AMD: R8xx Speculation

How soon will Nvidia respond with GT300 to upcoming ATI-RV870 lineup GPUs

  • Within 1 or 2 weeks

    Votes: 1 0.6%
  • Within a month

    Votes: 5 3.2%
  • Within couple months

    Votes: 28 18.1%
  • Very late this year

    Votes: 52 33.5%
  • Not until next year

    Votes: 69 44.5%

  • Total voters
    155
  • Poll closed .
I was talking about RV100 it had dual DACs as RV150 and RV250 had on RV100 was just one TMDS an RV150 they came out with two. RageTheater is neede only for VIVO functions not for TV-out since RV100.
R200 (just like R100) didn't have internal TV-Out hence those cards use Rage Theatre for this (a bit overkill I guess...). Dunno why it's on the pics (or maybe the hw was there but broken).
No idea what that HDCP block is doing there neither :). It may be possible that this really old chip could already do it in fact, in the same way newer radeons could do it - that is with an external crypto key chip which no card ever had except some late r5xx.

What is DC i thought you missed push A for DAC?
I guess DC = Display Controller.

Since R5xx series we had dual DL (dual-link) TMDS all before we had TMDS but just up to 1920x1080 or SL. First 7800GTX came out with that DL support afair.
Yes though some (quite rare) r4xx based cards used external dual-link TMDS encoder to get past that limit.

So by that diagram you send after R200 also didnt have dual DAC but RV100 and RV150 (RV200 :D) had cause you could connect two monitors via D-SUB.
Yeah, I'm not sure why both AMD and Nvidia skipped some features on their high-end hardware (8500/GF3) wrt display controllers which were already present on lower-end chips (radeon ve/GF2MX). Maybe because of parallel development of these chips, and since these features were not deemed absolutely necessary they just didn't make the cut.
 
Last edited by a moderator:
Just thought I'd post this for a lark. A fake 5870/50 "review" I made :p

I used these assumptions, that 5870 would be 1.6 times as fast as 4870 per the rumor, and then that 5850 would be 83% as fast as 5870 (based on the fact 4850 was clocked 83% of 4870). And I randomly used the benchmarks from this review (I used the stock clocked cards benchmarks only).

Also it's not particularly readable, as I've never made a graph before, and just used some online graph maker website.

20090904065937.jpg


If you assume the 399/299 price points, (and pretend these benches resemble reality) 5850 is probably the more interesting card. It's an easy sell next to GTX285, as it's faster, cheaper, and DX11. However it does get some competition from the $200 4890. The main drawback of the 4890 would be that it isn't DX11, and I wonder how many people will trade that off. Of course if it turned out 5870 was 299, then all bets would be off.

Edit, I may be lowballing 5850 too. If you assume it's 1600 shaders, and clocked conservatively at 600 mhz, then it has 41% more shader power than 4890. Yet my benchmarks it only averages ~22% faster (Of course it could be texture limited etc). So there may be upside to both these cards over these assumptions if anything I'd suspect.
 
Last edited by a moderator:
I'm half-expecting this graph to be "leaked" for an exclusive preview any moment now.

Maybe on chinese sites, but anywhere else they will see the fake review text at the top.

Hey rangers do you think you could repost it without the fake text at the top.
Please
;)
 
*Sigh*
1+ Ghz are needed with 1200 SP to achieve 2 teraflop (5850) + 20+% (5870). If there will be 1280, it will be different, but you'll anyway need more than 900 Mhz.

1200 SPs presuppose 20 SIMD clusters which isn't all that likely. 1280 is one case scenario that still fits the 16 SIMD cluster theory and I haven't seen anywhere even a hint that 5870 has to have at any price 20% more arithmetic throughput than 5850.

55 nm chip are representative of a manufacturing process that had not so big problems. Making a big volume of chips on a 40 nm process with good yields (failures also include parts that are not capable to reach a given clock) seemed until now a "bit" harder for both Nvidia and ATI.

There are no reported frequency problems with 40nm just crappy yields.

Of course 900 Mhz will be reachable. Now? Maybe, maybe not.
The distance between RV770 and RV790 could be representative of the improvement in we can expect to reach on a given process (in this case 55nm) in a certain time range. 55 nm seemed to give less problems than what 40 nm is rumored to do.

And?

And yes, RV740 should be compared to Juniper, but, even if we suppose an improvement like the one had with the Rv770->RV790 upgrade, and we consider that chips with the same basic architecture do not normally differ so much in term of clock speed (i.e. RV730-RV770) a 1280 SP chip will be a little off the mark.

Off the mark sure if you compare slippers with tractors.


And, anyway, this is called the speculation thread, so if we cannot speculate about something because we don't know exactly what it is has no reason to exist, can we?

I'm speculating myself; I'm granting you your theory. Can I have mine too not as something that's absolute but as a second possibility? If yes then we just added a bit more democracy to this rather useless debate.
 
Charlie's juniper

The 180mm^2 or so Juniper will basically equal the 4870's performance...

It has the same shader count as 4870 (according to charlie), and I dont see it having slower clocks, then why should it just equal 4870 in perf. There are no dx11 games to test it with, then why wouldn't it match in dx10 games(if not beat 4870)?
 
It has the same shader count as 4870 (according to charlie), and I dont see it having slower clocks, then why should it just equal 4870 in perf. There are no dx11 games to test it with, then why wouldn't it match in dx10 games(if not beat 4870)?

Bandwidth limited?
128bit w/ maybe 5ghz GDDR5 so 80GBps max, most likely it will have slower memory though.
 
1200 SPs presuppose 20 SIMD clusters which isn't all that likely. 1280 is one case scenario that still fits the 16 SIMD cluster theory and I haven't seen anywhere even a hint that 5870 has to have at any price 20% more arithmetic throughput than 5850.

I explained before why 5870 is likely to have 20% or more arithmetic throughput.


There are no reported frequency problems with 40nm just crappy yields.

Yields depend on defective units AND not reaching the desired frequency. There are also no reports of 40 nm parts on the market having very high clocks.


And this could (conditional) mean that to refine a process time is needed, probably more on a process that gives problems initially than another that executed almost flawlessly .

Off the mark sure if you compare slippers with tractors.

Not really, I find the example pertinent - it's comparing different chips , yes, but at the same process node and then we know that with a given architecture normally we have no big clock changes among chips sharing it. Yes, there could be a complete redesign of the architecture that allows higher clocks. Yes, there could be no more "RV770-style" SP. But until I see a very high 40 nm part going out in the market, I think I'll stick with the "lower clocks" guess. :smile:

I'm speculating myself; I'm granting you your theory. Can I have mine too not as something that's absolute but as a second possibility? If yes then we just added a bit more democracy to this rather useless debate.

Yes, everyone has his theory, and none of them is absolute until ATI gives out the official specs. Anyway, if you look at the thread posts, you'll se that I pointed out very explicitely that " I'll bet on xxxx SP" and that "I'm making a guess". So I never intended to pose my POV as absolute, but only to discuss about pluses/minuses of each theory.
 
Last edited by a moderator:
Same shader count doesn't necessarily mean same performance, architectural improvements/tweaks can go a long way.
 
Here's what MonicaAMD had to say.

Bring a camera, Flip video camera, whatever you like. The whole joint will have wifi, so you can post pics and vids live from the event.

:D
 
Bandwidth limited?
128bit w/ maybe 5ghz GDDR5 so 80GBps max, most likely it will have slower memory though.
Taken at face value, Juniper with 80GB/s has 56% more bandwidth than HD4770 (51.2GB/s). I would hope that Juniper's performance is close to that. Say 45% faster. Assuming HD4770 is 90% of HD4850's performance, that would make Juniper about 30% faster than HD4850, which is pretty much the same as HD4870 which is also around 30% faster than HD4850.

But 80GB/s does seem like a stretch. Though I think it's worth noting that HD4770 is less bandwidth sensitive than I was expecting - or if you prefer it isn't as heavily bandwidth-constrained as it seems HD4850 is. Almost as if the architecture dislikes splitting its work across MCs, i.e. the less the better (which could imply a general issue, perhaps hinting why a ring bus was chosen way back?)... Or maybe there's something in RV740's design that is just better at dealing with heavy bandwidth consumers?

Jawed
 
keritto: DC = display controller. You need as many display controllers as independent outputs. It's possible, that development of R200 began earlier than development of RV200 or RV100. Maybe it was a design decision - videophiles bought cheaper products, not the most expensive parts... Anyway, both reference boards (R8500/R8500LE) had the secondary DAC present. The last possibility is bug - R600 was said to be the most buggy design since the R200. So R200 is possibly even more buggy.
 
Could we somehow agree to stop the talk about DACs and R100/R200 generation of video cards from ATi? Or at least stick it in another thread where it may have more relevance?
 
I think it's realted to the R8xx, too. The R8xx prototype boards are equipped by 2x DVI, 1x DP and 1x HDMI. Notice, that TV-out is not present. It means, that only 2 connectors are utilizable for analog output (DVI via adaptor). It could signify, that the GPU (thought Triple-Head capable) could containt 2 DACs and only the number of display controllers and TMDS controllers was increased.
 
Bandwidth limited?
128bit w/ maybe 5ghz GDDR5 so 80GBps max, most likely it will have slower memory though.

IIRC HD4870 wasn't exactly bandwidth starving, so there's "spare" membandwidth when you go down before you start doing seriously bad things to your performance?
 
Question to all:

Do you think there could be a simultaneous launch of both a 5850X2 and a 5870X2 this time?

Is there going to exist a 5850X2 in the first place? Will it be a sinlge vendor exclusive deal like Sapprire's 4850X2 or will we see versions of it from most manufacturers?
 
Back
Top