I've read way too many “which graphics card should I buy” threads and there's one recommendation that pops up a lot. It's that you should buy a Radeon HD 6950 and unlock the extra shaders to turn it into a Radeon HD 6970. With all the hype surrounding the Radeon HD 6950's unlocking ability, I wanted to find out how much of a difference unlocking the extra shaders really makes. So began my weekend of running benchmarks and compiling the results.
AMD Radeon HD 6900 Series Specs
Here are the specs that matter for my tests:
Stock HD 6950
Test Setup
I'll be using my unlocked HD 6950 2GB to simulate a HD 6970. Everything that I have read suggests it should perform identically to a real HD 6970.
Below are the specs of my test system:
Unlocking the HD 6950 2GB
I'm not going to go into detail on how to unlock the HD 6950 as there are many threads on this subject already. Instead, I'm going to explain which method I used and why.
After reading several forum posts on many different sites I decided not to flash my HD 6950 using a HD 6970 bios. There seems to be a common agreement that problems could arise from using the HD 6970 bios as the memory voltage and timings are different. I don't have any data to back this claim up, but I still chose to go with the “safer” way of modifying my original bios to unlock the additional shaders. This method allowed me to keep the cards core speed, memory speed, timings & voltages totally stock while unlocking the additional shaders. Just a note, a friend of mine used the HD 6970 bios method and hasn't had any problems with it. Your mileage may vary.
Overclocking
For my tests I also wanted to provide benchmarks for the max stable overclock I could reach on my particular card. Stable is a relative term and could mean different things for different people. It could also mean something different depending on your goal. For me, stable means rock-solid, 24/7 usage, stock-like stability.
I set out to find the max core and memory speeds I could use at the lowest voltage possible and remain 99.9999% stable. While I was able to boot at 980MHz core and run through several loops of Crysis, it wasn't stable. Sure, I could run Furmark for 24H at 950MHz, but it would crash after a few hours of real gaming. The best test I've found for stability is running the Crysis benchmark 99 times on the highest settings. If it can pass this, I'm totally satisfied.
As a result, the max stable overclock for my card was 920MHz core @ 1.175v and 1375MHz memory. No matter how much voltage I threw at it (up to 1.3v), I couldn't get stability. Is it a coincidence that the voltage required for this speed was that of a stock HD 6970? It failed at 1.165V and passed at 1.170V but I added 0.005V just to be safe. Is it another coincidence that my max memory speed was that of a stock HD 6970? Anything over 1375MHz and I lost performance. Above 1400MHz and I got artifacts. Also, unlocking the shaders didn't affect my overclock. Whether it was 1408 shaders or 1536 shaders, the max core speed remained the same.
Summary: Max stable overclock is 920MHz core @ 1.175V and 1375MHz memory
Your mileage may vary.
3DMark Vantage Test
For this test I used version 1.0.2 on the performance preset with PPU disabled. I ran the test 2 times for each setting and if the results were close, I used the average of the 2 runs. If not, I re-ran the test until I found consistent results.
3DMark Vantage seems to favor the extra shaders more than any other test. As you can see, there's a 3-4% increase clock for clock by unlocking here.
3DMark 11 Test
For this test I used version 1.0.1.0 on the performance preset. I ran the test 2 times for each setting and if the results were close, I used the average of the 2 runs. If not, I re-ran the test until I found consistent results.
3DMark 11 gives slightly more modest results. Clock for clock the extra shaders provide 2-3% more performance.
Crysis: Warhead Test
For this test I used version 1.1.1.687. Gamer Quality seems to be the most playable, but Enthusiast looks amazing and begs for Crossfire 6900's.
To me, Crysis: Warhead represents more of a real world benchmark. The gamer quality setting provides more ideal frame rates for playability. The extra shaders give a 1-3% increase here.
Cranking up the quality to Enthusiast really brought the HD 6950 to it's knees. As a result, the extra shaders only bump up performance by 2%.
Battlefield: Bad Company 2 Test
I used the latest build as of 2/21/2011. I used Fraps to benchmark the first 145 seconds after the guy says "Hey".
The extra shaders give a 1-3% increase here.
Overall Performance
I took the % numbers from all the tests and averaged them out.
Overall the extra shaders still provide 2-3% of an increase.
Closing and 2GB vs. 1GB
So... Did unlocking the shaders really make a difference? Yes and no. Yes, a 2-3% is an increase, but does that small increase really matter? I say no, not really. Especially since the more intense tests and higher core speed resulted in a slightly less performance increase. It was also interesting to find that the extra shaders do not scale with clock speed. Whether at 800MHz or 920MHz, the extra shaders still only gave a 2-3% increase. In some cases, you actually got less of an increase.
When I bought my card, the going price for the HD 6950 was $300 and the HD 6970 was $370. My results show that you get the same, if not more performance, for 81% of the cost whether you choose to unlock the card or not. With current price drops, that number drops to 75%. If you opt for the 1GB version, it drops even further to 68% (rebates not included). In my opinion, the HD 6950 1GB is the best card you can buy for the money. Sure there are faster cards, but the performance you get for what you pay is great.
Unfortunately I do not have the 1GB version to test. All the reviews I've read show that the 1GB version performs virtually identical to the 2GB version, even at 2560x1600. It's my opinion that when having 2GB of video memory matters, maybe 2-3 years from now, this card will be irrelevant and you will have already upgraded. If the 1GB was available when I purchased my HD 6950, I would have definitely got it instead.
So why buy a HD 6970? The only reason I can see is for more overclocking headroom for both core and memory.
Summary
AMD Radeon HD 6900 Series Specs
Here are the specs that matter for my tests:
Stock HD 6950
- Unified Shaders: 1408
- Core Clock: 800MHz
- Memory Clock: 1250MHz
- Unified Shaders: 1536
- Core Clock: 880MHz
- Memory Clock: 1375MHz
Test Setup
I'll be using my unlocked HD 6950 2GB to simulate a HD 6970. Everything that I have read suggests it should perform identically to a real HD 6970.
Below are the specs of my test system:
- CPU: Intel Core i5 2500K @ 4.8GHz
- Motherboard: Gigabyte P67A-UD3
- Hard Disk: OCZ Agility 2 (120GB)
- Memory: Corsair Vengeance DDR3-1600 2 x 4GB (8-9-8-24 1T)
- Video Card: Sapphire AMD Radeon HD 6950 2GB
- Video Drivers: Catalyst 11.2 (8.821-110126a-112962C-ATI)
- OS: Windows 7 Professional 64-bit
Unlocking the HD 6950 2GB
I'm not going to go into detail on how to unlock the HD 6950 as there are many threads on this subject already. Instead, I'm going to explain which method I used and why.
After reading several forum posts on many different sites I decided not to flash my HD 6950 using a HD 6970 bios. There seems to be a common agreement that problems could arise from using the HD 6970 bios as the memory voltage and timings are different. I don't have any data to back this claim up, but I still chose to go with the “safer” way of modifying my original bios to unlock the additional shaders. This method allowed me to keep the cards core speed, memory speed, timings & voltages totally stock while unlocking the additional shaders. Just a note, a friend of mine used the HD 6970 bios method and hasn't had any problems with it. Your mileage may vary.
Overclocking
For my tests I also wanted to provide benchmarks for the max stable overclock I could reach on my particular card. Stable is a relative term and could mean different things for different people. It could also mean something different depending on your goal. For me, stable means rock-solid, 24/7 usage, stock-like stability.
I set out to find the max core and memory speeds I could use at the lowest voltage possible and remain 99.9999% stable. While I was able to boot at 980MHz core and run through several loops of Crysis, it wasn't stable. Sure, I could run Furmark for 24H at 950MHz, but it would crash after a few hours of real gaming. The best test I've found for stability is running the Crysis benchmark 99 times on the highest settings. If it can pass this, I'm totally satisfied.
As a result, the max stable overclock for my card was 920MHz core @ 1.175v and 1375MHz memory. No matter how much voltage I threw at it (up to 1.3v), I couldn't get stability. Is it a coincidence that the voltage required for this speed was that of a stock HD 6970? It failed at 1.165V and passed at 1.170V but I added 0.005V just to be safe. Is it another coincidence that my max memory speed was that of a stock HD 6970? Anything over 1375MHz and I lost performance. Above 1400MHz and I got artifacts. Also, unlocking the shaders didn't affect my overclock. Whether it was 1408 shaders or 1536 shaders, the max core speed remained the same.
Summary: Max stable overclock is 920MHz core @ 1.175V and 1375MHz memory
Your mileage may vary.
3DMark Vantage Test
For this test I used version 1.0.2 on the performance preset with PPU disabled. I ran the test 2 times for each setting and if the results were close, I used the average of the 2 runs. If not, I re-ran the test until I found consistent results.
3DMark Vantage seems to favor the extra shaders more than any other test. As you can see, there's a 3-4% increase clock for clock by unlocking here.
3DMark 11 Test
For this test I used version 1.0.1.0 on the performance preset. I ran the test 2 times for each setting and if the results were close, I used the average of the 2 runs. If not, I re-ran the test until I found consistent results.
3DMark 11 gives slightly more modest results. Clock for clock the extra shaders provide 2-3% more performance.
Crysis: Warhead Test
For this test I used version 1.1.1.687. Gamer Quality seems to be the most playable, but Enthusiast looks amazing and begs for Crossfire 6900's.
To me, Crysis: Warhead represents more of a real world benchmark. The gamer quality setting provides more ideal frame rates for playability. The extra shaders give a 1-3% increase here.
Cranking up the quality to Enthusiast really brought the HD 6950 to it's knees. As a result, the extra shaders only bump up performance by 2%.
Battlefield: Bad Company 2 Test
I used the latest build as of 2/21/2011. I used Fraps to benchmark the first 145 seconds after the guy says "Hey".
The extra shaders give a 1-3% increase here.
Overall Performance
I took the % numbers from all the tests and averaged them out.
Overall the extra shaders still provide 2-3% of an increase.
Closing and 2GB vs. 1GB
So... Did unlocking the shaders really make a difference? Yes and no. Yes, a 2-3% is an increase, but does that small increase really matter? I say no, not really. Especially since the more intense tests and higher core speed resulted in a slightly less performance increase. It was also interesting to find that the extra shaders do not scale with clock speed. Whether at 800MHz or 920MHz, the extra shaders still only gave a 2-3% increase. In some cases, you actually got less of an increase.
When I bought my card, the going price for the HD 6950 was $300 and the HD 6970 was $370. My results show that you get the same, if not more performance, for 81% of the cost whether you choose to unlock the card or not. With current price drops, that number drops to 75%. If you opt for the 1GB version, it drops even further to 68% (rebates not included). In my opinion, the HD 6950 1GB is the best card you can buy for the money. Sure there are faster cards, but the performance you get for what you pay is great.
Unfortunately I do not have the 1GB version to test. All the reviews I've read show that the 1GB version performs virtually identical to the 2GB version, even at 2560x1600. It's my opinion that when having 2GB of video memory matters, maybe 2-3 years from now, this card will be irrelevant and you will have already upgraded. If the 1GB was available when I purchased my HD 6950, I would have definitely got it instead.
So why buy a HD 6970? The only reason I can see is for more overclocking headroom for both core and memory.
Summary
- Locked HD 6950 overclocked to HD 6970 speeds performs within 2% of a HD 6970
- My locked HD 6950 overclocked to max performs about the same as a HD 6970
- My unlocked HD 6950 overclocked to max performs 3% better than a HD 6970
Last edited: