Anandtech 6850 comparison is up

[HurleyBird]

GaiaHunter, you are driving me insane!

I'm saying that some don't need wider buses to get higher bandwidth. If you believe that 96% at stock is not efficient, well...

Efficient? What are you talking about? I'm saying that no high end GDDR5 card has launched with it's rated memory and it's foolish to expect Cayman to be any different. Unless 96% is suddenly the same thing as 100% your comment is irrelevant.

And of course the Gigabyte 5870 Super Overclock comes with 5GHz from the factory, but I guess they don't run stable either.

AMD could have easily made the stock HD5870 have a 1250MHz mem clock as well. Hell, they could have set the stock clocks at 1GHz for core! Only problem is, hardly any chips would be able to make those qualifications. These things work good for low volume OC editions, but not for high volume stock products.

And you can't disregard other ways to reduce bandwidth requirements, like cache.

I actually went out of my way to regard that: "Obviously, with Cayman looking to be a different or at least highly modified architecture anything can happen"

Look at the GTX480 - it only has 11% more bandwidth than the GTX285.

Yes it is. And yes that is a chip that did make changes IRT cache. It's also a chip with thermal, size (512-bit + ecc might need a bigger die), and technological limitations (TSMC issues, Nvidia being less experienced with GDDR5). In any case, what does this have to do with Cypress, Barts, or Cayman?

 

[Idontcare]

Wrong thread, wrong sub-forum, first and last warning.

Moderator Idontcare

We are in the same boat. As a moderator, I'm not allowed to ignore you either.

You can, however, stop trolling and baiting though.

Gentlemen,

This manner of back-and-forth escalation is unproductive and entirely beyond the scope of this thread as well as this sub-forum.

Please take it outside (PFI, pm, or Mod Discussions).

We all thank you for being considerate and mindful of the rest of the thread's readers by taking this reminder under consideration when making posts in VC&G going forward.

Moderator Idontcare

 

[happy medium]

These results dont match the ones from Xbit at all.?
In fact the gtx460 768 mb model ties the 6850.
http://www.xbitlabs.com/articles/video/display/radeon-hd6870-hd6850.html

The 6850 needed a 1015 core clock to equal a 6870 and got beat easily by a overclocked gtx460.
http://www.hardwarecanucks.com/foru...d-6850-vs-gtx-460-1gb-overclocking-study.html

Here is the discussion.
http://forums.anandtech.com/showthread.php?t=2116158

Ryan must have used the wrong lower quality settings for the AMD cards.

 

[Throckmorton]

Yes, it has been applied. XFX provided that fix to us before the 6850 even launched. There's nothing "wrong" with the XFX 6850 - its fan scales up and down correctly. It's just tuned for cooling over noise.

Hmmm... is it still possible to adjust fan speeds? I remember Powerstrip had that ability but a lot of stuff in Powerstrip doesn't work anymore thanks to Windows 7/Vista.

 

[psoomah]

GaiaHunter, you are driving me insane!

The five word reply an opponent relishes beyond all others.

Well done sir!

 

[3DVagabond]

Good article, but no OC NVIDIA cards there?

In 1 article where AT review stock cards and don't even approach OC of that product at all you compare them to OCed competition cards.

In 1 article where AT review OC cards and goes an OC cards some more, AT sticks with NVIDIA stock cards.

Go figure.

Maybe he was figuring "turnabout was fair play" since in the original article they had stock 6800 against O/C 460. Besides, if the 6850 can be shown as faster than the 460 it'll make 460 pricing come down even more.

 

[GaiaHunter]

GaiaHunter, you are driving me insane!

Efficient? What are you talking about? I'm saying that no high end GDDR5 card has launched with it's rated memory and it's foolish to expect Cayman to be any different. Unless 96% is suddenly the same thing as 100% your comment is irrelevant.

AMD could have easily made the stock HD5870 have a 1250MHz mem clock as well. Hell, they could have set the stock clocks at 1GHz for core! Only problem is, hardly any chips would be able to make those qualifications. These things work good for low volume OC editions, but not for high volume stock products.

I'm saying it is a 4% OC over stock. Are you telling me that 5870 can't do 4% OC on memory?

I guess due to tsmc 40nm problems some of the/a few of the 5870 chips required more voltage than required and it was too problematic/time consuming to bin for 5GHz GDDR5.

Otherwise it seems the controller has few problems to reach 5GHz. Now try to get 5GHz with Barts or GF100/GF104 controller.

 

[ViRGE]

Hmmm... is it still possible to adjust fan speeds? I remember Powerstrip had that ability but a lot of stuff in Powerstrip doesn't work anymore thanks to Windows 7/Vista.

You would lose fan ramping though, wouldn't you? The last time I meddled with Afterburner on my 5850, you could only set a constant fan speed.

 

[Skurge]

These results dont match the ones from Xbit at all.?
In fact the gtx460 768 mb model ties the 6850.
http://www.xbitlabs.com/articles/video/display/radeon-hd6870-hd6850.html

The 6850 needed a 1015 core clock to equal a 6870 and got beat easily by a overclocked gtx460.
http://www.hardwarecanucks.com/foru...d-6850-vs-gtx-460-1gb-overclocking-study.html

Here is the discussion.
http://forums.anandtech.com/showthread.php?t=2116158

Ryan must have used the wrong lower quality settings for the AMD cards.

They aren't "wrong" its an option and it seems like MOST tech sites don't notice the lower quality.

It still isnt proven to show lower quality in new games. If it makes you feel better. The optimizations for nV hardware were most probably on too.

 

[happy medium]

You would lose fan ramping though, wouldn't you? The last time I meddled with Afterburner on my 5850, you could only set a constant fan speed.

Last I checked you could set many fan profiles with Afterburner so there was no need for ramping.

 

[happy medium]

It still isnt proven to show lower quality in new games.

This I didn't know. thanks

If it is a driver profile setting why would it not be for all games though?

 

[Powermoloch]

i am disappoint because sapphire was not included.

 

[GaiaHunter]

If it is a driver profile setting why would it not be for all games though?

Because the optimizations can work differently for different games and different games will have different optimizations.

I think this IQ quality will have to be investigated more extensively before declaring it is more that just a driver bug or an optimization going wrong.

Another thing to note is that there is a change on how the catalyst AI works for the 6800 series and it seems to be exclusive to it so far.

I'm running 10.10 with a 4850, no hotfixes, and I have a mipmap tab and an AI tab (well only on All tab now). The 6800 series has no mipmap tab and has no cat AI standard, advanced and off option.

I'm not sure how all these things translate when comparing quality settings in between series.

 

[HurleyBird]

I'm saying it is a 4% OC over stock. Are you telling me that 5870 can't do 4% OC on memory?

Why don't you actually read my last few posts to find out?

I guess due to tsmc 40nm problems some of the/a few of the 5870 chips required more voltage than required and it was too problematic/time consuming to bin for 5GHz GDDR5.

My 5870 is a higher voltage one, and it can't run at 1250MHz. It's not time consuming or anything like that, it's economics. If AMD would have been able to bin at 1250MHz without being too negative on their ability to put out 5870's they probably would have.

Otherwise it seems the controller has few problems to reach 5GHz. Now try to get 5GHz with Barts or GF100/GF104 controller.

I'm sure you'll find some exceptional ones that can make it all the way, but yes, there is no disputing the fact that they need to bin those parts lower since the memory controller is weaker. However, what I've been trying to get through to you for the last little bit is that AMD went with a weaker memory controller (despite having all of the R&D done on a Cypress quality controller) on Barts for a reason, despite it being the most bandwidth limited part they've made. Exactly how much sense does it make for that same company to put a memory controller on Cayman that can run 6 or 7Gbps memory at it's rated speed? AND then there's the historical perspective that no high end GDDR5 based GPU has ever run it's memory at the rated speed. I feel like I've pointed this out a dozen times already, and this entire debate you've created is a waste of time.

 

[Xzyrus]

Yea I was more interested in cooler comparisons than O/Cing since O/Cing is very much YMMV. Disappointed to see it missing the Gigabyte and Sapphire models which are at present much more readily available than MSI and ASUS.

Would someone knowledgeable about the 6850 dies say it is perfectly safe to run 6850s at 6870 voltage since it is essentially a binned 6870?

 

[GaiaHunter]

However, what I've been trying to get through to you for the last little bit is that AMD went with a weaker memory controller (despite having all of the R&D done on a Cypress quality controller) on Barts for a reason, despite it being the most bandwidth limited part they've made. Exactly how much sense does it make for that same company to put a memory controller on Cayman that can run 6 or 7Gbps memory at it's rated speed? AND then there's the historical perspective that no high end GDDR5 based GPU has ever run it's memory at the rated speed. I feel like I've pointed this out a dozen times already, and this entire debate you've created is a waste of time.

As I said before, I believe Barts controller spawns from AMD need to further make a cheaper GPU to go to the <$250 market.

Anyway, if the 6970 gets released on schedule we will be able to resume this argument in a couple of weeks, when we have more information.

Until then.