The AMD HD 6950 is the second best single-GPU card in the AMD graphics card lineup. It is possible to flash the card to use the 6970 BIOS, so the HD 6950 looks like it may be able to pack a punch and seriously rival its competitors.
Features and Specifications
The HD 6950 features a hefty 2 GB of GDDR5 RAM, an 800 MHz core speed and a 1250 MHz RAM speed. The 6900 series features several new features, but since these have been mentioned in depth in our launch article and review on the HD 6970, we’ve just briefly summarized them here (courtesy of AMD, from their Press Deck):
- New core design accounts for “10% improvement in performance per square mm”
- Enhanced Quality Anti-Aliasing (EQAA) which features better coverage per pixel without increased memory usage
- Improved power consumption efficiency through AMD PowerTune technology
- Major tessellation performance enhancements in the latest games and benchmarks
- Significant CrossFire scaling improvements over the 5800 series
- Dual BIOS
HD 6950 Product Specifications
The CrossFire scaling issue is a big one. Apparently, the 6970 should scale at 80%-100% – that is, two 6970 cards should achieve 1.8x – 2x of the performance of a single card. This compares favourably with the 60%-80% seen with the HD 5870 (two cards perform 1.6x to 1.8x as fast as one). This topic, specific to the 6950, will be explored in a future article where we will put the theory and the marketing to the test. Yes, we have two 6950’s!
With an MSRP of $299, the HD 6950 fills a hole in the marketplace. From a price perspective, NVIDIA does not have any products that compete directly with the HD 6950. The GTX 470 is $55 less and the GTX 570 $50 more than the HD 6950. This huge $100 gap in pricing leaves the door open for AMD to exploit the mainstream gamer market. Even with strong relative price positioning, the HD 6950 still needs to perform well to be an attractive package for gamers and overclockers.
|GTX 460 (2gb)||$229.99|
|Lowest pricing per model at Newegg.com as of Dec. 16, 2010|
This handy slide from AMD illustrates the above point. However, it is always prudent to do a bit of your own research and take manufacturer data with a grain of salt.
Based on the price and AMD’s data, the HD 6950 ought to outperform the GTX 460 significantly. The GTX 470 comes in at $249.99 at time of writing yet is conspicuously missing from AMD’s slide. The GTX470 should be a closer competitor, but still fall short of the 6950 if the marketing speaks truth. If the results reflect these inferences, we have a solid card on our hands.
Test Setup and Methodology
|Processor||Intel i5 655k @ 4.85 GHz|
|RAM||Kingston HyperX DDR3-2000 4GB|
|Video||AMD HD 6950|
|Power Supply||Corsair 650w|
|Operating Systems||Windows Vista x64 (3DMark11 benchmarks were run in Windows 7 x64)|
All benchmarks were run three times using the “performance” preset and the average result is displayed. The 3DMark scores presented are run using the Performance (‘P’) preset.
Important Note: None of the graph abscissa ranges begin at zero. This was necessary to be able to display the differences in the scores, which are typically very high numbers all round. The actual scores are printed on each graph.
Due to the PowerTune technology the card will automatically adjust the voltages based on the GPU load, which was immediately disabled by bumping the voltage control to “+20%” in Catalyst Control Center (CCC). Even though CCC has an ATI OverDrive tab, the overclocking was limited to 840/1315 (about a 5% overclock). However, using Sapphire TRIXX this card was overclocked to 880/1380 without any issues whatsoever. This is a 10% overclock, however, bumping up either the memory or core any further let to immediate instability when running 3D benchmarks or games. Though TRIXX does provide a setting for voltage adjustment, it was locked.
Achieving a 10% gain in both the memory speed and core speed is not too shabby considering that it was done with locked voltage. With a custom BIOS or volt-mod this card should have some extra potential, especially as the card temperatures never held me back. The cooling system is the same vapor-chamber design as the 6970, and proved up to the task of cooling this card. On the AUTO setting the fan was nearly silent, but luckily CCC allows for manual control of the fan. All of the tests were run with the fan at 50% which was fairly loud, but it certainly wasn’t annoyingly so. Turning the fant up to 100% requires some loud music or headphones to mitigate the noise. Most users will never need the 100% fan setting, but I am eager to turn the cooling system up to full blast once the card is capable of getting fed more voltage.
With the overclock maxed out, the HD 6950 achieved approximately 6% performance increase over the stock clocks.
Similar to the Vantage results, the card was able to perform 7% better at its maximum overclock. The performance scales pretty linearly with core and memory speeds, especially for the more demanding 3DMark11.
Without a slew of video cards to compare in the same machine, we came up with a different solution for comparing relative performance. This solution is not without its flaws, but it should give a relatively clear picture. My system score (or CPU/Physics score) trended on the low side, so this analysis will include my average system/CPU score from both 3DMark Vantage and 3DMark06. Using a static system score, I was able to sub in the varying GPU scores into the score calculation equations provided by Futuremark.
GPU scores were generated by taking an average of the top 5 scores of each of the respective video cards on HWBot that met the following parameters:
- Cooling: Stock or Air
- Cores/# of Cards: 1
- i5 CPU
- Cooling: Water or Air
This type of testing gives the clear advantage to HWBot scores for several reasons:
- These are the Top 5 scores that meet the above parameters, not just typical user scores. Many of these scores were achieved by some of the top overclockers on HWBot.
- On average my CPU score was significantly less than the HWBot scores. Though I removed the CPU scores from the results, the overall system still does impact final scoring.
- Scores on HWBot may or may not have had the advantage of custom BIOS or increased voltages, while the HD 6950 voltage is locked.
Note: All scores for the HD 6950 below were run at the maximum overclock I achieved, 880/1380.
Though the HD 6950 only outperforms the GTX 460 by 2.5% and the GTX 470 by a slimmer margin, the results are still pretty good.
As predicted, the HD 6950 outperforms both the GTX 460 and 470. Though the HD 6950 only shows a 4% performance increase over the GTX 470, this still is a victory in my opinion. Considering the type of competition on HWBot, even a small increase means the HD 6950 is capable of going even further with more voltage, new drivers and a more experienced benchmarker. Also, it is important to note that the HD 6950 has a solid 11% performance gain over the GTX 460.
Unfortunately, there were not enough submissions on hwbot.org to support my above guidelines for the GTX 570. However, by eliminating the CPU guidelines, I managed to do a piecemeal comparison of the HD 6950 and GTX 570. I averaged out thirteen results based on my guidelines (except restricting it to i5 CPUs) and got a Vantage score of 20922, a 2759 difference from the top HD 6950 score. This accounts for a 15% performance dip between the HD 6950 and GTX 570, but considering the 17% difference in price, this is a pretty reasonable disparity. The GTX 570 is intended to be a direct competitor to the HD 6970, rather than the HD 6950.
Flash to HD 6970 BIOS
After seeing what the card could do at stock, it was time to take the HD 6950 to the next level by flashing it to the 6970 BIOS. For more information on how to do this, see techPowerUp.com’s excellent article. Using their simple tutorial, this $299.99 piece of hardware became equivalent to a $369.99 GPU in less than five minutes. Unlocking the extra shaders and setting 6970 stock speeds allowed for a solid 9% performance boost from stock HD 6950 Vantage scores. The BIOS also yields more overclocking headroom, presumably because of additional stock voltage.
I was slightly disappointed that my particular card maxed out at 935/1425, but considering this was a 6950 and not an actual 6970, I had no complaints. The graph below compares my maximum achieved HD 6950 overclock (880/1375) to the card with 6970 BIOS at stock 6970 speeds and my max overclock. The card approaches, but doesn’t reach, the performance of a GTX480. I’d argue that the price/$ is better, however.
|Card||Vantage Score||Price||Vantage Pts. per $|
|HD 6950 (w/HD 6970 BIOS 935/1425)||19234||$299.99||64|
Value wise it is difficult to match the 6950, considering for $300 you are essentially getting a $370 card. From the chart above, it is clear that the 6950 provides superior performance per dollar when compared with the GTX 480, about 30% better value to be exact.
Ultimately, this card has a ton of potential: the possibility of major overclocking with the new high-tech cooling system (assuming voltage modifications or unlocking is possible), the potential to outperform competitors in the same or even a higher pricing bracket and of course the possibility to be a BOINT machine. At just $300, this card packs a major punch considering it can be unlocked to a 6970 with ease. When battling head to head, this card outperforms the GTX 470 with similar overclocks, exactly what AMD described in the press deck. Overall this card definitely lives up to expectations and should be high on any overclocker’s wish list.
It will be interesting what this card is capable of in a CrossFire configuration. This new line from AMD is said to scale near 90%, and I can’t wait to see the results. Stay tuned for my follow up article.