MSI HD 6950 Twin Frozr II Review: Cool, Quiet, and somewhat Beastly
From the HD 5770 HAWK to Now
MSI hits the scene with their own version of the AMD HD 6950. This features the Twin Frozr II cooling solution and has a price point slightly above where most reference HD 6950s sit. When combined with MSI's Afterburner software, this card becomes an interesting tool in a gamer's arsenal. We find out how it runs when combined with the Catalyst 11.4 Preview Driver, which delivers some significant improvements in performance for the Cayman family of chips.Last year I reviewed the MSI HD 5770 HAWK video card, and I came away impressed by the engineering that MSI brought to the table. The card was quiet, it was efficient, it didn’t build up any significant levels of heat, and it was pretty affordable as compared to a bone stock HD 5770 based on the reference design. The board could also overclock. It was a budget enthusiast board that wouldn’t empty the pocket, but still give a lot of DX11 bang for the buck.
Then on the other hand we had the MSI HD 5870 Lightning. This was a card that had a lot of promise. This particular card had a custom PCB design with high end power circuitry, quality components, and the TwinFrozr fan design. All of this came to naught. The board would not overclock any further than the reference HD 5870 that we had seen for some months before, and in fact the board appeared to pull a little bit more power at the same speeds as a reference board. This was almost the exact polar opposite of the HD 5770 HAWK.
The product I am looking at today is an interesting hybrid from MSI. MSI has taken the stock HD 6950 reference PCB, populated it with slightly higher rated components (though not up to their “Military Class” standards), and put on the Twin Frozr II cooling solution. This is more in line with the reference version of the HD 6950, but the addition of better cooling and advanced fan profiles gives it a boost above the reference, without going into the stratified air of producing another “Lightning” type of product. This has allowed MSI to get a differentiated product out in fairly short order, and still give consumers something extra to potentially make their buying decision on.
The HD 6950
Due to the lack of a 32 nm (or 28 nm for that matter) process node to produce a new generation of chips upon, both AMD and NVIDIA were forced to refresh their lineup using the same old 40 nm process we were introduced to around two years ago. Cayman is the code name for the 6900 series of chips, and it has the distinction of being the largest chip since the ill-fated R600. This is not exactly a simple refresh though, and AMD has put a lot of new engineering tricks into the Cayman series. AMD has increased the size of the chip pretty dramatically from the HD 5870. That smaller chip contained around 2.15 billion transistors, while Cayman now hosts 2.6 billion. All the while featuring fewer shader ALUs than the previous generation.
Some snappy packaging from MSI which gives a tantalizing glimpse of the cooling prowess under the... plastic cover.
The biggest change is going from the 5 wide vector unit in the previous generation, and going to a new 4 wide vector unit. In the 5 wide there was a single “T” unit which handles the more complex mathematics needed with some instructions. In the new 4 unit, two of the units now can handle this particular functionality. This is actually a pretty hefty change, and it requires a new driver level compiler to get the most work out of this new generation of chip. Other changes include increased tessellation performance as well as a new coverage sample AA functionality that compares well to NVIDIA’s several year old CSAA scheme. AMD also fixed the texture filtering bug which could be quite annoying in the HD 5000 series of products with large, repetitive textures (such as roads in some racing games).
Power Tune is the final big feature that AMD included, and this was designed to give the user maximum performance but still restrict the card to reasonable power limits. There are some applications such as Fur Mark which will push a GPU so that it hits or exceeds its rated TDP, causing the chip to crash or the card to fail to supply enough power to the chip. There are only a handful of applications such as this which will cause a GPU to writhe and scream in agony. Most other applications do not feature such heavy usage workloads. AMD has clocked their chips to a level which adequately covers 99% of all applications and their power usage. To protect against these 1% of applications which could cause the GPU to exceed its rated TDP, AMD has instituted Power Tune.
Everything is well protected, and the card is surrounded by a large amount of foam. Shipping damage will likely be rare here.
Dedicated circuitry inside the GPU constantly monitors the power being utilized by the GPU. When it starts to detect that the chip is approaching or exceeding its rated TDP, it then starts to clock the GPU down. This essentially allows AMD to clock their GPUs up to the highest level possible for the majority of applications out there, yet still protect the chip and card when it works on an application which will cause the GPU to exceed its TDP. This functionality should not have a negative effect on overclocking the chip though. In most applications Power Tune can be overridden when overclocking is enabled, but the functionality will still protect the chip when it determines that it is getting hammered a bit too hard. Users can also access the Power Tune functionality to raise and lower the TDP limit for their particular card.