Catalyst 13.11 V9.2

Have you had enough AMD driver releases for 13.11?  Hope not! They keep on rolling.

This new version changes things up for AMD's Radeon R9 290X and 290 cards by modifying how the fan speeds are set.  You will find a slider in the CCC that is set to 40% for the 290X and 47% for the 290. Both represent a maximum fan speed that the cooler is allowed to reach before starting to drop clocks. 

new-driver-settings.png

With this new driver, AMD is no longer fixating on the percentage but instead is targeting a specific fan RPM.  As the graph on our previous page shows, even when both the MSI and press sample R9 290 were reporting 47% fan speeds, they were actually rotating at very different rates.  Now, AMD will basically ignore this slider and instead attempt to nail the fan RPM to the new, correct settings.

(As an aside: it would make much more sense for AMD, going forward, to present the user with an RPM slider rather than a seemingly useless (and error prone) percent rating.)

In the email from AMD, it now claims that the target RPM for the Radeon R9 290X is 2200 RPM and for the R9 290 is at 2650 RPM.  Well, did it work?  I ran both cards through our now-standard Hawaii GPU testing methods that include a full 6 minutes of GPU "warm up" time in Crysis 3, Very High settings at 2560×1440.

fanspeedv9.png

Well look at that – both the press sample and the MSI R9 290 are reporting fan speeds at 2650 RPM.  We do see a bit more fan speed fluctuation, though, which needs to be controlled to avoid any kind of shifting noise levels or tones that could adversely affect experiences.  In the retail card's case, the fan RPM jumps from 2350 out of the box to 2650, an increase of 12%! 

clockspeedv9.png

As a result, during the same period of time, the clock rates of both the retail and press sample cards are sticking very close to the peak 947 MHz clock rate on our R9 290s.  Notice though that for the press sample we received, this change almost has no effect – the card was already running at 947 MHz and it continues to.  Only the retail card sees the improvement. 

Performance is affected very positively because of this change.

crysis3-2560x1440-ofps.png

crysis3-2560x1440-per.png

crysis3-2560x1440-plot.png

What a jump for the retail R9 290!  Average frame rates with a "warm" GPU go from 23 FPS to 27 FPS, an increase of about 15%.  The performance of the press sample remains unchanged, as you would expect based on the clock speed graph above.

But there are some issues with this change that lead in the wrong direction.

soundv9.png

Look, the R9 290 was already a loud graphics card, there is no getting around that.  With the jump to the 13.11 V9.2 driver, fan speeds are going from 2550 to 2650 (on the press sample) and from 2350 to 2650 RPM on the retail card, and that doesn't come free.  Sound levels in our testing went up to 50.8 dbA in our testing; up 1.3 dbA on the press sample and 4.4 dbA (!!) on the MSI retail R9 290.  Think about that in terms of a user who purchased an R9 290 and, after a simple driver update, you experience a 15% boost in performance in some games but also a noticeable increase in sound levels too. 

Power consumption will also increase with this change. By running at higher clocks the Hawaii GPU is going to be drawing move voltage and thus more power.  We saw the same effect in our GPU scaling article from earlier in the week with the R9 290X. 

 

Why is AMD doing all of this?

The primary reason for AMD's sudden reaction here was the growing concern of the possibility of "golden samples" being sent to the press; that AMD has purposefully seeded better-than-normal performing graphics cards to the media to paint a better picture of Hawaii than was accurate.  By attempting to get ahead of the problem with Catalyst 13.11 V9.2, AMD is hoping to prevent an online rampage from filtering through forums across the world.

Though this driver affects both the R9 290 and the R9 290X, the effect on the R9 290 is what I was able to test tonight.  I know others are working on stories around the 290X so look for that later this week.  The R9 290 has been transformed by AMD twice now, once before its release and once after, in ways that dramatically change its positioning.  Before the original launch of the R9 290, AMD has set a fan speed maximum of 40% for the card but after NVIDIA's GTX 780 price cut, they propped it up to 47%, increasing sound levels and performance.  At the time the move made very little sense to me as the R9 290, even at 40% maximum fan speed and at $399, was already aggressively attacking the GeForce product line.  Now, with the move to an even higher fan speed (even if AMD will choose to call it 47% going forward), the R9 290 changes again; with higher performance and a louder cooler.

powertune.jpg

I already mentioned that an increase in fan noise of 4.4 dbA was no small change, but to apply this to users AFTER reviews have been live and purchases have been made seems kind of crazy.  Clearly AMD was worried enough about the potential PR damage of the "Hawaii variability" issue that they were willing to adjust base settings of both the R9 290X and R9 290 post-launch.  Some users will likely appreciate the "free" performance while others will wonder why their PC just got louder after a driver update. 

Seeing R9 290X cards that were only running at fan speeds of 1850 RPM, which is 16% lower than what AMD wants us to believe they had planned, shows either a dramatic hardware variation or a lack of QA process to catch this before hand.  If that is true, then at this point AMD would have had to look for a new maximum fan speed for each GPU that would be universal on all cards yet still allow even the worst GPUs (from a binning perspective) to meet the expectations of consumers and media for sustained clock speeds.  As a result, the new fan speed would have to EXCEED the fan speed of the best case graphics cards (like our press sample appeared to be) in order to "bring up the rear" of cards that have high leakage GPUs. 

The result?  A fan speed RPM target of 2650 for R9 290s and 2200 for R9 290X. 

The first major issue I have with these changes is that AMD claims this driver simply removes fan speed variations by normalizing them to 2650 RPM (and 2200 RPM on the R9 290X).  But fan speeds on the R9 290 were never actually seen running at 2650 RPM on either the press samples or any retail samples (I am using a data set larger than my own thanks to friends in the media).  Not a single person that I referenced for this article saw fan speeds at 2650 RPM on the R9 290.  And no one ever saw fan speeds at 2200 RPM on the R9 290X.  And this is with a total sample size of about 12 units. (Correction: we now have data from 19 total R9 290X boards and 6 total R9 290 boards.)

img-1800.jpg

So, based on my data, AMD wasn't normalizing anything. Instead, they are raising the fan speed across the entire field.  When I contacted AMD for feedback on this though they were adamant that my sample size was not large enough to make a general assertion about the performance of the Hawaii GPUs.  In their testing, which is based on "thousands" of graphics cards, they claim there were indeed seeing fans at RPMs higher than 2650 on the R9 290 and higher than 2200 on the R9 290X.

At the end of the day, this doesn't change much for consumers though it does mean that reviewers like myself need to spend some time contemplating a reevaluation of the R9 290 and R9 290X in the AMD product stack.  Does the added noise or power draw of the Hawaii cards change our view on recommendations?  Or does the added performance make up for those changes?  I will be looking into these areas more closely over the next several days by my first reaction is that this change is only going to affect the landscape minimally. 

AMD again deserves some credit for reacting quickly to the media's concerns over this problem, and the company insists they were aware of the variance issues for "a couple weeks" and were already working on a fix.  Things just got accelerated a bit when outlets started questioning results in the field.

It does leave the impression that AMD has been playing this whole launch by ear rather than implementing a well laid out plan from day one.  That doesn't make the Radeon R9 290 or the Radeon R9 290X any more or less impressive of a graphics card though, which is why our recommendations from our recent reviews will stand. 

UPDATE: This driver is now publicly available on AMD.com.

« PreviousNext »