AMD Radeon HD 6970 Review

+1. BTW you shouldn't even bother with guest posts. This article alone shows that they are usually *****ic fanboys.

Not a horrible price, but only ~20% faster than my overclocked 5870. Meh.

It does really depend on the game. For example the Radeon HD 6970 was much faster than the HD 5870 in quite a few games...

World in Conflict = 29% faster.
Call of Duty: Modern Warfare 2 = 30% faster.
Aliens vs. Predator = 33% faster.
Metro 2033 = 73% faster.
STALKER = 77% faster.

That said even a 20% performance boost is still quite significant.

Read article. I was talking about ATI's plan as a whole. they probably plan on going after the 580 with their dual gpu card. The 570 matching the 6970 is plus points for nvidia but not the worst news in the world for ati since they can remain competetive with a simple price drop.

Hooray for looking stupid and not thinking about what i was saying before you post.

keep in mind the gtx 570 and 580 are pretty much maxed for overclocking, they dont like being nudged up even slightly. nvidia is capitalizing on every bit of performance that chipset can handle.

amds 6970 overclocks to 950 and shows no sign of not being able to go higher. at this clock level it approaches gtx580 performance, for quite a bit cheaper.

i wouldnt call this round from amd a total loss.

Isn't it funny how....
1) the one's defending the underwhelming comparison to Nvidia's best keep mentioning that new drivers will help improve the 6970's performance like Nvidia doesn't make drivers to improve their new cards ( that also have drivers less than a month old)

2) the defenders keep mentioning " the dual gpu 6990 will save us" like the dual gpu Nvidia 595 hasn't already been announced for Early 2011 release (with pics online a week after the 580 was released).

I replaced my awesome Asus 5870 which I overclocked to a fully stable 960MHz core. Earlier I replaced it with an MSI Geforce 580, which I overclocked to a fully stable 900MHz core. Both of those cards are incredible, but if this card can only be stably overclocked a mere 20-30MHz, that is very poor in comparison and doesn't sound like it will offer a noticable increase in gameplay performance.

That's it. I would like to see how does a overclocked 5870 compare with the 6970.

Quote:TomSEA said:
"This site has become biased toward Nvidia as of late."

Might as well say that about every other tech site that's done ATI card reviews, then. Simple fact of the matter is if you take ATI's current generation of cards as a whole, they're in second place right now (not by much, but they are). And that's what ALL the reviews are showing.

+1. BTW you shouldn't even bother with guest posts. This article alone shows that they are usually *****ic fanboys.


-1 Tommy forgets that the Radeon 5970 is still the fastest card available.(like Techspot which consistantly rubs the Geforce 580 in everyones face with no mention of it's superior). Also take into though the 6850 and 6870...both of those cards are well beyond their Nvidia counterparts...which are sub Geforce 460 cards.. you know that they both beatout the 460 too. How do you figure the 6970 is in 2nd place? It's just as fast and arguable faster than than it's intended counterpart the Geforce 470. The dual GPU 6990 is going to be king in Feburary or whenever it gets out and it's clearly set against the Geforce 580. So in reality, AMD is actually in 1st place as they have been for a while now.

See above regarding Nvidia's Geforce 595 dual gpu card being released early next year. In fact, type it in google and you'll see the pics that have been online for about a month already. Seems like comparing the single gpus to dual gpu cards won't be a valid argument a month from now...

Every HD 6970/6950 is hand finished !!!

If this is even remotely true, it constitutes one of the biggest tech Quality Control fails in recent history.
http://www.xtremesystems.org/forums/showthread.php?t=263668

While I'm here I may as well do some troll bashing...

Like this GTX 570 at 850/4200 (16%/10.5% OC) ?, or this GTX 570 at 822/4740 (12%/24.7%), or this GTX 570 at 825/4352 (12.7%/14,5%) or maybe this GTX 570 at 860/4500 (17.5%/18.4%)

...and that seems to be the hard limit in Overdrive ( Guru3D, TS, and HC etc)...which is a fabulous 8%!
Can the card go higher? who knows?
Can the max OC be lower...you bet!
Hexus (948MHz -7.7%), TPU (915MHz- 4%) and HT4U (910MHz-3.4%...note their GTX 570 sample oc'ed by more than twice that percentage) sure are proof of that.

Thanks for stopping by.

Our AMD sample had the back of the 8-pin connector shaved back. Will be interesting to see how many production cards have been modified.

If I had a dollar for every time someone said we are biased toward Nvidia, AMD, ATI or Intel I'd be a rich man!

But can it play System Shock 2?

Isn't that a big load of bull.

The FACT is that the GTX580 is a single GPU pushing almost the same performance or better than the 5970.

It's not an apples to apples comparison.

However, it does bring a big question, why don't Nvidia come out with a dual-GPU of their own. Why? Because of pricing. They'd have to overprice the thing just like the 5970 was. But if for some reason they decide to do something like a dual 560, then you'd bet your *** it will destroy the 5970.

But really, dual gpu or not, it does come down to overall performance and support. And overall support from Nvidia is excellent. You simply get more for your money whether it's features like CUDA, Physx, or driver updates. Say what you want about Physx support on games, the fact is that you get it anyway with an Nvidia card.

But enough of this banter with the silly GPU wars, the people who win are the consumers. Now we have more options, the market is moving forward in which we can buy GPUs for the same price as the older generation but yet it outperforms it.

I agree that is the only sensible way to look at it.

Its amazing how fanboys call things biased when its not favoring their brands...

Its amazing fanboys call anything biased in any case...

Its amazing how people can be that dumb even when there are figures and everything to back the information...

Rant over

I have read all of the reviews around, and do not understand the use of non current drivers for the benches, actually i do, im just being a smart ***. Toms may be of the worst using a 10.10d set of drivers that make the 69xx look like a low end card. here is a bench from a site that has a stellar reputation.
http://www.hardocp.com/article/2010/12/14/amd_radeon_hd_6970_6950_video_card_review/1
they get a slightly different result.

@red1776
Kyle and the crew at [H] I think are the exception that proves the rule for most reviews.
1.They mostly test to the "ragged edge" of playability- in fact with some of their framerates that are produced in their tests I don't think would actually be representative of actual gameplay. Most gamers I suspect would quite happily forego 8xMSAA (for example), drop down to 4xMSAA and get a smoother and more responsive experience.
2. Their game selection is very narrow. Five games (cherry-)picked for their graphical intensity will always compress the results across the same-segment cards reviewed ( driver bugs excepted). For the most part nvidia's shader horsepower is being equalled by AMD's 2Gb of faster VRAM.
3.Kyle is still using Cat 10.11, so the driver isn't the differentiator here. Sites using Cat 10.12P have still reported similar figures to reviews with earlier driver releases:

GTX 580.........HD 6970...... GTX 570......HD6950
100%...............85.5%.........85.4%..........77.0%.....Hardware Canucks (@ 1920 res)
100%...............86.8%.........84.8%..........77.4%.....Hardware Canucks (@ 2560 res)
100%...............89.4%.........93.2%..........81.4%.....Hardware Heaven (@ 1920 res)*
100%...............84.8%........ 88.2%..........77.1%.....Hexus (@ 1920 res)
100%...............89.1%.........85.5%..........80.3%.....Hexus (@ 2560 res)
100%...............82.5%.........91.6%..........73.8%.....Neoseeker (@1920 res)**
100%...............80.8%.........88.0%..........72.6%.....Neoseeker (@2560 res)**
and for the record...
100%...............95.7%.........84.6%..........85.3%....[H]OCP (@2560)

* Excludes the HAWX2 benchmark
** Includes the horrendous Street Fighter IV framerates which skew the percentages

The principle difference between [H] and these (and other sites) is that Kyle's methodology is producing 29-47fps with an average in the mid thirties, while the other sites have set 40-50 fps as their median- which to me seems more realistic (having a first-person-shooter preference).
The second difference is that a wider range of games were used in the other reviews (HC 7 games over 2 resolutions and 2/3 IQ settings, HH 8 games, Hexus 7 games o/ 2 resolutions, Neoseeker 9 games o/ 2 resolutions)
Third difference is of course the inclusion of games that show a greater seperation between nvidia and AMD's offerings (Lost Planet 2, Dirt 2, Call of Duty:MW2/Black Ops, MoH, Batman) to AMD's detriment- this is more a product of the relative driver teams. When AMD tanks on a benchmark it can be radical, nvidia's "losses" are usually on a much smaller scale. A case of relative benchmark consistancy providing a superior overall effect.
3. I'm personally not convinced that testing exclusively at 2560x1600 makes that much sense. There are no cheap 25x16 screens. If you're popping for one then likely you're also going Crossfire/SLI. The vast majority of gamers who will end up with these cards are likely to be going with 1920x1080/1200 or Eyefinity/Surround for a select few and those that opt for two or more cards.

I'm not knocking Kyle's testing, just placing it into context with the other reviews out there. I tend to flag inconsistancies in reviews (i.e. HAWX2 bench) in order to get a feel for what the "average" gamer is likely to experience with the product. Kyle's numbers at 30 fps for a FPS are to me a little too low, just as benching UT3 (with these cards) at 1024x768 with 0xAA/0xAF makes no sense at the other (very) extreme.

One last thing....
[H] reviewed the GTX 580 last month (November 9)...they used Forceware 262.99 beta drivers to achieve framerates of 32 (min), 43 (max), 39.4 (ave.) with 8xMSAA + TRMS AA and 16 x AF with F1 2010.
In the 6970 review, the GTX 580 running 263.09 beta drivers posted exactly the same framerates but with 8xMSAA 16xAF (i.e. no transparency AA)Possibly a co-incidence ?
Compare the in game framerate graphs from the two reviews. They look similar for the GTX 580...well, because they are identical. I printed off both and overlaid one over the other (should have used Photoshop I guess) on a light table and the runs are identical. I'm thinking that this might be stretching co-incidence just a tad too far....

So...either the GTX 580 is remarkably consistant in gameplay_AND_transparency antialiasing carried absolutely no penalty, or the performance gain from the 262.99 to 263.09 driver was remarkable, or...............

I really don't like how "Kyle" does his testing. Not only are most of the tests horribly unrealistic, but he cherry picked games that favor the Radeon platform.

But oh well, reviewers can be fanboys too.

Nice detective work there Chef. I don't know which it is either. I have grown disgusted with benchmarking on the internet. Everyone is beholden to one company or the the other. They cant possibly get results that vary this much. Kyle stated that AMD sent him a set of drivers to test the 69xx cards with. The implication being I suppose that they will become the 1.11 drivers. is he the only one they sent them to?...I think probably not. You have sites using 10.10, 10.10e,10.10e, and 10.11 drivers, everything but the current set. BTW Toms Hardware has developed into the biggest Nvidia ***** to date. I was reading a revoiew of the 6850& 6870 release where apparently they were performing a little to well. so they added to the game benches a specific 'map' in Civ V, and and tessellation scene in just cause and the like, and gave them equal weight o the overall game suite. The thing that pisses me off is that they rarely get the same result I get when I test bench the cards (both Nvidia and AMD) coincidence? I think not. I have a 6970 speeding to me via Newegg, so I will find out myself. Unfortunately few people have this luxury afforded them. I just want to know how these things perform, i guess I am not going to get that answer unless I do it myself.

Everyone was supplied with a driver for testing the Radeon HD 6900 cards. I have no idea why you guys keep talking about drivers? What do you know that we don’t? The current drivers that you can download from the AMD website do not even support the Radeon HD 6900 cards so what are these latest drivers that we are not testing with?

As for you saying that you never get the same results as certain websites when you get the same graphics card, well where do I start? There is so much that can influence the results and for a lot of our tests for example we use FRAPs to measure the performance of certain scenes. Still having said that is not even the frame rates themselves that are of vital importance.

For example if we get 34fps when testing with Crysis and you only get 29fps for whatever reason that’s not such a big deal. What is important are the performance trends where we compare a range of graphics cards. So if we say the GTX 580 is much faster than the HD 6970 when playing Crysis then you should also see that, especially when playing the section of the game that we tested.

As for the HardOCP review I would say they just took the set of results from the previous review so transparency anti-aliasing was still being used. So the graph is just incorrectly labeled. As for the driver issue I am sure the drivers brought about no performance changes, we see this all the time.

Agreed. I would tend to support the view that you bench withthe release driver (as most sites did). If you can bench with a newer release then well and good. The onus in this situation falls squarely on AMD. In recent weeks they have released a whole slew of 10.10 hotfix and 10.11 (and now 10.12) drivers- this I think has led to the confusion/consternation of readers- almost entirely from AMD-philes who seem to be looking for a "magic bullet driver"...."10.12preview was released 6 hours ago why aren't you guys releasing a full review of benchmarks using it!!!!!!"
I have yet to see an AMD-centric commenter anywhere ask why (with the relaxed release date of these cards) 1. performance in games such as BFBC2 and Battleforge seems to have regressed, and 2. What performance 10.12 was supposed to bestow? -isn't it more a CCC2/ Eyefinity update? Maybe Terry Makedon had to put the driver writing on the back bench when he got the job shaving the corners off PCIe power connectors.

This is new release series, and more importantly, a new release architecture...yet AMD's own website isn't even promoting them on it's front page...it's all about the HD 68xx !
Maximum Secret Squirrel mode for the last few months (including bs performance slide decks) ...build the hype and pre-orders, and at launch day turn Secret Squirrel into Maxwell Smart

True enough. System build, in-game settings and FRAPS v benchmark are all going to make an apples-to-apples direct comparison between reviews a non-starter. Personally I use weighted averages from as many relevant reviews as possible, using one card as baseline and its competitors as a percentage +/- which tends to offer a more precise indicator than wildly fluctuating framerates. A simple spreadsheet -which I email to customers who indicate they want to upgrade etc. takes a lot of guesswork out of the equation:

The beauty of collating the reviews as I usually do thankfully flags such anomolies. BTW Tech Report benches the game both early in the piece and when the game map fills up.

How would you know this from the review unless you have an uber-keen interest in hardware and an eidetic memory (both of which relate to me), or had the forethought to check bench results from previous reviews?

Granted. Brent & Kyle also state that the HD 6970 is offering the best gameplay based on the fact that both cards use 8xMSAA. If the GTX 580 (in this instance) is actually offering 8xMSAA + TRMSAA then the statement is erroneous since 1. TRMS is essentially antialiasing samples from all polygons ( a dumbed down supersampling) not just the edges as in multisampling and 2...

-emphasis added.

As I mentioned in a previous post, I'm not pointing the finger at [H]. Sh*t happens, errors creep in- especially if the manufacturer artificially constrains review time/deadlines as seems to be the case with the HD 6970/6950 release judging by some of the various reviewers comments...and yes, I'd agree with red that some reviewers load the dice (re benchmark choice) in favour of one camp or the other- which led me to thank you and the TS staff for supplying such a wide range of gaming benchmarks in my first post. Putting so much effort and diligence into the reviews, especially with the time constraints here, keeps me (and no doubt many other enthusiasts) returning to read each one for the quality and expansiveness they convey.

Ooooooooo...For yourself? or customer/review? Colour me jealous!
Only the Sapphire card available in New Zealand at present and pricing wise it doesn't stack up well against the GTX 570-especially the EVGA Superclocked/Black Ops (797MHz) which are still cheaper and have a much better warranty. Once availability is more widespread no doubt I'll have the opportunity to check it out.