AMD’s Raja Koduri Breaks Silence On Vega – Discusses Pricing, Perf/$, Perf/Watt & Untapped Hardware Features
AMD officially introduced its first high-end gaming RX Vega graphics card, the RX Vega 64, just over two weeks ago and followed with the more affordable Vega 56 just a couple of days ago. And while both products have generally been received favorably by the press, it’s also fair to say that the launch hasn’t been without its issues.
The cards have been subject to a significant shortage that has not only led to considerable pricing increases over the official MSRP of the products but also significant delays to product shipping by online e-tailers. If you ordered an AMD Radeon RX Vega 64 on launch day more than two weeks ago, there’s a good chance you haven’t received delivery of your card yet. Some RX Vega 56 buyers have even been told they won’t be receiving shipment of their purchase until early October.
The Vega Conundrum
Supply issues aside, Vega has managed to largely deliver on what the company promised. Which is partly why there’s a supply issue to begin with. The AMD Radeon RX Vega 64 delivers performance that’s competitive with NVIDIA’s GTX 1080 at a similar MSRP and the RX Vega 56 manages to do even better against its competition from the GTX 1070, again at a similar MSRP. Vega finally gives FreeSync monitor owners and Radeon fans something better than the RX 580 to upgrade to. It also gives potential GTX 1080 and GTX 1070 buyers an interesting alternative.
With all of that being said, many PC hardware enthusiasts expected better than just an alternative to NVIDIA’s GP104 based products. After all, Vega is a new architecture and it took AMD more than a year after NVIDIA introduced Pascal to bring it to market. So what exactly happened? Well, thankfully Radeon Technologies’ Group graphics guru Raja Koduri, who has mostly been silent since the Vega launch, is back from vacation. Hopefully we’ll find some answers in what he had to say.
Good to be back home after 2 weeks in India. Combination of vacation and visiting our engineering sites in Hyderabad and Bangalore.
— Raja Koduri (@GFXChipTweeter) August 29, 2017
Raja Talks Vega Pricing, Perf/$, Perf/Watt & Untapped Hardware Features
Raja reiterated AMD’s previous statements regarding supply and reinforced the company’s commitment to deliver as many Vegas as possible as soon as possible to retailers worldwide. He went on to briefly touch on all the media buzz that’s been surrounding Vega’s pricing issues. Many pointed the finger at AMD, accusing it of marking Vega chips up over the advertised MSRP, something that Raja emphatically refuted. Saying that doing that would only help the competition.
All students of economy know that pricing is subject to the forces of supply and demand, and as supply continues to be limited we have little doubt that the drama and blame game – miners gobbling up supply, retailers accusing suppliers of raising prices to take advantage and vice versa – surrounding the current situation will go on until supply normalizes.
Due to the nature of how Twitter arranges tweets in reverse-chronological order I’d recommend reading the tweets starting from the bottom.
The Balancing Act Of Power And Performance
Raja went on to discuss Vega’s performance per dollar and performance per watt. According to Koduri, Vega 10 has the largest performance per watt dynamic range of any GPU. In other words Vega’s design allows it to be setup in a way that makes it extremely frugal and power efficient if desired and the extreme opposite is also true if performance is the only concern, at the expensive of power. Also, according to Raja, Vega has the widest feasible operational power range in either direction of any GPU in existence.
Raja explains that due to this fact the RTG team had a very flexible range of perf/watt and perf/$ that they can optimize Vega around. They wanted to strike a balance that would please as many users as possible and their research indicated that perf/$ was more important than perf/watt for a larger portion of users than perf/watt. Although in an effort to please everyone, the team implemented three separate operational power modes for Vega. A Power Saver mode, a Balanced mode and a Turbo mode.
Turbo mode tells the GPU to run as fast as it absolutely can and to hell with power consumption. The Power Saver mode on the other hand makes sure that the GPU sticks to a very efficient power/frequency curve whilst not sacrificing much performance. Balanced mode, as the name suggests, is a middle ground between the two.
Vega’s power efficiency can only be exposed and truly appreciated in power saver mode. In the RX Vega 64’s case this mode can save up to 120 watts of power, cutting effective power consumption by over 35%, whilst only slashing performance by roughly 4%.
Vega’s Untapped Hardware Features
Last but not least Raja talked a little bit about Vega’s new architectural features and the role they played in Vega’s conception. According to Raja, Vega 10 was intended to be as versatile as possible from the get-go in order to compete with three different GPU designs from NVIDIA, GP100, GP102 and GPU104. This meant that the team had to make many trade-offs to pull it off. Unlike NVIDIA, which had the luxury of designing different GPUs for gaming and compute, AMD had to design a GPU that can do both.
When pushed for further detail Raja stated that a lot of the growth in transistors/power in Vega was related to the implementation of Infinity Fabric and Vega’s plethora of new features, some of which require further software enablement to be fully leveraged.
infinity fabric and many of the new features some of which still need software enablement explain most of the delta
— Raja Koduri (@GFXChipTweeter) August 30, 2017
AMD Radeon fans have long coined the term “fine wine” to describe the process by which AMD’s GCN based Radeon GPUs mature overtime and often significantly surpass their NVIDIA counterparts in performance. The HD 7950 Boost for example was originally a GTX 660 Ti competitor back when it came out in 2012, but over time we’ve seen it not only outperform the more expensive GTX 670 but also the GTX 680 and even the GTX 770 in many gaming titles as time went by.
Raja’s comments regarding Vega’s well of untapped hardware potential brings that term to mind, although he emphatically denies that he was referring to “fine wine” in his tweet. So, only time will tell whether the same story will be true for Vega, .