AMD Radeon RX 6600 XT review: the Digital Foundry verdict

A great RTX 3060 competitor, but pales in comparison to RTX 3060 Ti.

Review by Will Judd Deputy Editor, Digital Foundry

Additional contributions by Richard Leadbetter

Updated on Aug. 11, 2021

The RX 6600 XT is a fascinating graphics card, no doubt. Compared to the RTX 3060 at 1080p, it comes off pretty well - it's over 20 percent faster in games like Death Stranding, Borderlands 3, Dirt Rally 2.0 and Assassin's Creed Odyssey, with smaller but still significant gains in Hitman 2 and Assassin's Creed Unity. The wheels come off somewhat in Control, a game in which AMD GPUs suffer even with RT turned off, with a three percent deficit to the 3060, but it's the only non-RT title where the 3060 holds onto a slim margin.

If you're choosing between the RTX 3060 and the RX 6600 XT, both at their respective RRPs, AMD's offering is the better value card when it comes to pure rasterised performance - for 15 percent more money, you get more than 15 percent extra performance in seven of the eleven rasterised games we tested, with smaller advantages or small deficits in the remaining four. That's great efficiency for the RX 6600XT, a card with a rated TGP that's 10W less than the RTX 3060.

The Red Devil model we were supplied with is some distance above RRP, but it does provide good performance and minimal noise while only hitting around 73C at maximum during our sustained testing period. The build quality here looks good too, with a metal backplate and a suitably menacing red glow towards the rear of the card, so definitely a good choice for a red-themed build.

However, RT remains a weak point for Big Navi, with Nvidia's RTX 3060 commanding a massive performance advantage over the RX 6600 XT in the three games we tested - Control (41 percent faster), Metro Exodus (18 percent faster) and Battlefield 5 (58 percent faster). These tests were at 1440p, but at 1080p the RX 6600 XT still struggles - its Control 1080p average of 34fps puts it only a little ahead of the 29fps of the RTX 3060 at 1440p, a much harder workload. It's a bit better in Metro, where the 6600 XT manages a playable 48fps at 1080p, but much worse in Battlefield 5, where the RTX 3060 at 1440p outperforms the RX 6600 XT at 1080p by 12 percent. This doesn't even take DLSS into account, which could push the Nvidia card's frame-rates significantly higher; none of these games (yet) support AMD's FSR tech to claw back performance for the 6600 XT. You can at least play games with RT, something that wasn't possible last generation, so it's still a box ticked and a decent effort from what is a first-generation implementation of RT in RDNA2.

But perhaps you're not interested in RTX or DLSS - you just want to play rasterised games at 1080p. That's fair enough - it's AMD's pitch after all - and it makes the RX 6600 XT a much stronger contender than the RTX 3060. But again there's an issue. The RTX 3060 Ti is a much stronger card, pound for pound, than the RTX 3060, and costs just $20 more than the RX 6600 XT at $399 (as always, assuming cards can be purchased at RRP). How does it fare against the RX 6600 XT?

Well, Team Red doesn't get off lightly. At 1080p, we see advantages that range between plus or minus four percent in more CPU-limited games (Assassin's Creed Odyssey, Hitman 2 and Borderlands 3) and absolute blowouts that see the 3060 Ti leading the RX 6600 XT by 18 to 39 percent margins (Metro Exodus, Shadow of the Tomb Raider, Doom Eternal and Control). For an extra $20, the Nvidia card seems an obvious pick if it's available - and that's not even mentioning DLSS or hardware-accelerated ray tracing.

At 1440p, it gets even worse for the RX 6600 XT against the RTX 3060 Ti, with Team Green holding percentage leads ranging from around 12 percent (Borderlands, Far Cry, Death Stranding) to 45 percent (Control) and even 80 percent (Doom Eternal). That last result is a memory bottleneck that could be resolved on a more sane setting, but it shows that Nvidia's 3060 Ti offers a bit more flexibility to target higher resolutions with bigger textures. 4K gaming is also possible on the RTX 3060 Ti, but definitely off the table for the RX 6600 XT - here it's worse again, and I experienced pretty severe texture pop-in on games like Control and Assassin's Creed Unity.

Interestingly, it's not that Nvidia is particularly strong here. The old RX 5700 XT has the same amount of RAM but much more raw memory bandwidth - 448GB/s versus 256GB/s - and seems to handle 1440p and 4K gaming much better as a result. In fact, the RX 5700 XT outperforms the RX 6600 XT in a handful of titles at 4K, which is not what we expected going into this review. Clearly, AMD's Infinity Cache solution works wonders to augment limited memory bandwidth at lower resolutions, but it seems to face sharper performance drop-offs at higher resolutions. All told, the diminished 1440p and 4K performance makes the RX 6600 XT feel a little less flexible, a little less future-proof, than you'd hope for a $379 graphics card. Perhaps memory overclocking can let this card shine, but out of the box it feels a little underpowered.

So: the RX 6600 XT looks like better value than the RTX 3060 for 1080p gaming, but both cards are outdone by the RTX 3060 Ti. The further you stray from the rasterised 1080p box - say, by adding RT or upping the resolution - the RX 6600 XT gets comparatively weaker. If it were a bit cheaper - say $349 - then I think the AMD offering here would be much stronger. (And indeed, the UK price of £329 makes the 3060 Ti £40 more expensive, making the 6600 XT's performance more reasonable).

However, all of this supposes that graphics cards are actually readily available at or near RRP/MSRP, and as we all know we don't currently occupy a reality like that. As it stands, pretty much every card bought at RRP will be a better value than an ostensibly better competitor bought at an inflated price. If AMD's partners are able to produce an avalanche of these cards, then they could be on to a winner. If not, well - better luck next time?

AMD Radeon RX 6600 XT analysis

Read this next