Conventional wisdom, or perhaps military compulsion, says that if one were to have multiple GPUs in their rig that they should all be the same.
That’s not exactly true.
While you may want to have everything neat, tidy and all matched up inside your case, sometimes you need to frankenstein things together. Whether it is due to an unexpected replacement or simply a cost saving feature, at some point you may have to step away from that neat, orderly and part-identical build into something a little more complex, and fun.
Several months ago I reached out to AMD and asked about mixing parts from different vendors. While it makes sense that the underlying technology is the same, therefore the parts should all work together nice, it was still something of which I had no confirmation. Would my Gigabyte R9 280X perform reasonably close to other vendors, or would I see differences in quality control and manufacturing play out in unexpected ways?
I managed to get my hands on a MSI R9 280X video card and decided to test the two cards against each other under identical conditions. Specification wise, they are nearly identical with the Gigabyte having a 1100 MHz Boost Clock versus the 1050MHz of the MSI. Both are the R9 280X series GPU with 3 GB of VRAM each. Other than the shroud and heatsink, I can find no difference between them.
This is not exactly a scientific test; I am looking to see the performance difference, if any, between the two video cards, and then compare those results to the two cards in CrossfireX. Will there be any noticeable difference between the two, and will the two cards play well together when in a multi-GPU set up? The only way to test these cards under identical settings is to have them in the identical computer, so I would disable one and enable the other, and so forth until I was satisfied with the results. Just to level the playing field, I used AMD Catalyst Software to manually set the values to match one another, raising the Boost Clock of the MSI card to that of the Gigabyte. Because, why would you ever go lower?
For the test rig, I used my main PC consisting of the AMD FX-8350 overclocked to a modest 4.3GHz, running on a MSI 990FXA-GD65 Motherboard with 32 GB of Corsair Vengeance DDR3 1600 RAM. As stated, I alternated between disabling one GPU or the other, and ran the latest version of 3DMark’s Fire Strike Extreme benchmark utility. The graphics driver is the latest beta from AMD, and is the explanation for the “Graphics Driver Not Approved” warning on the tests.
Fire Strike Extreme is one of the most intensive benchmarking utilities, designed to push systems with high end and multiple GPUs as hard as possible in a rapid amount of time. It runs two separate graphics test, a physics test and a combined test measuring average Frames Per Second (FPS), and the temperatures of the GPU as well as the CPU.
Running Fire Strike Extreme on both GPUs we see a combined score of 3593 for the MSI card and 3724 on the Gigabyte. Both tests were run multiple times with similar results; while the difference seems rather high between the two, using a simple percentage formula with the Gigabyte as the denominator and the MSI as the Numerator, we find only a 3.5% difference. This is easily within explainable variations of batches, or even acceptable margins of error with the tests themselves. In all honesty, you likely would never see a difference.
The truly remarkable results come from a test of the two GPUs set to work together in CrossfireX. CrossfireX is AMD’s multi-GPU solution, in which two or more graphics cards work together to render the images on screen. Generally what you will see is a main GPU doing most of the work while the secondary picks up some of the slack and carries a little weight of its own.
In CrossfireX, the combined Fire Strike Extreme test jumps from the Gigabyte high of 3724 to a score of 6475, bringing the rig from the top 56% of all results to the top 16%. The same percentile formula shows the cards performing at 174% (rounded up) from the Gigabyte card by itself.
It’s unlikely that you will ever notice any possible difference in performance between identical series cards from different manufacturers; while the R9 series was tested here, you could reasonably interpret these results other mixed and matched cards as well. The variations in score are relatively minor, and will not affect the overall experience unless you have some sort of robotic cyber eye or a monitor that costs more than your computer. When using identical series cards from different manufacturers in a multiple GPU setup, the performance increase is substantial and noticeable in any application that takes advantage of it. Not every game is compatible with CrossfireX or even SLI, so at some point you will have to return to a single GPU for a brief time. Test your gear and choose whichever one you feel best.
This is good news for those of us who wait and watch for the perfect time to buy an upgrade. Yes, the inside of your rig will look prettier if everything is all matched up and identical, but will perform just as well if you mix things up a little, plus you might save a few dollars.