AMD RV770 refresh -> RV790

Not exactly now but probably in March or April...
I meant "5800 X2 series (R870 dual core package shared memory via sideport)" and "5800 X4 series (R870 dual core package x 2 connected with PLX chip)".....

"Dual-core package" can be read as "multi- chip module", right?:???:

Make it stop! Make it stop!
 
....If they however just die shrink the current gpus and clock them slightly higher , they will get more chips per wafer with better power draw characteristics adn be able to sell to a larger market as they reduce prices on tehse cards in comparison to the old line up

OK, but this "just die shrink" will not give us the desired performance improvement. ATi needs at least 20-30 percent performance improvement.
I will not buy such a die shrink. Never. It is not enough for me. What about you and the others?
 
Hopefully, you'll see this "Powered by ATiTM". Your account name has been changed to remove the "TM", as it gives a sense of officialness to your posts that is unwarranted. Unless you have something to share on that score.
 
Yep, since the relative pasting the 4800's put on GT200's, ATI has done nothing but squander much of that. Now Nvidia has become faster at nearly every price level, taken back the overall crown etc. Nvidia has been very aggressive since initially with the core 216, 285, 295, as well as tons of factory overclocked models of all those. ATI has done nothing, and most ATI cards at retail continue to run stock only.

Like I say, the initial victory was all ATI's, but since then they've done nothing and become the ATI of old as usual seemingly. Conversely Nvidia was "dealt" a terrible hand this round but is maximizing it.
The 4870 X2 is slightly slower than the GTX 295 but is significantly cheaper, the 4870 performs on par with the GTX 260 Core 216, and they're both at the same price level, the 9800+ GTX is slightly faster than the 4850 at around the same price level, the 4670 performs on par with the 9600 GSO at the same price level (it's cheaper if you include MIRs).

So really the only clear victory for Nvidia is the 9800 GTX+. But 4850s seem to be selling well due to name recognition, anyway, so it's not clear that ATI is doing badly in that segment, or that they need to drop prices further.

But what else did you expect ATI to do? They've given partners freedom to overclock, and most of them have released overclocked versions of the 4870 at least, but they can't enable more clusters like Nvidia, and they're already at 55 nm.
 
I'm not saying Nvidia isn't victorious in the high-end single GPU market, only that's not relevant to Ranger's criticisms because ATI chose not to compete in that market from the start.

As such if that's a failing, it's a failing of their strategy in general not their post RV770 support, which Ranger is criticising.

At the price points ATI is selling their cards at, they are very competitive with Nvidia's offerings; that's all I'm claiming.
 
True, but one aspect of them being as competitive as they are certainly is their choice not to compete against GTX280/5.
 
The 4870 X2 is slightly slower than the GTX 295 but is significantly cheaper, the 4870 performs on par with the GTX 260 Core 216, and they're both at the same price level, the 9800+ GTX is slightly faster than the 4850 at around the same price level, the 4670 performs on par with the 9600 GSO at the same price level (it's cheaper if you include MIRs).

So really the only clear victory for Nvidia is the 9800 GTX+. But 4850s seem to be selling well due to name recognition, anyway, so it's not clear that ATI is doing badly in that segment, or that they need to drop prices further.
Maybe it depends on the country you're living in. Around here, GF 9800 GTX+ is generally more expensive than a Radeon HD 4850. Close to cheapest HD 4870's I'd say. So from my point of view, nVidia doesn't have any clear winners except the GTX 280/285 - that ATI chose not to compete with, as you say.
Subtlesnake said:
But what else did you expect ATI to do? They've given partners freedom to overclock, and most of them have released overclocked versions of the 4870 at least, but they can't enable more clusters like Nvidia, and they're already at 55 nm.
Well, they could have designed a new chip with more clusters. RV770 is not exactly a monster chip and bigger dies can be done on 55nm.
 
Well, ATI may not have chosen to compete with GTX 280. But in most consumer minds the 4870 was a more attractive product, meaning that inadvertently the 4870 and GTX 280 were competeing with the GTX 280 losing out massively until its price was slashed. And even then uptake of GTX 280 is/was pretty low. GTX 260 on the other hand saw a huge rise in popularity once price was adjusted inline with the 4870, GTX 260/216 was release, and new drivers giving a perf boost to a few hand selected games came out.

Looking back at Q4 numbers, Nvidia was only winning in market share due to 88xx and 98xx series with the GTX 2xx losing out badly to the 48xx series.

So I'm not quite sure what ATI could have done "more" of. Other than perhaps slash prices even more. But that may not be a good idea for the bottom line in these economic times.

Regards,
SB
 
I don't believe the 9800 GTX+ beats the 4850 when AA is applied and/or at higher resolutions and the 4850 is generally slightly cheaper, according to Newegg.
 
Maybe it depends on the country you're living in. Around here, GF 9800 GTX+ is generally more expensive than a Radeon HD 4850.
Costs aside, since when is the 9800 GTX+ a clear winner against the HD 4850? That's certainly news to me...
 
So RV790 is a clock bump ala GTX285 and RV740 is 40nm?

Sort of - the RV790 is using a diff process that gives much better yields of 4870 class chips. If Mr Konkort's figures are accurate the current 4870 - 750Mhz @ 1.263V is been replaced by a chip at 750Mhz @ 1.24V which really doesnt sound like much progress. Think the GT has less leakage so hopefully the overall power figures will be better. Otherwise wont be able to put it into a 4850/4830 form factor at those clocks, and the point of the exercise is lost.

Speaking of leakage, the RV740 @ 40nm apparently hasnt hit required power levels. Not sure what they plan to do now, could release at lower clocks i guess or send it back for another spin. Might also be other problems with TSMC, there has been speculation their utilization could fall to 35% this quarter. Your customers are pretty much captive, why start up additional lines of production which will only make the situation worse?

Originally the 740 was supposed to come out in November last year, TSMC pushed this back to May 2009, then it came forwards again to around March as all the other 40nm orders disappeared, now its seems back to May perhaps.
 
Costs aside, since when is the 9800 GTX+ a clear winner against the HD 4850? That's certainly news to me...
I wouldn't say it's a "clear winner", but in some newer benchmarks, it beats the HD 4850 by quite a few percent. Which is, of course, offset by the higher price...
 
Sort of - the RV790 is using a diff process that gives much better yields of 4870 class chips. If Mr Konkort's figures are accurate the current 4870 - 750Mhz @ 1.263V is been replaced by a chip at 750Mhz @ 1.24V which really doesnt sound like much progress. Think the GT has less leakage so hopefully the overall power figures will be better. Otherwise wont be able to put it into a 4850/4830 form factor at those clocks, and the point of the exercise is lost.
Single-slot HD4870, so at least 20-30W less power? Not sure which ATI single-slot card had the highest power consumption, or what that was, though. 130W for single slot, is that do-able?

Speaking of leakage, the RV740 @ 40nm apparently hasnt hit required power levels. Not sure what they plan to do now, could release at lower clocks i guess or send it back for another spin.
"HD4750" at 600MHz launched soon, with 750MHz "HD4770" 3 months later would seem like a reasonable plan.

Originally the 740 was supposed to come out in November last year, TSMC pushed this back to May 2009, then it came forwards again to around March as all the other 40nm orders disappeared, now its seems back to May perhaps.
:rolleyes:

Jawed
 
My problem with Fuad isn't that he reports these things. It's that they're never accompanied by any sort of skepticism on his part. How in the world can RV630 drive a model number higher than the 46xx RV730 series?
 
4750 is based on RV630 if you believe what Fuad's telling.

Of course... :D
http://www.fudzilla.com/index.php?option=com_content&task=view&id=9512&Itemid=34

But why is DAAMIT so slow?

My highlight:
ATI to re-brand Radeon 3x00 to 4x00 Print E-mail
Written by Fuad Abazovic
Thursday, 17 July 2008 09:52

AMD says no renaming for HD 3000 Print E-mail
Written by Fuad Abazovic
Tuesday, 22 July 2008 10:36


RV770&RV730 have some interesting GPGPU capabilites, which are new and HD3000 don't have. It's pure BS to suggest, that RV670s will be used for HD4700s.
 
Last edited by a moderator:
This isn't a particularly large surprise that power gains were marginal.

Even the nominal gains in watts per mm2, without any actual problems, would not have kept pace with the doubling of transistor density.

That leakage might be worse than that something that would have some supporting evidence.
Leakage has been getting worse, TSMC has had issues before with the performance process and power numbers, and very few manufacturers have done 45nm and below without some power headaches.

One of the successful ones is Intel, and they have a hi-k dielectric and metal gates to help control leakage, which TSMC will not have for a long time. It should be noted that Intel has also indicated that these additions have also significantly helped with device variation.
 
Back
Top