Conclusion

After what seems to have been forever, AMD has finally revealed all the important details on RX Vega. It has been over two years now since we first heard about the Vega architecture as a flagship GPU replacement to the Fury X during which NVIDIA managed to release an entire new microarchitecture from top to bottom and to top again. RX Vega has come in not one, but two different versions with 4096 and 3584 stream processors, respectively, and the RX Vega 64 coming up at the $499 price point to tackle the GeForce GTX 1080 head on and the RX Vega 56 aiming to convince end users to justify spending $399 for it over the GeForce GTX 1070.

We still have to wait for independent reviews and benchmarks (you should keep your eyes on your favorite hardware site for that, by the way). However, AMD's Vega architecture does help us paint a revealing picture - one we are working furiously to put the last few brush strokes on for you. But there is no doubt that Vega is an entirely new architecture built around HBM, with a novel high-bandwidth cache controller and new "Rapid Packed Math" that enables high performance 16-bit operations that otherwise are handled as 32-bit operations taking up unnecessary resources and time.

Backing the hardware is AMD's now mature Crimson ReLive Edition software. What was once a weakness for the company and a bottleneck in unlocking performance out of the raw GPU core is now a feature-rich, on the eye pleasant driver that has only recently introduced multiple features to enhance the user experience with AMD Radeon GPUs, and the latest update for Vega is no exception.

All in all, it would seem that AMD's time on the Vega architecture was more about laying the groundwork for the future than answering the consumer's immediate needs. Vega seems to breathe AI and computing design philosophies, taking some tremendous strides forward when it comes to memory architecture and overall architecture balance. And even though it may well open up many possibilities for AMD's incursion on the AI and HPC markets, for the general consumer (and for the general gamer), it is just… competent.

At the same time, pure performance does not seem to be at the level of the competition's current flagship, and we know NVIDIA is already working on GeForce Volta as we speak. AMD's current halo product, by the company's own estimates, delivers performance that is comparable to the GeForce GTX 1080. At the time of writing, the least expensive GTX 1080 (an MSI custom with a triple-fan cooling solution through Newegg) costs $509. AMD's entry line Vega 64, which AMD says is the comparable graphics card, will retail for $499. As of now, the higher performing (out of the box) liquid-cooled edition may not even be available for purchase outside of a Radeon Pack at $699, putting it out of context for this comparison. Speaking of which, you can make a $100 upgrade to a Radeon Pack that throws immense savings your way, should you be looking for a whole new system, but if you are not, it ends up exposing a few flaws in the system. What will Vega 64 do with a reference blower design and a 225 W TDP against the GTX 1080's 180 W TDP with an aftermarket cooling solutions. Will AIB custom designs still be relevant when they finally hit the shelves a couple months later?

Vega is a step forward for AMD and delivers a much-needed architecture upgrade that will certainly bear fruits down the road, but the road still has hurdles in it. AMD's Radeon Packs are a likely disguise to the fact that they are unable to offer the Vega SKUs for lower prices due to HBM2 cost and availability; it disguises the fact that inventory will indeed be low due to that same design decision. Design decisions that certainly were well weighted for the markets AMD is trying to break into and claim for itself alongside NVIDIA, but which come at the cost of simple pragmatism in the design of a high performance graphics architecture. Time will tell if AMD has done enough with the product as it is today, but questions will continue to be asked of them until then.