R590 - Will we see this beast?

110nm wasn´t great for ATi, but please don´t compare it with 80nm, since they are entirely different half-nodes with entirely different options, from a design- and manufacturing standpoint.

 

you misunderstood me, i was referring to the R580 or R590 fabrication co-existing with R6XX. IE R420 and R360 production or R300 and R200 production, having 2 different cores from different architectures and featuresets being manufactured at the same time, which really doesnt happen.

I know the R600s are a ways off, i dont doubt the possability of a new card or a die shrink depending on how long they have before new products arrive, i just think people are letting their imaginations run rampent. The odds of seeing a fully modified SKU from ATI in the next month or two just dont seem that great to me. And absolutly not with GDDR4. The R580 just launched, on time, less then a month ago, cant we just relax a bit cause i really doubt anything is right around the corner.

 

I predict.

R590 = X1950XTX = late July 06

R600 = X2800XTX = Late January 07

 

Actually, it does happen. The reason for that is also quite simple: Transistor count and production capacity (yields, availability) were all in very different ballparks. R360 production wasn´t halted completely when R420 took over, they´ve continued with those lines, since they´ve also been on 150nm FSG and R420 was on 130nm Low-K. R360 had a fairly long life - when taking into account it was actually their high-end part back then and featuresets didn´t differ that much between R300/R420.

However, when speaking of R580 production and a possible R590, this situation can be very different. I wouldn´t even wager on a possible R590/R600 scenario just yet.

Well, it certainly will be modified in some ways (on a transistor level), but some mentioned modifications also don´t seem very reasonable to me, so i can agree with you here. "Fully modified" sounds like it´s very far from R580, which i think is a fairly long shot. GDDR4 is a possibility (Q2/06).

 

It depends on the 7900GTX

If nVidia can pull off a victory with its G71, then ATI will release R590. Otherwise, don't hold your breath and focus on R600.

 

G71 WILL be a very adequate performer, Nvidia is allocating the proper of amount of resources to see to it. Additionally, Vista will not be commerically available until at least late Q4, So market demands for a R6XX Product is a year away.

Therefore, market dynamics, competitive pressures and product development cycles seem to indicate that ATi is already working on releasing a interim product for early Q3 availability.

What is not clear is what features or manufacturing process will be utilized.

My guess is that Ram specifications and interface types will be solidified when the final G71 specifications are published, since these are realitively flexable in the final process, since the GPU has these features built into its design.

 

I think it's worth bearing in mind that R480 was a December/January release, just over one year ago (back when a launch was followed one month later by availability) and R520 was due to follow in May/June.

So it seems entirely feasible to me that R580 will have 4/5 months in the spotlight before being replaced by R590.

Jawed

 

Please correct if I am wrong.

How 7900GTX will be any powerfull than X1900XTX?, from what I understand that 7900GTX will have 32 coupled TMUs with 32 Pixel Pipelines, but still 16 ROP's , so output will be 16 Pixels to the framebuffer, just like the case in X1900XTX, it has 16 ROP's, so in reality only 16 Pixels are being written to framebuffer.

ATI clearly mentioned you need more Shading power with current games then Texture Units, and from what I heard that Multi Texturing is almost over.

The only thing which will help 7900GTX is more Memory speed and Core speed, I assume if Nvidia wants to give stiff competition, then they have to clock the 7900GTX to around 700/1800, because X1900XTX still has the edge as far Pixel Shader ALUs are concerened.


And can anyone tell me why X1900XTX ROP's aren't double pumped?

 

Well, 6600GT has only 4 ROPs for 8 pipes and it doesn't seem to hurt it.
days of raw fillrate have ended, and ROPs don't do anything for the shading power.

As for the 32 TMUs, if there are so many of them it's because they are still coupled to the pixel pipes, changing that would mean redisigning the architecture to have something similar to ATI R5xx. Much easier to coninue using the updated NV4x design.

 

ROPs usually don't matter much these days, at least if there are enough to somewhat saturate bandwidth.

Surely ATI's claim has nothing to do with their products.
"Multitexturing" in the usual meaning is over, yes. But that doesn't mean texturing performance is not important.

Assuming G71 is more than a G70 die shrink, G71 will have the edge in MULs per clock, while R580 will be able to perform more ADDs per clock.

Maybe because ATI doesn't think the no-AA Z-only case is important enough to justify the costs and changes to the rest of the architecture that would be required.

 

16 ROPs are plenty 700Mhz x 16pixels = 11200000000 pixels/second. 11200000000/1600/1200 = 5833 frames per second at 1600x1200. I think it safe to say that the numbers of ROPs are not the bottleneck.

Multitexturing is not "almost over", the texture use is not dropping it is increasing. The thing is that the use of math is increasing faster - changing the optimal ratio in hardware too.

A 32 pipe GF7900 will have 33% more pixel shading power than the 24 pipe chip per clock, even if the increased number of texture units has zero effect at will still be faster.

The edge might not be so big. Each of NVIDIAs pixel shader processors do more work than ATIs per clock (unless for instance dynamic branching are used, then ATI has a big advantage) so a 48 vs. 32 advantage is not equal to a 50% advantage.

They are, but only when AA is in used.

PS.: I am not sure the 7900 will be a 32 pipe chip yet, a 24 pipe chip is IMHO still very much a possibility.

Edit: Forgot a not

 

Thanks for your time.

While I understand Nvidia have designed their Shader ALUs in such a way that both ALUs in a Shader Unit has the capability to MADD Operations, but still Texture Units aren't decoupled yet. So, don't you think it decreases the performance?, as the only Secondary ALU will be helpfull if there's no Texture calculations. So, logically I think there's still Primary ALU has the main capability to do MADD, while Secondary ALU is Texture or MADD

I am sure this applies to X1900XTX also.

 

http://www.digitimes.com/mobos/a20060214PR206.html

 

Yes, it sure looks like Samsungs GDDR4 will rock. Let's just hope that they start massproduction soon and allocate enough capacity so graphics cards can use it and still be available in volume. Let's also hope that the chips will work with the current R580 core.
0.6ns (~3.2GHz) could give almost double the bandwidth that we have now with GDDR3.

 

Crikey, that's one hell of a lot of bandwidth!

 

I wonder what effect that would have on performance for R5XX. Anyone care to comment.

 

yes...

its still going to be all about the R600 series. They can say the switch will be fast all they want but i will bet bottom dollar that mid and low end parts will continue to use DDR3 into early 07. Concerning the R590, ATI would have to be threatened enough by the G71 to warrent a speed bump and destroy heat and power requirments (although GDDR4 should be a little more kind), and not only that but they would have to see GDDR4 shipping in enough quantities that they can infact secure enough for the parts of the R600 that require them, as Nvidia will for their counterparts.

Time tells all but i wouldnt hold my breath for a "beast". We should or will see/hear news of a taped out R590 and new card very soon if this will infact meet summer demand. I think many again ignore the fact that ATI was conservative on clocks for the R580 for a purpose, the R590 wont help with that much, with or without GDDR4. Its an optical shrink to give better yields, it wont do anything for power, heat, and quite problably core clock ranges compared to current shipping R580s...i cannot say that too many times i guess.

 

X1999 XTX?

Party like it's 1999!

 

Seems like one of the R590 models (if not the one and only) will appear as the X1800 GTO.

http://www.dailytech.com/article.aspx?newsid=804

 

Makes little sense to call it X1800GTO rather than X1900GTO if it's based on the R580 core. And where does Kristopher get X1800 Pro?

If R590 is indeed a mid-range part (R570 code name would have made more sense in this case) and R580 has to hold the high-end fort until October as the source suggests, ATI will be in for a long summer.