AMD: R8xx Speculation

[willardjuice]

Have you ever overclocked a Phenom II? I have, and they do quite well.

Because clearly that's what I was referring to? (think sales numbers)

How many products that aren't R8xx-based are you going to mention in the R8xx thread?

How many trivial points are you going to raise?

We all know the downsides of multi-GPU rendering. The upsides tend to outweigh the downsides in most titles.

But again not my point. With R700 when cf doesn't work at least one can get performance near a 4870 (I say near because one has to disable Cat AI to disable cf, thus a "normal 4870" will still be slightly faster). With R800 when cf doesn't work one is going to have performance ~10-20% lower than a 5870.

How so? I haven't seen that happen in a long time.

Saints Row 2. Or pretty much any demo or beta.

A2 cannot even hit the targeted shader clocks.
They also have ~100w to work with going off the ~190w typical consumption, TDP of 210-225w(?).
Hopefully A3 clocks better and makes up for that 18% clock deficiency.

And you know this for sure how?

 

[ShaidarHaran]

This "discussion" is going nowhere. Nice job ignoring my perfectly valid point about motherboards that have only one PEG slot, BTW.

FFS, the product's not even been launched yet and we have no idea just how overclockable it will be. Your worst case scenario may be a moot point if every 5970 hits 5870 speeds. Even if they don't the best case is still a much greater difference from a 5870 than the 10-15% performance you may lose in a worst case.

 

[willardjuice]

This "discussion" is going nowhere. Nice job ignoring my perfectly valid point about motherboards that have only one PEG slot, BTW.

It was irrelevant to my point so I disregarded it. I'm not arguing about whether or not X2/GX2 configurations are valid products. I am pointing out a regression. Last cycle AMD was able to produce a 4870 X2 part, this cycle they could not produce a 5870 X2 part. In addition they added $50 to this cycle's high-end card. I found this humorous.

we have no idea just how overclockable it will be.

I don't care what it overclocks to. Most users won't overclock and AMD's warranty doesn't cover it. Thus the theory: "but it overclocks to LOL GHz!!!111" is invalid.

 

[digitalwanderer]

I think they may be telling the truth about it OCing very nicely this time Willard, imho...

 

[Psycho]

With R800 when cf doesn't work one is going to have performance ~10-20% lower than a 5870.

So, we should have a Nehalem style turbo mode, when CF is not working?
(loads of cooling and power headroom for one gpu).

 

[psolord]

I think they may be telling the truth about it OCing very nicely this time Willard, imho...

Yeah I do too! I mean, my 5850 is already running perfectly fine at 1Ghz! Here are a couple of videos of it operating


ATI 5850 @1.0GHz Q9550 @4.0GHz CRYSIS BENCHMARK DX10 HIGH 1920X1080 16XAA

ATI 5850 @1.0GHz Q9550 @4.0GHz CALL OF DUTY MODERN WARFARE 2 1920X1080 4XAA

Unlocking the 5970 is a very smart move. Honestly I couldn't see that coming.

My major issue with ATI cards, seems to have been resolved. The DX10 HDTV support/borders issue. There's only one that remains though. Will they add a "disable the darn crossfire" option in the driver? And no, I don't like disabling AI on bit. It doesn't work anyway.

On my 4870X2, Prototype, Ghostbusters, Killing Floor, all were running with like 40fps or so, no matter what I did! I mean seriously, having an uber expensive and ultra performing card, not being able to run a silly game, is just stupid.

Also temperature and noise will be an issue. I'm pretty happy with my 5850 and I am actually thinking of getting a second one, if the 5970 is a let down!

Any info on that PLX chip?

 

[Iron Tiger]

So, we should have a Nehalem style turbo mode, when CF is not working?
(loads of cooling and power headroom for one gpu).

As long as it also comes with double the bandwidth.

 

[DegustatoR]

Plain AFR again? Meh.

 

[Malo]

I don't care what it overclocks to. Most users won't overclock and AMD's warranty doesn't cover it. Thus the theory: "but it overclocks to LOL GHz!!!111" is invalid.

I'd assume that most users who would spend $600 on an enthusiast gaming GPU will a) know that it is unlocked and probably very overclockable, at least to default 5870 speeds, and b) will be more than willing to try. You can't really count the average joe into this class of purchase.

 

[LordEC911]

Saints Row 2. Or pretty much any demo or beta.

Ah, your argument is that it doesn't automatically turn of CF if the performance is not there plus no user profiles.
This is something that AMD/ATi is working on addressing per some recent interviews.

And you know this for sure how?

Those whitepapers from Nvidia, stating 768GFlops of DP, plus a certain favorite tech journalist posting the target clocks.
Coupled along with the recent article about the ~630GFlops from GF100 @ 1230mhz on the shaders.

 

[fbomber]

It was irrelevant to my point so I disregarded it. I'm not arguing about whether or not X2/GX2 configurations are valid products. I am pointing out a regression. Last cycle AMD was able to produce a 4870 X2 part, this cycle they could not produce a 5870 X2 part. In addition they added $50 to this cycle's high-end card. I found this humorous.


I don't care what it overclocks to. Most users won't overclock and AMD's warranty doesn't cover it. Thus the theory: "but it overclocks to LOL GHz!!!111" is invalid.

AMD is doing this time exactly what NVidia always does with their x2 cards: cut something off (clocks or disable parts of the chip).

 

[LunchBox]

http://www.techpowerup.com/reviews/ASUS/EAH5970/

very nice review on 5970!

http://www.techpowerup.com/reviews/HIS/Radeon_HD_5970_CrossFire/

5970 on Xfire!

 

[Bob]

Why are two 5870s more than 2x faster than a single 5870?

 

[itsmydamnation]

Why are two 5870s more than 2x faster than a single 5870?

im betting drivers, also from what im gathering techpowerup are using newer drivers then a lot of reviewers.

 

[digitalwanderer]

Why are two 5870s more than 2x faster than a single 5870?

I'm betting they used different drivers. Probably used the 5970 drivers that came out today, (well, the same set went out with the sample 5970s...just a bit older version of CCC) for the 5970 and the 9.10s for the 5870. They've improved performance 'tween the two sets.

 

[ChrisRay]

http://www.techpowerup.com/reviews/ASUS/EAH5970/6.html

This is where they listed the drivers they tested with.

 

[DegustatoR]

http://www.legionhardware.com/document.php?id=868&p=1
http://www.tgdaily.com/hardware-features/44804-ati-clarifies-5970-furmark-benchmark-stats

 

[anaqer]

Sorry if this has been discussed elsewhere - any word on sub-Juniper parts yet? Stuck at 1280x, even the 5750 is a huge overkill for me so I'm thinking something around a 300-400core part would be best... sort of a 4670 replacement, actually. Are there any rumors about such a part coming out in the next few months?

 

[neliz]

Sorry if this has been discussed elsewhere - any word on sub-Juniper parts yet? Stuck at 1280x, even the 5750 is a huge overkill for me so I'm thinking something around a 300-400core part would be best... sort of a 4670 replacement, actually. Are there any rumors about such a part coming out in the next few months?

I think AMD is confident these will launch before Furby..

(actually.. january for them is a good time, they should've launched next month but.. were delayed.

 

[Dave Baumann]

(actually.. january for them is a good time, they should've launched next month but.. were delayed.

In every press and analyst session we've done since September 10 we've showed a strategy / roadmap slide that showed they are 2010 parts.