AMD: R7xx Speculation

[Wesker]

Wise words.

Thanks!

Thanks also for the '[high end]' addition. I can't edit posts (yet...?), so my messages are kinda rough around the edges.

But yeah; I don't remember either ATI or NVIDIA being well off by simply marketing their solutions as price/performance mainstream parts.

 

[Freak'n Big Panda]

NV did a pretty good job of selling NV30 (& derivatives) despite their shitty-ness compared to R3XX. Although I think that has more to do with the fact that they were the undisputed performance leader from the GF1 days right up until the release of R300 in 2002. It takes more then one generation of under performing parts to break that sort of reputation.

 

[Wesker]

Indeed. 1.5x is not exactly a significant lead when competition is factored in (performance wise). They need at least 2x to become serious in that playing field. The 512bit bus coupled with GDDR5 is promising though.

Still, if the price is right and there are other features to be offered, its a decent boost in speed overall. I'm not saying it sucks, but if that becomes true when we see final benchmarks, well, I may still wait for GT200.

I hate speculating... and waiting.

Even if this all just pure speculation, I highly doubt ATI would both a 512 bit bus on R700 with GDDR5. The bandwidth would be ridiculously high (~ >200GB/s), not to mention the COST of such a board.

Unless R700 actually makes use of that massive bandwidth, I doubt ATI would go ahead with it. They would at least cut back on either of the two--stick with the 512 bit bus and drop GDDR5 (in favour of GDDR3/4) or vice versa.

As for my personal thoughts on R700: I'm still hoping that R700 (the actual high end chip) ends up being something like the 800 ALU rumour--a traditional monolithic GPU design:
- 800 ALU's
- 32 TMU's
- 24 ROP's (+hardware MSAA resolve?)
- 512 bit bus
- 1024MB of GDDR3/4
- ~200w TDP

Basically, something that R600 should've been.

Anyway, we'll just have to wait and see...

 

[AnarchX]

800 ALUs and 32 TMUs would be a "worser" ALU:Tex-ratio than on R600, which is imo not really appropriate in the near future.

More likely would be that we see 800SPs on two dies, but with 2x32 or 2x20TMUs.


At the moment I would think, we can be happy if RV770 is able to reach G92GTS/GTX in normal usage scenarios.

 

[DJ12]

But yeah; I don't remember either ATI or NVIDIA being well off by simply marketing their solutions as price/performance mainstream parts.

I don't disagree, but for the sake of argument.....

Before the top line has been the aim, maybe a change of approach from ATI/AMD marketing the card along the price v performance will change that.

 

[ShaidarHaran]

800 ALUs and 32 TMUs would be a "worser" ALU:Tex-ratio than on R600, which is imo not really appropriate in the near future.

I agree

More likely would be that we see 800SPs on two dies, but with 2x32 or 2x20TMUs.

That would be nice, but I don't know if either configuration you've listed is truly likely for an R7xx-product.

At the moment I would think, we can be happy if RV770 is able to reach G92GTS/GTX in normal usage scenarios.

That would be a major disappointment. If RV770 "Pro" doesn't surpass all existing single-GPU solutions it is a failure in my opinion. ATi has to know the cause of their performance "problems" and how to fix them. They did it with R300 and again with R580, I think we're about due with RV770.

 

[Mat3]

800 ALUs and 32 TMUs would be a "worser" ALU:Tex-ratio than on R600, which is imo not really appropriate in the near future.

But R600 performance takes a big nose dive every time AA is used, and it goes through the SPs. More SPs could be their idea of fixing that.

 

[AnarchX]

But R600 performance takes a big nose dive every time AA is used, and it goes through the SPs. More SPs could be their idea of fixing that.

Only a minor part of AA is made in SPs and there was a great improvement over time trough driver team:
http://translate.google.ch/translat...nd_afskalierung&langpair=de|en&hl=de&ie=UTF-8
The bigger problem is that R6xx only came with 2Z per ROP, but RV7xx seems to have 4.

That would be a major disappointment. If RV770 "Pro" doesn't surpass all existing single-GPU solutions it is a failure in my opinion. ATi has to know the cause of their performance "problems" and how to fix them. They did it with R300 and again with R580, I think we're about due with RV770.

Since we are probably talking about a GPU still at 55nm and ~250mm², which gives ~30% more transistors and there are some rumors that AMD has shorten RnD budget for ATi, I see no big room for wonders.

 

[Lukfi]

=>AnarchX: As I wrote a few pages back, RBE improvements are quite probable. Generally, ATi wants to improve performance with AA, so I expect some appropriate steps to be taken...

 

[ChronoReverse]

That would be a major disappointment. If RV770 "Pro" doesn't surpass all existing single-GPU solutions it is a failure in my opinion. ATi has to know the cause of their performance "problems" and how to fix them. They did it with R300 and again with R580, I think we're about due with RV770.

If it didn't surpass all existing single-GPU solutions (currently really still a G80) after this much time, it would be time to write off ATI as a competitor for quite a while.

 

[Berek]

Even if this all just pure speculation, I highly doubt ATI would both a 512 bit bus on R700 with GDDR5. The bandwidth would be ridiculously high (~ >200GB/s), not to mention the COST of such a board.

Unless R700 actually makes use of that massive bandwidth, I doubt ATI would go ahead with it. They would at least cut back on either of the two--stick with the 512 bit bus and drop GDDR5 (in favour of GDDR3/4) or vice versa.

As for my personal thoughts on R700: I'm still hoping that R700 (the actual high end chip) ends up being something like the 800 ALU rumour--a traditional monolithic GPU design:
- 800 ALU's
- 32 TMU's
- 24 ROP's (+hardware MSAA resolve?)
- 512 bit bus
- 1024MB of GDDR3/4
- ~200w TDP

Basically, something that R600 should've been.

Anyway, we'll just have to wait and see...

That's true, it would provide a huge amount of bandwidth, but we've seen announcements of great DDR5 yields, so that seems to be coming along smoothly (and thus at a lower cost).

Looks good to me either way. I just like the sound of a 512bit bus, so I'll take two of those please! .

 

[Chris123234]

Looks good to me either way. I just like the sound of a 512bit bus, so I'll take two of those please! .

When I can read a post, see it mentions "I'm going to get so and so card", and instantly know who the poster is before I look, there is something wrong.

I could easily quote 15 instances of this, and in fact I did before my PC bluescreened (go Vista!).

(OT I know...)

 

[Shtal]

We've learned that ATI's new chip might end up 1.5 to 1.6 times faster than the current RV670 solution.

I'm still confuse about the word might end up ~1.5 faster, what do they mean by might ??????

 

[Slyne]

According to some rumours, the GDDR3 (I can't believe Fuad is still writing DDR3 and DDR5 after all this time) version of RV770 is supposed to appear late May, yet we still have no idea what the clocks are going to be, or what the size of the chip REALLY is.

At this point, I believe the partners should have received some chips already. I'm starting to think May will come and go and we will still be hooked on Fudzilla's all-over-the-map prediction for lack of better input. What was GEO's horizon for accuracy of release dates, 6 weeks away or 3 months away? I can't remember.

 

[ShaidarHaran]

Most reliable rumors were that chips would ship to board partners this month and we'd see products next.

 

[Sound_Card]

I don't think they are just going to let them trickle into the market without a decent press event to kick it off. Amazing that we have no word on the date this close to suppose time frame.

 

[ShaidarHaran]

Just because it hasn't been announced (or leaked) yet doesn't mean there won't be a press event

 

[Kowan]

Lets hope this thing blows the bell curve off the map.

 

[CarstenS]

Just because it hasn't been announced (or leaked) yet doesn't mean there won't be a press event

But they should at least tell the press they going to have there - don't they?

 

[OICAspork]

But they should at least tell the press they going to have there - don't they?

o.o can invitations include NDA`s on the contents of the invitation (time/place/etc)? XD Or would they have to send out a statement ahead of time saying that you have to sign an NDA to get an undefined future invitation?

Basically, is it possible that invitations have been discretely sent out already without anyone blabbing about them?