AMD Radeon HD6950/6970(Cayman) Reviews

[zalbard]

Found something interesting...

Finally, the 6800 series also introduced support for HDMI 1.4a and support for color correction in linear space. HDMI 1.4a support is fairly straightforward: the 6000 series can drive 3D televisions in either the 1080p24 or 720p60 3D modes. Meanwhile support for color correction in linear space allows AMD to offer accurate color correction for wide gamut monitors; previously there was a loss of accuracy as color correction had to be applied in the gamma color space, which is only meant for use for display purposes. This is particularly important for integrating wide gamut monitors in to traditional gamut workflows, as sRGB is misinterpreted on a wide gamut monitor without color correction.

So 5xxx series incorrectly worked with wide gamut monitors, and now it's fixed? Or does it only apply to monitors connected via HDMI, and those connected via DVI have no gamut issues with either card?

[flyck]

Regarding efficiency (or lack thereof), I just want to make it clear that we do everything possible to exclude outside variables from our power consumption test.

The power supply is plugged in to a Tripp-Lite line conditioner that regulates input voltage to a constant 120V. If this isn't done, input voltage WILL impact the consumption of the system and let's be honest; no one likely lives in a house with constant live voltage.

In addition, the 3DMark test we use puts a very minimal load on the CPU. This is important since the CPU's fluctuating load patterns in most of the apps out there can really mess up results.

I am also surprised to see quite a few sites still using FurMark, etc to test power consumption. AMD admits to that program being capped by PowerTune.

But every program that will exceed a certain power will be capped on the 69xx series. So either you tested with powertune set on +20% or something else is scewed in your test.

[SKYMTL]

Are there ANY reviews with the Catalyst 10.12 drivers??

Well, considering that in most of reviews they used 10.10 or 10.11 drivers, don't you think that the real performances will be clear when AMD releases 10.12? I mean, the Big Cayman could run nearly as fast as Gtx580 with some (5+10%) driver encrease!

Most people seem to be dreaming in technicolor here. If you think that AMD released drivers to reviewers that wouldn't feature the full performance, you are sadly mistaken. This these hopes and dreams seem to happen with EVERY AMD and they NEVER come to pass.

The drivers given to the press (both 10.11 and 10.12 RC2) are "full speed" drivers, make no mistake about that.

Having used both, I can tell you that they are identical in performance other than a small increase in 3DMark 11 scores and a patch that fixes some random crashes I was having in F1 2010.

For all of you harping on the drivers used, I think that is in poor form simply because there won't be one lick of difference between the two packages in terms of overall single GPU in-game performance.

The one to blame should be TSMC at the end for this "new generations".

I want to address this right away because placing the blame on TSMC shouldn't be done once you understand the situation.

As I said in my article, AMD had already begun taping out their cards on 32nm when it was realized that the power and space savings incurred were minimal whicle the pricing was too high. This meant AMD decided to use an existing 40nm process for their entry and mid-level cards. This left Ibiza at 32nm but since the volume from the lower-end cards would no longer be there, TSMC decided to cancel the 32nm node altogether. This left Ibiza which then needed to be ported to 40nm and eventually became Cayman.

So, it wasn't TSMC's fault but rather a combination of economic decisions by both parties which sunk 32nm.

I can't believe alot of you guys were slamming SKYMTL last week for trying to control your expectations and not to expect so much. He was probably the closest to giving accurate performance for the 6970 and he had an NDA.

I tried.

[SKYMTL]

But every program that will exceed a certain power will be capped on the 69xx series. So either you tested with powertune set on +20% or something else is scewed in your test.

Nope. PowerTune only caps when the infered TDP reaches a certain point. If a program doesn't allow the core to each that point, PT won't kick in.

It is completely possible that the program I use pushes TDP closer to the PT barrier that what Anand and some other sites use.

[Kai Robinson]

Seriously - did no one READ what i wrote? The only reason these cards are disappointing, is because you CREATED all the Hype, with rampant, pointless speculation, and expected a giant killer - which is NOT what AMD have been trying to make. They wanted Cayman at 28nm, TSMC let them down, and they had to re-configure for 40nm - what about that DON'T you get? Sometimes i swear you people ask for too much.

[migunani]

Wooww.... Cayman... all the way...
Two Caymans... oh My God....
Two Caymans All the way.... So beautiful....So Powerful.....
wWwooooohhhoo..hoo..hooo.....
O.. my godd... Oh.. my goood..... Whoooooo.......
What does this mean? OOoooohhh.... Ooooohhhhh......

Sooo Powerful..... so beautifullllll..... Like double Rainbowww....

Oooooohhhh..... OOoohhhhhhh....... oooh.. my gooddd.....

Two Caymans.... oooohhh...... ooooooohhh.....

/JK



Performance wise, cayman is a Failure.

AMD is getting HATTONed this round.

[[XC] gomeler]

Holy . What were you guys expecting AMD to do being stuck on 40nm? They clearly weren't going to create a 300w monster as that would go against what they've been doing since the HD3870. The HD6970 looks to me like a test of the architecture while they wait for the 28nm node to mature for mass production. On top of everything it isn't like the HD6970 is a GTX 480. It is a minor improvement on the HD5870 and it paves the way for a killer 28nm core.

Also, it is slightly faster than the GTX570 with a little over half a gig more memory and priced slightly higher in what I feel is the proper performance/price slot. What do you want, AMD to bleed money so you can get your high-end GPUs for $200?

I'm so disappointed in you guys, not AMD. AMD kept their lips sealed, you guys worked yourselves into a frenzy, and then turn on AMD when they don't deliver what you had hyped yourselves into expecting.

Just so we are ready for the HD6990.. "omg it comes with 2560SP, 128TMUs, two men/women(depending on the bundle you purchase) to wait on hand and foot and it consumes negative 400 watts!."

I will agree though that this naming scheme is horrible. What a trainwreck that is.

[flyck]

Nope. PowerTune only caps when the infered TDP reaches a certain point. If a program doesn't allow the core to each that point, PT won't kick in.

It is completely possible that the program I use pushes TDP closer to the PT barrier that what Anand and some other sites use.

That is what i'm saying. But your result indicate a 150W difference between idle and load for the 6950 which has a power cap of 140W. Add the idle statement and you are getting close to the 200W for the 6950, which is the +20% on powertune. So either you set it at +20% or it is not kicking in when it should be.

Or the cpu is loaded more then you expect. But that won't change the strange difference between the 5850 and 6950. If the 6950 would be capped 140W, that would mean the 5850 would use 90W during 3dmark. which seems a bit optimistic.

note that most numbers look to be correctly when you watch the idle-load of the same card. I do however see something unexpected for the 6950 and the 58xx-68xx. The difference is to big while their power consumption should be in the same range.

[flopper]

6950 crossfire beats the 580 sli in 5760x1080 resolution, so not sure why people think Nvidia is faster?
http://www.hardwareheaven.com/review...ty-vs-sli.html

so for me, playing at that resolution, I save half the cost chosing amd 6950.
2 amd cards for one one 580gtx if they were in stock............................another CARD isnt cost effective buying two nvidia cards for more money for worse performance????
twice the failure wouldnt you think?

[Gilgamesh]

A good review from Xbitlabs:
http://www.xbitlabs.com/articles/vid...70-hd6950.html


Impressive HD6970 vs HD6950 same clock



vs GTX570:



Power consumption:




others power consumption test (HD6970 = GTX570):
http://www.computerbase.de/artikel/g...stungsaufnahme

http://www.hardware.fr/articles/813-...ossfire-x.html

[Eagleclaw]

Battlefield Bad Company 2 @ 1920x1200 6fps advantage over a 5870, you gotta ing be kidding me.

I am really disapointed

[Vardant]

6950 crossfire beats the 580 sli in 5760x1080 resolution, so not sure why people think Nvidia is faster?

That seems weird, really weird. I mean everybody has Crossfire or SLI these days. Not to mention three monitors.

[-=DVS=-]

And only playing F1 , we already knew 6970 does very good in F1 game ,now try playing more games.

[RedBull78]

According to Anandtech, the 6970 goes head to head with the GF480...I think this is pretty impressive..I don't know why some of you folks in here find this card to under-achieve.....this card was intented to compete against the 480 and so it does......and remember, in CF mode these cards are very powerfull....what did you most expect to see from these cards anyway? they are still under the 40nm technology....blame the semis...for the delay of course.

[flopper]

That seems weird, really weird. I mean everybody has Crossfire or SLI these days. Not to mention three monitors.

dont care about your setup.
I care about value for my money.
there 6950 kick 580 gtx.

[Dezmen]

// Just woke up... All reviews added from topic and PM. Special thanks to PM guys, makes adding easier :P
Also start adding video reviews for those who hate read bazzilion pages... like meh xD

[flyck]

A good review from Xbitlabs:
http://www.xbitlabs.com/articles/vid...70-hd6950.html


Impressive HD6970 vs HD6950 same clock

basically 10% more shading power results in 1-2% performance increase? also those 10% shaders apperently cause a 80W difference??

Why doesn't the math work out?

[Vardant]

there 6950 kick 580 gtx.

I'm not disputing that. But your wording was, hmmm, let's say unfortunate.

The point is, it's probably one of the rarest cases you can think of. So asking, why everybody else sees things in different light could be considered trolling.

[SKYMTL]

That is what i'm saying. But your result indicate a 150W difference between idle and load for the 6950 which has a power cap of 140W. Add the idle statement and you are getting close to the 200W for the 6950, which is the +20% on powertune. So either you set it at +20% or it is not kicking in when it should be.

You have your numbers wrong.

HD 6950 PowerTune Max = 200W

"Typical" Gaming Power (whatever that is...) = 140W

So yeah, I AM getting close to 200W which is the upper limit of power tune and basically justifies the methodology.

[SKYMTL]

basically 10% more shading power results in 1-2% performance increase? also those 10% shaders apperently cause a 80W difference??

Why doesn't the math work out?

Adding geometry horsepower for high-level DX11 rendering adds die size and increases TDP.

[flyck]

Adding geometry horsepower for high-level DX11 rendering adds die size and increases TDP.

Considering 6950 and 6970 have the same die, same tesselation and only minor differences as in less shaderclusters. i don't understand your comment :-)

but an 6950 ran at 870MHz and slightly higher memory clock than 6970 (so little less core clock and bit more memory bw, 10% less shaders etc) causes a a performance difference of 1-2% and a difference of 80W on average (between 6970 and the oced 6950) makes me wonder what is wrong.

You have your numbers wrong.

HD 6950 PowerTune Max = 200W

"Typical" Gaming Power (whatever that is...) = 140W

So yeah, I AM getting close to 200W which is the upper limit of power tune and basically justifies the methodology.

If that is the case then yes.

[flippin_waffles]

Looks like an awesome card. A few fps here and there are meaningless, especially since AMD targeted these price points directly.
Pretty good engineering IMO. Also PowerTune looks like killer technology. Imagine the implications for mobile ( ala fusion, mobility 6900) and Antilles. Like anandtech says, 'this is just the beggining for Cayman.

[Vardant]

They didn't. There were last minute price drops. GTX 570/580 surprised them quite a bit.

[SKYMTL]

Considering 6950 and 6970 have the same die, same tesselation and only minor differences as in less shaderclusters. i don't understand your comment :-)

Ah. Well, I am not shure what reviews you have been reading but here is the rundown.

Basically, the move to VLIW5 to VLIW4 did save ALU space on the die but AMD didn't just use this to shrink the core. Basically, they used the "saved" space and expanded upon it by adding additional SIMD engines. The engines with their associated TMUs, cache, etc naturally take up more die area than the reduction to VLIW4 allowed.

The graphics engine iteself (which contains the fixed funtion units like the tessellator) was cloned so instead of one, there are now two. This also adds to the transistor count.

Finally, there are the GPGPU compute changes that necessitated an additional direct memory access engine be added along with some other bits.

All of this of course added to the transistor count and thus increased overall TDP.

[Gilgamesh]

basically 10% more shading power results in 1-2% performance increase? also those 10% shaders apperently cause a 80W difference??

Why doesn't the math work out?

That math don't works also for HD5850 vs HD5870, 1 years ago



3% for 320-5D vs 288-5D.

And power consumption: