Official AMD Radeon 6000 Series Discussion Thread

[bear jesus]

here you go, just for you Tom, thought I would really go to town, GPU set at 900mhz, fan set manually to 50% only (still cannot hear it), Furmark for 8 minutes, forced 8xMSAA, 97% GPU load, temps don't go above 66C Now who's gonna tell me these DirectCU coolers don't do their job!

I would love a couple DirectCU coolers over the referance ones, my ear still hurts

 

[Tatty_Two]

I would love a couple DirectCU coolers over the referance ones, my ear still hurts

To be fair, Tom (TrT) caught a poor one, he replaced it with a Zalman, just goes to show that even the best fails at times.

 

[CDdude55]

Furmark for 8 minutes, forced 8xMSAA, 97% GPU load, temps don't go above 66C

That's awesome.

 

[trt740]

Well my Zalman doesn't do aswell but this gpu is hotter than most 5850 and this is in a solid sided case with no side vga fan. I have the same video card you have. My stock CU cooler didn't handle the heat well at all and would have been a bunch hotter. Having said this, this chip will do well past 1000 core.


5870 speeds

well past 5870 speeds

 

[trt740]

I would love a couple DirectCU coolers over the referance ones, my ear still hurts

mines for sale. My Direct CU cooler.

 

[Tatty_Two]

Still not bad Tom, maybe I got lucky with mine, I should have kept the first test I ran with furmark, GPU at 930Mhz, forced 8xMSAA, ran the fan manually at 100% (loud), temps never got above 58C

Edit: the max i can get out of mine is 1040mhz and thats on 1.3V, it is stable and temps are decent but I would not go beyond that personally on air. I do have an extra spot cooler fan blowing air down the PCB over VRM's etc plus I have a 250mm fan on the side case literally 2 inches from the card. As for memory, I have never tried beyond 1200mhz, there is so little performance improvement with memory beyond a certain point anyways and with this error checking business at a point things will actually slow down.

 

[trt740]

Still not bad Tom, maybe I got lucky with mine, I should have kept the first test I ran with furmark, GPU at 930Mhz, forced 8xMSAA, ran the fan manually at 100% (loud), temps never got above 58C

look at my ram ( at 1250 it will heat your cooler up not sure why but it does mine.) and sometimes hotter chips overclock better and in my case it is true. It isn't my stock cooler it's my chip that runs hotter.

 

[bear jesus]

mines for sale.

I would want 2 for the pair of 6870's i have but i am hoping that these are just holding me over until i can get a single card to game at 5040x1050, really i intend to leave these as they are and let the second owner/s do what they want to keep them cool.

look at my ram and sometimes hotter chips overclock better and in my case it is true.

I'm jealous of your ram speed

 

[trt740]

example with ram at Asus default for top

so slightly higher than a default 5850 4 c drop in temp

 

[mastrdrver]

damn you mail man! I kep clicking the post reply in your image, wondering wtf was wrong.

You could call me an "industry analyst". Like when I knew stuff was going on @ HP before it was public, etc...I know when memory manufacturers started producing memory, what bins they have, and availability of those bins. THAT is publically available info(DRAMEXchange). I subscribe to industry websites with "insider" info(pay-to-view websites). Add in a bit of common sense, and you've got the source of alot of my comments.

So any of that info you see say anything about whether TSMC has their stuff together finally for 40nm? Sorry if I missed it. Just kind of wondering what the "markets are saying" about it because at 400mm2 Cayman will get expensive quick if the yeilds are not there.

Another unrelated question: Are all your 5870 XXXs the low voltage ones? I've got one of those (regular XFX 5870 but low voltage) and found it to run slightly hotter than my higher voltage 5870. About 5C seems right. Just a thought.

But what does this actually mean?

Like how does this affect polygon per second?

Because TBH I've no clue seeing as I still see nvidia/ati cards being very close most of the time, so the extra triangles nvidia get per clock cycle most do shit all

Tessellation. You know the wire frame view in the Heaven benchmark? As you increase tessellation from moderate to extreme the GPU must generate more and more triangles. Aside from needing the shader power for all those triangles, you also need a GPU that must be able to turn out more than 1 triangle/clock otherwise you get what you see currently where ATI cards are brought to their knees the higher you go in that bench compared to nVidia. Of course this is an extreme example but it is where tessellation is heading toward. As such the up come Cayman is what I expected to see in Barts. That is ATI's version of nVidia's GPC (graphics processing clusters) where the tessellator (and some other pieces) are grouped together with shaders and texture units. So the more GPCs an nVidia card has the more triangles/clock it can generate and thus the higher the triangle count can be without slowing down the game.

cliff notes: The more triangles/clock a GPU can generate, the higher the tessellation levels can be without a performance hit (assuming the shaders can handle the increase without becoming overloaded).

Read

And if that screenie is accurate.... I was right in my prediction that everyone argued about (and mostly were against) for a week....... 32 ROP's Where's that Cheezburger character when you need him

I was thinking about that too when I saw that posted.

 

[wahdangun]

hd 6900 series is based on 4-http://img.donanimhaber.com//images/haber/23095/amdcayman4d_dh_fx57.jpgway arc

wow, is that true a VLiW4

so the cayman will use 4D shader architecture after all

and 2X previous gen, so it seems its really have same performance like HD 5970 and where is chezzburger guy, because i win the bet

 

[CDdude55]

What is exactly the benefit of having 4D shaders?

 

[erocker]

and 2X previous gen, so it seems its really have same performance like HD 5970 and where is chezzburger guy, because i win the bet

Wrong, I win. The GT 240 stays with it's rightful owner.

 

[wahdangun]

Wrong, I win. The GT 240 stays with it's rightful owner.

WTF, you said its 384 bit, so its not fair, i win the bet , so give me the damm card muahahahahaa

 

[Lionheart]

http://img.donanimhaber.com//images/haber/23095/amdcayman4d_dh_fx57.jpg
wow, is that true a VLiW4

so the cayman will use 4D shader architecture after all

and 2X previous gen, so it seems its really have same performance like HD 5970 and where is chezzburger guy, because i win the bet

Only 32ROPS I thought having less ROPS would starve the stream processors from being use to their full potenial or am I smokin weed

 

[erocker]

WTF, you said its 384 bit, so its not fair, i win the bet , so give me the damm card muahahahahaa

You were never part of the bet. Sorry.

 

[mastrdrver]

What is exactly the benefit of having 4D shaders?

Nothing in of itself.

Someone correct me if I'm wrong but doing some reading on some of the more technical forums I think the current 5-way goes under utilized. This is the architecture change that has been rumored now for a couple years at least. At least part of the change. The new 4-way shaders can do everything that the old 5-way could do so no functionality is lost, but with now they will have a higher utilization.

At least that's what I've come to understand. They talked a lot about this on the B3D forum thread a few months back.

Wrong, I win. The GT 240 stays with it's rightful owner.

Are we limited to this thread because I made a bet a while back in one of the news threads.

 

[btarunr]

No offense, man, but that's exactly what I said in the following paragraph...high-end is about performance, #1. High-end users will go that extra mile and spend yet more on cooling, if needed...

I then went on to say it's only disappointing becuase the process is holding the design back. I don't get what YOUR point is, other than reinterating what I already said.

However, you've revealed your own disappointment in Fermi...but the only thing wrong with Fermi was the management behind it. Is that your point?

No, I meant that big die and high power doesn't always mean fail. Core i7 lived up to it, GTX 480 did not.

I'll put it into perspective:

Radeon HD 5970, which has two Cypress GPUs, 4 billion transistors between them, and 16 GDDR5 memory chips, has lower power draw and higher performance compared to GTX 480, which has one GF100 GPU, 3 billion transistors, and 12 GDDR5 memory chips. Both Cypress and GF100 are built on TSMC 40 nm process.

Core i7 Bloomfield and Lynnfield have some of the largest dies among quad-core chips, even the slowest Bloomfield Core i7 has 130W TDP, yet it has the highest performance among quad-core chips and beats AMD in performance per Watt.

 

[cadaveca]

You cannot compare Intel's process to TSMC's. Intel got 32nm working fine...TSMC gave up and skipped it. It's all about the silicon quality. And that's MY point. Whatever AMd comes up with, no matter how good, and how well they deal with the issues, will still be held back by TSMC's silicon. That ahs nothing to do with AMD...the fact they've brought so much success over the past 13 months shows how capable they can be, but they need new tools before that true skill can shine.

If it was 32nm....like it was meant to be...

 

[mdsx1950]

No, I meant that big die and high power doesn't always mean fail. Core i7 lived up to it, GTX 480 did not.

I'll put it into perspective:

Radeon HD 5970, which has two Cypress GPUs, 4 billion transistors between them, and 16 GDDR5 memory chips, has lower power draw and higher performance compared to GTX 480, which has one GF100 GPU, 3 billion transistors, and 12 GDDR5 memory chips. Both Cypress and GF100 are built on TSMC 40 nm process.

Core i7 Bloomfield and Lynnfield have some of the largest dies among quad-core chips, even the slowest Bloomfield Core i7 has 130W TDP, yet it has the highest performance among quad-core chips and beats AMD in performance per Watt.

+1

Couldn't agree more.

 

[Steevo]

Mine stays less than 40C even at 1050 core, 1.35 volts.


I will never air cool again.

 

[20mmrain]

Unfortunately, this is not true. I'd say anything over 90c would be hot. My reference 5-series cards ran @ 90c++, and I had many of them.


Only my newest non-reference runs lower than 90C, and not by much. Claiming anything over 75C is hot, is spreading FUD.

well you had bad luck my friend or i have much better airflow in my case. I owned several of the 5000 series cards. Including 5870's 5850's and even a 5970.
They were all reference for the most part except the last couple I had.

The only card I saw get up to 90c with stock fan was the 5970 and that was during Furmark.

The 5870 never got above 75c in my case with reference cards. The only times I saw my cards get as hot as your talking about is when overclocking.

I have also owned fermi cards.... GTX 460's and GTX 470's. While the GTX 460's weren't that bad they were warmer on average then my 5000 series cards. (W/ exception of the 5970)

The GTX 470's in SLI were very warm. Until they helped that with Driver updates and the like. But even after all that they still ran very very warm.... it was not uncommon seeing the top GTX 470 to reach 85c to 95c depending what you were doing.

 

[wahdangun]

well you had bad luck my friend or i have much better airflow in my case. I owned several of the 5000 series cards. Including 5870's 5850's and even a 5970.
They were all reference for the most part except the last couple I had.

The only card I saw get up to 90c with stock fan was the 5970 and that was during Furmark.

The 5870 never got above 75c in my case with reference cards. The only times I saw my cards get as hot as your talking about is when overclocking.

I have also owned fermi cards.... GTX 460's and GTX 470's. While the GTX 460's weren't that bad they were warmer on average then my 5000 series cards. (W/ exception of the 5970)

The GTX 470's in SLI were very warm. Until they helped that with Driver updates and the like. But even after all that they still ran very very warm.... it was not uncommon seeing the top GTX 470 to reach 85c to 95c depending what you were doing.

i think cadeva was just having extremely bad luck with Ati

 

[mdsx1950]

i think cadeva was just having extremely bad luck with Ati

Agreed. NO ONE can have so much of bad luck. Cadeva, you're clearly cursed lol.

 

[wolf]

Only 32ROPS I thought having less ROPS would starve the stream processors from being use to their full potenial or am I smokin weed

could be both, I'm going with both.