Radeon RX Vega – Sold out before anyone could buy it

The rapid influx of reviews for the Radeon RX Vega 64 and 56 have hit the web today, and the results have been rather mixed.  The high level summary of reviews across multiple sites results in the following information:

• AMD Radeon RX Vega 56 vs. nVidia GeForce GTX 1070 – performance edge to the RX Vega 56 in many use cases with the caveat of considerably higher operating temperatures and power draw.
• AMD Radeon RX Vega 64 vs. nVidia GeForce GTX 1080 – performance edge to the GTX 1080 in many uses cases; power draw and operating temperatures on the RX Vega 64 make it a tough pill to swallow

Within minutes of being available from retailers such as Newegg and Amazon, the card was already sold out.  The baseline cryptocurrency benchmarks demonstrated somewhat competitive performance that are quickly offset with operating costs that nullify some of the profitability of mining.  Keep in mind, the limited supply of cards was being sold at a $100 USD price premium without the “Radeon Pack” that was supposed to help ration some portion of the available inventory for “gamers”.  While Ryzen and Threadripper have received solid reviews that recognize the value proposition and capabilities of these processors, the amount of recommendations for AMD’s high-end GPU offerings were muted and specifically conditional.  It appears that the missteps and mistakes encountered during the Radeon Fury launches did not provide the necessary lessons to prevent the same mistakes from happening yet again.  The innovation of the Fury line was offset in part by the initially high cost of entry and manufacturing or production level issues early in the cycle.  The optimizations and binning for the Fury Nano resulted in a better product that consumed less power and operated with better performance than the full-sized, air-cooled Fury line of products.

The comments section on many popular review sites offer up various viable theories as to how AMD made a rather significant misstep after successful launches of multiple CPU lines built around a fairly solid foundation.  Aspects ranging from finite budgets, inadequate resources to develop the same caliber of value-rich solutions on multiple fronts, and even some finger pointing at issues with the fabrication process at partners with whom AMD is contractually obligated to do business with have been bandied about.  The larger concerns that may be extrapolated or “theorycrafted” in the quantified results of the extensive analysis performed upon the Vega variants are as follows:

• The insistence on utilizing HBM2 + silicon interposer in lieu of a traditional, high bandwidth bus that may be paired with high speed memory has significantly increased the power envelope and thermals beyond expectation.
• If the reduced iteration of this architecture is still capable of drawing over 200W, the implementation in an APU model will either require extensive compromises to the compute allocations of the module or will demonstrate that the memory configuration is indeed the culprit of the extraordinary power draw and associated cooling requirements.
• An improved yield on optimized bins of this part will provide the capability to launch a mass-market equivalent of the Fury Nano.  AMD cannot afford to get the price, position, and power incorrect for this “hinted” SKU.

It is certainly just the start of driver optimization for the Vega line.  Some may pin hope to the past successes of AMD’s capability to wring every last ounce of performance out of a given architecture.  While it may be true that the raw compute numbers quoted for these GPUs provide hope that there will be a major revision that turns the tables and improves performance across numerous applications, doing so while drastically reducing power may not be as feasible as it was with the Polaris family of GPUs.  The consistent message across hundreds of reviews provide a better understanding of what analysts and AMD have respectively forecast with respect to near-term growth prospects related to revenue and market share.  This lack of competition doesn’t force nVidia’s hand in releasing the consumer variants of Volta.  Without viable competition, we may be looking at GPU cycles and innovation that will slow to the point where CPU cycles and associated innovation rested until this year.  Here’s to hoping that AMD can right the ship in the GPU space, especially in advance of the release of their mobile APUs.