NVIDIA GeForce RTX 2080 & 2080 Ti 4K & Ultrawide Gaming Performance

[PAGETITLE=A Look At GeForce RTX & Our Test Suite][Top Ad] The moment PC performance seekers have been waiting for (this month) has arrived. NVIDIA has lifted the embargo on its GeForce RTX GPUs, so we’re finally able to share what many of you have been earnestly waiting for in recent months: benchmark results. I had travel overlap precious RTX time, so as of right now, the only testing I've been able to conduct is straight-forward apples-to-apples performance. That means I haven't yet fiddled with things like DLSS or the new NVIDIA Scanner auto-overclocking tool. That, along with light overclocking, will come later. I'd usually try to fill this page with a look at the architecture and new features, but Jamie did that legwork for me last week. I'd highly recommend digging into that article if you want a detailed overview of what NVIDIA's new Turing architecture brings to the game. Mentioned a moment ago, DLSS is one of the highlighted new features of this launch. The initialism stands for deep learning super-sampling, and as you may have guessed, the feature uses deep learning to accelerate anti-aliasing. The end goal with DLSS is to deliver high-quality anti-aliasing at a reduced performance cost against comparative AA modes. As for NVIDIA Scanner, it's an automated overclocking engine that's actually worth giving a shot. Wtih it, NVIDIA spawns threads that look for specific bit failures to determine whether or not a given overclock is stable. As it runs, the GPU clock will be gradually increased until failures can be seen. If you manually adjust the voltage and temperature threshold settings, this scanner will reach different conclusions. NVIDIA Scanner automated overclocking tool When a run is completed, the overclock the scanner settles on should be guaranteed stable, and shouldn't leave any notable headroom left over. One way to look at the scanner is this: if you were to spend an hour figuring out a quick and dirty stable overclock, this tool could do it quicker, and without your help. One limitation to this scanner is that the memory is left alone, so if gains could be seen overclocking that, you will have to go the manual route. Overclocks higher than what the scanner decides on can surely be seen as well should you decide to begin cranking voltages and other secondary dials. If you have questions that don't get answered in this article, please leave a comment. A Look At The Hardware. Since its Pascal GeForce launch, NVIDIA has offered 'Founders Editions' of its high-end graphics cards which have been kitted out with great looking industrial-inspired coolers. Unfortunately, looks are not everything: these same coolers have been found to throttle clocks due to the high temperatures, an issue not seen on other readily available options (or the same cards if you crank the fan speed). It's refreshing, then, that the Turing architecture has ushered in a brand-new Founders Edition card design. The card's overhauled cooler is wrapped in machine-finished aluminum, which makes for the most physically stable graphics card you've likely ever laid your hands on. If you manage to bend this card with your own hands, or by throwing it off of a building (don't do that), please provide proof. If you're already familiar with NVIDIA's Founders Editions, then you know what's seen here is a massive improvement over the previous generation. We move from a less than ideal blower-style fan to dual axial fans equipped with 13 blades each. This card shouldn't be starved of airflow, and at the same time, it shouldn't be audibly obvious even when the going gets tough. I haven't been able to put proper testing into thermals or noise yet (it's been so humid this week!), but I will evaluate it more later. First impressions tell me that this FE design will do well to negate the clock throttling issues of the previous gen, where some cards could lose 100MHz at the top-end. In a quick test, I hit 80°C on the 2080 Ti during Epic's Infiltrator demo, but I didn't break 70°C during Far Cry 5 gameplay. At the back of the new FE, some other changes can be seen with a matte black finish replacing the typical glossy silver, as well as a first for gaming cards: a USB Type-C port for VirtualLink (for use with VR headsets). Fortunately, the addition of the USB Type-C port didn't impact the number of total video out ports, as there remain 3x DisplayPorts and 1x HDMI. Test System & Methodology. Historically, I've always used Intel's current-gen enthusiast platform (eg: Core X-series) as the base for Techgage's gaming GPU test PC, but I changed things up for this launch. After having conversations about it here and there over the past couple of months, it's felt increasingly safe to say that clocks matter more than core counts for most gamers. The ultimate reason I've chosen enthusiast platforms for game testing in the past was because more PCIe lanes were made available. The reality, though, is that single-GPU testing is going to dwarf the amount of multi-GPU testing I'll want to do. Beyond that, it doesn't appear that Intel's Z370 reduced PCIe lane count impacts dual-GPU performance to any significant degree anyway. A platform comparison in the future could be fun to dive into, since it's something Techgage has never tackled. For my purposes, which is to have a PC that simply lets me benchmark a graphics card reliably, the Intel Z370 platform isn't going to hold anything back. As a higher-end enthusiast who wishes to pile on NVMe drives or other PCIe-bound peripherals as well as a second GPU, a beefier platform would be warranted. Techgage Gaming GPU Test PC Processor Intel Core i7-8700K (3.7GHz, 6C/12T) Motherboard EVGA Z370 FTW CPU tested with BIOS 1.09 (August 20, 2018) Memory G.SKILL TridentZ (F4-3400C16-8GSXW) 8GB x 2 Operates at DDR4-3400 16-16-16 (1.35V) AMD Graphics AMD Radeon RX Vega 64 (8GB; Radeon 18.8.2) AMD Radeon RX Vega 56 (8GB; Radeon 18.8.2) AMD Radeon RX 580 (8GB; Radeon 18.8.2) AMD Radeon RX 570 (4GB; Radeon 18.8.2) AMD Radeon RX 550 (2GB; Radeon 18.8.2) NVIDIA Graphics NVIDIA GeForce RTX 2080 Ti (11GB; GeForce 411.51) NVIDIA GeForce RTX 2080 (8GB; GeForce 411.51) NVIDIA TITAN Xp (12GB; GeForce 399.07) NVIDIA GeForce GTX 1080 Ti (11GB; GeForce 399.07) NVIDIA GeForce GTX 1080 (8GB; GeForce 399.07) NVIDIA GeForce GTX 1070 Ti (8GB; GeForce 399.07) NVIDIA GeForce GTX 1070 (8GB; GeForce 399.07) NVIDIA GeForce GTX 1060 (6GB; GeForce 399.07) NVIDIA GeForce GTX 1050 Ti (4GB; GeForce 399.07) NVIDIA GeForce GTX 1050 (2GB; GeForce 399.07) Storage Kingston SSDNow V310 960GB (SATA 6Gbps) Power Supply Corsair RM650x (650W) Chassis NZXT S340 Elite Mid-tower Cooling Corsair Hydro H100i V2 AIO Liquid Cooler (240mm) Et cetera Windows 10 Pro (64-bit; build 17134) Games Tested & Vendor Neutrality. A total of eight games are included in our current test suite. Some have appeared here before, while others make their first appearance: Monster World Hunter, Fortnite, and F1 2018. I had planned to include PlayerUnknown's Battlegrounds as a ninth title, but the results were too sporadic to inspire any sort of confidence (an issue not seen in Fortnite, by comparison). Here's the full list of tested games and developer allegiances, as well as synthetic tests also used: Battlefield 1 Deus Ex: Mankind Divided - AMD partner F1 2018 Far Cry 5 - AMD partner Fortnite Monster Hunter World Rise of the Tomb Raider - NVIDIA partner Tom Clancy's Ghost Recon Wildlands - NVIDIA partner UL 3DMark & VRMark Unigine Superposition For our apples-to-apples testing, the graphics settings seen above apply to every one of our tested resolutions so as to deliver standard apples-to-apples results. In most cases, each configuration is tested twice, with more runs added if the initial results make the extra testing necessary. Fortnite is the only game tested three times by default due to its variable nature. Shadow of the Tomb Raider will replace Rise of the Tomb Raider in our suite as soon as its RTX enhancements come along. Similarly, those same RTX enhancements can't be taken advantage of until Microsoft releases its DXR API to the wild, something expected to happen with the Redstone 5 fall update. For now, we're largely stuck to traditional testing. Speaking of, let's get on with it. [PAGEBREAK][PAGETITLE=Battlefield 1, Deus Ex: Mankind Divided, F1 2018 & Far Cry 5] Battlefield 1. As it's tested in our suite, Battlefield 1 isn't as intensive as I'd like it to be. The best way to stress any game in the game's series would be to hop online, but with varying match settings and net latencies, it's not a reliable way to test. My wish is that DICE will eventually include a built-in benchmark that's focused on large-scale battles, since it's with online play when the going gets tough. From a single player perspective, any higher-end card of this or the previous generation will handle the game at 2560x1440 with ease. Even the GTX 1080 Ti peaked at 127 FPS on average, and it along with its higher-end brethren delivered 100+ FPS at the ultrawide resolution of 3440x1440. At 3840x2160, the RTX 2080 Ti delivers almost 100 FPS on average, which should be a welcomed sight to anyone who's considering the purchase of a high-refresh 4K monitor. Changing to the RTX 2080, it manages to place ahead ever so slightly of the GTX 1080 Ti at all three resolutions. Deus Ex: Mankind Divided. Most of the games in this test suite have been benchmarked with "near max" graphics settings, but Mankind Divided is different. Applying a blanket 'High' or 'Very High' profile with this game can quickly lead to crippling performance, so this is one title where the graphics settings have been massaged to deliver as much eye candy as possible while still retaining realistic frame rates. With those as-tested settings, NVIDIA's newest top-end GPU delivered 60 FPS on average at 4K resolution. Put another way, the 2080 Ti delivers a 27% boost in performance over the TITAN Xp. At the ultrawide resolution of 3440x1440...

NVIDIA GeForce RTX 2080 & 2080 Ti 4K & Ultrawide Gaming Performance

Page 1264 –

Rob Williams