ATI Radeon 4890 in April?

[chizow]

Originally posted by: MegaWorks
Didn't you say that AMD needs a fast single GPU to compete with the nVidia, there you have it. The 4870X2 is a lot faster than 30%, where did you pull that number from.

Yep, I did, but what does that have to do with you grossly overestimating the benefits of a 13% core clock increase? As for relative performance, how much faster would you say a 4870X2 is over a 4870? You can look at any of the GTX 285 reviews though and see that extra 10% boost over GTX 280 brought it awfully close to a 4870X2, and even manages to beat it in many cases.

Originally posted by: thilan29
Why is the 4870 at the limits of the design? How could you possibly know that unless you have intimate knowledge of the architecture?

Its really quite simple, the lack of factory OC'd variants and relatively poor results from user overclocking, along with the power required and heat generated to operate at even 750MHz offers little optimism for extending the performance of RV770 beyond what we've already seen. That and the continued rumors of faster binned RV770s that simply never materialize offer good indication that a faster RV770 on a 55nm process is unlikely. At 1.3V I can see 850MHz happening, it'll just be really hot, but those extra SP and TMU units, less likely imo.

 

[dguy6789]

The GTX 285 is no more competitive with the 4870x2 than 4870 is with the GTX280.

The 4870 doesn't even run hot..... The stock fans all just run at around 20% speed by default so that the card is relatively silent. Bring the fan speed up to even 50% and the card's full load temps drop by something like 30C.(This is with the reference design)

A lack of factory overclocked parts is nothing new. Even when ATI was making cards that overclocked extremely well, nobody sold factory overclocked ones.

The 4890's clock speed increase alone obviously isn't enough to increase performance by 20%, but they could also couple the release with performance increasing drivers or activate more ALUs on the cards. The 4890 is obviously not a card that will be sold in terribly massive quantities, so they probably don't need insanely good yields anyway. Judging by the name, the 4890 will probably be only a clock speed increase and nothing else hardware wise. If there are any added performance beyond that, it will probably be through drivers. That's my guess.

 

[thilanliyan]

Originally posted by: chizow
Its really quite simple, the lack of factory OC'd variants and relatively poor results from user overclocking, along with the power required and heat generated to operate at even 750MHz offers little optimism for extending the performance of RV770 beyond what we've already seen. That and the continued rumors of faster binned RV770s that simply never materialize offer good indication that a faster RV770 on a 55nm process is unlikely. At 1.3V I can see 850MHz happening, it'll just be really hot, but those extra SP and TMU units, less likely imo.

The power required (and hence the heat generated) is not unmanageable. It draws as much power as the 65nm 260 and if you look on newegg there are OCed versions (I think the lack of OCed versions is more due to the crappy partners not the card itself...look at the 3870 which was a very cool running chip but had very little if any OCed versions).

And I know this has been said many times (and certain people keep perpetuating this) but the card DOES NOT put an enormous amount of heat out...it runs at a higher temperature because the fan was set to a lower speed. Ask some of the people that have cards with non-stock cooling such as the Sapphire, Powercolor, or HIS 4870s and they'll tell you their cards don't run that hot.

EDIT: dguy6789 hit some of the points before me.

 

[chizow]

Originally posted by: dguy6789
The GTX 285 is no more competitive with the 4870x2 than 4870 is with the GTX280.

Yet that doesn't stop certain folks from making comparisons between the latter.

The 4870 doesn't even run hot..... The stock fans all just run at around 20% speed by default so that the card is relatively silent. Bring the fan speed up to even 50% and the card's full load temps drop by something like 30C.(This is with the reference design)

It draws as much power and generates as much heat as a much larger chip on a larger process (GT200), which indicates its already inefficient or suffers from significant leakage to reach its current clockspeeds.

A lack of factory overclocked parts is nothing new. Even when ATI was making cards that overclocked extremely well, nobody sold factory overclocked ones.

Rofl, ATI made cards that overclocked extremely well? I guess if you count underclocked parts meant to fill lower performance market segments.

The 4890's clock speed increase alone obviously isn't enough to increase performance by 20%, but they could also couple the release with performance increasing drivers or activate more ALUs on the cards. The 4890 is obviously not a card that will be sold in terribly massive quantities, so they probably don't need insanely good yields anyway. Judging by the name, the 4890 will probably be only a clock speed increase and nothing else hardware wise. If there are any added performance beyond that, it will probably be through drivers. That's my guess.

And older RV7XX parts wouldn't benefit from these magical driver improvements? If no, why not? If yes, it'll just be closer to a 13% improvement in-line with a linear core clock increase. Nice, but nothing specatular and certainly not enough to overtake a GTX 285, much less GTX 280.

I just find it funny the people so concerned with "misinformation" and "trolling" are so willing to believe reports of 20-30% performance gains from a simple 13% core clock boost.

 

[chizow]

Originally posted by: thilan29
The power required (and hence the heat generated) is not unmanageable. It draws as much power as the 65nm 260 and if you look on newegg there are OCed versions (I think the lack of OCed versions is more due to the crappy partners not the card itself...look at the 3870 which was a very cool running chip but had very little if any OCed versions).

The 4870 draws the same power as the 65nm 260, the 4890 at 1.3V will undoubtedly draw more power and run hotter as a result. While that doesn't mean it'll be unmanageable, it'll certainly generate more heat, meaning a hotter GPU or steeper cooling requirements to dissipate that heat.

As for lack of OC models, again I'd say its a lack of positive results and lack of significant risk-benefit-reward for selling and warrantying marginally overclocked products. The lack of overclocked AMD parts certainly helps explain some of the wildly optimistic guesses about performance gains from a 13% core clock boost though, its clearly a mystery or well-kept secret how an overclocked 4870 performs.

And I know this has been said many times (and certain people keep perpetuating this) but the card DOES NOT put an enormous amount of heat out...it runs at a higher temperature because the fan was set to a lower speed. Ask some of the people that have cards with non-stock cooling such as the Sapphire, Powercolor, or HIS 4870s and they'll tell you their cards don't run that hot.

Where's the professor when you need him. Actually the 4870 does draw as much power as its Nvidia counterparts, which should be much hotter and power hungry given their much larger die size and process (and certain people keep perpetuating this). The 4870's cooler is still inferior to the GT200 cooler, which means higher relative core temps or higher RPMs and more noise to compensate. A 4890 at higher core voltages and frequencies will mean even more leakage and TDP, which will obviously mean more heat and/or more noise. Again, I can see either higher clocks or additional SPs/TMUs, but not both. But I guess we'll see.

 

[SlowSpyder]

Isn't a GTX260 just an underclocked version of a higher binned card too? I just don't get this with you, you constantly say AMD cards don't overclock well, and I just don't believe it. And this is from personal experience as well as reading tech sites. Nvidia makes some damn good cards and often they have a lot of head room... the GTX260 is a good example. It's a crippled GTX280, it is meant to run faster then it's sold at. Kind of like the 4850. I've seen a number of 4850's on this site approaching 800MHz... infact I think dug777 is over 800Mhz. My 2900Pro went from 600MHz factory to over 800Mhz... that's not poor overclocking. My current 4870 is overclocked about 10% on the core and my memory is running 1GHz above reference.

I guess we'll just continue to disagree on this... just because AMD's partners aren't releasing highly overclocked cards doesn't mean the 4870 doesn't have head room.

 

[chizow]

Originally posted by: SlowSpyder
Isn't a GTX260 just an underclocked version of a higher binned card too? I just don't get this with you, you constantly say AMD cards don't overclock well, and I just don't believe it. And this is from personal experience as well as reading tech sites. Nvidia makes some damn good cards and often they have a lot of head room... the GTX260 is a good example. It's a crippled GTX280, it is meant to run faster then it's sold at. Kind of like the 4850. I've seen a number of 4850's on this site approaching 800MHz... infact I think dug777 is over 800Mhz. My 2900Pro went from 600MHz factory to over 800Mhz... that's not poor overclocking. My current 4870 is overclocked about 10% on the core and my memory is running 1GHz above reference.

I guess we'll just continue to disagree on this... just because AMD's partners aren't releasing highly overclocked cards doesn't mean the 4870 doesn't have head room.

The GTX 260 is underclocked by 25MHz, but the main difference in performance still comes from actual core/bandwidth/vram differences rather than just clockspeed scaling. Both the GTX 260 and 280 overclock significantly beyond their similar stock speed and are offered at those higher clock speeds as factory OC'd parts.

AMD parts on the other hand have little headroom at the high-end and can't be found with significant factory OCs, which leads to the general perception AMD parts do not overclock well as clock thresholds were pushed in order to achieve the high-end part to begin with.

I guess one way to look at it is Nvidia allows their vendors to call OC versions SC, SSC, FTW, etc. while AMD gives their OC versions a completely new model number, like 4850, 4870, etc. 4890 is just another example, instead of 4870 FTW or 4870 XXX they give it a completely new model number. What do you think the chances of seeing a 4890 variant with a significant OC are?

Again, the ones in this thread who are best positioned to dispel any "misinformation" seem like the ones most intent on defending it. What kind of performance gains are you seeing from your 10% overclock? Is it closer to 10%, or 20-30%?

 

[Denithor]

Originally posted by: chizow
The 4870 draws the same power as the 65nm 260, the 4890 at 1.3V will undoubtedly draw more power and run hotter as a result. While that doesn't mean it'll be unmanageable, it'll certainly generate more heat, meaning a hotter GPU or steeper cooling requirements to dissipate that heat.

Yep.

Originally posted by: SlowSpyder
Isn't a GTX260 just an underclocked version of a higher binned card too?

Actually, no. The core has fewer stream processors enabled (192/216 vs 240) so the cores are probably binned (those that pass with the full 240 working go into 280 cards). However, the cards built around those cores are actually different - remember the 260 features a 448-bit memory interface while the 280 has a full 512-bit interface.


The way I see it - this new 4890 will probably feature higher clockspeed and increased stream shaders. ATi has probably been binning higher count chips for months in preparation of a final card for the series. They already bin the cores for use in 4830/4850/4870 so why not just test the top tier for even higher performance? They'd only need a small percentage to pass in order to launch this card, which is in essence a stop-gap measure until the new 5800-series launches later this year.

 

[SlowSpyder]

I know it's crippled, I just meant the clock speed portion of what I said in regards to the discussion Chizow and I were having. The 4850 is just a lower clocked RV770, not cripped. The GTX260 is a slightly different animal, it's lower clocked and has some internals disabled. I was just trying to point out that both are lower clocked versions of a higher binned card.

Chizow, I don't believe that the 4890 will be 20%+ faster with 850MHz clocks. My guess (only a guess) is that it'll be a higher clocked part with more shaders, if it exists at all. That's where the extra performance will come from.

And for the record, I have a 4870 Toxic edition. That is Sapphires's equivalent to say an SSC, FTX, etc. Asus makes a 'Dark Knight' edition or something like that for their card. Sapphire and Asus don't call them a 4880.

 

[Cookie Monster]

Id like to note that die sizes have nothing to do with power consumption but rather alot to do with yields. GT200 is quite big because its layout is loosely packed compared to RV770 but dont let the size fool you since when it comes to transistor count, its only ~1.4billion (GT200) vs ~0.9 billion (RV770).

I agree about slowspyder and ATI in general with OCed parts. In the past 5~6 years, it was a rare thing to see ATi partners with OCed parts. I dont think one can assume that for 6 years
they have created chips without headroom.

Originally posted by: Denithor
Actually, no. The core has fewer stream processors enabled (192/216 vs 240) so the cores are probably binned (those that pass with the full 240 working go into 280 cards). However, the cards built around those cores are actually different - remember the 260 features a 448-bit memory interface while the 280 has a full 512-bit interface.

GPU cores used for GTX260 is just the ones that didnt make the cut. These have one of its ROP/TMU/IMC section disabled along with fewer SPs.

edit - If this RV790 has an increase in logic units, i.e shaders/TMUs for example then there is simply no need for upping the core voltage (1.3V sounds illogical seeing as the highest advertised core voltage for the 55nm process is 1.26V iirc) because the added performance increase will be bigger without the need for upping the clockspeeds.

 

[Wreckage]

Originally posted by: Cookie Monster


I agree about slowspyder and ATI in general with OCed parts. In the past 5~6 years, it was a rare thing to see ATi partners with OCed parts. I dont think one can assume that for 6 years
they have created chips without headroom.

They have generally always run their cards at max. Check out reviews going way back to the x1800 and you will see that heat and power have always been an issue.

http://techreport.com/articles.x/8864/16

Even with the caveats, I think we can draw some provisional conclusions. Systems based on the new Radeons uniformly draw more power than the NVIDIA competition, which isn't entirely a shock given the higher clock rates for the ATI GPUs.

 

[thilanliyan]

Originally posted by: chizow
The 4870 draws the same power as the 65nm 260, the 4890 at 1.3V will undoubtedly draw more power and run hotter as a result. While that doesn't mean it'll be unmanageable, it'll certainly generate more heat, meaning a hotter GPU or steeper cooling requirements to dissipate that heat.

The 4870 is already running at 1.26v...you think an extra 0.04v is gonna make a large difference? Also, if these cores are binned and considering they've had time to work out any yield kinks, they may actually have the same power draw as regular 4870s.

Actually the 4870 does draw as much power as its Nvidia counterparts, which should be much hotter and power hungry given their much larger die size and process (and certain people keep perpetuating this). The 4870's cooler is still inferior to the GT200 cooler, which means higher relative core temps or higher RPMs and more noise to compensate.

Power draw is also related to the architecture, not only process and die size, as Cookie Monster mentioned above me. As I've said before unless you have intimate knowledge of both architectures you can't claim one is inefficient or power hungry, etc. compared to the other.

This is nowhere near conclusive proof but AT did an "efficiency" test when RV770 first launched:
http://www.anandtech.com/showdoc.aspx?i=3341&p=7

From that it seems RV770 uses its transistors effectively.

 

[Cookie Monster]

Originally posted by: Wreckage
They have generally always run their cards at max. Check out reviews going way back to the x1800 and you will see that heat and power have always been an issue.

http://techreport.com/articles.x/8864/16

Even with the caveats, I think we can draw some provisional conclusions. Systems based on the new Radeons uniformly draw more power than the NVIDIA competition, which isn't entirely a shock given the higher clock rates for the ATI GPUs.

You picked a very bad example. R520 was plagued with leakage issues, delayed for 6 months, pretty much a shortlived product and plus, it was quickly replaced with the successful R580.

I still dont get this fixation about this myth, that ATi cards run at max with no headroom..

 

[Wreckage]

Originally posted by: Cookie Monster


You picked a very bad example.

Actually I picked a perfect example of what we are talking about. That review was from 2005 and clearly illustrates that it uses a lot of power to run a very high clock (leaving little or no headroom). A similar situation as seen today.

 

[chizow]

Originally posted by: SlowSpyder
I know it's crippled, I just meant the clock speed portion of what I said in regards to the discussion Chizow and I were having. The 4850 is just a lower clocked RV770, not cripped. The GTX260 is a slightly different animal, it's lower clocked and has some internals disabled. I was just trying to point out that both are lower clocked versions of a higher binned card.

Chizow, I don't believe that the 4890 will be 20%+ faster with 850MHz clocks. My guess (only a guess) is that it'll be a higher clocked part with more shaders, if it exists at all. That's where the extra performance will come from.

And for the record, I have a 4870 Toxic edition. That is Sapphires's equivalent to say an SSC, FTX, etc. Asus makes a 'Dark Knight' edition or something like that for their card. Sapphire and Asus don't call them a 4880.

Yep, the 4850 is similar to the 3850, 2900pro and various other AMD parts underclocked to meet lower performance market segments, so yes they'll have headroom up to the high-end clocks, that's expected. The problem is there's very little headroom on the high-end as they've already been stretched to their limit.

Surely you don't think AMD was leaving performance on the table when they could've clocked their parts higher, especially in the case of parts like the R600 and RV670 and even in this case with RV770/790 where they were trailing the competition and any additional performance would've certainly been welcomed......

As for the Toxic and Dark Knight being equivalent to an SSC or FTW....lol doubtful...a 30MHz max OC on an Nvidia part would be like a FTL or NoC edition....

Originally posted by: Cookie Monster
Id like to note that die sizes have nothing to do with power consumption but rather alot to do with yields. GT200 is quite big because its layout is loosely packed compared to RV770 but dont let the size fool you since when it comes to transistor count, its only ~1.4billion (GT200) vs ~0.9 billion (RV770).

Actually power consumption should directly corrolate to die size, transistor count and process, as historically these factors have resulted in better efficiency for the smaller/newer chips. The RV770 bucks this trend of course as it requires more power to run at its rated speeds, which again, isn't anything new for ATI chips, which have historically run hot near their max thresholds.

 

[dguy6789]

If you want to argue efficiency, consider that the GT200 core has 40% more transistors and is much less than 40% faster.

http://www.anandtech.com/showdoc.aspx?i=3341&p=2
http://www.anandtech.com/showdoc.aspx?i=3341&p=7

It should be very clear which chip would be faster if they were the same size. The RV770 would use significantly more electricity and generate more heat in that case though.

 

[chizow]

Originally posted by: thilan29
The 4870 is already running at 1.26v...you think an extra 0.04v is gonna make a large difference? Also, if these cores are binned and considering they've had time to work out any yield kinks, they may actually have the same power draw as regular 4870s.

Possibly, given temps and voltage scale non-linearly, especially as you reach higher clocks where leakage and temps become a greater concern. As for the 1.3V, that was actually quoted from the article, but given certain people are just injecting whatever conjecture they feel appropriate, I guess you might as well believe they can achieve 850MHz at stock volts when apparently no one else with a 4870 can.......

Power draw is also related to the architecture, not only process and die size, as Cookie Monster mentioned above me. As I've said before unless you have intimate knowledge of both architectures you can't claim one is inefficient or power hungry, etc. compared to the other.

Again, all historical indicators would lead you to believe the 4870 would draw less power than GT200, yet it doesn't. It uses more power per transistor and performs on par or worst. Not only is it less efficient in terms of power, it leaves the possibility of a higher clocked refresh in doubt due to the already high power requirements of the 4870.

This is nowhere near conclusive proof but AT did an "efficiency" test when RV770 first launched:
http://www.anandtech.com/showdoc.aspx?i=3341&p=7

From that it seems RV770 uses its transistors effectively.

Actually the one comparison they got right showed the RV770 was inefficient with power as it draws more in order to run at its rated speed. Again, this just shows ATI is already overclocking/overvolting RV770 to achieve the desired performance level to compensate for fewer transistors.

The biggest flaw in that comparison is that AT doesn't take clock speed differences into consideration. RV770 has fewer transistors but runs 25% faster to compensate, requiring similar power to achieve those clocks. To show how flawed their comparison is, you'd come to a similar conclusion comparing 4850 to 4870 even though they're the same exact core. Clearly a flawed comparison.

 

[chizow]

Originally posted by: dguy6789
If you want to argue efficiency, consider that the GT200 core has 40% more transistors and is much less than 40% faster.

http://www.anandtech.com/showdoc.aspx?i=3341&p=2
http://www.anandtech.com/showdoc.aspx?i=3341&p=7

It should be very clear which chip would be faster if they were the same size.

Seems like you're forgetting clockspeed.......once you clock that GT200 up (See GTX 285 FTW) it does look awful close to 40% faster, and still uses less power than the 4870.

 

[dguy6789]

In some ways, RV770 reminds me of Prescott.

 

[ronnn]

Originally posted by: SickBeast


They're still saying that it will be clocked higher. They're not going to name it the 4890 and have it be the same speed or slower than a 4870. I'm actually shocked that you would defend Wreckage's post. :Q

Vrzone is a nv fud factory, so they may be wrong. Anyways what is Tos? Always thought it was a polite way of saying upchuck.

 

[Cookie Monster]

Originally posted by: Wreckage
Actually I picked a perfect example of what we are talking about. That review was from 2005 and clearly illustrates that it uses a lot of power to run a very high clock (leaving little or no headroom). A similar situation as seen today.

Are you blindly ignoring the fact that the 90nm process R520 faced serious leakage issues hence the absurdly high power consumption? They had no choice at the time, since the product was delayed even further so that brought forward R580 in no less than 3 and a half month after the launch of R520. The successor pretty much fixed this issue while performing alot faster than its competition.

Originally posted by: chizow
Actually power consumption should directly corrolate to die size, transistor count and process, as historically these factors have resulted in better efficiency for the smaller/newer chips. The RV770 bucks this trend of course as it requires more power to run at its rated speeds, which again, isn't anything new for ATI chips, which have historically run hot near their max thresholds.

Care to clarify this? I agree that power consumption does somewhat correlate to transistor count, and the process technology, but how does the die size correlate?

Im going to cut to the chase here. Without knowing/having the data that the engineering teams behind RV770 have access to, statements like "requires more power" to run at its "rated speed" (not sure what you mean by here) is just pure misinformation. Do you somehow have access to these test datas of RV770 or basing your statements on articles on the web?

The last statement of yours is abit puzzling. Do you know the max thresholds for each and every chip designed by ATi? To my knowledge, most GPUs are generally very hot, and depending on the HSF used basically determines its heat characteristics (when looking from a retail product perspective). Take a look at the 8800GT. This card had serious thermal problems because of the single slot HSF, but temps dropped almost 30~40C when paired with a better aftermarket cooler.

Somehow chips that are branded ATi are magically hotter at their supposed "max thresholds" for some unknown reason is just pure fantasy that holds no real value in my book.

And no, a FTW edition of GTX285 by EVGA does consume more power than the HD870. Taking the SSC version as an example found here Im sure the FTW version will consume extra 10~20W more power.

 

[Zstream]

Rarely do I agree with CM but damn nice post.

 

[thilanliyan]

Originally posted by: chizow
Possibly, given temps and voltage scale non-linearly, especially as you reach higher clocks where leakage and temps become a greater concern. As for the 1.3V, that was actually quoted from the article, but given certain people are just injecting whatever conjecture they feel appropriate, I guess you might as well believe they can achieve 850MHz at stock volts when apparently no one else with a 4870 can.......

I myself can get to 830MHz (others have stated similar findings) at stock volts (this was tested before I switched to watercooling) so imagining that with a few tweaks and better yields they could get to 850 is not a stretch but I really can't guess whether they would actually go to that speed.

Again, all historical indicators would lead you to believe the 4870 would draw less power than GT200, yet it doesn't. It uses more power per transistor and performs on par or worst. Not only is it less efficient in terms of power, it leaves the possibility of a higher clocked refresh in doubt due to the already high power requirements of the 4870.

See Cookie Monster's post above...the 2 are completely different architectures so you can't do direct comparisons as you are doing.

The biggest flaw in that comparison is that AT doesn't take clock speed differences into consideration. RV770 has fewer transistors but runs 25% faster to compensate, requiring similar power to achieve those clocks. To show how flawed their comparison is, you'd come to a similar conclusion comparing 4850 to 4870 even though they're the same exact core. Clearly a flawed comparison.

There's a flaw in that test as there's a flaw in you making generalizations of 2 different architectures of which you have no intimate knowledge.

 

[SickBeast]

Originally posted by: ronnn

Originally posted by: SickBeast


They're still saying that it will be clocked higher. They're not going to name it the 4890 and have it be the same speed or slower than a 4870. I'm actually shocked that you would defend Wreckage's post. :Q

Vrzone is a nv fud factory, so they may be wrong. Anyways what is Tos? Always thought it was a polite way of saying upchuck.

Yeah but to say that the next high-end AMD part will merely "compete" with NV's current midrange part is asinine at best. To say that the 4870 is not in the same ballpark as a 260+ is just stupid, sorry, there's not much other way to put it.

TOS = terms of service

Basically it's what people agree to when they sign up here. You're not allowed to post false information or anything defamatory. IMO what Wreckage has said in this regard is both false and defamatory.

 

[chizow]

Originally posted by: Cookie Monster
Care to clarify this? I agree that power consumption does somewhat correlate to transistor count, and the process technology, but how does the die size correlate?

Because given similar process, die size is going to scale with total transistor count and larger dies typically take more power to operate than smaller ones.

Im going to cut to the chase here. Without knowing/having the data that the engineering teams behind RV770 have access to, statements like "requires more power" to run at its "rated speed" (not sure what you mean by here) is just pure misinformation. Do you somehow have access to these test datas of RV770 or basing your statements on articles on the web?

Pure misinformation? Really? How much power does your 4850 take to run at 625MHz? How much does a 4870 take to run at 750MHz? How much is this article claiming in order to run at 850MHz? Do you think the single slot cooler is sufficient in cooling the 4870 at 750MHz? What kind of overclocks are end-users seeing with the 4870 and what kind of factory OC variants are available for purchase?

You can probably start your research here.....

As you can see, people aren't undervolting their 4870s to achieve higher clock speeds.... there's no need for top sekret test data for RV770 to verify how live samples out in the wild are behaving. I guess you can believe ATI has some magical solution to extract more performance out of an RV770 die on the same process and have really just been holding back for the last 7 months since launch.

The last statement of yours is abit puzzling. Do you know the max thresholds for each and every chip designed by ATi? To my knowledge, most GPUs are generally very hot, and depending on the HSF used basically determines its heat characteristics (when looking from a retail product perspective). Take a look at the 8800GT. This card had serious thermal problems because of the single slot HSF, but temps dropped almost 30~40C when paired with a better aftermarket cooler.

Again, I don't need to know the max thresholds from any white paper, I just have to look at what's available on the market mixed with end-user reports that show very little success overclocking to the expected 4890 speeds of 850MHz. Simply put, if ATI was able to get more performance out of their RV770, why haven't they, why would they leave performance on the table when they've trailed GT200 since launch and Nvidia in the market in general since G80?

As for the 8800GT, again, a poor example, as that card didn't have thermal issues beyond the stock fan spinning at only 29%. The difference is it still ran within reasonable temps at both idle and load, even while overclocked, which is a stark contrast compared to similar fan speed problems with the 4850 and 4870 where idle/load temps were hitting 80-90/100C respectively. 8800GT OC Thread at launch. You'll see plenty of users, including myself, had no issues overclocking the 880GT with the stock cooler, especially once fan speed was increased to 40-50%.

Furthermore, the 8800GT and G92 are a poor choice as comparison because this architecture went on to produce the 650MHz GTS, 675MHz GTX and many 700MHz+ OC variants, so clearly Nvidia wasn't hitting any thermal or design walls with the 600MHz GT.......

Somehow chips that are branded ATi are magically hotter at their supposed "max thresholds" for some unknown reason is just pure fantasy that holds no real value in my book.

Rofl, but dreaming up a 850MHz with 1000SP and 48TMU based on the same core and same process isn't pure fantasy, especially given the lack of retail or user feedback as proof of concept for even the clockspeed bump? Again, just look at ATI's design and product decisions and overall market position over the past few years. Its obvious they scale their parts based on clockspeed and leave very little room for clock increases on the high-end. If validating and achieving such clocks were so simple, it begs the question as to why ATI isn't selling a faster part given they've trailed Nvidia in performance since G80. Clearly they would've benefitted from a faster part, yet they were unable to produce one. Honestly its just common sense at this point, no need for "intimate knowledge" of their technical data.

And no, a FTW edition of GTX285 by EVGA does consume more power than the HD870. Taking the SSC version as an example found here Im sure the FTW version will consume extra 10~20W more power.

Actually I was going by the published 189W TDP for GTX 285, but as Denithor linked, its much closer than it should be given the GTX 285 is faster, larger, and has 40% more transistors than the 4870.