Fake Rumors Suck: Will the Real RX 6000 Please Stand Up?

There's a ton of excitement and anticipation surrounding AMD's upcoming RX 6000, aka Big Navi. We get it: We really want to see how AMD stacks up to Nvidia's new Ampere GPUs. Could AMD nab the top spot on the best graphics cards and GPU benchmarks? Maybe! But let's douse the flames a bit. Here's why we urge caution and not getting too caught up in all the hype.

First and most importantly: Every benchmark 'leak' so far is in question. We don't have confirmation of GPU core counts, clock speeds, or even model names. Perhaps more critically, final drivers are not yet available. We're not saying AMD doesn't have work-in-progress drivers, but if AMD has even half a clue of how to launch a big new GPU without actual verifiable leaks getting out, the AIBs do not have anything close to final RX 6000 drivers.

The best information we have right now is that AMD will be hosting its Radeon Gaming event on October 28, which will be the GPU equivalent of AMD's Zen 3 reveal from October 8. AMD also teased RX 6000 performance at that event, showing results from three games: Borderlands 3, Call of Duty Modern Warfare (2019), and Gears 5. Based on our own testing, the unnamed RX 6000 GPU came out tied with the RTX 3080 in Borderlands 3 but was 8% behind in Gears 5. As for Call of Duty, there's no built-in benchmark, so we have no way of knowing how AMD tested and thus have no comparison point.

The good news is that AMD at least has some RX 6000 graphics card coming that looks like they will be fairly competitive with the RTX 3080. But we have lots of disclaimers: AMD could have cherry-picked the tests to show scenarios where it does relatively better. The test platforms were not the same (i9-9900K vs. Ryzen 9 5900X). Drivers are not yet finalized. And we still don't know which GPU AMD showed.

Regarding that last point, one take (my take) is AMD would show its highest-performance option, meaning the RX 6900 XT. Others think AMD is sandbagging and showing off a GPU that's a step down from the rumored RX 6900 XT, i.e. the RX 6800 XT or even RX 6800. The only people who actually know which GPU was used aren't talking (yet).

But now we have a veritable flood of performance leaks and benchmark results, none of which can be fully confirmed. The results look good! Except, they look almost too good, considering AMD's own performance tease. Most of these rumors and supposed benchmarks fall into the category of stories amounting to, "My best friend's sister's boyfriend's uncle knows this guy who has an RX 6800 XT, and he says..." Again, what drivers are these cards using? What GPU are they running? Take a few spoonfuls of salt, in other words.

Here's the thing: We've been here before. Let me recap the most recent three high-end GPU launches from AMD.

AMD's Fiji architecture back in 2015 was going to crush Nvidia. It was the first GPU to use high-bandwidth memory (HBM), packing a whopping 512 GBps of bandwidth and 8.6 TFLOPS of theoretical compute performance. At the time, it was going up against Nvidia's GTX 980, which was only a 5 TFLOPS card with 224 GBps of bandwidth. The rumor mill said AMD was even going to beat Nvidia's then-fastest Titan X! Except, just before the R9 Fury X launched, Nvidia released the GeForce GTX 980 Ti.

The Fury X had moments of greatness, but overall didn't quite match Nvidia's similarly-priced card. There were select cases where it could even beat the Titan X, but mostly it was slower than the 980 Ti. Today, five years later and running more recent games, the Fury X is still about 3-5% slower than the 980 Ti while using 30W more power.

A similar story unfolded two years later, except it was even worse in many ways. RX Vega 64 was again going to take down Nvidia's top GPU, the GTX 1080 Ti, or so the rumors suggested. 12.7 TFLOPS and 8GB of HBM2 with 484 GBps of bandwidth, versus 11.3 TFLOPS and 11GB of GDDR5 with 484 GBps. That seemed a stretch, but AMD talked up its high-bandwidth cache controller and architectural enhancements, and there was at least some hope.

This time, there wasn't even a last-minute update from Nvidia to spoil the launch party … and the competition wasn't even close. Not a single game favored the RX Vega 64, and even the year-old GTX 1080 managed to claim quite a few wins. AMD's best was only able to match Nvidia's second best. Today, RX Vega 64 is about 5% faster than the GTX 1080 FE across our gaming test suite, but it's nowhere near the 1080 Ti.

Last year's Navi 10 and RX 5700 XT continued the pattern. One year after Nvidia launched the RTX 2080 and other Turing GPUs, AMD showed up with a new RDNA architecture promising significant performance improvements. It also used TSMC's 7nm (N7) node, over a full step ahead of the 12nm 12FFN Nvidia used for its Turing GPUs. Rumors and leaks suggested "better than RTX 2080" performance, with a significantly smaller chip. Some of those (which in retrospect were obviously 100% fake) claimed AMD's new high-end parts would cost as little as $250.

AMD did choose to go after the 'mainstream' market, eventually, with the RX 5600 XT. However, AMD didn't even try to beat the RTX 2080 Ti, and at best comes in just ahead of the RTX 2070. Navi 10's performance was good, but efficiency was only a match for Turing, and it didn't have any ray tracing or deep learning hardware. And of course, Nvidia launched the 2070 Super and 2060 Super to counter Navi 10, with the 2070 Super beating the 5700 XT by about 5-7 percent. That resulted in AMD dropping the launch pricing by $50, and street pricing dipped as low as $330 (but now hovers in the $380 range).

I present these past launches not as evidence that AMD will underperform, but as evidence that the rumors that come out are often extremely skewed, often in favor of AMD. Consider the source and the track record of that source. There are prolific Twitter users that seem to be wrong about 90 percent of the time — with the 10 percent coming either via luck or actual leaks right before the official reveal. Again, the best indication of Big Navi's performance is from AMD, and it doesn't tell us much. Next week, we'll find out a lot more, and I for one will be happy to put the rumors to bed.

So like we were saying, we've been here before. AMD has new and improved GPUs coming, and the rumor mill is spinning full tilt. We'd love to see AMD stick it to Nvidia for a change. Realistically, even matching Nvidia's RTX 3080 performance for a slightly lower price would be great to see. By the time RX 6000 cards arrive, Nvidia will have launched at least the GeForce RTX 3090, GeForce RTX 3080, and GeForce RTX 3070, so that's AMD's competition. AMD has to beat Turing, because Nvidia's own RTX 3070 is apparently set to match the RTX 2080 Ti with a launch price starting at $500.

Actually, what would really be great to see is a supply of AMD cards that's actually able to meet the demand. That's sort of dependent on the performance and price, but right now, it remains extremely difficult to buy an RTX 30-series GPU. Nvidia supposedly delayed the RTX 3070 launch by two weeks to help build up inventory, but we still fully expect the cards to sell out. Nvidia's CEO even says Nvidia expects demand to outpace supply until early 2021. Unfortunately, while AMD has released guidelines to its retail partners on how to avoid scalpers and bots buying up inventory, we also expect AMD's RX 6000 cards to sell out for at least several weeks after launch.

That's just the way these things go. We're excited, and so are lots of other gamers. Being cooped up with COVID-19 still running rampant doesn't help, as that means even more people are thinking about PC upgrades. We've seen shortages on CPUs, GPUs, PSUs, webcams, and more over the past six months. A highly anticipated GPU from AMD isn't going to somehow be immune from shortages, no matter what AMD tells it's retail partners.

Our advice, then, is to be patient, sit back, and watch things develop. We'll have full benchmarks and a review of Nvidia's GeForce RTX 3070 next week, and we'll also get AMD's official RX 6000 announcement at roughly the same time. That will likely include some preliminary AMD performance numbers for RX 6900 XT and RX 6800 XT (or whatever they end up being called). We'll have our own reviews of AMD's cards, with independent numbers using our standard testbeds, by the time those cards launch as well. Once the dust settles, we'll know how the cards truly perform and how much they cost, with no messy rumors or speculation muddying the waters.