Official AMD Radeon 6000 Series Discussion Thread

[pantherx12]

On topic what for me is the most important that i hope ati gives the lower models back the double precision capabilty since all models below 58XX lack them now.
Making them useless for my personal use, i hope the 6XXX series will get a bit faster memory or 256 bit wide bus as was leaked but time will tell

Yo!

Actually they all still do aside from 5750 and 5770, weird eh?

 

[Bronan]

LoL

preying they've improved crossfire scaling...

I think scaling does work for 2 cards the problem i see with scaling is using more cards the more the less performance gain.
For now we see very good result when using 2 cards, at 3 cards you get most of the time less then 10% gain and with 4 about 5% or less performance boost (i can't give exact figures because all hardware setups are a bit different)

But my gut feeling says its not the hardware vendors who are the blame for not scaling good, i think the game developers do not really work on this issues yet or do not have a solution to solve it.

Because fact is software development leaps far behind hardware improvements many of us use x64 windows but we hardly have any x64 programs which make full use of the given extra power.

So on topic even though most of those pictures can be photoshop results fact is that amd have allready given certain information themselfs, the new cards will be 40nm not 28 so it would be a huge surprise if they do.
Second amd have stated this is a new design
So what means a new design is what we can ask ourselfs, do they mean by that the 256 bit memory bus or have they improved something on the core or on the pcb
Everything is blurred till we finally will see the real ones appear on the test beds of all the major hardware sites.
Because even if the get released this year i seriously think you won't be able to buy one before next year or at a very steep price tag.

Actually they all still do aside from 5750 and 5770, weird eh?

What do you mean only the 5830,5850,5870 and 5970 have double precision non of the lower models can do DP

 

[pantherx12]

Whoops, was trippin' the new 6 k cards supposedly all do XD

 

[Bronan]

[opens wishfull thinking] I really would be impressed [closing wishfull thinking] xD

 

[Berlusconi]

All of these discussions are making me anxious.

 

[Berlusconi]

Test

 

[cadaveca]

LoL
I think scaling does work for 2 cards the problem i see with scaling is using more cards the more the less performance gain.
For now we see very good result when using 2 cards, at 3 cards you get most of the time less then 10% gain and with 4 about 5% or less performance boost (i can't give exact figures because all hardware setups are a bit different)

I'm currently experiencing NEGATIVE scaling...adding second card gives LOWER framerates than a single card.

It's a cpu limitation, for sure...I only get negative scaling due to each card being under 50% load or less...and clearly, one card can easily handle the 50% from each gpu.

And this is even up to 8xAA...

So, I do not hold the same confidence you do that this situation will improve. To me, scaling, and ahrdware-based cursor corruption...are the two things that MUST be addressed. I think they can fix the cursor issue... I think that's why 6-series cards now have 2xDP connectors...but the scaling issues...I can only see that being fixed if they have changed the shader arrangement.

 

[Berlusconi]

Test

Sorry for that, I suppose it was the easiest way for me to see if my signature was showing or not.

 

[pantherx12]

Sorry for that, I suppose it was the easiest way for me to see if my signature was showing or not.

For future reference, that's precisely what the preview button is for. it's next to the submit button when your on the sig section of the CP.

 

[Berlusconi]

For future reference, that's precisely what the preview button is for. it's next to the submit button when your on the sig section of the CP.

I'm new here, please forgive me.

 

[wolf]

...scaling, and hardware-based cursor corruption...are the two things that MUST be addressed

agreed, Nvidia can afford to make slower cards if they scale better. just like how a single GTX460 reference model is clearly behind a 5850, but two together touch on 5970/5850CF performance. pre overclocked models tend to match or exceed a 5970 where it would take a GTX460 overclocked to the brink (~850 core ~2000 mem) to match a 5850 by itself.

...but the scaling issues...I can only see that being fixed if they have changed the shader arrangement.

If by arangement you mean the architecture, I'm not so sure. changing them may well help all around though.

I think, as shown by many threads, they need to increase the ratio of ROPS to SP's. 32 ROPS to 1600 sp's seems out of proportion to me.

 

[pantherx12]

I'm new here, please forgive me.

Ahh I wasn't having a pop, just letting you know

 

[cadaveca]

I think, as shown by many threads, they need to increase the ratio of ROPS to SP's. 32 ROPS to 1600 sp's seems out of proportion to me.

As far as I understand ATI's tech, the only reason ROPS and such are increasing in 6770 vs 5770, is the move to 256-bit. Thier design dictates that the only way they can increase memory bus size is to increase ROPs, etc.

With this in mind, 6770 seems like 4890, with a few more shaders, a different tesselation engine, and DX11. I do not see this gpu being that much faster than 4890.

On the same thinking, ROP numbers are unlikely to increase in 6870, unless memory bus increases as well. To me, it only makes sense for them to boost shader numbers, and tehn tweak the dispatch processor to keep the shaders filled.


The dispatch processor is where most of teh performance improvements will come from, IMHO.


But really, I'm straight out guessing. But most options present in rumour seems realyl far-fetched, especially considering R600 hype. I'll not get bitten by the hype again.

 

[pantherx12]

As far as I understand ATI's tech, the only reason ROPS and such are increasing in 6770 vs 5770, is the move to 256-bit. Thier design dictates that the only way they can increase memory bus size is to increase ROPs, etc.

I disagree


If you consider the this.

4890 pixel fillrate 13.6
5770 Pixel filrate 13.6

Texture filrate is 34 on both cards

The only difference is memory bandwidth supposedly. 124.8(4890) vs 76.8

Sure the extra rops/shaders AND bus-width should equate to at-least a 30% difference? ( I expect a massive improvement in tessellation though)


Also, just for fun 5770 @ 960/1445 =15.4 P.F.R 38.4 T.F.R Bandwidth 92.5

 

[wolf]

As far as I understand ATI's tech, the only reason ROPS and such are increasing in 6770 vs 5770, is the move to 256-bit. Thier design dictates that the only way they can increase memory bus size is to increase ROPs, etc.

Yeah ROP's are very closely linked to the memory bus in both cards from ATi/AMD and Nv. This is why when you see a cut down memory bus you also see ROP's cut too, like for example;

GTX480, 48 ROPs, 384-bit bus
GTX470, 40 ROPs, 320-bit bus

GTX460 1gb, 32 ROPs, 256-bit bus
GTX460 768mb, 24 ROPs, 192-bit bus

and so on and so forth, twas exactly the same with GT200.

However consider that the whole GPU can always be redesigned to make ROPs and memory bus link differently, for isntance they could make 64 ROPs and a 256-bit memory interface if they wanted to. at least I think they can...

The dispatch processor is where most of teh performance improvements will come from, IMHO.

An interesting thought, as it seems they will also tweak shader structure, they may well have to tweak the dispatch processor to better take advantage of them too.

 

[cadaveca]

I thought adding this here now might prove interesting:

http://www.anandtech.com/show/2937/1


What 6870 might be, is the original 5870.

Eric was telling me about how they trimmed down 870 from over 400mm2 down to 334mm2 and how wonderful the end product was. I stopped him and asked for more detail here. I wanted an example of a feature that they had to throw out but they really wanted to keep in. Manufacturers rarely tell you what they threw out, marketing likes to focus on what’s in the chip and make everything sound like a well calculated move. Thankfully, marketing wasn’t allowed to speak at my dinner.

 

[Super XP]

Great link. Perhaps that explains the rumours with the names HD 6980 & HD 6990 along with what that's familiar to us the HD 6870 & HD 6970.

 

[cadaveca]

That's probably one of the best articles done in quite some time, actually. I wish there were more, and hope to see the same thing done after the high-end 6-series launches.

I read this article the day it was written, and today, it seems even more relevant.

 

[20mmrain]

This was said to be a hoax.... but it was reported to wikipedia..... If these do turn out to be true. There is no way on god's green earth I will be buying one of these cards. At $999 for a 6970 or $599 for 6870 ATI could put their cards were the sun don't shine.

I do think this is a fake though. I don't see the 6870 having a 512 Bit memory bus. It also just looks crummy put together. But still take it for what you will.

I would stay with Nvidia for right now. But anyway here is this for what it's worth and were I found it from....

http://forums.guru3d.com/showthread.php?p=3711619

If this is real I will really have lost allot of respect for AMD/ATI. AMD/ATI used to be the company for the underdogs. If this is true they have forgot were they came from and all of those people who supported them through tough times *Cough* "2900XT"

 

[erocker]

This was said to be a hoax.... but it was reported to wikipedia..... If these do turn out to be true. There is no way on god's green earth I will be buying one of these cards. At $999 for a 6970 or $599 for 6870 ATI could put their cards were the sun don't shine.

I would stay with Nvidia for right now. But anyway here is this for what it's worth and were I found it from....

http://forums.techpowerup.com/attachment.php?attachmentid=38066&stc=1&d=1285020246

http://forums.guru3d.com/showthread.php?p=3711619

If this is real I will really have lost allot of respect for AMD/ATI. AMD/ATI used to be the company for the underdogs. If this is true they have forgot were they came from and all of those people who supported them through tough times *Cough* "2900XT"

That Wiki page is utter crap garbage. It has since changed. I just don't understand why people even bother with something that can be edited by any crazy person. As a matter of fact I edited it today and put "?'s" next to the 512 bit bus values. Wiki is worth nothing.

 

[20mmrain]

That Wiki page is utter crap garbage. It has since changed. I just don't understand why people even bother with something that can be edited by any crazy person. As a matter of fact I edited it today and put "?'s" next to the 512 bit bus values. Wiki is worth nothing.

Well I bother with it because.... it is a rumor and speculation. I understand it is fake most likely. I was just more bothered buy the prices.

That's all ....

I just report what I find.... people can choose to believe it or not. But I do apologize I didn't realize this graph has been discussed already or has been floating around for awhile.

But it does make me feel better that you have recognized it as a fake Erocker. I trust your judgement on it.

I also did notice the ???? on the graph as well. It did make me suspect something was up with it. I probably posted it before I really thought it through.

 

[erocker]

I just checked it again and the prices aren't even in there. I didn't do it though.. It seems to get edited quite often.

 

[20mmrain]

I just checked it again and the prices aren't even in there. I didn't do it though.. It seems to get edited quite often.

Feeeww.... I know these cards will be a little pricey and I can deal with that. I just saw these prices and it nearly drove me insane! It was my anger speaking.

I just took a look too.... (Quickly mind you) I don't even see the graph anywhere. So nothing to worry about thanks bud.

agreed, Nvidia can afford to make slower cards if they scale better. just like how a single GTX460 reference model is clearly behind a 5850, but two together touch on 5970/5850CF performance. pre overclocked models tend to match or exceed a 5970 where it would take a GTX460 overclocked to the brink (~850 core ~2000 mem) to match a 5850 by itself.

I would look again with that performance ratio compared to the 5850. With the latest drivers the GTX 460 1Gb is just a little lower then a 5850 and the 768 MB version is about on par with a 5830 or a little lower in some games.

Those new drivers bump up those cards a bit bud. Plus these drivers are still maturing. I wouldn't be surprised by the time they are done to see these cards 5850 and GTX 460 going neck and neck against each other. Of course the GTX 460 beating the 5850's in a multi card set up too..... shoot they already do that.

With my EVGA GTX 460's 768MB Super Clocked cards... I already get a GPU score of over 15K at my stock speed of 763 Mhz and in SLI I get a GPU score of 27K.

Not to get off the subject... I just wanted to correct the situation.

 

[Zen_]

Those wiki MSRP's were laughable.

Barts is targeting the 460. If it delivers on the performance, the XT being somewhere between a 1GB 460 and 5850, the Pro being better than a 768M 460 and 5830, you have to figure the Pro will be priced @ $150-170 and the XT priced @ $190-$210.

With those prices there is a bit of a hole where the 5750 and 4770 had been. In the short term I'm sure that 5770/5750 prices will but cut but in the long term maybe we'll see another 4770 type part (~$100) as AMD's first 32nm GPU, before Northern Islands. It would make sense based on precedent and what the Anandtech article said about AMD wanting to get process refinement before moving to the larger and more expensive dies. That's a long ways out though.

 

[pantherx12]

That Wiki page is utter crap garbage. It has since changed. I just don't understand why people even bother with something that can be edited by any crazy person. As a matter of fact I edited it today and put "?'s" next to the 512 bit bus values. Wiki is worth nothing.

I might edit it later and add "pics or it didn't happen" to the end of each table.

That ought to end any future edits!

lol