 64 = 320 stream processors.

Or to break it down some more:

3870= 4 arrays of 16 shaders totaling 64 shaders, each shader with a 4+1 MADDs, or 80 MADDs per array = 320 sp total
4870 =5 arrays of 32 shaders totaling 160 shaders, each with a 4+1 MADDs, or 160 MADDs per array = 800 sp total

With g92 you'd be talking:

128 shaders, each shader having a 2+1 (MADD+MUL) = 384 units.

Of course that's layman's and not taking into account the shader clockspeed is higher (~double) on an nvidia part, and that the MUL is not used in general shading, but rather for special function which (iirc) ATi does with their MADDs. I would say look at g92 as 256 MADDs, but that wouldn't be giving it fair credit for the work the MUL is doing, or the fact it's used in CUDA (and perhaps will be used for physX.)

At any rate, it's a serious boost.

So 5 arrays...Are we thinking 20 ROPs?

 

[fellix]

ROP count isn't tied to the MIMD width of the ALU array in this case.