Today's Hard|Forum Post
Today's Hard|Forum Post

AMD Radeon R9 290X Retail Performance Variance Review

The AMD Radeon R9 290X arrived recently with a high level of performance, and a high level of controversy. There have been reports of performance variance between Radeon R9 290X video cards. We have two purchased retail cards today with stock cooling that we will test and see if performance variances exist.


The AMD Radeon R9 290X was introduced on October 23, 2013 and upped the ante for GPU performance. It has also been praised for its low introductory MSRP, which has recently been raised by retail and etail outlets due to high demand. Beyond the demand, and pricing the Radeon R9 290X has been the center of a heated debate with regards to its thermal profile, reference cooler effectiveness, and operating temperature.

As with any change to the status quo, there can be a lot of fear uncertainty and doubt spread around causing folks to jump to a quick conclusion prior to understanding how everything works. One of those conclusions that has been reached by other review sites is that there are performance variances that exist among retail Radeon R9 290X video cards and those retail video cards performing differently than the press samples sent to reviewers from AMD.

There are some facts that we have in regards to this issue. On November 7th, 2013 AMD issued a response to this issue with a forthcoming driver fix.

Hello, We've identified that there's variability in fan speeds across AMD R9 290 series boards. This variability in fan speed translates into variability of the cooling capacity of the fan-sink. The flexibility of AMD PowerTune technology enables us to correct this variability in a driver update. This update will normalize the fan RPMs to the correct values.

The correct target RPM values are 2200RPM for the AMD Radeon R9 290X "Quiet mode", and 2650RPM for the R9 290. You can verify these in GPU-Z. If you're working on stories relating to R9 290 series products, please use this driver as it will reduce any variability in fan speeds. This driver will be posted publicly tonight.

That driver was released, and we used that driver, and newer for all reviews. We also had a very lengthy forum thread about this performance variance issue which was an extremely active discussion.

We thought the issue was fixed, as per the driver, and did not have any retail video cards at the time to test it. Well, we do now.

We have purchased a pair of retail GIGABYTE Radeon R9 290X video cards. These are all stock reference cooling and we will be pitting them against the pair of press sample Radeon R9 290X video cards that we have in the underground bunker. We will focus on the gameplay performance of each of the cards in both quiet and uber mode as well as measure whether each card is throttling the GPU clock speed and if so, determine the impact to the gameplay experience. These cards were purchased anonymously from an online retailer, while these are retail GIGABYTE video cards, these are all reference design, no custom cooling, and no custom PCB design.


We previously went into detail about the new concepts that were introduced with PowerTune 2.0. The frequency is affected dynamically based upon the card’s load, power usage, and temperature. The clock speed of the GPU is adjusted every 30ms based upon these parameters and is constantly trying to achieve the top operating frequency that is specified.

PowerTune 2.0 also introduced two different fan modes on the flagship R9 290X, Uber mode and Quiet mode. Uber mode caps the fan speed at 55% of its maximum capability while Quiet mode caps the fan speed at 40% of its maximum capability. AMD also advertises the R9 290X as having a GPU clock of UP TO 1000MHz. AMD does not make any assertions that you will achieve that clock speed at all times.

We initially found that in quiet mode, the press sample cards that we evaluated would throttle its GPU clock and perform lower than a card configured in Uber mode. While we initially decided to review the R9 290X in quiet mode during the launch evaluation, after spending more time with it and comparing our noise level observations between Uber mode and Quiet mode, we decided that all future evaluations involving the Radeon R9 290X would be conducted in Uber mode.

Given the nature of the variability between electrical components, silicon quality, and pure luck of the draw, we expect to see some variability in average GPU clock speed on the Radeon R9 290X much like there is variability on Kepler based NVIDIA cards using GPU Boost.

The key difference in marketing between the Red and Green teams is that NVIDIA advertises a minimum GPU clock and a minimum Boost speed for its GPUs whereas AMD advertises the clock speed as an "up to" value. With GPU Boost being around for a couple of years, it has certainly set the expectations of video card buyers that clock speeds should start at the specified clock speed rather than be limited to the specified clock speed.

Test Setup

For today’s test, we have mustered up four AMD Radeon R9 290X video cards آ– two cards that are [H]’s press sample cards and two GIGABYTE reference cards that we purchased in etail with our own cash. We did not request any assistance or notify NVIDIA, AMD, or GIGABYTE about the purchase of the cards used within this evaluation.

For each card in the evaluation, we will be playing four games in our review suite at the highest playable settings that we have previously determined for the R9 290X in both Uber mode and Quiet mode. We will record FPS using FRAPS over the course of each runthrough as well as collect operational data from GPUz to determine the throttling, if applicable, for each GPU.

Each card was warmed up for a minimum of 30 minutes to ensure that a normalized "hot" operating temperature and fan speed were reached prior to proceeding through the set of games. For each card, we ran through the games in the same order in rapid succession to reduce the potential for variability. Testing was conducted in the underground bunker at a constant 68 degree Fahrenheit ambient temperature.

We will be using a ASUS Sabertooth Z77 motherboard, an Intel Core i7 3770K Overclocked at 4.6GHz, and 8GB of Corsair RAM.

Note that Catalyst 13.11 Beta 5 is exactly the same as 13.12 WHQL, the 13.12 WHQL driver is basically 13.11 Beta 5 with WHQL status. No differences exist between the drivers.

Article Image