R590 - Will we see this beast?

[kemosabe]

Seems the negative from ATI's point of view is that R520 yields were obviously marginal enough to provide plenty of reject cores for a higher-volume midrange SKU. The positive is that this time they seem to be releasing their 256-bit stop-gap simultaneously with Nvidia's natural midrange 90nm part. Could be more effective this time in preventing the 7600 from sweeping up the mind/market share like the 6600GT did, at least until RV560 is readied on 80nm and with the advantage of knowing what to aim for.

The bigger challenge for ATI will be to deliver RV560 performance at least as good as X1800GTO and to avoid a repeat of the 9500 Pro 9600XT disappointment.

 

[Geo]

Seems the negative from ATI's point of view is that R520 yields were obviously marginal enough to provide plenty of reject cores for a higher-volume midrange SKU. The positive is that this time they seem to be releasing their 256-bit stop-gap simultaneously with Nvidia's natural midrange 90nm part. Could be more effective this time in preventing the 7600 from sweeping up the mind/market share like the 6600GT did, at least until RV560 is readied on 80nm and with the advantage of knowing what to aim for.

The bigger challenge for ATI will be to deliver RV560 performance at least as good as X1800GTO and to avoid a repeat of the 9500 Pro 9600XT disappointment.

Well, I think it's unclear yet on what percentage of GTO are yield-fallouts. I would certainly expect that *some* of them are, but even a majority? X800 was out for quite a bit longer before we saw its GTO. So I tend to think that some significant portion of X1800GTO are going to be 16-quad enable-friendly.

On the "potential disappointments" front, it seems NV will get their chance to clear that hurdle first --will 7600GT be able to take 6800GS lunch money, give it a wedgie, and send it home sniffling?

 

[BrynS]

Before the slew of 256-bit X800 GT/GTO/GTO2/GTO16/etc variants released in the latter half of 2005, ATI did release the 256-bit X800SE (R420 with 2 PS quads disabled) in Q4 2004 and then the R430-based (with 1 PS quad disabled) X800 128MB and 256MB to compete with the 6600GT, after the cancellation of the X700XT and the sub-par performance (relative to the 6600GT) of the X700 Pro. Saying that though, I don't recall these 256-bit stopgaps being as competitively priced at the time as the recent GTO's et al.

 

[kemosabe]

Well, I think it's unclear yet on what percentage of GTO are yield-fallouts. I would certainly expect that *some* of them are, but even a majority? X800 was out for quite a bit longer before we saw its GTO. So I tend to think that some significant portion of X1800GTO are going to be 16-quad enable-friendly.

Argghh......the thought of thousands of full-fledged R520 cores selling for peanuts is unbearable - go ahead and plant that knife deep in ATI shareholders' backs, Geo.

Of course the lower-income gamers will be thrilled, but ATI's margins are already taking it on the chin. I guess that was the price to pay for bunching up the R520 and R580 releases like that, since R520 would not have commanded its launch price for very long after R580 anyway.

 

[neliz]

and to avoid a repeat of the 9500 Pro 9600XT disappointment.

What dissapointment? they pulled the plug out of a high cost unit and replaced it with a low cost one.

 

[Fodder]

What dissapointment?

9600 Pro was usually slower than 9500 Pro, at the same price. Bit of a downer for Captain Consumer.

 

[Moloch]

the problem was the 9500 was too expensive for ati to make.
And wasn't the 9600XT close to the 9500 pro?

 

[kemosabe]

Yes, I'm talking about disappointment from the consumer's point of view. Obviously RV350 was more viable from an economic perspective, but once you get the masses accustomed to that 256-bit goodness in the midrange, it's no piece of cake for the marketing team to evangelize that slower 128-bit part in the same price range. I seem to recall the 9500 Pro leaving the 9600 Pro in the dust and keeping ahead of the XT in most benchmarks, at least with the eye candy turned on. It's not like the sky will be falling, but I can't see how ATI will avoid a similar dilemma with X1800GTO and RV560.

 

[neliz]

9600 Pro was usually slower than 9500 Pro, at the same price. Bit of a downer for Captain Consumer.

We're talking about the 9600XT here.

And well yes, I too became a victim when my 9500pro crashed and burned and I had it replaced (BY A MUCH CHEAPER) 9600XT.
the 9600XT wasn't generally slower and had some nice bonuses and performance improvements (overdrive.)
Granted, the Pro, as a first incarnation was underwhelming but mostly because people expected to get a 9800Pro on the cheap, like the 95/97.

 

[neliz]

Yes, I'm talking about disappointment from the consumer's point of view.

Not really, the 9600xt was a good $50 cheaper than the 9500

looking at anandtech, you can see.. the 9500Pro being faster in older titles (where the Geforce 4600 also shun) but the 9600 being faster in most things "new"
http://www.anandtech.com/video/showdoc.aspx?i=1812

It did show it's strenghts however with AA/AF.


So in general, the mid range market got a card that was comparable to the product it replaced in most situations, yet costed a lot less

Tom's VGA chart three (the last one with the 9500Pro)

http://www.tomshardware.com/2003/12/29/vga_charts_iii/

tom's quality benches even show the 9600Pro winning in all three benchmarks (UT2003, Nascar 2004 and call of duty.)

Don't forget that driver maturity also comes in to play

 

[snk]

Yes, I'm talking about disappointment from the consumer's point of view. Obviously RV350 was more viable from an economic perspective, but once you get the masses accustomed to that 256-bit goodness in the midrange, it's no piece of cake for the marketing team to evangelize that slower 128-bit part in the same price range. I seem to recall the 9500 Pro leaving the 9600 Pro in the dust and keeping ahead of the XT in most benchmarks, at least with the eye candy turned on. It's not like the sky will be falling, but I can't see how ATI will avoid a similar dilemma with X1800GTO and RV560.

9500 Pro was a 128-bit part with 8 pipelines. Nevertheless it was still faster than 9600 Pro with AA and AF.

There were few vanilla 9500's with 256-bit memory bus, but they had only 4 pipelines and were slower than 9600 Pro.

 

[neliz]

Although the Pro was slower than the 9500, it definitly had more bang for the buck
http://www.tomshardware.com/2003/12/29/vga_charts_iii/page17.html

And the XT was a much better investment overall, the 9500pro was only quicker in BattleField and Halo, the rest of the benches are well within the grip of the XT, even with AA/AF

 

[kemosabe]

9500 Pro was a 128-bit part with 8 pipelines. Nevertheless it was still faster than 9600 Pro with AA and AF.

Indeed, my bad for forgetting it was the pipe count and not bus width. Anyway, if RV560 can reach the 650+ MHz range and with the R580 shader architecture, it might fare quite well against theX1800GTO in many titles after all. Has anyone in the know been able to confirm 8-1-3-1(2)?

 

[CJ]

8-1-3-1/2 sounds likely for the RV560 while the RV570 could be a 12-1-3-1. If you use 3DM05 as a basemark with the improvement percentages I provided a few pages back, you'll get the following 3DM05 scores:

RV570XT - ~10000 (10% faster than a X1800XT @ 9100 3DM)
RV570XL - ~8500 (20% faster than X1800XL @ 7100 3DM)
RV560 - ~6800 (assuming it's a mix up and it's 35% faster than a X1600XT @ 5100 3DM and not the 6800GS)

As you see the RV560 would be able to top the performance of the X1800GTO at ~6600 and the RV570XL/XT look like worthy replacements for the R520XL/XT. If the rumor is true that these new cores will be paired with high speed GDDR4, even a 128-bit memory interface could be enough to provide sufficient bandwidth and performance in eyecandy modes. And ATi did claim that we would really see the new memory controller shine if it would be paired with highspeed GDDR4.

But that still leaves the question... what is R590 if it's not the X1800GTO like DailyTech would like to make us think? Just a R580 die-shrink or a speedbump?

 

[Fodder]

There were few vanilla 9500's with 256-bit memory bus, but they had only 4 pipelines and were slower than 9600 Pro.

They could however often be modded into a 9700 Pro, which was certainly nothing to sneeze at.

But that still leaves the question... what is R590 if it's not the X1800GTO like DailyTech would like to make us think? Just a R580 die-shrink or a speedbump?

As everything but that DailyTech article has said, an optional 80nm shrink of R580.

 

[neliz]

They could however often be modded into a 9700 Pro, which was certainly nothing to sneeze at.

Not all pro's could, I remember looking in every nook and cranny for the black pcb version (produced by Sapphire.) actually, I asked a supplier if he had any 9500pro's left. and the reply was "Yes, and it's a sapphire one, thus unlockable"

9500Pro=x800ProViVo

 

[CJ]

They could however often be modded into a 9700 Pro, which was certainly nothing to sneeze at.
As everything but that DailyTech article has said, an optional 80nm shrink of R580.

But this time lowk won't be lost in the die-shrink process like it did with the 130nm+lowk -> 110nm transition. Shouldn't this make higher clockspeeds possible especially if ATi puts some effort into it (which I doubt they did with X700)?

 

[Geo]

http://www.theinquirer.net/?article=29800

Okay, can we just stick a fork in dailytech's "X1800GTO=R590" nonsense?

 

[neliz]

http://www.theinquirer.net/?article=29800

Okay, can we just stick a fork in dailytech's "X1800GTO=R590" nonsense?

I'll bring the steak knife

 

[Pete]

But that still leaves the question... what is R590 if it's not the X1800GTO like DailyTech would like to make us think? Just a R580 die-shrink or a speedbump?

IT's fairly obvious that Kris didn't have all the facts, or wasn't in full command of them, when he penned that article. (I'm not knocking him in general, just on that one specific piece.) I also think the more obvious comparison for R590 would be R430. If 80m doesn't clock as high as 90nm (I have no idea how low-k or yields figure into this), the safest choice would seem to be to aim for yields at X1900XL speeds.