Graphics HD7970 - My Review (BF3 Eyefinity vs HD7970)

Hi all.

After receiving my Powercolor HD7970 at 9:30 this morning I decided to work a half day and get some BF3 Eyefinity benchmarks done and write a "brief" review putting across my opinion on the card and how it compares to my "old" HD6970.

First things first, we all know the architectural changes that have happened between the HD6970 and this card (move from VLIW4 to GCN), therefore I won't bore you with going into the details here. The likes of Anandtech and Bit themselves can do a much better job in this area.

Anyway, onto the card itself.

The HD7970 I ended up going for is a Powercolor model which I managed, by a stroke of luck, to get for under £420 from Ebuyer. The same card is still available for £444 if anyone does want one for a decent price. Considering all of the cards are reference design and basically the same, I didn't see the point in spending more for the same card with a difference sticker on it.



The card itself is literally the reference design with a sticker on the fan itself. There isn't much more to say in this area as in the end it looks like every other HD7970 for sale bar the XFX DD model.



Size wise it is basically identical to a HD6950/HD6970 reference card. It measures just under 11" nose to tip.



Connectivity consists of a single DVI port backed up by a single HDMI port and 2 mini DP connections. The mini DP connections can be connected to a hub to provide more connections for Eyefinity if you so wish.




In the box you get the normal user guide and driver CD (Very useful in my case as the 12.1 drivers I downloaded didn't work... ). You also get the following:

- Crossfire bridge
- DVI-VGA adaptor
- 6 pin to 8 pin PCI-e adaptor
- HDMI to DVI adaptor
- Active mini DP to DVI adaptor
- Mini DP to DP cable

Basically you have the ability to run Eyefinity right out of the box assuming you run 3 DVI equipped monitors. Considering the active adaptors still cost £15+ this is a pretty good bundle and not doubt goes some way in explaining the relatively high ticket price of the card itself.

You also get a voucher for a copy of Dirt 3, great if you don't already have the game (like me). Interestingly the voucher has a validity date that runs out at the end of December 2011. I wonder if it will actually work?!

Overall I think the card itself looks good, connectivity is good especially as the relevant adaptors are included in the box.

To close out this portion of the "review" I thought I would post a picture of the card in my case. The case is a Fractal Design Define R3.



(Yes I know I am still running the stock Intel cooler. I really need to get an better one but for now it does its job admirably).

Eyefinity performance

I should point out that I have only run performance numbers for BF3 & Arcania (Gothic 4) at the moment. I will look at some other games at a later point. I will not be running any synthetic benchmarks as quite frankly I find them a waste of time.

BF3 Eyefinity Performance

Benchmark is a 5 minute FRAPS run through ( I actually play the game) taken from The first real level - Operation Sword breaker. I have found performance in this level to be nigh on identical to performance online so it should be a good indication of what you can expect in "real world" scenarios.

Resolution is: 3560x1920 (technically 3588x1920 but I won't split hairs)
Settings: All high bar Shadows and Decoration which are on medium. No Deferred AA, Medium Post AA no HBAO/Motion blur (the latter makes feel very queasy).

System setup: Per Signature

Drivers: HD6970 - 11.11a / HD7970 - "On disc" - 11.12

"HD6970 Playable Settings"

HD6970

Min - 26.00
Avg - 35.54

HD7970

Min - 34.00
Avg - 46.63 (+31.2%)

First up I was genuinely surpised at how well my HD6970 held up at this resolution / settings combo, 36FPS isn't exactly bad all things considered. Look at the HD7970 fly though, 31.2% faster at the same settings running the "on disc" drivers (which appear to be Cat 11.12).

Ultra Settings without AA (SBAO not HBAO)

HD6970

Min - 16.00
Avg - 27.83

HD7970

Min - 22.00
Avg - 36.96 (+32.8%)

Interesting results here. Firstly the performance delta has increased to ~33% which is excellent. The biggest suprise was the fact that I can now play at "Ultra" settings without AA at roughly the same level of performance as my HD6970 managed at High/Medium. Great stuff TBH.

Ultra Settings with 4x AA - AKA "LOL" Settings

HD6970

Min - 8
Avg - 14.68

HD7970

Min - 15
Avg - 20.21 (+37.7%)

Well, I have nicknamed this "LOL" for a reason. My HD6970 simply ran out of VRAM and despite posting a average FPS of 14.68 it was completely unplayable (hard to play to even get the result TBH). The HD7970 ends up 37.7% better at this setting but in minimum terms it is 87% faster. Overall though this is academic as neither card can manage this setting. VRAM usage on the HD7970 peaked at 2400MB.

Overclocked Performance:

"HD6970 Playable Settings"

HD6970

Min - 26.00
Avg - 35.54

HD7970

Min - 34.00
Avg - 46.63

HD7970 OC @ 1075/6000

Min - 38.00
Avg - 52.98 (+49%)

HD7970 OC2 @ 1125/6600

Min - 40.00
Avg - 56.59 (+59.2%)


***Update regarding CPU Overclocking results:***


"HD6970 Playable Settings"

HD7970 OC @ 1075/6000

Min - 38.00
Avg - 52.98

HD7970 OC @ 1075/6000 with 2500k at 4GHz

Min - 38.00
Avg - 52.97

Finally we can see the HD7970 start to stretch its little legs. At the this OC level it ends up being ~ 50% faster than the HD6970 at the same settings. More importantly, the minimums are now higher than the average on the HD6970!. It is amazing to see over 50fps at this resolution on a single card and what is basically "high"settings.

**Update* - I have found during my own testing that the VRM'S become less tolerant to heat the higher the card is clocked (there is more to it than this but this is how I can easily explain it). In essence to clock higher you have to keep the card cooler. By doing exactly this I can now run 1125/6600 at 1.175v. The results are shown above as "HD7970 OC2". The card is now just shy of 60% faster than the HD6970 at the same settings and coming very close to the all important 60fps barrier.

Overall BF3 Eyefinity performance has left a massive smile on my face which at the end of the day is what really matters. I can finally really enjoy Eyefinity in my favourite game currently which is just fantastic.



Arcania - Eyefinity Performance

I have also run some Eyefinity benches on a game that no-one is probably playing but I am. I got this game for under a fiver in the Steam holiday sale along with the rest of the Gothic line-up and was amazed when it literally brought my HD6970 to its knees.

I thought I would include the results as a point of interest so here they are:

HD6970

Min - 19.00
Avg - 27.30

HD7970 OC @ 1075/6000

Min - 30.00
Avg - 38.76 (+42%)

Another instance where the minimum on the HD7970 is higher than the Average on the HD6970. You may note a lack of stock HD7970 benches for this game. The reason is simple - it took a number of attempts to get the game to play (thanks to a small change in resolution due to bezel management) and once it finally opened I couldn't see FRAPS sigh.

Overall the HD7970 (OC) is 42% faster than the HD6970 in this game.


Now the benches I have run had been put to bed I can move onto the conclusion.

Based on my own experiences I am very pleased with the performance that the HD7970 has offered in Eyefinity. It has shown itself to be much faster than the HD6970 it replaces whilst pulling similar levels of power and outputting similar levels of heat. My old PSU issues aside it also clocks like it has 100 weight of TNT up its bottom.

The price is obviously the biggest sticking point for a lot of people. At £500 there is no way I would purchase this card. At £450-£475 (where alot of them are residing) I would be very reluctant to part with my own money but at £450 and under I can see the attraction. Now, I got my card for a relative steal but can still see why it has been priced like it has. Who is to blame for the pricing is still up in the air with alot of fingers pointing to the UK retailers taking a nice chunk. Pricing in the US seems much more reasonable at $479-$500, so it does look like a case of "rip off Britain" once again.

Would I recommend the HD7970 to anyone potentially interested in it? Yes but only if they are running a resolution that actually requires it and only if they don't pay more than £450.

Anywhoo I have come to the end of my "little" (LOL) review of the HD7970. I hope you have found the content interesting. If you have any queries please feel free to ask in this thread or my PM.

EDIT - Update including overclocking and BF3 1080p "Ultra" results


Small update as I have been spending some more time fiddling with the overclocks and playing around with 1080p with my monitors in PLP.

First up; overclocking

Based on my testing so far I have managed to achieve the following clocks:

OC1 - 1075/6000 @ 1.112v & custom fan profile (tops out at 45%).
OC2 - 1125/6600 @ 1.175v & fan fixed at 60%.

OC1 is what I consider to be my 24/7 gaming clock as it quite happily churns away at my cards "stock" volts. The fan is not very intrusive either which is quite nice. The card under this profile tops out at around 75 degrees. Further testing has shown the memory to have much more "go" in it so this OC will be updated to 1075/6600 when I remember.

OC2 is my highest stable OC from my testing so far. BF3, Crysis 2 & Arcania are all rock solid at this clock speed, core voltage is set at 1.175 and the fan manually fixed at 60%. At this fan speed the card is very audible and gaming is only enjoyable with headphones (quite enjoyable actually - I think I will start to game like this a bit more).


Now for 1080p BF3 gaming benchmarks

After having some fun with BF3 in Eyefinity I decided to place my monitors in PLP setp so I could try some BF3 @ 1080p. Given that the HD7970 mullers BF3 @ 1080p at my normal settings, I decided to only test at Ultra settings with deffered AA applied (4x). Every setting was maxed in game with no ammendments made in CCC.

(Test is identical to my Eyefinity runs with the only adjustment being the resolution and this case the visual settings).

The results were:

HD7970 @ Stock

Min - 34
Avg - 48.8

HD7970 @ 1075/6000

Min - 39
Avg - 54.4

HD7970 @ 1125/6600

Min - 40
Avg - 58.8

So close to 60fps Avg I can almost taste it.

I need to do more testing and tweaking on the clocks but I am loving the way this card scales with its clocks at the moment.

 

Nice review. Thanks OP.

 

Awesome review dude, + rep, but do I spy NO MOLEX TO 6 PIN ADAPTOR? PRAISE BE TO JESUS

 

Indeed.

 

Great read

 

i dont think the retailers are the problem with prices, the prices at supply are also high, if i wanted to sell at close to the £420, i would be making less than 10%, ie it wouldn't cover business running costs.

edit: just checked suppliers price, and i would be in negative equity at £420

 

great review

 

Great stuff! Very interesting.

 

Great review.

I would suggest doing tests with the CPU overclocked, as it's almost certain that your stock CPU clock is bottlenecking the 7970 (and 6970).

I'm pretty sure the 2500K will do 4.5GHz on the stock cooler for "light" benching seeing that it doesn't require a high level of stability for that. Perhaps others could chip in?

 

Nice review... agree (obviously) with the pricing, it does seem to be a case of money grabbing. The biggest shame is that a major plus point with these cards, is that they scale VERY well.

If you're running a 2 card configuration at present, then buying one at this price isn't justifiable IMO. Tis a shame, but I'm sure they will come down in price shortly...it's been awhile since the crown was moved so it'll come off it's high horse soon hopefully

 

I spy an IBM / Lenovo thinkpad.

Great choice

 

I was tempted by a bit of a CPU OC but I want to get a new PSU first. Both GPU's were showing 99% load in BF3 which should indicate a minimal CPU bottleneck, especially at this resolution. I will be testig HD7970 CPU scaling in BF3 at a later date though as it is interesting.

The price is definately the biggest stumbling block but IMO it isn't as big a deal as some people are making it out to be (HD7970 costs £500 etc). OC Scaling is for want of a better word incredible. It is nearly linear.

It is my X201 Work Laptop

i5 520m IIRC with 4GB of memory. Enough for large Excel spreadsheets and general use. I don't like the location of the FN key though. It always mucks me up for keyboard shortcuts in Excel.

 

Good to hear. And fwiw, GPU usage doesn't indicate anything about a CPU bottleneck; it just means that BF3 is highly optimised and makes good use of all the graphics processing resources. It's likely that the GPU usage would increase exponentially with a CPU overclock, meaning that at 5GHz you'd still get 100% load on the GPU, but significantly better frame rates.

 

Good read, I'll refer back to posts like this when I decide whether Kepler is good enough to go for or not.

 

I think it is a case that I will have to try and see.

At 99% load it is most likely that my GPU isn't being held back by the CPU but the game itself could benefit from more CPU grunt.

Other tests have shown BF3 to simply not care about CPU overclocks so it could be a bit of a dead end but it will still be interesting to test.

It is something that I have always found hard to test as even comparisons to fundamentally identical PC's can be flawed due to driver issues or those little game quirks that annoy us all (like Arcania not loading at all due to a slight change in resolution).

Tonight I may try a quick OC to 4GHz on all four cores to see if the framerate budges at all.

 

BF3 really isn't that fussed by CPU grunt so I'd expect minimal gains. As far as recent games go, Skyrim is very CPU dependent, might be worth noting differences there

 

+1 to this, BF3 will not be held back by a stock 2500k at all.

 

Don't have it unfortunately. Ummed an Ahh'd in the Steam holiday sale and missed out.

TBH I am dumping enough time into BF3 at the moment I don't want another game sat on the shelf doing nothing - especially when I can pick it up later, patched and cheaper.

As a quick aside I didn't properly bench Multiplayer BF3 last night but I did take note of the rough Framerate range.

"HD6970 Playable settings" - Metro 50 player (full server) conquest mode.

HD6970 - Framerate between 30 and 40fps with an average closer to 35.

HD7970 @ 1075/6000 - Framerate between 45 and 60fps with an average closer to 55.

Ultra settings without AA & SBAO instead of HBAO should be close to 40fps average which at this resolution is pretty sweet.

The HD7970 was MUCH smoother and at times I forgot I was actually running such a high resolution. On my HD6970 the framerate was good but the game did have the occassional stutter when the action got really heavy (lots of smoke).

The minimums certainly hold up much better and the framerate is overall more consistant.

At this point it was 23:50 and I had to get up at 5:20 for work so I couldn't test anything else.

I will run some proper multiplayer benches later on the HD7970 only. I know the HD6970 always pushed around 30-35 regardless of map so I do have a reasonable comparison to work from.

 

It's not the game that's being held back, it's the hardware - that's what a bottleneck is, lol.

Considering the performance guys get out of graphics cards by overclocking CPUs to 6GHz, I think it's safe to assume that GPU performance in general scales with CPU frequency, irrespective of the sofware/game. I've done tests myself which corroborate this, and it'll be interesting to see what xaser finds with his system.

 

I think they're just saying that people tested like 3 GPUs in a rig, with stock and overclocked i7's and they gained literally next to nothing by doing it, something like 32 against 33fps or something like that.