Just as I did with the Ryzen series announcement earlier in the month and the RTX 3000 cards last month, the Radeon announcement has come and gone and I’ve had a chance to watch some more videos from AMD, look at articles and videos from tech outlets, and get a better read than my immediate, just watched the reveal thoughts reflected.
First up, the obvious must be restated – while independent testing is needed to verify performance claims, assuming AMD’s claims are within 5-10% of accurate, the cards shoot above what many rumors had initially suggested. Some rumors and less-reputable “leakers” had suggested performance on-par with the 2080 Ti, which, in a way, was true, but for the bottom card of the announced stack. The 3 cards competing directly with Nvidia is a fantastic uplift over what many of expected, with sharpened cynicism from years of botched launches from AMD when they promise a high-end competitor.
There are a few points that are good, but some worthy of further interrogation, so we’ll jump right in.
Pricing Is Sort of All Over The Place: Similar to the Ryzen 5000 lineup, AMD is asking more up-front for parts this time around, with cards starting at $579 (to compete with the 3070), $649 (fighting the 3080) and $999 (competing with the 3090). The pricing of the top two, similarly to the Ryzen lineup in a way, make sense. The RX 6800 XT is a slightly reduced cost competitor to the RTX 3080 and meets the challenge very well, with performance shown beating the 3080 nearly all the way around. Likewise, the RX 6900 XT meets and beats the 3090 in many benchmarks, with an important caveat we’ll get into later, so its pricing much lower than the 3090 fits with both the potential performance and the Radeon card’s reduced VRAM in that comparison. However, the RX 6800 sort of makes me scratch my head. It is ahead of the RTX 2080 Ti (and thus the RTX 3070) but comes in $80 higher than the 3070. It does have what appears to be a greater margin over the 2080 Ti compared to the 3070, and has twice the VRAM of the 3070, but that performance also comes with an asterisk. Speaking of that!
Benchmarking With Smart Access Memory and Rage Mode Will Be Seen as Disingenuous: I noticed quite handily that both the RX 6800 and the RX 6900 XT benchmarks were done with varying combinations of Smart Access Memory and Rage Mode on. Now, these features don’t provide a night and day difference, with the combined average being around 5%, but some will complain that this is “unfair” and fanboys have already been drumming on this one incessantly. I find myself somewhere in the middle – it is clearly an obfuscating manuever to show benchmarks in this way with inconsistent usage of features, but at the same time, they’ve shown the difference that results from these features in the games they showed data for, and if you extrapolate that out, you can figure where the cards sans the features roughly are. The other side, of course, is that these features are pretty easily available. AMD has a dedicated fanbase that will match Ryzen and Radeon as a matter of course, so provided it is the right era of Ryzen and a 500-series chipset, it’ll probably be on for a large number of Radeon buyers right away. Even better, Rage Mode is a thing in the software right out of the box, so unless you’re building a small form factor PC with limited cooling capacity or noise tolerance, or have a gimpy power supply, this is also a no-brainer auto-enable feature. Sure, there will be people using this Radeon on an Intel platform (or with an older Ryzen/non-500 series chipset motherboard) and they’ll miss out on Smart Access Memory, but neither boost seems that substantial and if you take them away from the benchmarks presented, it either gives Nvidia a narrow lead or simply reduces the margins on the AMD victory, so I would struggle with calling it misleading.
High VRAM Across the Board: All 3 cards announced today have 16 GB of VRAM, which is great. I say great because I do think that the potential for use in DirectStorage might be there, but at the same time, it isn’t necessarily a requirement. Like I broke down in my post about VRAM last week, the usage (actual usage vs. allocation) of VRAM points towards 8 GB remaining a gold standard even as the consoles have 16 GB, because by the time OS memory reservation and CPU RAM utilization hit that 16 GB in the PS5/XSX, you’re left with pretty much 8 GB! But, in my opinion, 16 GB is a strong marketing number – it says that these cards can match and exceed the consoles readily, and by committing to 2 GB per memory controller, it means the benefit will travel down the stack (rumors are that AMD’s 5700 XT replacement will come in at 12 GB of VRAM and the stack pretty much bottoms out at 8 GB) so that potential customers are going to be offered something that creates the perception of added value. Already, the $80 price difference of the RX 6800 over the RTX 3070 is being justified by the doubling of VRAM in addition to the better performance, which is not a bad case to make. I do think, however, that 16 GB remains a bit of an overshoot for most folks, but it should provide a reasonable future usage – as games grow in VRAM usage thanks to the new consoles and high-resolution texture packs for those games start to hit PCs, having a card with such a large amount of VRAM at lower price points will prove its value – potentially. For now, it is an overshoot, but it has a clear marketing purpose and since the supply of GDDR6 isn’t really constrained in the same way that G6X is for Nvidia, AMD can get away with this handily.
Ecosystem Partnerships Are Great: One thing AMD has is a few consistent development partners that optimize for Radeon, and using those partnerships again to tout the card is a positive. However, as someone looking to get an RX 6900 XT and get into custom loop watercooling on their next build, the announcement that EKWB was already partnering with AMD and showcasing a custom kit with a full-cover RX 6000-series waterblock is fantastic news. One of the major weaknesses of Nvidia in this regard is that they often don’t provide watercooling vendors or other third parties with board diagrams for their reference and Founder’s Edition designs in advance, which means that if you want a watercooled Geforce card you have to wait for a vendor you like to buy the card you have (and if you have a really great partner card with a custom board layout, that wait is longer!) and then to test their designs before they release. As of now, a full month into the RTX 3000 series lineup, and most watercooling vendors don’t even have Founder’s Edition blocks, including EKWB. EKWB has a coming soon Strix waterblock and an EVGA FTW3 one, but nothing for higher-quality custom cards outside of those two. If this EKWB partnership means day one block availability (coupled with the performance characteristics of the cards), then AMD is going to sell more cards on the virtue of that arrangement. Maybe not substantially more, but it speaks to their growing confidence in their market position and also to the effect that has on third-parties wanting to get into business with AMD.
Raytracing – A Lot Of Meh?: Generally speaking, I like a lot about ray-tracing at a fundamental level. It is exciting technology and seeing the ways in which GPU companies are trying to engineer around the resource challenges it brings is endlessly fascinating to me, a nerd who used to read detailed breakdowns about memory compression in mid-2000s GPUs. However, AMD’s solution seems sort of interesting, but the company is being extremely cagey about it, which, coupled with benchmarks that have been shared, indicates that it isn’t really going to be all that great on these cards. While the RX 6900 XT cleanly beats the RTX 3080 in a variety of benchmarks, in a very simple DXR raytracing demo, the 3080 beats the RX 6900 XT. This may change as FidelityFX support for denoising, other various raytracing tech, and the like come into play, but for right now, AMD seems to know they’re weak on that front, so they basically said “we have it and that’s all about it for today.” To be fair, raytracing, while a very interesting technology, is also incredibly niche and doesn’t have a lot of developer support. It probably also won’t have a lot of developer support until the fidelity of it fully surpasses that of highly optimized and worked over rasterization, and until hardware adoption catches up. Right now, only Nvidia cards that cost $300 or more have it, and they generally don’t do well enough at it for standard play as is. With AMD entering the market on RDNA2, they’ll only be offering it on cards that cost $580 or more at launch, and it will likely be a full year before we see the full stack including it. Similarly, it probably won’t be until near Summer 2021 where Nvidia will have a full Ampere stack that goes down into a xx50 tier card with RTX, and even then, they could still do what they did with Turing and just gut raytracing to make something silly like a GTX 2660 and 2650 or some other weird shit like that. Given all of this, I am disinclined to ding AMD too hard on this one, but I did find it funny how quickly it was mentioned and then moved on from!
Overall Impressions: Great News Almost All The Way Around, And Now I’m Getting a Radeon RX 6900 XT: My biggest impression is that I want an RX 6900 XT, a thought that didn’t enter my mind with a high degree of seriousness until the presentation was over. In my mind, AMD had a list of criteria they could meet to win my upcoming GPU purchase away from Nvidia and they absolutely smashed that out of the park. Pricing, while high, remains competitive at these higher-tier offerings which bodes well for a full stack replacement over the next year (a Radeon RX 6500 that matches the PS5 in graphical performance at a sub-$300 cost is going to be bonkers, and I suspect that is where such a card will land!). For a lot of halo-chasing enthusiasts (I count myself in this group) it really does impress that the RX 6900 XT reference card is doing so well against the RTX 3090, because that means a board partner card from a high-quality AMD-only vendor like Sapphire or PowerColor will have even greater performance and still be cheaper than the RTX 3090 FE, much less a partner card on the Team Green side of things. For those who buy at more mainstream pricing, this news is even better – today, there isn’t a new RDNA2 card for those tiers yet, but the announcement today puts forward a promise of higher performance in those tiers.
The competitive response from Nvidia will be interesting to see, as well. Rumors have it currently that they have cancelled their 16 GB 3070 and 20 GB 3080 and are instead fighting with the GA102 silicon to get a 3070 Ti and a 3080 Ti out of it, bookending the 3080 with cards that will offer slightly less and slightly more performance, with the 3080 Ti functionally being a 3090 with less VRAM. The challenge is that Nvidia priced themselves into a corner, clearly not expecting AMD to show up with this much fight, and so a 3070 Ti would need to beat the RX 6800 cleanly while being similarly priced or cheaper (a difficult proposition) while the 3080 Ti is going to have to drastically undercut the RX 6900 XT to have a chance of being the value champion in that tier. It may, for once, come down to mainstream competition – Nvidia has yet to announce their 3060-tier and below lineups, and I suspect that the next year will be a fascinating tug-of-war between the two companies as they trade blows with each new member of the product stack.
For me though, as I mentioned above, I am pretty firmly now on the RX 6900 XT hype train. It offers me what I ideally wanted out of the RTX 3090, but does so at a lower cost. I was prepared to spend near $1,000 for an EVGA FTW3 Hydro Copper RTX 3080 as-is, so adjusting that to an RX 6900 XT and getting a waterblock for it isn’t a huge stretch in my upcoming system budget, and the partnership announcement with EKWB means that doing this on day 1 of launch is pretty likely. I still have open questions on driver performance, optimization, and potential issues there, as software is consistently AMD’s weak spot on the Radeon side of things, but I am very glad that the RX 6900 XT launch is after the 6800 family, as that launch should be illustrative of any potential issues.
Whichever way you go, it is a great time to be a PC gamer and it is cool to finally see AMD back in the fight at all price tiers on both CPU and GPU!
2 thoughts on “Sidenote: Analyzing the Radeon RX 6000 Announcement”
Haha, the Amiga parallels for SAM continue – if I had a nickel for every tome some Intel or Mac fanboi complained that using Amiga-specific video modes in comparison to their less modern designs, I’d have … well, a lot of nickels.
The same applies now as applied then: compare the cards when used as intended. If SAM / RAGE is intended to be on and active during normal use, that’s what your base comparison is. I mean, we don’t turn off DMA for testing, and that’s “a feature” as well.