Subtitles section Play video Print subtitles (upbeat music) - Sometimes you don't realize what you're missing until it's staring you in the face. But when you notice it, it's just ... So sweet, isn't it? Well, that's what we finally have here today with Nvidia, the opportunity to have our cake and eat it too, because they have finally enabled resizable bar on their desktop GeForce GPUs. Get it? Resizable bar, extra sweetness? (laughs) In theory, this is a simple driver update that will unlock extra performance on your existing GPU. Assuming you're lucky enough to own one. But what we're gonna find out is if it is truly something for nothing, or if it's just an illusion. (upbeat music) I'll give you the segue to our sponsor for nothing though, Smart Deploy can help you manage every Windows endpoint in your environment from one centralized golden image. Grab your exclusive free software worth over $800 at smartdeploy.com/linus (upbeat music) Resizable bar has been a big deal ever since AMD announced their implementation of the feature called smart access memory. If you haven't been keeping up with all the buzz, it basically allows a CPU to access all of a PCI Express device's memory at once, rather than in small, 256 megabyte chunks, as has been the case until now. Resizable bar has existed in the data center for some time now, but it couldn't find momentum on the desktop until AMD kicked off this scramble to implement it industry wide. Which kind of raises the question. Why now? Well, in the early days of PCI Express, it wasn't a problem for gamers, because game assets, like textures, were sized proportionally to the video frame buffers of that time. But with the complexity of today's games, and with modern graphics cards having as much as 24 gigabytes of onboard memory, you can easily imagine a situation where the CPU could waste a lot of cycles, getting the data it needs, if it can only address 256 megabytes at a time, ultimately eliminating this waste is where our performance boost comes from. The only problem is that in order to support it, every device in the chain needs to be capable. The CPU, the motherboard, and the PCI Express device, which, in this case is a GPU. Now we've already looked at Nvidia's preliminary implementation of resizable bar on mobile GeForce GPUs, with an MSI GE76 Raider laptop. And at the time we concluded that there was more to the sometimes impressive performance improvements that AMD was getting, than just enabling resizable bar. And we also concluded that Nvidia had some work to do before it was ready for prime time. But even with it fully cooked, it's important to know that not every game or application will benefit, and how the driver handles memory management appears to play a major role in what kind of performance uplift, or fall you can expect. That's where Nvidia's latest drivers, and the free chocolate come in. Our previous investigation showed that the impact differs depending on CPU performance. So we grabbed both, a Ryzen 5 5600X and a Ryzen 9 5900X to represent the mid-range and the high end. We'll be using a GeForce RTX 3080, and for comparison against team red, we've got a Radeon RX 6800 XT. Something to note is that because we're focused on their respective improvements, over stock performance, and not on completely re-reviewing these cards, all of the results for both sides will be in relative percentages. At 1080p, things are already really interesting. Shadow of the Tomb Raider doesn't change much, but where Radeon loses some performance in the 5% lows on Ryzen 9, GeForce gains hinting that Nvidia might be doing a better job of reducing die to die latency, on AMD's own CPU, than AMD's graphics team is. GTA V on the other hand, gives GeForce a hard time, with Radeon gaining a bit with Ryzen 5, which further hints that AMD's implementation favors single die CPU's. As for F1 2020, this is one of Nvidia's call-outs for performance uplift and it brings the first major victory for resizable bar. We ended up with slightly higher average frame times across the board for team green, which is bad, but significantly lower, 1% minimum frame times that reach toward a 10% improvement on the 5,900 X. That's good because in almost any game improving performance during challenging scenes is way more important than that when animations are already smooth. This is especially true in competitive titles when performance tends to fall in the heat of battle because of all the character models and effects that are on screen. Again, Nvidia seems to be taking better advantage of the extra course on the Ryzen 9. Forza Horizon 4 is another call-out and another win for Nvidia. Although the pattern flips here, with the Ryzen 5 pulling better improvements across the board than the Ryzen 9. This suggests that memory access patterns play a significant role in whether the die arrangement causes issues. Radeon on the other hand struggles, especially with the Ryzen 9, where minimum frame rates are as much as 6% lower then with resizable bar disabled, which brings us back to our chocolate analogy, doesn't it? It's perhaps unsurprising that there's little change on Flight Sim 2020, given how CPU-bound that game, but Invidia does manage to eek out a slight improvement with Ryzen 5, while Radeon manages a bit better frame rates on Ryzen 9. And Assassin's Creed Valhalla was the call-out title that we found improvements for in our previous video, and that continues today. Nvidia pulls a respectable six to 9% improvement across the board, but then again that pales in comparison to team red, pulling way ahead of stock with up to 18% higher performance. Given the way that both teams are going back and forth here, it seems like additional optimizations can still be made for resizable bar on both sides, which is really promising. Finally, CS GO, it's an older title that already runs well on hardware of this caliber, but we still manage to see a slight improvement in minimum frame rates, at the cost of some average FPS, which is a price that most players should be willing to pay for the reasons that I outlined before. There is no price, by the way for the new sticker packs that lttstore.com, free with every order, checkmate Nvidia. Moving on to 1440P. You might not expect things to get better here, but you'd be wrong. While GeForce stays more or less the same in Shadow of The Tomb Raider, Radeon pulls two to 4% ahead with Ryzen 5, and where GeForce loses in minimum frame rates for no gain on Ryzen 9, Radeon manages to gain about as many average frames as it loses in minimums, which in a non-competitive title like this one might be an okay trade-off. GTA V is a bit all over the place, with Ryzen 5, more or less on par with stock performance on both teams, with a slight tip to Radeon, and Ryzen 9 pulling as much of a win on GeForce as Radeon loses. It's almost like Nvidia stole the frames right out Elisa Sue's pocket. And, F1 2020, again, shows improvements as high as 13% with the only maybe regression coming from Nvidia's average frame rates on Ryzen 9, but we're talking a mere 1% difference, Nvidia comes out ahead in performance improvements for minimum frame rates here, while Radeon pulls better averages. As for Forza, there is no other way to put it, it is a staggering win for Radeon, their worst improvement here is 13%, and yes, your eyes are working fine. That is a nearly 20% gain for Radeon's minimum frame rates. Now, to be clear, Nvidia still wins. Everybody wins when they go over 100% of stock performance, they just don't win as hard. Now with a little extra emphasis on the GPU, thanks to our higher resolution, Flight Sim 2020 shows us near universal improvements, with Radeon actually pulling a healthy 8% improvement in minimum frame rates on Ryzen 9. And then Assassin's Creed is once again a bigger win for AMD than it is for Nvidia, with improvements reaching 16% for both CPUs, and Nvidia managing a very respectable six to 9% lead overstock performance. Isn't it nice when everybody wins, guys? Too bad it doesn't always work out that way. CS GO at 1440p resulted in straight up performance regressions across the board with the sole exception being minimum frame rates on Ryzen 9 for Radeon. GeForce takes the worst of it though, here, with up to a 20% performance loss with resizable bar enabled. That's as big a loss as the largest gains we've seen today. Maybe that's why Nvidia's driver has an easy to find indicator for whether the feature is active. I guess AMD felt it was less likely that users with it enabled would be troubleshooting performance problems. Looking at the big picture here, though. Radeon has the best average gains, running roughly 6% faster than stock at 1440p, and 2% faster at 1080p. While Nvidia's gains hover around 2%, regardless of the resolution. That means one of two things. Either AMD's memory management for Radeon is worse than Nvidia's for GeForce, or Nvidia has some more optimization work to do. The work they've done already since our previous video has clearly paid off handsomely though, so either way it goes, we'll hopefully keep on seeing improvements with future driver revisions on both sides. On that subject, by the way, get subscribed because Nvidia finally rolled a driver that fixes the code 43 error, and we are gonna be all over that in a future video. Changing gears a bit, by now we know that productivity tends not to be affected much by resizable bar, at least for AMD. But for fun, we did some runs so that we could validate that for Nvidia, as well. Blender, well, it's 100%, yup. 100% of stock performance. So how about that? PugetBench shows some dips, especially in AfterEffects, although which CPU causes the dip changes depending on which team GPU you're batting for, and then LuxMark shows basically flat with our Ryzen 9 GeForce combo managing to lose some performance. This clearly isn't the norm, but it's worth taking into account that some applications may suffer for those sweet, sweet gaming gains.