AMD's Next Gen. GPU; A Recipe for Success
Make sure to visit the forum comment thread to discuss this article.
AMDís next generation graphics will be announced via livestream on June 16th sometime between noon and 1PM Eastern time. To say there are high expectations is an understatement of vast proportions given the gaming market has been waiting for 20 long months since their last flagship cores were released.
As my excitement for next generation Radeon cards mounted, I took a step back and started wondering about what AMD needed to accomplish to ensure Fiji (or whatever the architecture will ultimately be called) is a success. This quick blog post puts all of my thoughts down to paper so to speak.
This is not intended as a piece thatís critical, nor is it meant to be a definitive recounting of ďmust haveĒ goals for AMD. I just wanted to have a bit of fun crystal ball gazing and the end result is the few thousand words you see below.
Make no mistake about it though: AMD needs a runaway hit right now. I donít want to sound overly dramatic here but the situation has been looking a bit grim for the better part of two years now. Their APUs have thus far failed to gain any meaningful market share and, despite some sales successes due to crypto currency mining, the long wait between new graphics architectures has allowed NVIDIA to take a commanding market share lead. AMD needed a Hawaii refresh that never materialized but Fiji should have the capability to move the company forward in a meaningful way.
Announcing a flagship graphics card amidst a livestream, thumping music and marketing speak is relatively easy compared to what happens afterwards. So letís get on with this.
Is It All About Performance and Watts?
Regardless of the breathless rumor mills churning out a mix of educated guesses and complete fabrications, the only people who really know how Fiji is meant to perform are preparing to announce it at E3. Drivers are still in their pre-beta form and clock speeds are undergoing final tuning.
Even though NVIDIA has a significant generational lead on current AMD cards, the situation Fiji finds itself in is not all that different from Hawaiiís launch nearly two years ago. Back then, the R9 290X proved to be a surprising competitor, beating the TITAN and causing NVIDIA to lower prices while also launching the GTX 780 Ti. The only difference this time around is the GTX 980 Ti actually preempted AMDís schedule instead of being released as a reactionary measure.
That 980 Ti may represent a bit of a challenge for AMD since it outperforms the R9 290X by a good 50% while also consuming 50W less power. With Maxwell NVIDIA found a way to successfully blend the fairly ancient 28nm manufacturing process with reasonable efficiency levels and the Radeon teamís engineers need to find a way to do the same thing with Fiji. To match them any Radeon card would have to take a very different path than Hawaiiís brute force approach. AMD just cannot afford to have another card that consumes significantly more power than the competition without the ability to offer better performance as well.
Image courtesy Johan Andersson / DICE
As an enthusiast it is very easy to turn a blind eye towards the performance per watt equation but Iíd argue it is becoming increasingly more important. With the burgeoning popularity of small form factor systems, users are looking to a way to maximize performance out of a compact layout and that layout comes with obvious constraints. PSU capacity, available fan mounting locations and of course space all contribute to some very unique challenges when looking to shoehorn a high performance graphics card into an SFF chassis. The volume-focused system builder market (a segment AMD desperately needs to appease) also face these challenges but in an altogether different way and if anything, they represent a stronger voice than SFF gamers. As such, tightly constrained TDPs are the way of the future.
How AMDís next generation architecture lines up against NVIDIAís current lineup is just the tip of the iceberg though. Fiji also represents AMDís best chance to throw up a medium-term bulwark against Pascal, a GPU which promises to be a leap forward in performance and efficiency.
Luckily, AMD has shown they can implement innovative ways to mitigate some of the concerns Iíve expressed above so the ball is in their court right now.
The Higher the Price the Better
Performance is one of the major cogs within the machine for success but an acceptable pricing structure is arguably even more important. From an end-user standpoint this might sound counter-intuitive since we all want to pay the least amount for the best possible performance but AMD needs Fiji to be a champion and charge accordingly for it. Simply put, Iím saying that AMD needs a GTX 980 Ti beater and their card needs to be priced above $650.
Before you start writing angry messages in our forums, hear me out. At this point we donít know much about what will be announced later this week but what has been shown by AMD points to a large, complex GPU core which sits astride an interposer for communication with a quartet of onboard HBM modules. Thereís a significant cost associated with such an undeniably complex next generation design, not to mention potential yield limitations and those costs need to be recuperated somewhere. That should lead to a suitably high price since this AMD is facing a fight for its very survival and taking write-downs on more quarterly balance sheets for a loss leader isnít realistic at this point.
This is about more than just BOM costs though; AMD not only has to sell Fiji for more than it costs to make but they also have to recuperate the substantial development costs poured into the architecture. That could lead to a higher cost for the flagship part but thatís something I wouldnít cry over.
From my standpoint, an expensive card would be a sign for hope rather than derision since it would prove AMD has something good enough to charge a premium for. Also, there should be other chips in the lineup to fill the need for that perfect price / performance combination. This brings me to the next point....
The Cascade Effect Leads To Success
As AMD has seen in the past, thereís more to a successful graphics architecture launch than flagship products. It will take a full court press to dislodge NVIDIA from their current position and that means implementing an architecture that can scale down into a successive number of lower-priced SKUs.
The HD 7000-series is a great example of how a well implemented, highly scalable core design could lead to broader recognition among the gaming community and, as a result, a larger slice of the GPU market pie. That generationís HD 7900, HD 7800 and HD 7700 cards ushered in a year or so of continual pressure against competing GeForce cards.
Unfortunately, Hawaii changed that equation in a big way. Missing in action were (at the time) next generation cards that could push volume and profit rather than raw margins on low volume parts. Any successful lineup needs a blend of high margin and volume yet without those two elements operating alongside one another, finding success becomes increasingly difficult. The Tonga-based R9 285 was a great example of what could have been but in many cases it was too little, too late.
Now AMD finds themselves in almost the same position as when the R9 290 series launched. NVIDIAís GTX 980 Ti, GTX 980, GTX 970 and GTX 960 represent an entrenched lineup but theyíre not infallible and thereís plenty of opportunity to slide a newly designed core into an open price bracket. With the GTX 970 at $329 and the GTX 960 at $199 thereís a yawning chasm thatís just begging for some fresh blood, not a rebrand educated buyers have seen for the last 20 months. I feel myself willing AMD to make a move here.
Buyers are smarter and more educated than ever. That means rebranded or relabeled or repurposed SKUs could be glossed over if their feature sets arenít up to todayís standards. If thereís ever been a golden opportunity for AMD itís beckoning within the wide and infinitely more affordable $175 to $375 range.
Availability; An Announcement Isn't A Launch
This is a sore point for me because I make a clear distinction between an announcement, a launch and product availability. Announcing a new architecture is one thing, getting it into the hands of buyers is a completely different ballgame. No one wants a carrot dangled in front of their face only to be told its still weeks or months away from widespread distribution.
The large-scale cores with integrated HBM AMD has shown will undoubtedly be complicated to produce but I know that lessons were learned from the launches of Hawaii and Tahiti. Both of those featured promising new technology followed by long waits for cards to materialize at retailers. That gave the competition ample time to formulate a response which is the last thing AMD wants or needs at this critical juncture.
Itís my hope that AMD can deliver a competitive product in the volumes needed to appease their fans and feed potential ship-jumpers from NVIDIAís boat. Only time will tell whether or not that wistful thinking comes to fruition but getting products onto retailersí shelves is absolutely essential.
Drivers, Drivers, DriversÖ.DRIVERS!
Any graphics architecture lives and dies by the ability (or lack thereof) of driver development teams to properly write supporting software. Without those critical performance optimizations for the latest games, a graphics card will run into a wide range of potential issues from missing multi card profiles, to rendering artifacts, to low framerates.
Truth be told both NVIDIA and AMD have exhibited their fair share of driver development hiccups in one way or another. However, what I want to focus on here isnít necessarily the actual quality of each companyís software (a whole article could be written about that!) but rather the frequency of those key releases and patches.
Simply put, if AMD wants Fiji to be a success they need to look well beyond launch-day drivers. The last year has shown a noticeable drop in the quantity of their driver revisions and thatís a major problem when, at times, games were being launched at a break-neck pace.
Taking a look back at the last six months AMDís new beta drivers were introduced on January 18th, March 20th and May 29th with a minor update between the latter two for FreeSync implementation. While seemingly frequent, these dates align perfectly with NVIDIAís introduction of their GTX 960, TITAN X and GTX 980 Ti respectively.
Indeed the previous AMD driver roadmap looks purely reactionary to NVIDIAís releases rather than software being rolled out to address optimizations for newer games. It should also be mentioned their last WHQL signed driver (a key component for many system builders) was unveiled in December of last year. Thatís a disturbing trend that cannot be allowed to continue if AMD wants Fiji to compete against Maxwell, let alone Pascal.
As with my other points, I have hope this one will be addressed as this new generation rolls out since it has been a longstanding bone of contention between AMD and the gaming community as a whole.
A Few Parting Words
Like so many of you, I like cheering for the underdog. Itís always nice to have a David versus Goliath story come true and if the stars align just perfectly this time, AMD could pull off a resounding upset win. However, they are facing an uphill battle to claw back their position within the graphics card market. It may be a tough situation to be in but we have to remember that adversity usually breeds success.
What the market needs is competition and a thriving, profitable AMD will help drive prices downwards while also pushing innovation forwards. The next few days and weeks will ultimately determine how many of my points above are met and how that leads to the success or failure of an architecture that may ultimately dictate AMDís future in the graphics card market. One way or another, it is going to be a wild summer!
|Latest Reviews in Video Cards|