So, with reviews out, we can see why I dislike Nvidia marketing – the performance details offered were cherry-picked extremes more than a median representation of the actual performance improvements offered by the RTX 3080 card. That isn’t to say that the card isn’t impressive – it is! However, what is important here is to contextualize this card in the broader history of Nvidia, the GPU market, and evaluate why Nvidia made some of the choices they did.
Firstly, let’s start with something of an observation – absent any external pressure, Nvidia would have been more than happy to Turing this launch. When I use Turing as a verb in this context, it is to deride Nvidia for the worst generational leap they’ve ever given gamers. Turing had exactly 1 mainstream and 1 Titan card that broke free of the prior Pascal GPUs from the GTX 10-series – the 2080 Ti which was a measly 30% better than the 1080 Ti at best, and the Titan RTX, which was marginally better than the 2080 Ti at nearly double the price.
Since the GTX 680 in 2011, Nvidia has always used lower-bin silicon for their “top end” xx80 cards. In Nvidia’s naming scheme for chips, the codename is XX10x, where XX is the family (GK for Kepler in the 680, GA for Ampere in the current cards) and the last digit marks the performance category, with closer to 100 being better. In the past, prior to the GTX 680, the 100 silicon was the xx80 series, but starting with the GTX 680, Nvidia made a sharp cutdown to the xx80 cards, using the 104 silicon (this is only two steps down, as Nvidia uses even-numbers only). This is how it has been ever since, with xx80 cards on 104 silicon, xx80 Ti cards on 102, and Titans on the full 102 silicon, with 100 generally reserved for data center GPGPU usage.
Right away, when Nvidia announced that the 3080 was built on GA102, that raised my hackles. It was good – they were being forced to bring a better product to that pricing tier – but it also told me that absent that motivation, this generation would have represented the same paltry leap we got in Turing. And sure enough, the benchmarks back that up.
In the world Nvidia probably wanted, the 3080 would be marked as Ti, with the 3090 being a Titan GPU and sold for nearly double its already-astronomical cost, and the 3070 coming next month being the 3080 instead, with the whole stack down from there seeing similar modifications. The benchmarks confirm this by showing the same leaps – TU102 to GA102 is a paltry 30%, the same as GP102 to TU102 was before it.
So why’d they move to reduce their margins on sales in this way?
It’s easy actually, a single acronym – AMD.
The thing to understand about AMD in context in this market is that they haven’t been competitive at the high end since the time of the GTX680 or thereabouts. In the early 2010s, AMD made a strategic decision on how they wanted to compete in the GPU space – no longer would they make massive dies or even offer a card outside of the upper-midrange based on a single GPU. Instead, their top-shelf silicon designs would be firmly in the midrange, with reasonable die sizes, lower part costs, and the constraints that came from that decision, with any high-end competition being multi-GPU solutions. For a while, they did this and it worked – the Radeon HD 5870 and later 6970 were phenomenal products that matched or even exceeded Nvidia’s top-end offerings for cheaper and with lower power consumption. When I built my monster dual-socket CPU system in 2011, the choice of video card was easy – the 6970 won and I ended up with 3 of them for a Crossfire setup (not in the context of this article, but I’ll put a brief aside to say this – yikes, don’t do that kids). The move to the 104 silicon from Nvidia was posed as a competitive response – the generation Nvidia followed with after the 6970’s dominance was the GTX 680, which, similarly to AMD’s strategy, had a dual GPU card in the GTX 690.
Since Nvidia snatched back the performance and power crown in that era, AMD hasn’t really directly competed at the high end. The HD 7xxx Radeon cards were good, great even, but as time wound on, AMD only competed in niche offerings when they really wanted to – the Fury Nano was a great idea and competed with the GTX 980, Vega was a decent lineup that was too compute-focused and too late to the dance to be at the high end, and the Radeon VII was clearly an attempt to sell off datacenter GPUs to gamers while pretending later that it was never intended as a gaming GPU (despite the reveal featuring gameplay performance demos in multiple titles on stage and in the official stream).
This time, however, the rumors about RDNA2 are that AMD may have something that, even after the switchover of silicon in the stack, will rival or beat the RTX 3080. The 3090, eh, that is going to be the best single graphics card you can buy for gaming, with a price increase over the 3080 that easily outsizes its performance gains, but below that, the performance crown might be up for grabs. Despite several rumored disadvantages (standard GDDR6 memory with much lower bandwidth, raytracing on the shader cores instead of full dedicated RT hardware), AMD appears to be bringing a card that is capable of directly competing with Nvidia’s higher end enthusiast hardware, which they haven’t done in a while.
Now, here’s the thing. AMD doesn’t have a great track record in modern times on the GPU side of things. CPUs? They’ve absolutely turned that business around and are doing some of the best work the company has ever done, but the Ryzen spirit has yet to infect the Radeon Technologies Group within the company. The RX 5700XT launched last summer was a great card, but AMD continues to be plagued by reports of driver issues (it is anecdotal, but compare the front page of the AMD and Nvidia subreddits on new GPU day and you’ll see a stark contrast, and that is even considering that the AMD subreddit also is for CPU and game console hardware conversation!), poor reference cooler designs, and ambivalent add-in board partner manufacturers who often don’t put the same thought into their designs that Nvidia’s AIBs do.
Despite all of that, however, AMD has remained competitive in the mainstream, sub-$500 graphics card market, which is probably the most important one as most people buy a card below that $500 mark. Where competing in the high end helps, though, is in perception. Nvidia (and AMD and Intel, for that matter) seed the press with review samples of what are usually the halo products, and less commonly the mainstream stuff people actually buy. This is why Nvidia, for example, made sure to have the 3090 as a solution and will be likely to ensure that reviewers get their hands on them – if the high end market is Nvidia on their own, then that perception can bleed down to the mainstream. For years, this is how Nvidia maintained mindshare even when they were fighting a losing battle on the merits. If their cards could be faster than the top AMD offering, no matter how loud the cooling, hot the card, or poor the power draw, they “won.” A lot of nerd competition on the internet is won in this way – not on the actual merits of what point in the product stack you are purchasing, but instead the battle way up at the top of the stack. To some people, the Xbox Series X has more raw horsepower than the PS5 by certain means of measuring, and for that reason, people may end up evangelizing for it over the PS5 just on that measure. It’s literally the whole reason the Titan line from Nvidia exists – introduced because they can both sell the cards for a handsome profit margin in low volume directly to consumers, but also because in benchmark suites from tech publications, there will be a gap between the top end AMD part and the Titan from Nvidia.
Make no mistake, RDNA2 is the primary reason that the RTX 3000 series graphics cards aren’t maintaining the same cash-grab power play that Nvidia ran with the last generation. Also, let it be stated clearly – if AMD scores a win over the 3080 or 3070, there are already rumors that they have lined up versions of those cards with twice as much VRAM to better meet what AMD is expected to have (16 GB 3080 competitor and a 12 GB 3070 competitor). If AMD’s offering does better at high resolutions than the RTX competition, Nvidia will absolutely unleash 16 GB 3070 cards and 20 GB 3080 cards, and if the rumors are true, they’re already ready to do exactly that. They may not even claim the crown back from AMD, either – but they’ll lean hard on existing community perception – AMD driver issues, poor AIB support, etc – and they’ll likely win unless AMD absolutely destroys them in performance.
But, for right now, the nice thing is that regardless of which manufacturer you want, we have a better generation shaping up with a competitive, improved 3080 over what we would have been sold otherwise, and a Titan-tier card for less than the last two Titan releases (although still gimped on compute compared to an actual Titan card). If you believed Nvidia’s marketing hype of “two times the performance of 2080!” well, unless you play an obscene amount of Quake II RTX or Minecraft RTX (the only titles where that is true), you might be disappointed today. Likewise, if you went to buy one and ran into the absolute shit show that the launch availability ended up being online, you’re probably also disappointed. For everyone else waiting on the sidelines or for less in-demand cards (as I rub my hands together thinking about the 3090), well, you might actually be the winner this generation.
Regardless, competition is good, and even without actually launching a product as of yet, AMD merely existing and working on a thing to compete has made this generation better than Nvidia would have wanted it to be.
The way it was meant to be played, indeed.