WEBVTT 00:00:00.080 --> 00:00:07.200 The RTX 3090 or BFGPU as NVIDIA's CEO 00:00:04.400 --> 00:00:12.800 called it has been nicely toppled by a card that costs a nice 23 as much. At 00:00:10.320 --> 00:00:18.960 least that's what the nice guys at AMD would like us to believe. And this is 00:00:16.240 --> 00:00:27.119 it. The RX6900 XT. It looks nice. It feels nice. It 00:00:24.160 --> 00:00:32.480 even smells nice. But can it really be that simple? Just tack a few more 00:00:29.439 --> 00:00:35.440 compute units onto the 6800 XT and 00:00:32.480 --> 00:00:41.559 suddenly you're the top dog. Well, that would be nice, but it's also not quite 00:00:38.239 --> 00:00:41.559 that easy. 00:00:47.920 --> 00:00:54.160 After the announcement, we hoped that AMD was doing more under the hood than 00:00:52.160 --> 00:01:01.199 just increasing the number of compute units because the RX6900 XT, at least on 00:00:58.239 --> 00:01:08.000 paper, is otherwise exactly the same card as the RX6800 XT. Same RDNA2 00:01:05.439 --> 00:01:14.799 architecture, same core clocks, same memory, same power draw. Hey, wait a 00:01:11.119 --> 00:01:17.360 minute. AMD pulled a fast one here. The 00:01:14.799 --> 00:01:23.920 rated graphics power is the same, but the recommended power supply is 100 00:01:20.479 --> 00:01:26.320 watts higher for the 6TN00 XT. So, it 00:01:23.920 --> 00:01:31.360 seems that in spite of those extra compute units and the cooler being 00:01:28.799 --> 00:01:37.920 identical, barring what AMD calls minor tweaks, the sixtyn00 XT is in fact being 00:01:35.040 --> 00:01:42.799 pushed quite a bit harder in boost than its sibling. Well, okay, mystery solved 00:01:40.479 --> 00:01:48.240 then. That's exactly what AMD says is happening. And the sixtyn00 XT is in 00:01:45.759 --> 00:01:53.280 fact a highly binned part capable of overclocking much more reliably than the 00:01:50.640 --> 00:01:58.479 6800 XT can. It's an impressive feat considering just how fast some of those 00:01:55.360 --> 00:02:01.040 cards can go. Of course, we can't just 00:01:58.479 --> 00:02:05.600 take AMD at their word. This is Lionus Tech Tips. We've got to test it for 00:02:03.200 --> 00:02:09.440 ourselves. So, we grabbed the three fancy new Radeons and their 00:02:07.439 --> 00:02:14.160 corresponding GeForce cards and took them all for a spin. As you might expect 00:02:11.599 --> 00:02:17.599 by now, AMD crushed it in traditional rasterization, coming close to the 00:02:16.080 --> 00:02:23.040 RTX3090 in Shadow of the Tomb Raider despite 00:02:19.440 --> 00:02:26.080 costing $500 less. Though, they fell 00:02:23.040 --> 00:02:28.160 well short of even the RTX3080 when it 00:02:26.080 --> 00:02:32.800 came to real-time rayraced lighting. Minecraft RTX tells a similar story with 00:02:30.480 --> 00:02:37.519 AMD choking hard enough here that the sixt 00:02:34.400 --> 00:02:40.080 managed only half of the RTX3080s 00:02:37.519 --> 00:02:44.400 performance. To be fair though, ray tracing wasn't AMD's focus for this 00:02:42.319 --> 00:02:49.840 generation. And as much as we keep pointing it out, and it is the future, 00:02:46.800 --> 00:02:52.080 path tracers like this aren't common 00:02:49.840 --> 00:02:55.440 yet. So, turning back to more traditional rendering, Wolfenstein Young 00:02:54.080 --> 00:02:59.920 Bloodood looks like a win for the RTX3090, 00:02:57.120 --> 00:03:05.120 except that given the six-t managed numbers halfway between the 3080 and 00:03:02.400 --> 00:03:09.599 3090, the price difference makes AMD's offering look very compelling here. 00:03:07.519 --> 00:03:16.159 Microsoft Flight Simulator didn't do AMD any favors. Yeah, the SixT is a better 00:03:13.200 --> 00:03:20.560 value than the 3090, but then the 3080 makes both of them look like a ripoff, 00:03:18.319 --> 00:03:25.519 and Counterstrike is nearly a win across the board for Team Red with the RTX3090 00:03:23.519 --> 00:03:30.799 only pulling ahead in minimum frame rates. Of course, all of this changes if 00:03:28.319 --> 00:03:36.159 you're willing to accept DLSS upscaling as a viable option. And in these cases, 00:03:33.280 --> 00:03:43.440 the NVIDIA cards unquestionably dominate AMD in terms of raw FPS. Love or hate 00:03:39.840 --> 00:03:46.720 it, DLSS is here. Customers are using it 00:03:43.440 --> 00:03:48.879 and it looks pretty darn good. So AMD, 00:03:46.720 --> 00:03:54.159 you guys need to get on Fidelity FX Super Resolution and stat. In our review 00:03:51.440 --> 00:03:59.599 of the RX6800 series, we noted that AMD's productivity performance lagged 00:03:56.239 --> 00:04:01.920 behind, and it's no different here. AMD 00:03:59.599 --> 00:04:06.959 says they focused first and foremost on gaming performance, and that's fair 00:04:04.319 --> 00:04:10.959 enough, but it did cost them. It's not like it's bad at productivity if you 00:04:09.120 --> 00:04:15.760 find the right workloads. I mean, just look at how it handles the more memory 00:04:12.720 --> 00:04:18.479 intensive Specview perf subtests. But 00:04:15.760 --> 00:04:24.240 there is one glaring exception where AMD firmly falls short, and honestly, I 00:04:21.440 --> 00:04:29.120 consider it unacceptable. Radeon Pro Render. We called out AMD's poor 00:04:26.800 --> 00:04:32.960 rendering quality in the 6800 series review. And in response, they gave us a 00:04:31.280 --> 00:04:39.280 tweaked version of the tests to play around with, and okay, it's fast, but 00:04:37.280 --> 00:04:45.440 the problem is that it still looks like hot garbage. This is useful maybe as a 00:04:42.960 --> 00:04:50.560 preview, but it is not good enough for a final render. So, sorry AMD. Until you 00:04:48.560 --> 00:04:55.120 get hardware ray tracing support baked into the Blender cycles renderer like 00:04:52.479 --> 00:05:01.199 NVIDIA's optics engine, this is not even remotely a fair fight. Unlike our prices 00:04:58.320 --> 00:05:04.479 at ltstore.com, they're always fair. Oh, and we've got the ABCs of gaming in 00:05:02.960 --> 00:05:08.960 stock. Once again, smart access memory, while 00:05:06.800 --> 00:05:12.960 arguably the star of the show for gaming that pushed them into their competitive 00:05:10.479 --> 00:05:18.240 position, continues to do little or nothing for productivity. AMD reasons 00:05:15.759 --> 00:05:22.080 that while NVIDIA and Intel are both working on enabling the resizable bar 00:05:20.320 --> 00:05:26.320 feature that makes up smart access memory's core functionality, there is 00:05:24.400 --> 00:05:30.800 far more that goes into the special sauce. Specifically, they claim that 00:05:28.880 --> 00:05:35.120 significant driver optimization is required to help smart access memory 00:05:32.880 --> 00:05:39.520 reach its full potential. If that's true, then there is significantly more 00:05:37.039 --> 00:05:45.120 work ahead for team blue and team green than just toggling a switch. Or at least 00:05:42.479 --> 00:05:49.840 in theory. ASUS has apparently already enabled this functionality in their 00:05:46.960 --> 00:05:54.720 Intel Z490 motherboards, and early performance looks about as good as what 00:05:52.320 --> 00:05:57.840 we might see on an AMD platform. So, I suppose we're going to have to 00:05:55.520 --> 00:06:03.840 investigate that. Get subscribed so you don't miss our coverage. Thermally, the 00:06:00.400 --> 00:06:05.919 RX6900 XT is similar to, but on average 00:06:03.840 --> 00:06:10.560 a smidge hotter under load than the RTX3090 with the notable exception of 00:06:08.639 --> 00:06:15.199 those memory bandwidth intensive tests at the end that we pointed out in our 00:06:12.479 --> 00:06:20.800 6800 series reviews. Compared to the 6800 XT, the bigger card doesn't take as 00:06:18.240 --> 00:06:25.600 long to heat up, but they both end up in roughly the same thermal envelope by the 00:06:22.880 --> 00:06:30.080 end of the test. At 2.6 6 GHz. The sixtyn00 XT's core clocks are 00:06:27.680 --> 00:06:35.199 consistently 100 MHz higher than the 6800 XT. And like that card, these 00:06:32.880 --> 00:06:39.759 clocks appear to be quite stable. This hints that tinkerers might get great 00:06:37.440 --> 00:06:44.240 results from an undervolt, either to reduce power consumption and heat or to 00:06:42.160 --> 00:06:48.720 potentially improve thermal or power headroom for better boost behavior. As 00:06:46.240 --> 00:06:53.360 for power consumption, this is becoming a bit of a pattern this generation. On 00:06:51.199 --> 00:06:59.199 average, power consumption is usually lower than the RTX3090, 00:06:55.840 --> 00:07:02.240 but the sixtyn00 XT experiences spikes 00:06:59.199 --> 00:07:05.680 that can go much higher. Check out this 00:07:02.240 --> 00:07:07.759 462 W spike. These kinds of 00:07:05.680 --> 00:07:11.280 instantaneous loads can trip up sensitive power supplies, and we 00:07:09.599 --> 00:07:16.880 actually had to swap out our test bench's CSIC 1,000 W unit for a Corsair 00:07:14.800 --> 00:07:22.160 one to avoid tripping the overcurren protection during benchmarking. Compared 00:07:19.199 --> 00:07:26.720 to the RX6800 XT, at least, it looks to average just about 50 watts higher under 00:07:24.400 --> 00:07:30.639 load at any given point in time. This is pretty crazy considering that it's got 00:07:28.479 --> 00:07:36.319 exactly the same cooler design as the 6800 XT, although it does have a beefier 00:07:34.240 --> 00:07:40.479 16-phase power delivery setup to go along with the 14 layer PCB that they 00:07:38.479 --> 00:07:45.360 share. Of course, AMD goes out of their way to point out that they're using 00:07:42.479 --> 00:07:49.280 standard sizes for their coolers and standard dual eight pin power 00:07:47.039 --> 00:07:53.120 connectors, which sounds like it shouldn't be something to brag about, 00:07:50.639 --> 00:07:58.080 but turns out to be a sick burn on the 3090 with its behemoth size, not to 00:07:56.080 --> 00:08:05.520 mention proprietary 12pin power connector. Oh, and the sixtyn00 XT has 00:08:01.120 --> 00:08:07.680 RGB. Checkmate, NVIDIA. Before I close 00:08:05.520 --> 00:08:12.400 out, I can't talk up AMD's promise of overclocking without actually doing some 00:08:09.280 --> 00:08:13.840 overclocking, can I? Yes, I can. You'll 00:08:12.400 --> 00:08:19.840 just have to wait for us to pit the RX6900 XT against the RTX3090 in a 00:08:17.440 --> 00:08:25.599 head-to-head overclock off to see who can pull off 8K gaming the best. Stay 00:08:22.479 --> 00:08:29.680 tuned. For now though, make no mistake, 00:08:25.599 --> 00:08:32.800 the RX6900 XT is a powerful card, and at 00:08:29.680 --> 00:08:35.120 $999, it undercuts NVIDIA's best by a 00:08:32.800 --> 00:08:40.640 significant margin, even if it doesn't always beat or even meet it, and falls 00:08:38.159 --> 00:08:45.760 well short in productivity. This card looks like another classic AMD look for 00:08:43.519 --> 00:08:51.519 the gap in the market and fill it move. And given how expensive the RTX3090 is 00:08:48.800 --> 00:08:56.160 and how relatively uncommon ray tracing still is in gaming, having a 00:08:53.519 --> 00:09:02.800 ballsto-wall fast card that while not optimized for real-time ray tracing, can 00:08:58.720 --> 00:09:04.720 do it is still a pretty big deal. It 00:09:02.800 --> 00:09:08.720 would just really be nice if there were any to buy because if you've waited this 00:09:06.800 --> 00:09:14.640 long to pull the trigger, I can almost guarantee that they are gone by now. How 00:09:10.720 --> 00:09:16.399 do I know? Well, the RX6800 series was 00:09:14.640 --> 00:09:22.320 weaponsgrade unoptanium when it launched. And uh 3 weeks later, supply 00:09:20.000 --> 00:09:27.920 still hasn't caught up with demand. Then given that 6900 XTs are pretty much 00:09:25.040 --> 00:09:31.040 cherrypicked 6800 XTs, I think it's pretty safe to say that for the few of 00:09:29.760 --> 00:09:39.200 you who will be able to get your hands on it at MSRP in 2020, well, enjoy it, 00:09:35.360 --> 00:09:41.200 you lucky so. For the rest, maybe next 00:09:39.200 --> 00:09:44.080 year will be better. Thanks for watching, guys. If you 00:09:42.480 --> 00:09:50.080 enjoyed this video, go check out our review of the RX6800 series to learn a 00:09:47.279 --> 00:09:53.760 little more about the whole RDNA2 thing that makes this card tick.