Mr. K6's 6950 -> 7970 Overclocking Review

[MrK6]

1/4/11 - Update - 6950 data collection complete, analysis to follow
I had some free time today so I ran all the benchmarks discussed and collated the data here:

6950 to 7970 Overclocking Review - Data and Analysis

The raw data is there, but no analysis yet (literally just finished the last Shogun 2 bench). Looking at the different sheets I've prepared as well as my notes in the margins, you can see where I'll be going for analysis. I'd not only like to look at raw performance, but also do performance/W and performance/$ indices. I'll also post up the FPS graphs from those benches that produce them and do some detailed overlays on what exactly overclocking gets as far as min, max, and average FPS. I've noticed some interesting trends already and I think the analysis is going to be very interesting, if not informative. Once complete, I'll put up a summary and conclusion in post 2 as well as some commentary on the benches and results. Also, if anyone is really dying to see something added, now is the time to ask as next Tuesday/Wednesday the 6950 will be out for good. Enjoy!


For the time being, I'm using the first post as announcements and an idea board. I'd like to do a review that also helps out others, so for the time being I'd like people to chime in on what they'd like to see compared, and I'll try to create a setup that covers most of it.

[Test Setup]

System Setup:
CPU: i5-2500K @ 5.0GHz
Motherboard: Gigabyte Z68XP-UD4
RAM: G.Skill Sniper 2x4096MB DDR3-2133 @ 2134MHz 10-11-10-30
SSD : 160GB Intel X25-M G2
Graphics Cards:
Sapphire AMD HD 6950 2GB @ stock, 1408 shaders, 800MHz/1250MHz, 11.12 Cats
Sapphire AMD HD 6950 2GB @ unlocked and overclocked, 1536 shaders, 1000MHz/1400MHz, 11.12 Cats
TBD AMD HD 7970 3GB @ stock, 2048 shaders, 925MHz/1375MHz, TBD Cats
TBD AMD HD 7970 3GB @ overclocked, 2048 shaders, xxxxMHz/xxxxMHz, TBD Cats

Water Cooling:
Radiators: Swiftech MCR220, XSPC X240, 2x Gentle Typhoon AP-15 120mm fans each, no shrouds
Pump: Swiftech MCP350 w/ XSPC acrylic top
CPU: XSPC Rasa
Graphics cards: MCW80 on 6950, MCW82 on 7970
Barbs and Tubing: Bitspower 1/2" High Flow True Silver and 7/16" Masterkleer

Official Benchmark Suite:

3DMark11 Performance Preset -
Unigine "Heaven" Extreme - 1920x1080, 4xAA; 2560x1600, 4xAA
Skyrim custom benchmark - 1920x1080 4xMSAA; 2560x1600 4xMSAA
Witcher 2 Official benchmark - 1920x1080; 2560x1600
Metro 2033 Official benchmark - 1920x1080 4xMSAA; 2560x1600 No AA
Crysis 2 Adrenaline benchmark - 1920x1080 Extreme; 2560x1600 Extreme
Crysis: Warhead Frost Bench - 1920x1080 4xMSAA; 2560x1600 No AA
BF3 custom benchmark - 1920x1080 Ultra, 4xMSAA; 2560x1600 No AA
Shogun 2 In-game benchmark - 1920x1080; 2560x1600

12/25/11 - Original Post/Mission
Right now, I'm looking at testing the 6950 at completely stock (1408 shaders, 800MHz/1250MHz, etc.) and unlocked and overclocked (1536 shaders, 1000MHz/1400MHz) and the 7970 at stock and completely overclocked (whatever that turns out to be). I also want to look at power consumption for both. Noise and temps I feel will be irrelevant since I'm on water cooling, but I'll make note of any global changes I see (overall temp changes, squealing, etc.). The biggest part I'm trying to figure out is what games suite to use. I'll basically use whatever settings are playable at 2560x1600 (even barely) on the stock 6950 and then keep these settings consistent throughout the 7970 and overclocks. That way, it'll stay apples-to-apples. I'll also make comments on my overall gaming experience and how it changes probably a week or two after I've had the card and time to really get to tweak it and feel it out.

 

[MrK6]

6950 OC Results and Commentary:

I'll throw up some nice, concise results here once I finish all of the analysis, but for now just some tidbits I think are interesting as well as some general commentary on the whole process and results.

As far as benchmarking goes, I'm pleased with the way my custom benches turned out. I think a true test of their merit will be seen once I get the FPS graph comparisons up. My current plan is to see how synced each run is. If they are very off, I will overlay and resync the graphs, cutting the time demo in the process, to make more comparable results. If this is confusing right now, I think it will be cleared up once I show the process.

Some comments regarding the suite chosen:

[ranting]Crysis 2 support is pretty ridiculous, the website is crap and NVIDIA's website is the only way to get patches/updates. Don't mind my soap box, but I think it shows just how much Crytek sold out on this one, I'm glad I only paid $10 for the game. The game itself looks good, but it lacks are certain sharpness/fidelity that the first game had. I think it's the use/overuse of Adaptive AA, but really, there's little difference visually between running 1080p and 1600p, which should never be the case. The game really doesn't interest me either, I actually fell asleep during the intro.

Shogun 2 is another game that was surprising. I have a friend who loves it, as he's really into strategy games and Japanese culture (not a weeaboo, just likes sushi and what not), so I was interested in checking it out. As I mentioned before, $7.50 is a pretty lenient entrance fee. The story and ambiance of the game are interesting, but the graphics are mediocre, and certainly don't warrant the performance seen in the benchmarks. The only reason I could see performance being so poor is draw distances, which really is just poor coding in my book.

BF3 was another game I was disappointed with. I played multiplayer for about 3 weeks and never even cracked up the single player, and I'm glad I didn't waste my time. This game is consolized crap and I can't believe I wasted money pre-ordering it (at least I got Karkand for free, if I ever play it). Everything from the "Press Start" main menu to unskippable intro movies and no quicksave screams that this was a console port and DICE was talking out of both sides of their mouths. They're not getting another pre-order from me, and definitely not $60.[/ranting]

For results so far, I think they're very interesting. As n7 pointed out, it's surprising that overclocking gives higher yields at 1080p than 1600p, considering AMD's penchant for higher performance at higher resolutions. My guess is there's a performance wall there beyond clock speeds. An overall 18.6% return on a 25% overclock is pretty decent though in my book (+ shaders unlocked, which is ~3% of that or so), especially since it's free. Looking at the benchmarks, that's ~GTX 580 speeds for only $230.

Looking at the vRAM usage results, I think in games like the Crysises a vRAM wall might be the culprit. The 3GB 7970 will hopefully shed some light on this. One thing to note is that vRAM usage for Skyrim increases the longer the game is played. Where as most benchmarks level off, Skyrim just keeps adding more, so don't take those results as indicative of real world gameplay. Overall, I think my results show that extra vRAM is always put to use, even at 1080p. Whether or not there is a performance penalty yet for only using 1GB remains to be tested.

The power consumption results are great, and I think they just prove what I've recommended time and time again on these forums: it matters what you get for your watts. Using an efficiency factor of 80% (which is generous, considering the age of my PSU), my system peaks at ~267W (334W from the wall) during typical gaming loads. The average is substantially less, probably ~85% of that to give a rough estimate. This is with the CPU at 5.0GHz and the GPU at 1.0GHz. Furthermore, you can see through a quick index analysis that my overclocked 6950 actually has higher performance/watt than a 6970 or a GTX 580, while being much cheaper (obviously).

More to come as I work on this.

 

[MrK6]

Reserved 2 (7970 Impressions, Stock performance, and Air Overclocking)

 

[MrK6]

Reserved 3 (7970 Water Overclocking and Conclusion)

 

[njdevilsfan87]

If you could throw in some 1080p with 4x/8x AA I'd really appreciate it, just because I that's what I use plus it's a popular gaming resolution.

 

[(sic)Klown12]

If you could, I'd like to see some numbers for Batman Arkham City in scenes that are heavily tessellated(missions with Poison Ivy should work well). Interested in how much real-world improvement there is in Gen 9 tessellators(the 6970 has a large 18% drop from off to high compared to 6% for GTX 580).

 

[Pulsar]

Do not test it with skyrim. Skyrim has some very significant issues with processor bottlenecking. Bethesda (morons that they are) did not compile the executables with the normal instruction sets (because it's a port from consoles).

I'm running eyefinity across 3 24 inch lcd's, max settings on everything with a single 6950 and I'm STILL cpu limited with a brand new overclocked sandybridge.

There are far better graphically intensive games, like metro.

 

[OVerLoRDI]

I'm also curious about The Witcher 2. Sadly it doesn't have an included benchmark...

Are you watercooling this card? If so I'd love to see some full load temps, in a torture test like Furmark, OCCT.

If I remember correctly you aren't going full coverage right? Can you report back some VRM temps as well? Have to see how hot they get to see if full coverage blocks are worth it on these cards.

19x12 benches would be appreciated as well. I won't be looking for at 2560x1600 screen until they are 120hz, if that ever happens

 

[MrK6]

If you could throw in some 1080p with 4x/8x AA I'd really appreciate it, just because I that's what I use plus it's a popular gaming resolution.

Sure, I can do that as long as the game list isn't too long. I'll probably do 1080p and whatever MSAA mode was most playable on the stock 6950.

If you could, I'd like to see some numbers for Batman Arkham City in scenes that are heavily tessellated(missions with Poison Ivy should work well). Interested in how much real-world improvement there is in Gen 9 tessellators(the 6970 has a large 18% drop from off to high compared to 6% for GTX 580).

I have the game but haven't gotten to it yet (Skyrim and SWTOR = all my time). If I get to those points before the card is released, will do. If not, it'll be whatever benchmark they have with the game .

Do not test it with skyrim. Skyrim has some very significant issues with processor bottlenecking. Bethesda (morons that they are) did not compile the executables with the normal instruction sets (because it's a port from consoles).

I'm running eyefinity across 3 24 inch lcd's, max settings on everything with a single 6950 and I'm STILL cpu limited with a brand new overclocked sandybridge.

There are far better graphically intensive games, like metro.

I hear you, I'm CPU limited with my 2500K @ 5.0GHz D:. Point well made, I'll probably take it off the list.

I'm also curious about The Witcher 2. Sadly it doesn't have an included benchmark...

Are you watercooling this card? If so I'd love to see some full load temps, in a torture test like Furmark, OCCT.

If I remember correctly you aren't going full coverage right? Can you report back some VRM temps as well? Have to see how hot they get to see if full coverage blocks are worth it on these cards.

19x12 benches would be appreciated as well. I won't be looking for at 2560x1600 screen until they are 120hz, if that ever happens

Yes! Witcher 2, totally forgot about that one, I'll definitely add it. The 7970 will be watercooled, but I'm not sure what load temps will tell you beyond the efficiency of my individual loop. I actually just ordered an MCW82 for the 7970, so comparing it to the MCW80 on my 6950 wouldn't be fair since the flow rates will be higher with the MCW82.

To be honest, I think full cover blocks are a waste of money unless you want them for the bling or are going for the most insane clocks you can on water. They add extraneous heat to the loop that is easily taken care of by case airflow. For example, with some basic copper mosfet sinks and a 92mm Panaflo @ 7V pointed at them, my 6950's mosfets don't go above 60C @ 1.3V 1000MHz/1400MHz. The other thing is full cover blocks are never released with the cards, and I hate waiting .

 

[Silverforce11]

Anno 2070. No review site i know of has done this new dx11 game, its very demanding.

 

[OVerLoRDI]

To be honest, I think full cover blocks are a waste of money unless you want them for the bling or are going for the most insane clocks you can on water. They add extraneous heat to the loop that is easily taken care of by case airflow. For example, with some basic copper mosfet sinks and a 92mm Panaflo @ 7V pointed at them, my 6950's mosfets don't go above 60C @ 1.3V 1000MHz/1400MHz. The other thing is full cover blocks are never released with the cards, and I hate waiting .

I prefer full coverage blocks for the bling. So logically, you are correct, but I frown on your logic and raise you pretty nickle plated blocks

Also are we seriously going to have an argument about watercooling being worthwhile financially? That is not the point of such toys. So again,

Also the wait should be minimal this time around:
"Radeon HD 7970 is now officially released! EK already has prototype water blocks and is optimizing them for best... "
http://twitter.com/EKWaterBlocks

 

[badb0y]

I am going to be on air cooling so we could compare our temperatures/overclocks. Hopefully the Shaman fits on the 7970....

 

[amenx]

Do not test it with skyrim. Skyrim has some very significant issues with processor bottlenecking. Bethesda (morons that they are) did not compile the executables with the normal instruction sets (because it's a port from consoles).

Yep, so you're aware of the TESV Application Layer mod that takes care of this I presume.

 

[MrK6]

Anno 2070. No review site i know of has done this new dx11 game, its very demanding.

Any way to get a benchmark without purchasing the game? I googled quickly and found nothing.

I prefer full coverage blocks for the bling. So logically, you are correct, but I frown on your logic and raise you pretty nickle plated blocks

Also are we seriously going to have an argument about watercooling being worthwhile financially? That is not the point of such toys. So again,

Also the wait should be minimal this time around:
"Radeon HD 7970 is now officially released! EK already has prototype water blocks and is optimizing them for best... "
http://twitter.com/EKWaterBlocks

Water cooling is actually pretty inexpensive if you buy correctly. My entire setup costs about $350 or so, and that's if you get everything new like I did. For me it's worth it because I can get great clocks out of my gear while having a silent case. This is especially helpful since I like to run single GPU setups. Lots of money can be spent on things like compression fittings and full cover blocks. Personally, I never look inside my case's window except to check fluid levels once a month, so I won't pay extra for "bling" I'm not going to enjoy. That said, I'd rather save my hobby money for expensive graphics cards and the like. Just a different approach, neither is less correct than the other :thumbsup:.

EK usually hypes their parts up at launch, but it will be at least 2 weeks and usually a month before we can get them here. And they'll probably be at least $100-120. I'm not knocking anyone who wants one, but I'm not getting one.

I am going to be on air cooling so we could compare our temperatures/overclocks. Hopefully the Shaman fits on the 7970....

I generally include global data such as idle temp, load temp while gaming, Vreg temps, etc. just for general information. I can run OCCT like OverLordi requested, what else would you guys like to see?

Yep, so you're aware of the TESV Application Layer mod that takes care of this I presume.

I just read up on. People seem to be reporting anywhere from 4-20FPS boosts so screw it, I'll run Skyrim too and we'll see if a 7970 makes any difference.

 

[skipsneeky2]

Orginal crysis .

Be sorta neat to see how the 7970 fares despite being old its still a gpu hungry game and a true pc port.

 

[chimaxi83]

Thanks for this OP, I was going to look for something like this once the cards were in first buyers' hands and you've got something in motion already. Awesome, subbed :thumbup:

Edit: I second the request for 1080p, I haven't moved into 1600p yet along with a bunch of us I'm sure. Gotta see if a card like this is worth it at this resolution.

 

[RussianSensation]

MrK6, sounds like you on the right track. Looking forward to your review. I think the performance increase from HD6950 @ 1000 to HD7970 @ 1165+ is going to be FAR greater for you than it was when you went from 5850 to 6950.

Before you disassemble the HD7970, it would be helpful if you overclocked the card on air so we would have reference point since your 6950 went to 1000mhz on water, which is way higher than the reference cooler .

I think you should add Crysis 1 / Warhead (I know those are ancient games), but it could be fun to see if you can break 60 fps at 1080P with a heavy overclock.

Also, what about Dragon Age II or Shogun 2? Those are very demanding.

 

[MrK6]

Orginal crysis .

Be sorta neat to see how the 7970 fares despite being old its still a gpu hungry game and a true pc port.

I'll probably end up doing Warhead Frost Bench for the simple reason that that's what Anandtech has benchmarked and it would be nice to compare my results with Anandtech's. My thoughts are that with two different "stock" cards to adjust a baseline, it should be easy to draw/extrapolate comparisons with the overclocked versions.

Thanks for this OP, I was going to look for something like this once the cards were in first buyers' hands and you've got something in motion already. Awesome, subbed :thumbup:

Edit: I second the request for 1080p, I haven't moved into 1600p yet along with a bunch of us I'm sure. Gotta see if a card like this is worth it at this resolution.

That will double the amount of benchmarking, so I'll probably keep the games list short, but I agree that it will be much more helpful to everyone. Would everyone prefer 1080p or should I use 1920x1200 so it can be directly compared to Anandtech's results like the 2560x1600?

MrK6, sounds like you on the right track. Looking forward to your review. I think the performance increase from HD6950 @ 1000 to HD7970 @ 1165+ is going to be FAR greater for you than it was when you went from 5850 to 6950.

Before you disassemble the HD7970, it would be helpful if you overclocked the card on air so we would have reference point since your 6950 went to 1000mhz on water, which is way higher than the reference cooler .

I think you should add Crysis 1 / Warhead (I know those are ancient games), but it could be fun to see if you can break 60 fps at 1080P with a heavy overclock.

Also, what about Dragon Age II or Shogun 2? Those are very demanding.

I'm interested as well, I think it will be higher than the ~%15 I got from the 5850 -> 6950 jump. This is all the more reason I'm going to do a very detailed stratified analysis. I also think I'm going to include FPS graphs and throw up everything on a Google Doc sheet so everyone can see it and access at will. I don't own either of those games, but Shogun 2 does have a built in benchmark (which is what Anandtech used I imagine). It's also only $15 on Steam right now, maybe less with a daily sale, so I'll probably pick it up and throw it in. I'll also do a quick overclock on air to see how far I can push it (especially if no voltage tweak applications are available), and will run a quick benchmark (metro?) to compare, but I'll run the full suite once the card is on water with its final, 24/7 clocks.

 

[Pulsar]

Any way to get a benchmark without purchasing the game? I googled quickly and found nothing.

Water cooling is actually pretty inexpensive if you buy correctly. My entire setup costs about $350 or so, and that's if you get everything new like I did. For me it's worth it because I can get great clocks out of my gear while having a silent case. This is especially helpful since I like to run single GPU setups. Lots of money can be spent on things like compression fittings and full cover blocks. Personally, I never look inside my case's window except to check fluid levels once a month, so I won't pay extra for "bling" I'm not going to enjoy. That said, I'd rather save my hobby money for expensive graphics cards and the like. Just a different approach, neither is less correct than the other :thumbsup:.

EK usually hypes their parts up at launch, but it will be at least 2 weeks and usually a month before we can get them here. And they'll probably be at least $100-120. I'm not knocking anyone who wants one, but I'm not getting one.
I generally include global data such as idle temp, load temp while gaming, Vreg temps, etc. just for general information. I can run OCCT like OverLordi requested, what else would you guys like to see?
I just read up on. People seem to be reporting anywhere from 4-20FPS boosts so screw it, I'll run Skyrim too and we'll see if a 7970 makes any difference.

Yeah, I'm running the TESV sk loader to fix that issue, unfortunatelly with eyefinity the processor still bottlenecks handling the AA and AF. Drives me a bit nuts. I can turn on my crossfire and not see an increase of a single frame =/.

 

[chimaxi83]

Didn't see this mentioned, but where possible, can you document your VRAM usage too? I think it would be informative and maybe even helpful in buying decisions

 

[palladium]

Are you doing a CrossFire review by any chance..?

 

[MrK6]

Yeah, I'm running the TESV sk loader to fix that issue, unfortunatelly with eyefinity the processor still bottlenecks handling the AA and AF. Drives me a bit nuts. I can turn on my crossfire and not see an increase of a single frame =/.

I just read about "SkyBoost R3" which supposedly does even better than TESV sk loader, so I'm going to give that a run. I'm also going to turn uGrids down to 9 to take some stress of the CPU and hopefully find a good balance.

Didn't see this mentioned, but where possible, can you document your VRAM usage too? I think it would be informative and maybe even helpful in buying decisions

That's easy enough to do, I'll throw up the highest usage seen by each benchmark on each card. One thing I'll hypothesize right now is I think you'll see the 7970 use more vRAM than the 6950 under the same testing situations simply because it's there. That may not necessarily mean that the 6950 was vRAM limited if it does go over 2GB on the 7970, but that the programs are taking advantage of the extra RAM on the 7970 to do more caching or whatever.

Are you doing a CrossFire review by any chance..?

Lol, no. Not only would that put a pretty big dent in my wallet, I'm not a fan of multi-GPU in the least. I'll give it a shot again when it's a little cheaper, but merely just to try it out again and see if there are improvements.

For an update, as I hoped, Shogun 2 went down to $7.49 so I snagged that and added it to the suite. I'm also updating the first post to flesh out more of what I'll be covering and what to expect in regards to data and setup. Thanks for all the interest so far and helping flesh this out.

 

[MrK6]

I also forgot to mention that I already received the MCW82, RAM sinks and MOSFET sinks, so I should be ready to go when I get the 7970. Unfortunately, I'll still have to completely pull apart my loop, scrub everything down, put in new tubing, and re-fill/assemble to put the 7970 on water. What I'll most likely do is put the card in to verify it works and I'll run some quick benches (but not the full suite) at stock and at maximum air overclock just to get some numbers. I also will post a few videos of the process, probably at least after I've put the block on the card. When I first started water cooling there were few videos showing what had been done and where as far as setting up hardware, so I thought it might help some people.

 

[MrK6]

Alright, collected the data on the 6950 both stock and overclocked and have collated it into a Google Doc for convenient viewing. I'm going to start the analysis, so you'll see it added in real time. I'm pretty excited to see what comes out of it as I've already noticed some interesting trends. See the OP for the full update.

 

[Pulsar]

I have a vested interest since I'm running dual overclocked 6950's.

What voltages are you running, and did you have any stability problems?

How long were your benchmarks (for instance, in skyrim?)