While it may seem like I'm the perpetual debbie-downer when it comes to Advanced Micro Devices (AMD), I'd like to point out something that the company did supremely right - the Radeon R9-290X chip, codenamed "Hawaii." With the official reviews out now, it seems that graphics cards based on this chip (which sell for just $549) offer performance that rivals that of the much higher end NVIDIA (NVDA) GeForce 780 and TITAN at 4K resolutions. While this answer to NVIDIA's flagship cards is pretty late, this will likely help AMD regain market segment share in the very high end of the GPU space.
The Good - Performance and Performance/Dollar
Starting off with the good, it looks as though the AMD Radeon R9-290X competes quite favorably with the highest end NVIDIA GeForce GTX 780 and TITAN ($650 and $1,000 cards, respectively) - coming within a few percent of NVIDIA's $1,000 beast in terms of performance at high resolutions such as 1920x1080 and 2560x1600. At 4K resolutions, it seems that AMD's cards enjoy a more marked lead over NVIDIA's best due likely to meaningfully higher memory bandwidth. This is a niche resolution, but the kinds of people buying these cards are niche players to begin with.
The incredible thing here is that AMD is willing to sell these parts for ~$549. Interestingly enough, AMD does have a die size advantage as the "Hawaii" GPU sports 6.2B transistors in its full configuration while the NVIDIA GK110 sports 7.1B transistors (albeit it's important to note that the NVIDIA cards ship with at least one SMX disabled, so more of the Titan/780 capable chips show up per wafer). While AMD is no stranger to playing it fast and loose with margins, especially since it's backed into a corner, NVIDIA has enjoyed steadily rising gross margins over the last several quarters - no doubt due to its success in the HPC/workstation/high end gaming GPU markets. AMD going this aggressive on pricing in the high end of the consumer segment makes this more difficult.
The Not As Impressive - Performance Per Watt
While it looks like AMD has designed a great chip from a raw performance standpoint (i.e. it ties the GTX Titan more or less), it doesn't look as impressive if viewed from a performance/watt perspective. Indeed, see this power consumption chart from the recent Anandtech review of the R9-290X:
In the R9-290X's nominal performance mode (i.e. the fans stay reasonably quiet) it consumes a full 11% more power. At its "Uber" mode (where the clocks can be run high and the fans get loud), it consumes 24% more power - at roughly comparable performance levels:
So, while AMD is catching up (and largely trading blows with) where NVIDIA's $1000 flagship has been for months on performance, it does so at the expense of performance/watt. Now, do most gamers picking a GPU care about a few dozen watt differences? Probably not - gamers buy big fat power supplies and the type of people buying these cards will be overclocking the snot out of them anyway, completely throwing themselves off of the optimal performance/watt curve. That being said, it is likely that the NVIDIA parts have room to overclock further while the AMD's are topped out, but strictly speaking, we should talk about "stock configuration" against "stock configuration" as overclocking results may vary.
The Investment Takeaway
AMD will probably gain back some market share at the very high end of the discrete GPU market or it will pressure NVIDIA's pricing on its very highest end parts, which could prove to be a slight gross margin headwind for NVIDIA. That being said, NVIDIA's real gross margin winners - the Quadro and Tesla professional cards - are still safe.
Further, a potential course of action that NVIDIA could take to counter this would be to actually release a fully enabled GK110 GPU into the gaming GPU market (these are currently sold in the high end workstation market). This would be enough to win back the performance crown at the high end (although by how much remains to be seen) and could replace the GTX Titan as the new $1000 halo card. The problem is, these fully-enabled dies sell for multiple times more as "Quadro" parts than they do as "GeForce" parts.
Now, a final thing to note. The AMD reference boards/coolers that the currently shipping R9-290X parts come with look pretty cheap, while the NVIDIA 780/Titan cards come packed with fairly expensive and slick looking cooling solutions. Presumably, this is part of the reason that the 780/Titan cards are priced at a premium, and it's likely that they can continue to be if the cooling solution quality is something that matters to the enthusiast base. I know it matters to me, but I don't pretend to speak for the enthusiast community.
Bottom line, though, look for AMD to take back some share at the very high end of the desktop GPU market. How much share it takes back and how aggressive NVIDIA will be in fighting with price drops remains to be seen, but it looks as though AMD has finally stricken back at the very high end of the GPU market with a competitive product.
The real question, however, is just how much this really matters to the top/bottom lines of the graphics divisions of each company. These are niche, high end halo parts, but they do carry great margins and high ASPs. NVIDIA's mobile GPU share is still very safe as it won the vast majority (>95%) of the Haswell-based gaming notebook designs, and NVIDIA and AMD have always been really competitive with each other up and down the product stack even on desktop dGPU. There's much more to GPU share shifts than the performance of the halo desktop GPU product, but at the very least, this will be good for AMD's mind-share among gamers.