Radeon HD 6970 And 6950 Review: Is Cayman A Gator Or A Crock?

Page 9 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

cangelini

Contributing Editor
Editor
Jul 4, 2008
1,878
9
19,795
[citation][nom]grinderifz[/nom]why oh why dont they use two 6870's in these performance tests ?They are much better performers and cost only a little bit more than the run of the mill 6850'sThis would make much better compertition than what is shown here !I also agree with using two 460 gtx's[/citation]

GTX 460s are in there. 6850s were a better price match. Thanks!
 

grinderifz

Distinguished
Dec 19, 2010
6
0
18,510
Really?

According to my calculations a GTX 580 cost's 400 pounds and an ATI 6870 cost's 200 pounds

So in my opinion anyone who will be interested in crossfire will go for 2 6870's rather than 2 6850's because of the better performance and price compared to a high end single card solution.

I,m also sure these will top most of your chart's in most games except F1.
 
First off, kudos to AMD for finally improving their CrossFire scaling, which has always lagged behind NVidia until now.

However. I honestly must say that the 6xxx series was a disappointment for me. I expected AMD to use the head start it had in the DX11 department and leapfrog NVidia. That did not happen, and instead AMD spent a year finetuning the architecture, with little performance improvement for their single-card offerings.

I wholeheartedly agree with Chris' assertion in the Conclusion page - the fact that AMD targeted the second fastest card from NVidia instead of aiming for the top was a serious mistake in my opinion.
 

kiddagoat

Distinguished
Mar 14, 2006
52
0
18,630
I really like how Tom's is really inconsistant with the reviews I have read.

Looking at bit-tech, guru3d, hardware canucks, and other sites, they all show the 6970 beating the 570 and 480 clearly and even manages to edge the 580 in higher resolution.

In a multi-gpu solution the 6000 series really shines.

Guess if I paid Tom's enough I could have a really good card/product too... *couch*nvidia*cough*apple*cough*
 

Tc17

Distinguished
Oct 31, 2007
23
1
18,515
AMD didn't fail. Typical Nvidia fanboys. Also, any game that uses the Physics engine of Nvidia, needs to have that option turned off when you benchmark in order to have a fair comparison. ATI/AMD always has been ahead of Nvidia when it comes to technology, whether the card is faster or not. Lets see the Nvidia fanboys run multi-monitor setups off of a single Nvidia card.
 

Undermoose

Distinguished
Sep 7, 2008
12
0
18,510
As an owner, I can say I am very happy with my 2 6970 in Crossfire. Techreport.com, as well as other sites, demonstrate that 2 6970 in Crossfire beat the 580 SLI often. Tom's review is thorough to a point, but they leave out the NVidia flagship 580 SLI in comparison, and had they included it, I believe, AMD's offering would have impressed people a bit more.
 

mapesdhs

Distinguished
(I wrote this reply before xmas, wasn't able to post then due to access
issues... sorry if the pricing data I quote is out of date now)


Zark Strife writes:
> ... So what im really asking is that once
> OC is applied will the 2x 460's eat the 2x 6850's alive?

See my results here:

http://www.sgidepot.co.uk/misc/pctests.html
http://www.sgidepot.co.uk/misc/stalkercopbench.txt
http://www.sgidepot.co.uk/misc/uniginebench.txt
http://www.sgidepot.co.uk/misc/uniginebench2.txt
http://www.sgidepot.co.uk/misc/uniginebench3.txt
http://www.sgidepot.co.uk/misc/x3tcbench.txt
http://www.sgidepot.co.uk/misc/ptboats.txt
http://www.sgidepot.co.uk/misc/callofjuarez.txt

You should be able to extrapolate/compare to 6850 CF numbers.


cangelini writes:
> These are reference clocked boards, and we're using the lowest price
> available. Buying higher-clocked cards will also increase cost

As I've posted before, that's wrong, at least in the UK anyway. Many
sellers either no longer offer stock-speed 1GB 460s, or overclocked
versions of the card are cheaper (usually by a significant margin).
Please stop saying oc'd cards are more expensive, because that's flat
out not the case now and hasn't been for some time.


Here's the pricing data (obtained 20/Dec/10) from several typical UK
sellers for the normal 336-core version of the 1GB GTX 460, where
'stock-speed' refers to the card's original 675MHz reference clock (all
prices UKP):

Scan: the cheapest stock-speed GTX 460 is 149.77; the cheapest oc'd
card is 139.19 with a core clock of 700MHz. The Palit/800 is 162.69.
The EVGA FTW/850 is now only 165.04.

Aria: the cheapest stock-speed GTX 460 is 146.86; the cheapest oc'd
card is 135.11 with a core clock of 700MHz. The Palit/800 is 152.74.

Microdirect: the cheapest stock-speed GTX 460 is 146.88; the cheapest
oc'd card is 141.00 with a core clock of 700MHz. The Palit/800 is
159.80.

DABS: the cheapest stock-speed GTX 460 is 144.99; the cheapest oc'd
card is 129.99 with a core clock of 700MHz.

tekheads: the cheapest stock-speed GTX 460 is 156.98; the cheapest
oc'd card is 139.45 with a core clock of 700MHz. The Palit/800 is
169.95.

QED, proved, end of discussion.


I mentioned the Palit/800 and EVGA/850 (where a store sold them) to
show how shopping around is definitely worthwhile.

Ian.

 

dragonsqrrl

Distinguished
Nov 19, 2009
1,280
0
19,290
[citation][nom]tc17[/nom]AMD didn't fail. Typical Nvidia fanboys. Also, any game that uses the Physics engine of Nvidia, needs to have that option turned off when you benchmark in order to have a fair comparison. ATI/AMD always has been ahead of Nvidia when it comes to technology, whether the card is faster or not. Lets see the Nvidia fanboys run multi-monitor setups off of a single Nvidia card.[/citation]
lol... tc17, you are an ultimate fanboy, and a decent hypocrite as well. It always amazes me how some people can be so incredibly judgmental, pointing the fanboy finger left and right, while obliviously flailing around their own fanboyism.
 

nottheking

Distinguished
Jan 5, 2006
1,456
0
19,310
The #1 thing that still irks me, really, is that AMD had trumpetted the supposed "dual-precision FP support" for the 6900 series... For some of us, that was rather interesting, since an improvement here could truly mean a victory for very, very cheap compute power.

Currently, GPGPU solutions aren't as efficient (capable of reaching their peak) as CPUs in heavy-duty compute (supercomputing) applications; a real-world throughput of 30-35% of their theoretical peak is typical for a GPU, while CPUs (be it PowerPC, Intel, or AMD) range from 75-90%. This makes GPGPUs not QUITE as appealing as their theoretical numbers would suggest.

An obvious solution would've been to simply engineer the GPU design to get better DP FP performance; as the article mentioned, the older 58xx/68xx VLIW5 designs had a DP:SP ratio of 1/5. I was eagerly hoping that, perhaps in a nod to those seeking compute power, AMD might've gone all the way, and brought their GPU to the maximum, making the ratio 1/2: that'd have meant a whopping 1.35 Teraflops DP (theoretical) from the 6970; even with a 30% efficiency rate, that'd be 1.125 gigaflops per dollar of real-world performance, making it the FIRST piece of hardware to pass a gigaflop per dollar. The previous record-holder, the PowerXCell 8i, 121.6 theoretical gigaflops (91.2 at 75% efficiency) for about ~$1,000US per chip, making it only 91.2 MEGAFLOPS per dollar. (note that both only specify the price of the processor itself, not the computer needed to house it)

Sadly, I see that AMD has not done this, instead being content to only put 64-bit MAD/FMA support on HALF of the cluster, leaving the other half all but useless for DP, while still, apparently, guzzling power. AMD will have to do far better than that if they want to make a truly stand-out general-purpose compute solution, in the face of nVidia's more-mature Tesla cards, IBM's potent PowerXCell, Intel's hyper-efficient Xeons, or even AMD's own Phenoms.
 

R3N3GAD3

Distinguished
May 12, 2011
4
0
18,510
this is a kind of unfair benchmark because those ati cards weren't made or meant to compete in the level that specific nvidia card was made, i'd like to see a comparison of that nvidia card with the 6990, that'd be a fair comparison. the above ati cards were meant to compete with the cards around nvidia 460 gtx level
 

Hotobu

Distinguished
Jun 27, 2011
41
0
18,530
I'm a bit late to the party, but it would have been nice to see the non crossfired versions of the cards in the graphs as well so we can see how well they scale.
 
G

Guest

Guest
The 2nd chart must be wrong. I have a Sapphire HD 6970 2GB with:
P22160 (3dmark score)
21862 (GPU score)
feature test1: 73.41 GTEXELS/S (!!!) ,when GPU-Z says: texture fillrate: 84.5GTexels/s
feature test2: 8.77 Gpixels/s
feature test3: 72.51fps
feature test4: 47.90fps
feature test5: 66.87fps
feature test6: 152.21fps
 
Status
Not open for further replies.