ATI Radeon HD 4000 Series discussion

[zerazax]

I have a hard time believing they'd randomly always gray the second best card... where is the logic there?

Sorry but I don't think so.

The logic is that it shows that hey, a cheaper *mainstream* card (the 4850 is priced awfully close to where the 2600xt was on release) can easily outperform a expensive CPU by a huge margin

[annihilat0r]

Sorry, but you are the one missing things. My point is, the mere existence of a shader clock in a separate domain would throw a wrench in the works of the calculation you just made, the number of SP's and also the slide itself being valid.

I'm telling you there is a separate clock domain

Figure the rest out for yourself.

How do you know there is a separate clock domain? Where'd you get that from? Information directly from AMD? Are you an insider?

[zerazax]

Sorry, but you are the one missing things. My point is, the mere existence of a shader clock in a separate domain would throw a wrench in the works of the calculation you just made, the number of SP's and also the slide itself being valid.

I'm telling you there is a separate clock domain

Figure the rest out for yourself.

Please then show us some proof or evidence besides going off whether a slide is aligned properly

[perkam]

Please then show us some proof or evidence besides going off whether a slide is aligned properly

When VR-Zone did their previw of the 4850 by showing us CCC info on it, there were no shader clock options, but that doesn't mean there won't be on future driver releases.

Chiphell had a GPU-Z shot of the 4870 but it was covered up for the most part and until we get a clear shot of that, we won't know for sure.

Perkam

[zerazax]

I'll offer one more piece... and this was posted in this very own thread but dismissed as unlikely it went from 64 * 5D to 160 * 5D...

Now that perkam mentioned it... remember that GPU-Z shot that showed RV770 at 256mm^2 and people were saying it might be fake since how can a database program know the area size unless its programmed within the BIOS? Well we all know that w1z gets a lot of this knowledge before we often do for GPU-Z (and the GPU-Z shots of GTX280 all have info there as well that might not be in the card).

Anyways, on the techpowerup forums almost an entire month ago, w1z posted this:

http://forums.techpowerup.com/showpo...9&postcount=36

see the yellow thingies inside the red shader blocks? the big question is how many of these are in the rv770 per red block.

if it's 5 then 800 is correct (160 * 5 = 800)
if it's 3 then 480 is correct (160 * 3 = 480)

He already hinted at 160 with that post since at the time everyone was saying it can only be 96 * 5D for 480... now all of a sudden 160 * 5D is very possible

[Extelleron]

Rather than arguing over whether that slide is true, (let it rest people, you will know for sure in two weeks) we can at least see the logical side of this.

At this point nobody knows for sure, but I would think that RV770 either has a seperate shader clock or more than 480SP. If RV770 had only 480SP clocked at 750MHz, that would give it 720 GFLOPs, only 45% more shading power than the HD 3870. I highly doubt that AMD would launch a card with such weak shading power compared to the previous generation.

While it's true that RV670 needs more texture power more than it needs additional shading power, it does not dominate nVidia's G92 parts in shader power. AMD's 5-way shaders are not as efficient as nVidia's. So I think AMD needs more shading power as well, especially since games will continue to rely more on shaders as time advances and DX10 becomes more common.

It would also be a complete departure from ATI's mindset since the launch of RV530 and the subsequent launch of R580 in January '06. They have always prioritized shading power over texture power, and I can't see this changing now.

I also can't see R700 performing the way it does (~X5500, beating GTX 280) if it did not have a significant amount of shading power compared to the HD 3870 X2/RV670 parts. 3D Mark Vantage is very reliant on shading power; this is why the HD 3850 can equal the 9600GT in Vantage. The 9600GT's texture advantage makes it faster elsewhere, but in Vantage its 64SPs are a big bottleneck. So how would performance have increased by more than 2x (comparing HD 3870 X2 -> HD 4870 X2) if shading power only increased by 45%?

[annihilat0r]

Extelleron: But the 800 SP case doesn't make much sense, too. It means a 2.5x increase in shading performance, while texture fillrate only increases less than 2x. This makes no sense, because R600 and RV670 were mainly texture fillrate bound.

[Extelleron]

Extelleron: But the 800 SP case doesn't make much sense, too. It means a 2.5x increase in shading performance, while texture fillrate only increases less than 2x. This makes no sense, because R600 and RV670 were mainly texture fillrate bound.

It might be possible that 32 TMUs @ 750MHz is enough texture power for modern games and additional texture power would prove unnecessary.....I don't know if that makes any sense, but it's a possibility. Remember that 32 TMUs @ 750MHz gives it significantly more texture power than G80 parts.... so is it really not enough?

Also remember that AMD developed the specifications for R700 parts years ago.... perhaps at that time they thought that games would be extremely shader bound by 2008 and that they would not need as much texture power.

It could be that the shader power is for another purpose as well. Additional shader power could be there to make physics processing possible without hurting performance that much. It's also likely that AMD wants to have good performance in apps that use the shaders.... like F@H and other commercial applications for GPUs. These would not care about texture units, only about stream processors.

[madcho]

The die is very small, and 800SP is too much.

[Extelleron]

The die is very small, and 800SP is too much.

That's my #1 problem with the 800SP rumor, but it's not entirely impossible.

Comparing RV635 to RV670 (same 55nm process) RV670 is 63% larger but has 2.67x the number of SPs, 2x the number of TMUs, and 4x the number of ROPs, in addition to larger caches, plus twice the ringbus size (512-bit internal for RV670, 256-bit internal for RV635).

It's not entirely impossible that RV770 is 34% larger than RV670 with 2.5x SP/2x TMU/ same number of ROPs/same ringbus.

[AliG]

The die is very small, and 800SP is too much.

see that's the thing, its not 800 SP, its 160 shaders that can do 5 operations, so the shaders actually don't take up all that space (especially not with the improved 55nm process), just look at the rv670, 170mm*2 with "320" shaders and the same 256 bit memory controller and is so small even g92b is supposed to bigger than it. Meaning that ati's shaders flat out don't take up much space (compared to nvidia's anyways).


But regardless, I stand by Extelleron, either there is a shader domain/difference in shader and core clocks, or the rv770 has "800" shaders, as it is confirmed the 4870 will have 1 tflop+ of processing power. Also g80 had 32 TMUs that were lower clocked, and yet the 8800ultra still outperforms just about all single gpu setups with max details (and even the 9800gx2 and 3870x2 in some cases). So while no where near what gt200 will have, 32@750mhz may be enough for current games.

Who knows why it has 32 TMUs (though I hear the r600 architecture was designed to only have a max of 32), as the r600 was in a designing process for something like 4 or 5 years and thus we can assume the r700 also was planned out years ago. You always build for the future, so who knows, maybe we were supposed to have shader heavy games but due to nvidia finishing g80 first we didn't

[mascaras]

R700 will be in about same price as one gtx280...

according to last rumors :

HD4870X2 = 499$
GTX280 = 649$



GTX260 = 449$
HD4870 = ~329$
HD4850 = 199$



regards

[NH|Delph1]

Since I was the one who posted the news you're arguing about maybe I should say something.

The source of the slide has been 100% in the past. I see no reason to doubt him.

Everything else on the slide looks legit.

Those of you who claim the slide to be fake because of the slant in the last column ... that's just hilarious
The colors make sense to me since 4870 is low-high-end, 4850 mainstream, and 3870/s870X2 last generation and then we have CPU. White, gray, white/twhite, gray.

According to the source, AMD planted the stories about 480SPs. If you go back you will see that 480SPs has been posted by the German site who first claimed to have Radeon HD 4000 series specifications (most were wrong), Fuad and Theo. I've never confirmed 480SPs, but been fed this information from several Western sources, but at the same time my Asian sources have told me 800SPs. Those of you who follow my rants should know this.

I don't think there is a separate shader domain, but I can't provide evidence for that.

//Andreas

[annihilat0r]

according to last rumors :

HD4870X2 = 499$
GTX280 = 649$



GTX260 = 449$
HD4870 = ~329$
HD4850 = 199$



regards

I wonder how two 4870's will cost only $500 where one costs $350?

[Extelleron]

I wonder how two 4870's will cost only $500 where one costs $350?

The rumored prices put HD 4870 at $299, meanwhile HD 4870 X2 would be $499.

That's a bit different than what happened with the HD 3800 series... HD 3870 X2 debuted at $449 and HD 3870 at $219.

It's likely that AMD wants to undercut the GTX 280 and wants the high-end market.... if HD 4870 X2 @ $499 outperforms the $649 GTX 280, then which one will people buy?

[cegras]

I wonder how two 4870's will cost only $500 where one costs $350?

X2 isn't selling two cards, it's two chips on one pcb. You're saving money there .. not sure how much. Besides, I'm sure the 350 price tag is already netting them some sort of profit, and they're just narrowing it with the X2. You don't have to preserve a specific % of profit for each product .. that doesn't make sense.

[annihilat0r]

Ignorant people will buy 4870x2 because they have no idea about the stuttering issues of CF; and will drool at the FPS numbers whereas their gaming experience is no way near that FPS.

That's of course, unless ATI has magically solved the problem, made all frames nicely sync'ed and with no loss of FPS in comparison to regular AFR.

[Extelleron]

Ignorant people will buy 4870x2 because they have no idea about the stuttering issues of CF; and will drool at the FPS numbers whereas their gaming experience is no way near that FPS.

That's of course, unless ATI has magically solved the problem, made all frames nicely sync'ed and with no loss of FPS in comparison to regular AFR.

Will you ever stop spreading this BS?

Good luck convincing all those people with CF/SLI that their improved gaming experiences are BS, and you are right despite the first hand experiences of so many others.

And Anandtech is completely lying to us, too:

In fact, during our testing it was very easy to forget that we were dealing with a multi-GPU board since we didn't run into any CrossFire scaling or driver issues.
The Radeon HD 3870 X2 is the most elegant single-card, multi-GPU design we've seen to date and the performance is indeed higher than any competing single-card NVIDIA solution out today.

[STaRGaZeR]

Will you ever stop spreading this BS?

Good luck convincing all those people with CF/SLI that their improved gaming experiences are BS, and you are right despite the first hand experiences of so many others.

Have you ever used SLI/CF?

[AliG]

Since I was the one who posted the news you're arguing about maybe I should say something.

The source of the slide has been 100% in the past. I see no reason to doubt him.

Everything else on the slide looks legit.

Those of you who claim the slide to be fake because of the slant in the last column ... that's just hilarious
The colors make sense to me since 4870 is low-high-end, 4850 mainstream, and 3870/s870X2 last generation and then we have CPU. White, gray, white/twhite, gray.

According to the source, AMD planted the stories about 480SPs. If you go back you will see that 480SPs has been posted by the German site who first claimed to have Radeon HD 4000 series specifications (most were wrong), Fuad and Theo. I've never confirmed 480SPs, but been fed this information from several Western sources, but at the same time my Asian sources have told me 800SPs. Those of you who follow my rants should know this.

I don't think there is a separate shader domain, but I can't provide evidence for that.

//Andreas

I as usual will stand behind you all the way Andreas, and you're sources generally have been right. But don't forget, you forget to mention that ati has generally been tenetive with initial launches, but goes all out with the upgraded version. What I mean by this is look at the x1800-->x1900 transition. The x1800 failed to compete with the 7800gtx, but the x1900 spanked it by tripling the number of pipelines. I think 800 shaders is very possible. But either way, I'm certain we can assume that which ever is true, the other was put out there as a plant to catch the leaks

[Extelleron]

Have you ever used SLI/CF?

No, but a lot of other people have.

And I don't see them complaining about performance or issues. Like always, it's a few people who have problems against the silent majority that are happy with their experience.

If SLI/CF were so bad, and there was no real world benefit, then we would have seen complaining beyond a few people. SLI/CF solutions are a solid way to improve performance; they are not always ideal and do not scale 100%, but that is just something you have to deal with. Plenty of games see near 100% scaling with the 3870 X2 or the 9800 GX2.

[Rammsteiner]

Ignorant people will buy 4870x2 because they have no idea about the stuttering issues of CF; and will drool at the FPS numbers whereas their gaming experience is no way near that FPS.

That's of course, unless ATI has magically solved the problem, made all frames nicely sync'ed and with no loss of FPS in comparison to regular AFR.

So 2x HD 4870 wouldn't suffer from this?.

You could go on a rampage about everything, audiophiles could do so about X-Fi. But if we really notice, or at least notice enough to complain, I dont know. I hardly heard anybody over it anyway.

[STaRGaZeR]

No, but a lot of other people have.

And I don't see them complaining about performance or issues. Like always, it's a few people who have problems against the silent majority that are happy with their experience.

If you have not used it I don't know why are you saying X or Y about SLI/CF is BS or not. Let those people speak by themselves.

The problems IS there for EVERYBODY, the truth is that. If 99 of 100 users of SLI/CF tell you they are happy with their system that makes the one who's saying there is a problem and has proven it, etc. wrong?

And like almost all in this world the majority is stupid. SLI/CF users here know what they are doing, but most users don't. Also lots of them know nothing about this kind of things, they only know that a bigger number reported by fraps is better (like you by your comments, sorry if I'm wrong). If you and the majority can't see micro-stuttering, congrats, but it's there, it's not BS and has to be fixed. I bet you whatever you want that if I put two rigs, CF and non CF without telling which of them is which, and then I tell them what micro-stuttering is, they will see it in the CF rig. Misinformation is the rule in the computer world.

Seriously, you won't believe it until you try it. I was the same as you. Never trust what the majority says.

Also, micro-stuttering is only one of the problems of current multiGPU solutions.

[ThatGuy16]

I just sold one of my cards to start saving up for the 4870, or two. From the looks of it, i can't wait to get my hands on one of these.

I have never had the so called "shuttering" issues with any of my games, and have been more than pleased with its scaling and results.

[mascaras]

So 2x HD 4870 wouldn't suffer from this?.

according with this rumors :

Radeon HD 4870 X2 card looking good


Got some inside info from a source who doesn't wish to be named.

HD 4870 X2 has 1024MB GDDR5, 2x 256bit memory interface, 1050MHz shader clock, 1800MHz memory clock. Core clock is defined by the AIB partners and ASUS will have the highest clocks.

While Radeon 3870X2 relys on the PLX chip to communicate between the GPUs, 4870X2 GPUs will comunicate with each other through the memory. Since the GDDR5 is clocked at 1800, the total bandwidth will be roughly 160 GB/s at 1 Gigabyte through the 256bit-bus, compaired to the 8 GB/s of the PLX chip of the HD3870x2. Also 4870x2 WILL NOT have micro studdering.

http://forums.vr-zone.com/showthread.php?t=285719

but i think HD4870X2 also comunicate between the GPUs through PLX chip (like HD3870X2 ) , however has more bandwith ... lets wait and see !

regards