Two months back, AMD launched the AMD Radeon RX 6600 XT, the first card to use the Navi 23 GPU. Today, AMD follows up with its first truly mainstream priced RX 6000-series card, the Radeon RX 6600. Take the same GPU but with four of the CUs (compute units) disabled, clock it a bit lower and you get the RX 6600 non-XT.
Will it be one of the best graphics cards, or will it come up a bit short? A lot of that will depend on retail pricing and availability, as GPU prices remain inflated, but supply has been a bit better on the RX 6600 XT than on other RDNA2 graphics cards, so hopefully for gamers, AMD can supply a reasonable quantity of GPUs for this launch.
I talked about the lack of a vanilla RX 6700 prior to the Navi 23 launch, and that previously widely rumored card remains MIA. Presumably that's because any of the Navi 22 chips that aren't fully functional can be sold as one of the various mobile RX 6000M-series solutions. AMD isn't taking that same approach with Navi 23, though, and along with trimming off some of the performance, the RX 6600 reduces the power requirement to just 132W and also cuts the official starting price to $329 — the same price as Nvidia's RTX 3060, though with 'only' 8GB VRAM. That's basically mainstream pricing in today's market — actually, it's less than you'll pay for most actual mainstream GPUs — though we suspect AMD's partners and the various retail outlets will jack up the price as long as GPUs remain in short supply.
Besides reducing the CU count and reducing the GPU clocks — by a relatively large 315MHz if you look at the Game Clock — AMD also reduced the GDDR6 speed from 16Gbps to 14Gbps. Note that the 'maximum' boost clock of 2491MHz (technically the GPU can exceed even the boost clock) is quite a bit higher than the game clock, so we'll have to see how it all plays out in the benchmarks. But overall we'd expect the RX 6600 to be 10–25 percent slower than the RX 6600 XT, depending on whether a game needs more GPU power (up to 25% slower in theory) or more memory bandwidth (about 12.5% slower). Here's the rundown of AMD's latest RX 6000-series GPUs and their specifications.
|Graphics Card||RX 6600||RX 6800 XT||RX 6800||RX 6700 XT||RX 6600 XT|
|Architecture||Navi 23||Navi 21||Navi 21||Navi 22||Navi 23|
|Process Technology||TSMC N7||TSMC N7||TSMC N7||TSMC N7||TSMC N7|
|Die size (mm^2)||237||519||519||336||237|
|Infinity Cache (MB)||32||128||96||128||32|
|Game Clock (MHz)||2044||2250||2105||2424||2359|
|VRAM Speed (Gbps)||14||16||16||16||16|
|VRAM Bus Width||128||256||256||192||128|
|TFLOPS FP32 (Boost)||7.3||20.7||16.2||12.4||9.7|
|PCIe Slot Interface||x8 Gen4||x16 Gen4||x16 Gen4||x16 Gen4||x8 Gen4|
Paper specs don't always match up with real-world performance, so we'll have to see how the RX 6600 fares against its main competition — which includes not just the RTX 3060 and RX 6600 XT, but also previous generation cards like the RTX 2060, RTX 2060 Super, and RX 5600 XT. Considering this card replaces the previous generation RX 5600 XT, it's unfortunate that generational pricing has gone up quite a bit, but then there's no sense in expecting AMD to launch at a price that few people will ever see. Again, we hope there will actually be a fairly decent supply of RX 6600 cards, both for the initial launch and going forward.
We mentioned the issue with AMD's game clocks vs. boost clocks already, and we've used AMD's game clocks for the above TFLOPS numbers. However, given the way things have changed with boost clocks on RDNA2 (i.e., RDNA2 GPUs often reach and exceed boost clocks while game), it might be better to compare performance using boost clocks rather than game clocks. If we do that, the RX 6600 can provide about 8.9 TFOPS of compute, while the RX 6600 XT delivers 10.6 TFLOPS of compute. That's a much lower 16% drop in theoretical performance, and it matches up better with the reduced memory bandwidth.
The Navi 23 architecture uses the same general formula as the other Big Navi and RDNA2 GPUs. It supports DirectX Raytracing (DXR) and implements the full DirectX 12 Ultimate features list, including Variable Rate Shading (VRS), mesh shaders, and sampler feedback. The smaller 32MB Infinity Cache on Navi 23 represents a compromise that mostly benefits 1080p and maybe 1440p, but mainstream GPUs generally aren't used at higher resolutions so that should be okay.