ATI 4xxx Series Thread

[Extelleron]

Originally posted by: bryanW1995
yeah, but oc'ing with CCC sucks. 625 to 700 is extremely weak compared to what MANY nvidia users get on their cards with rivatuner.

Previous rumors said that RV770 Pro (HD 4850) would be capable of overclocking to 1GHz or so. But that might be BS (if true then that would be amazing).

From what I see AMD has made a move for lower clocks in an effort to control TDP, and likely the RV770 core is capable of much more than 750MHz. So I think these cards are going to be good overclockers. I don't know about 1GHz, but I can't see why the HD 4850 wouldn't hit ~800MHz or so and the HD 4870 maybe ~900MHz.

 

[Nemesis 1]

Hi! Guys I am a little confused. I believe we have all read the same available info on the up and coming 4850/4870. Why this talk of set clocks. As I understand it ATI/AMD are allowing the card makers to finilize their own specs. So the750 MGz is ATI only cards .

Or did I misunderstand this. If its true and the 4870 can infact O/C really well. Some card makers will jump all over it. We well see cards clocked @ 850MGz if the chip can actually handle it

 

[Quiksilver]

Originally posted by: Nemesis 1
Hi! Guys I am a little confused. I believe we have all read the same available info on the up and coming 4850/4870. Why this talk of set clocks. As I understand it ATI/AMD are allowing the card makers to finilize their own specs. So the750 MGz is ATI only cards .

Or did I misunderstand this. If its true and the 4870 can infact O/C really well. Some card makers will jump all over it. We well see cards clocked @ 850MGz if the chip can actually handle it

...because 99% of all the vendors are going to release the cards at stock reference speeds. Then after a few weeks/months on the market companies like sapphire and powercolor will put out their vaporX and extreme overclock version of the cards. They are also talking because most enthusiasts will overclock their video cards too so their just speculating.

 

[taltamir]

Originally posted by: Extelleron

Originally posted by: bryanW1995
yeah, but oc'ing with CCC sucks. 625 to 700 is extremely weak compared to what MANY nvidia users get on their cards with rivatuner.

Previous rumors said that RV770 Pro (HD 4850) would be capable of overclocking to 1GHz or so. But that might be BS (if true then that would be amazing).

From what I see AMD has made a move for lower clocks in an effort to control TDP, and likely the RV770 core is capable of much more than 750MHz. So I think these cards are going to be good overclockers. I don't know about 1GHz, but I can't see why the HD 4850 wouldn't hit ~800MHz or so and the HD 4870 maybe ~900MHz.

A massive overclock means that the manufacturer underclocks and undercharges for the item. Either in error, or on purpose.
I doubt AMD will do so for either reason.

 

[firewolfsm]

The card has 480SP, NOT 800. There aren't enough transistors on the die to cover that many units. The R770 isn't much bigger than the last generation.

 

[Extelleron]

Originally posted by: firewolfsm
The card has 480SP, NOT 800. There aren't enough transistors on the die to cover that many units. The R770 isn't much bigger than the last generation.

I've posted this @ xtremesystems twice already (because people don't seem to believe) but here is why that argument is faulty:

As I've shown before, die size & processing resources do not scale linearly at all.

Case in point, RV635 (55nm/118mm^2) -> RV670 (55nm/192mm^2)

ATI fits 2.66x SPs, 2x TMUs, 4x ROPs, increased cache, plus twice internal ring bus size (512-bit vs 256-bit) in a die only 63% larger.

So it is not unlikely they can fit 2.5x SPs/2x TMUs, with all else the same, in a die 34% larger.

Comparing RV620 & RV670 just exaggerates the point.

RV670 is approximately ~3x RV620 and it features 8x SPs / 4x TMUs / 4x ROPs / 4x ringbus (512-bit vs 128-bit) / vastly more complex cache structure

The guy who runs NH has said that the source of this rumor has been 100% accurate in the past and he has no reason to doubt him on this.

I'm 95% sure that RV770 either has 480SP + split clock domains or 800SP, and right now it is pointing to 800SP.

 

[Extelleron]

Originally posted by: taltamir

Originally posted by: Extelleron

Originally posted by: bryanW1995
yeah, but oc'ing with CCC sucks. 625 to 700 is extremely weak compared to what MANY nvidia users get on their cards with rivatuner.

Previous rumors said that RV770 Pro (HD 4850) would be capable of overclocking to 1GHz or so. But that might be BS (if true then that would be amazing).

From what I see AMD has made a move for lower clocks in an effort to control TDP, and likely the RV770 core is capable of much more than 750MHz. So I think these cards are going to be good overclockers. I don't know about 1GHz, but I can't see why the HD 4850 wouldn't hit ~800MHz or so and the HD 4870 maybe ~900MHz.

A massive overclock means that the manufacturer underclocks and undercharges for the item. Either in error, or on purpose.
I doubt AMD will do so for either reason.

You have to remember that this is not the old ATI, AMD is at the helm now, and AMD is a different beast than ATI was.

ATI was concerned about raw performance with disregard to power consumption, but AMD is different. Power usage is very important in the CPU business and it is likely they are extending these ideals to their GPUs as well.

I think it is likely that AMD found a "sweet spot" with RV770... a clock that would result in good performance but relatively low TDP. Increasing core clock beyond that point would result in higher performance, but also significantly higher power consumption.

The other reason is that RV770XT is not the high end... R700 is. AMD probably doesn't want to clock R700 lower than RV770XT, but they cannot hit 800-900MHz with a dual RV770 card and stay within a ~250W thermal envelope. If HD 4870 was clocked at 950MHz, but HD 4870 X2 was 775MHz or so, then the HD 4870 would be cutting into its performance (and thus sales). So to keep RV770XT below R700, they are probably keeping the clocks lower than necessary.

I could be wrong, but I think these parts will have a decent amount of overhead in them. The only reason they wouldn't is if ATI pulls some BS like with the HD 3870 and locks the card at a certain frequency (for the 3870 it was ~860MHz) or the card has a very low voltage.

 

[taltamir]

that is what low end parts are for, to be underclocked and underpriced versions of the high end...
Their highest end part should be as fast as they can make it.

The reason why intel CPUs are not like that at the moment, is because intel figures they are "sufficiently faster than the competition" and don't want them to compete with their future products.

 

[golem]

Extelleron - I'm not sure about that... AMD has processors that span energy usage from low (BE chips) to very high (faster phenom and x2 chips). Their high end phenoms and x2 suck quite a bit of power, why would the be shy about making their high end GPUs do the same? Especially since according to most reports it would still be lower than what their competition draws?

 

[dv8silencer]

http://www.nordichardware.com/news,7845.html

 

[taltamir]

Originally posted by: golem
Extelleron - I'm not sure about that... AMD has processors that span energy usage from low (BE chips) to very high (faster phenom and x2 chips). Their high end phenoms and x2 suck quite a bit of power, why would the be shy about making their high end GPUs do the same? Especially since according to most reports it would still be lower than what their competition draws?

not to mention that doing one does not prevent the other... remember the GTX and the Ultra? the ultra was slightly overclocked, more power hungrey, mild benefit... higher price tag.

As I said, if you can push it further it is underclocked and underpriced.

 

[Sylvanas]

Originally posted by: dv8silencer
http://www.nordichardware.com/news,7845.html

Radeon HD 4870 will not suffer from microstuttering since the communication will be done through the memory instead of a bridge chip. This should also have a positive effect on the CrossFireX performance.

This would be pretty huge for all the opponents to Multi-GPU to which there are a few on this forum. It looks like the 4870X2 will beat the GTX280 by quite a margin and as Nvidia will probably not make a GTX280GX2 (due to ridiculous heat and power consumption of two of those chips) the 4870X2 will become the new Highest end offering.

I guess communications through the memory is a big reason for the use of Higher speed GDDR5, another 2 weeks .

 

[taltamir]

it is pretty huge, if this means what I think it does, then I see no reason not to get the 4870x2... now, it should still suffer from the same problem when doing 2 x 4870x2... but still, very nice.
It just remains to be seen how well it performs compared to the GTX 280...

 

[Extelleron]

Originally posted by: golem
Extelleron - I'm not sure about that... AMD has processors that span energy usage from low (BE chips) to very high (faster phenom and x2 chips). Their high end phenoms and x2 suck quite a bit of power, why would the be shy about making their high end GPUs do the same? Especially since according to most reports it would still be lower than what their competition draws?

The HD 4870 X2 is that chip. It will have ~250W TDP and be more concerned with performance than power consumption.

But HD 4870 is not high end, it is mainstream, and I'm pretty sure AMD didn't want a ~200W TDP for it. Higher power consumption increases board costs and would require an 8-pin connector instead of 2x 6-pin. Also, higher clocks lower the yield because not all chips will be able to hit those clocks.

And as I said, AMD would not be able to put out a HD 4870 X2 with 850-900MHz clock if HD 4870 @ 750MHz is 157W already. So if HD 4870 was 900MHz+ and HD 4870 X2 was <800MHz, the 4870 would be cutting into the X2's performance.

 

[bryanW1995]

they are going to sell a LOT more 4870's than 4870 x2's, however. if they could charge, say, $399 instead of $299 for the 4870 they'd be foolish not to.

 

[Extelleron]

Originally posted by: bryanW1995
they are going to sell a LOT more 4870's than 4870 x2's, however. if they could charge, say, $399 instead of $299 for the 4870 they'd be foolish not to.

If they priced the HD 4870 @ $399, nobody would buy it... they would buy the GTX 260 for $50 more or the HD 4870 X2 for $100 more.

Even HD 4870 @ 900MHz would not compete with the GTX 260. So they cannot price it so close. $349 would be the max they could price it unless they were clocking it at 1GHz+, which would probably not give good yields as I doubt AMD could get that many RV770's to run 100% stable at that speed.

At $299 though, the HD 4870 should give excellent performance and I have a feeling it will sell a lot of units. nVidia really won't have anything to compete in that price range, and from what we are seeing HD 4870 X2 will dominate the high end as well.

 

[Keysplayr]

Originally posted by: Extelleron

Originally posted by: bryanW1995
they are going to sell a LOT more 4870's than 4870 x2's, however. if they could charge, say, $399 instead of $299 for the 4870 they'd be foolish not to.

If they priced the HD 4870 @ $399, nobody would buy it... they would buy the GTX 260 for $50 more or the HD 4870 X2 for $100 more.

Even HD 4870 @ 900MHz would not compete with the GTX 260. So they cannot price it so close. $349 would be the max they could price it unless they were clocking it at 1GHz+, which would probably not give good yields as I doubt AMD could get that many RV770's to run 100% stable at that speed.

At $299 though, the HD 4870 should give excellent performance and I have a feeling it will sell a lot of units. nVidia really won't have anything to compete in that price range, and from what we are seeing HD 4870 X2 will dominate the high end as well.

What'd I miss? What is it that we are seeing? (Or that I'm not seeing?)

 

[manko]

Originally posted by: keysplayr2003

What'd I miss? What is it that we are seeing? (Or that I'm not seeing?)

You missed the speculation and rumors that the 4870X2 will beat the GT280 (in August?) at $500 vs. $649 for the GT280. So, as the story goes, ATI might possibly have the fastest card for a bit, until Nvidia's refresh/die-shrink or if they can come up with a GT260X2 or GT280X2.

 

[Keysplayr]

Originally posted by: manko

Originally posted by: keysplayr2003

What'd I miss? What is it that we are seeing? (Or that I'm not seeing?)

You missed the speculation and rumors that the 4870X2 will beat the GT280 (in August?) at $500 vs. $649 for the GT280. So, as the story goes, ATI might possibly have the fastest card for a bit, until Nvidia's refresh/die-shrink or if they can come up with a GT260X2 or GT280X2.

Oh, ok. I thought somebody had something good to check out.

Tanx.

 

[BFG10K]

800 SPs? Okay, now I'm quite interested.

Radeon HD 4870 will not suffer from microstuttering since the communication will be done through the memory instead of a bridge chip

I can't see how this will help micro-stutter. Micro-stutter seems to be caused by the timing of the frames in relation to each other, not by inter-GPU communication.

it is pretty huge, if this means what I think it does, then I see no reason not to get the 4870x2

Multi-GPU issues. Even if the 4870X2 is faster, cooler and cheaper than a GTX280 it won't be as robust as the latter is a single card solution and doesn?t rely on multi-GPU scaling.

That and we still don?t know what (if any) new AA modes the R[V]7xx brings to the table; ATi are generally behind nVidia in this area at the moment.

 

[taltamir]

that is why I am saying "if"... I don't believe they solved the major issues. Unless I see proof that it has I will take all those "perfect CF" claims with a bucket of salt.

 

[trajan2050]

Many reviewers have found major issues with Crossfire and refused to endorse the 3870x2 for those reasons. It does work well in some games, however.

 

[allies]

I don't think there's anyway in hell that the R770 has 800 SPs. That "AMD" slide looks photoshopped.

 

[Quiksilver]

Originally posted by: allies
I don't think there's anyway in hell that the R770 has 800 SPs. That "AMD" slide looks photoshopped.

It most certainly is...

 

[Extelleron]

Originally posted by: allies
I don't think there's anyway in hell that the R770 has 800 SPs. That "AMD" slide looks photoshopped.

What makes you think it will not have 800SPs?

The die size is enough to accomodate it. AMD's SPs do not take up much die area.