Last week, AMD made a big announcement at the Chinajoy festival – a new GPU.
RDNA 2 continues its march down the product stack, moving from the 6700 XT to the 6600 XT, a new card set to launch from AMD’s board partners on August 11th. The relevant details are as follows: a new GPU die featuring 32 compute units comprised of 2,048 stream processors, a 32 MB Infinity Cache paired with 8 GB of GDDR6 memory on a 128-bit wide memory bus, running a pretty high “game clock” of 2,359 MHz with a maximum boost near 2.6 GHz, all of that at 160W of board power, meaning only a single 6-pin PCIE power connector is needed along with the 75w of slot power available from the motherboard. This means the card has both fewer CUs and a smaller memory bus than its predecessor 5600XT, but due to faster GDDR6 memory chips and improvements to throughput at an architectural level as well as the much faster clock speed, the 6600XT is near 25% faster in theoretical compute performance – which won’t translate 1:1 in actual graphical performance, but is something we’ll see more about in benchmarks.
Lastly, the price – $380 is the MSRP of the card.
AMD’s positioning on this one is positively bizarre, in my opinion. The card is being posed as the best 1080p resolution gaming card, which may very well be true. However, this is undermined by a few factors. Firstly, the pricing is insane for a 1080p card in 2021, but at the same time, given that the state of the GPU market remains insanity in 2021 and thusly the positioning of the 6600XT fits in the current market, sadly. Compared generationally, the card is around $100 more expensive at MSRP than the 5600 XT’s $280. Some of that value comes at the increase in VRAM, from the 6 GB of the 5600 XT to 8 GB in the 6600XT. Some of the cost comes down to tariffs and their continued impact on the GPU market and the costs of these new products, as parts must be imported and a launch at this point means it is unlikely that AMD had pre-imported any cards to save on those tariffs.
The price creates a challenge in 2021 even still, though – for most 1080p gamers, their current GPUs, provided they are still working, are more than sufficient for 1080p gameplay, as 1080p is now realistically the baseline resolution that most “gaming” rigs would hit as a baseline, and it no longer is a stretch as it once was nearly a decade ago. If you have a reasonably good card purchased in the last 5 years, 1080p isn’t going to get that much better. Even then, while the 6600XT is likely to end up being a strong 1080p contender, will the difference in pricing be worth the gain over something like a GeForce GTX 1060 or a Radeon RX 580?
My big disappointment with the card as an idea, though, is that the pricing and compute unit change make it less ideal as a console replacement. The 6600XT, had it followed the CU count of the 5600XT, would have matched the PS5 in RDNA2 compute units 1:1, with a higher clock speed and the Infinity Cache to help mend the memory bandwidth discrepancies. That would have made it, at a normal x600 XT price point, a great entry point for PC gamers to be next-gen ready, with a card featuring capable raytracing, sufficient rasterization performance, and a good match on modern features. Instead, the card has fewer CUs than expected, a compromised memory arrangement that could negatively harm performance, and while the higher clock speeds compared to the PS5 get it very close to that performance, the PS5 is faster and can be had as a whole system for only $20 more than this card (in theory, provided you can ever find one).
So there is a handful of challenges in front of this card – it means that current-generation console replacement PCs still start at around $1,000, it offers less raw performance than the PS5 GPU, much less the Xbox Series X one, and with new features, the card may be suitable for certain applications of 4k resolution, but it will be constrained by low memory bandwidth, the already-low hit rate of Infinity Cache at the full 128 MB size for 4k resolution visuals, and the lower ROP counts. The card may benchmark very well, and that may carry it ahead, but for now, in theory based on the configuration, it seems doomed to be a card for no one – not an appealing upgrade path unless your card is very old, not an appealing new option at the price it is offered for a new system, and potentially underpowered compared to Nvidia’s RTX 3060, whose MSRP is $330 and also includes more and faster VRAM, with 12GB on a 192-bit bus. (At 1080p, for what it’s worth, 12 GB is overkill and future-proof to an absurd degree – if you’re running 1080p in like 2030, you might fill that VRAM at last!)
All of this is hypothetical, of course, based on a reading of the spec sheets in comparison and considering the pricing and market positioning. AMD is counting on a feature carrying the card forward in another way, though.
A few months ago, AMD launched their DLSS competitor, FidelityFX Super Resolution, or FSR for short. AMD’s competitive message with this card in particular leans on FSR as a selling point, that while it is a 1080p beast, it can also run higher resolutions through FSR with great image quality. Similar to how Nvidia positions cards by how DLSS can elevate them, FSR here is being used to inflate the potential of the card.
There is a complication, however, or really two. Firstly, as FSR is new, it requires implementation from developers to support it in their games. The roster of titles supporting FSR is…slim. Currently, just 13 titles are listed on AMD’s site for the technology, and of them, only 3 are anywhere near popular. Support for using FSR with rescaling tools is becoming a thing, which may allow savvy tech enthusiasts a way to force it on for more, but for your average gamer buying a 1080p card, that won’t be realistic. Secondly, however, is a blessing and a curse. Like all of FidelityFX from AMD, FSR is an open technology. AMD does not lock it to just AMD hardware, as Nvidia does with DLSS on their hardware, and so you can use FSR in supported titles on any GPU – Intel integrated or upcoming discrete cards, Nvidia cards, and AMD’s own. So where Nvidia can promote DLSS as a unique feature and selling point, with a growing library of titles, AMD is stuck in the worst part of the DLSS rollout from 3 years ago, where no titles support it short of a select few, and even then, they have it worse still because you can just keep your GeForce GTX 1060 and run FSR on it. In fact, AMD showed data on how FSR helped the aging GTX 1060 keep pace as a part of the FSR rollout! I admire AMD for making open technology and trying to contribute to the broader ecosystem, but at the same time – that means you have no competitive use for the technology!
So I’m honestly quite confused at the rollout strategy in use here by AMD. They’re offering a card with a minimal generational performance leap at a massive generational price leap, targeting a resolution that is largely considered baseline in 2021, and with a feature to extend the card’s usefulness that is also not unique to it and can help the cards you might want to replace with a 6600XT instead extend their lives. For a company that has been so spot-on with their CPU messaging, yet again, the Radeon Technology Group within AMD just doesn’t have the marketing savvy to deliver a coherent message on this card, and it falls apart with any amount of examination. Yet, it also doesn’t have to be that smart for now, because even given all of that – there is a market that is still eagerly buying up GPUs as they become available, and thus, it will very likely still sell out even though it makes so very little sense as a purchase, because no current GPU makes sense at current pricing and availability, and yet, here we are.
Hopefully, conditions in the silicon market stabilize more in the coming months so that such products are rightfully shunned (pending reviews, of course). But, given the state of affairs right now…I suspect that a 1080p GPU in 2021 at $380 will look like a steal for a little while yet, sadly.