I'm actually thinking RIP AMD...
Lets look at what AMD
actually showed us:
Power consumption: AMD quoted 250w for RX 6800 XT but then actual benchmark shows 300w in order to pull a little ahead of the RTX 3080 which is 320-330w. The difference is fairly insignificant, though not irrelevant. At current, they could be rated as being nearly identical but this may not play out to be so irrelevant in the long run depending on overclocks, driver improvements, etc. which can have a huge impact (as seen with the Ampere line with the initial crash bug then driver fix which actually improved performance). Big Navi's example is overclocked while Ampere's is not... otherwise bigger power consumption difference but also greater performance increase for Ampere.
Infinity Cache could be a really move for the company's future GPUS to cut corners on hardware while keeping them competitive but what we are seeing in terms of cost and specs isn't exactly that cost has gone down a lot or that saved costs were invested elsewhere in the GPU's specs. It could be a good move going forward but is proving to be questionable at current. Unfortunately, the sheer impact or lack thereof is unknown as we don't know anywhere near enough about how it is being utilized.
VRAM: AMD has noticeably more memory at 16GB GDDR6 while Nvidia has 10GB of the superior GDDR6X. AMD has Infinity Cache which will also help, though exact usage is unknown so this is still very up in the air. Most of the time 10GB should easily be enough so this difference will be nearly non-existent, at least when excluding the usage of the still uncommon 4k texture packs for games. In the case of the 4k texture packs or rare game scenarios where the streaming of assets and other elements benefit from more than 10GB in a meaningful way AMD lacks DLSS and thus may in fact, despite having more VRAM, actually end up not using it in its best case scenario against the competitor Ampere. The benchmarks offered by AMD, already to be questioned until we get 3rd party testing from DigitalFoundry and others, are keeping up with the RTX 3080 when not factoring DLSS or ray-tracing, and obviously 4k texture packs. If we include ray-tracing it simply wont be able to keep up with Ampere which I will delve into more below. If we include 4k texture packs then odds are DLSS is still going to pull significantly ahead, especially with ray-tracing which effectively mandates DLSS in games to date. We could say that DLSS isn't common but the RTX 3080's 10GB is enough for any current game and DLSS support is, finally, beginning to quickly grow. Last, a very important topic regarding next-gen IO. AMD lacks a solution while Nvidia has RTX IO + Microsoft's DirectStorage to solve this problem which would virtually single handedly negate the major benefits from Big Navi's VRAM lead and may even render Infinity Cache moot (take with a grain of salt because we don't know its exact usage and even being L4 cache it would be faster than typical memory fetches which can matter meaning Infinity Cache's value may not be negated by this). As for next-gen intense IO demanding games the VRAM difference and Infinity Cache will be totally irrelevant. Only a next-gen API solution will help (see below for more).
Next Gen IO Solutions: AMD's Big Navi has no contender at all for this. There could, and seriously I hope there is or they really will be in for an uphill battle, something in the works they just haven't publicly disclosed or been leaked but with IO being an incredibly relevant factor for next gen and the fact that it fundamentally alters how games may be designed and any system without sufficient IO, unless designed with a fallback measure, may be simply incapable of running such a game at all this matters... a lot! Nvidia offeres RTX IO + DirectStorage (Microsoft's API which is part of Xbox Series X' IO solution as well) to help with decompression and utilizing DirectStorage. Without RTX IO the expense of decompression on PC, due to lack of specialized hardware for this like the PS5 & Xbox Series X have, would bring even an AMD Threadripper CPU to its knees and prevent it from doing virtually anything else due to the staggering demand the decompression would place on the entire CPU in worst case scenario. Even in best case scenarios the performance hit would be immense in IO intense games. Therefore RTX IO which is a decompression solution using the GPU helps solve this problem for Nvidia GPUs allowing it to keep pace with games. This isn't a matter of simply waiting two seconds for a loading screen. There are going to be games built in a different way where it must be instant otherwise your going to deal with unplayable stuttering, excessive load screens, disconnects and possibly even processing issues or crashes without a next gen IO solution due to how the nature of next-gen IO performance impacts game design. Nvidia's own solution isn't even really available until 2021 and until developers and accompying games even really begin truly utilizing it this wont matter, but it eventually will result in a divide of what IS playable and what simply CANNOT be played on a platform that doesn't have appropriate IO performance.
Ray-tracing: AMD hasn't shown it despite everything else. Don't you find this odd? They show a host of benchmarks but no ray-tracing related benchmarks or examples. They state 10x faster than software shader implementation. What does this mean? Well, it means it isn't good enough. Turing, the RTX 1080 Ti is quoted by Nvidia as its RTX cores offering a baseline, minimal, and potentially higher (10x+ to be precise) ray-tracing over a software shader solution. This puts Big Navi's ray-tracing at last gen and we know Ampere's is a massive step up. In fact, early benchmarks and tests by GamerNexus and others have indicated we haven't even tapped the full extent of RTX cores (see GamerNexus' test with ray-tracing on the Quake II) indicating as ray-tracing matures it will only widen the gap between Big Navi and Ampere significantly. Ampere is achieving its current performance while also maintaining dedicated RTX and Tensor cores while AMD will have to make continued improvements to their specialized hardware and more space going forward which indicates it will likely not mature as adeptly as NVidia's GPUs will. Fortunately for AMD and its owners of Big Navi going forward ray-tracing is still immature, to put it mildly. Even the more ray-tracing oriented marketing is for games that may just have basic reflection or the occasional diffuse (
cough Watch Dogs Legion, and Cyberpunk 2077 which does much better but is still relatively ray-tracing lite).
DLSS: AMD still has no publicly disclosed solution for DLSS which is a technique that will only continue to mature. DLSS 1.0 was awful no matter how you look at it. DLSS 2.0 has shown to be an immense improvement and can often provide not only outright incredible performance gains, especially with ray-tracing which is largely considered to require DLSS at higher settings to even be reasonably playable, but it can even potentially offer better than native results in many cases. At the same time it still isn't perfect. It does produce the occasional artifacts, though over at GamersNexus they found that aside from someone intensely focused on discovering them and not moving at all... aka almost no one, the negatives are typically nearly impossible to notice. None the less, it still has room to improve and that is precisely what the nature of this type of technology that relies on trained artificial intelligence via neural networks (a system for, quite literally, training/improving AI) entails. We've even seen it allow the RTX 3090 to achieve incredible results with 8K which should otherwise be unplayable on demanding modern titles. That said, not only does it have room to improve but it also has to be implemented with the game so support will need to be more widespread to truly matter. Most of the notable major titles coming out later this year and next year will feature it, however, so it seems that it is finally starting to gain traction with the games that are actually potentially demanding enough to necessitate it. As AMD didn't show comparisons of performance with DLSS, or ray-tracing which it is weaker with regards AND basically requires DLSS even with Ampere's more potent ray-tracing capability, it should be noted if ray-tracing is included or DLSS is supported AMD couldn't even be said to honestly even be competing at that point. With DLSS and Ampere's superior ray-tracing performance in games that support both even weaker GPUs like the RTX 3070 may very well outperform AMD's higher offerings, and possibly not even by a small margin. This also means in terms of long-term value Nvidia's Ampere cards are going to mature more gracefully than AMD's Big Navi.
Performance and Cost: The reality is AMD's GPUs don't really compete in a meaningful way at the high end be it the RX 6800 XT, RTX 3080, or above. At that point AMD may be a tad cheaper for virtually identical performance in default scenarios, but odds are if your spending that kind of money you want ray-tracing as well. In addition, if the game supports DLSS its a pure one sided stomp in Nvidia's favor, especially if ray-tracing also gets factored in, but also as it matures and starts producing superior to native results on details like billboard text, emblems, background wiring and cables, etc. At the RTX 3070 and below tiers for the more budget oriented build is where AMD will fight its real fight.
Facebook
Twitter
Instagram