This episode of Ask GN focuses on some more technical topics – and ones which we were happy to address in more episodic fashion. The first question asked us to address why a framerate in excess of the refresh rate appears to allow smoother gameplay. It's a good question, too; it'd seem like the excess doesn't actually add anything, if just thinking about a 60Hz target, but there's a lot more to it than that.
This topic leads to discussion on monitor overclocking and VRAM measurements and consumption. We've previously covered monitor overclocking, for the curious, and we've also previously talked about how GPU-Z doesn't accurately measure VRAM consumption. It's more a measurement of VRAM requested by the game, and because some games will just ask for all the memory, it's hard to know how much is actually utilized (rather than reserved). We talk about that more in the content below:
It’s a great weekend for buying graphics cards, as AMD is dropping prices in preparation for the release of the GTX 1050. Today is the day those price drops take effect, although third-party manufacturers may drag their feet a bit. Some AIB partners are already offering discounts on the RX 460 and RX 470 cards, especially Sapphire.
SAPPHIRE Radeon RX 460 2GB ($100): MSRP for the RX 460 is now $100, and Sapphire Tech has wasted no time dropping their card’s price to match. This sale only lasts a few more hours, but it’s a safe bet that the price become permanent in the near future, making this a safe buy for budget builds. We reviewed the RX 460 here.
Part 1 of our interview with AMD's RTG SVP & Chief Architect went live earlier this week, where Raja Koduri talked about shader intrinsic functions that eliminate abstraction layers between hardware and software. In this second and final part of our discussion, we continue on the subject of hardware advancements and limitations of Moore's law, the burden on software to optimize performance to meet hardware capabilities, and GPUOpen.
The conversation started with GPUOpen and new, low-level APIs – DirectX 12 and Vulkan, mainly – which were a key point of discussion during our recent Battlefield 1 benchmark. Koduri emphasized that these low-overhead APIs kick-started an internal effort to open the black box that is the GPU, and begin the process of removing “black magic” (read: abstraction layers) from the game-to-GPU pipeline. The effort was spearheaded by Mantle, now subsumed by Vulkan, and has continued through GPUOpen.
The goal of this content is to show that HBAO and SSAO have negligible performance impact on Battlefield 1 performance when choosing between the two. This benchmark arose following our Battlefield 1 GPU performance analysis, which demonstrated consistent frametimes and frame delivery on both AMD and nVidia devices when using DirectX 11. Two of our YouTube commenters asked if HBAO would create a performance swing that would favor nVidia over AMD and, although we've discussed this topic with several games in the past, we decided to revisit for Battlefield 1. This time, we'll also spend a bit of time defining what ambient occlusion actually is, how screen-space occlusion relies on information strictly within the z-buffer, and then look at performance cost of HBAO in BF1.
We'd also recommend our previous graphics technology deep-dive, for folks who want a more technical explanation of what's going on for various AO technologies. Portions of this new article exist in the deep-dive.
AMD sent us an email today that indicated a price reduction for the new-ish RX 460 2GB card and RX 470 4GB card, which we've reviewed here (RX 460) and here (RX 470). The company's price reduction comes in the face of the GTX 1050 and GTX 1050 Ti release, scheduled for October 25 for the 1050 Ti, and 2-3 weeks later for the GTX 1050. Our reviews will be live next week.
Battlefield 1 marks the arrival of another title with DirectX 12 support – sort of. The game still supports DirectX 11, and thus Windows 7 and 8, but makes efforts to shift Dice and EA toward the new world of low-level APIs. This move comes at a bit of a cost, though; our testing of Battlefield 1 has uncovered some frametime variance issues on both nVidia and AMD devices, resolvable by reverting to DirectX 11. We'll explore that in this content.
In today's Battlefield 1 benchmark, we're strictly looking at GPU performance using DirectX 12 and DirectX 11, including the recent RX 400 series, GTX 10 series, GTX 9 series, and RX 300 series GPUs. Video cards tested include the RX 480, RX 470, RX 460, 390X, and Fury X from AMD and the GTX 1080, 1070, 1060, 970, and 960 from nVidia. We've got a couple others in there, too. We may separately look at CPU performance, but not today.
This BF1 benchmark bears with it extensive testing methodology, as always, and that's been fully detailed within the methodology section below. Please be sure that you check this section for any questions as to drivers, test tools, measurement methodology, or GPU choices. Note also that, as with all Origin titles, we were limited to five device changes per game code per day (24 hours). We've got three codes, so that allowed us up to 15 total device tests within our test period.
The Nintendo “Switch” was announced this morning, the next-generation half-portable, half-docked console. To reduce confusion, the Switch was previously referred to as the Nintendo “NX.” It is the same device.
Nintendo's new Switch is built in partnership with nVidia and leverages the Pascal architecture found in current-generation GTX 10-series GPUs. At least, based on this text from nVidia's blog: "[...] NVIDIA GPU based on the same architecture as the world’s top-performing GeForce gaming graphics cards." Tegra SOCs include ARM processors alongside the nVidia graphics solution, and also host all of the I/O lanes and memory interfaces. This is a complete system, as indicated by “system on chip.” We've asked nVidia for details on which ARM devices are used and which memory will be supported, but were told that the company is not revealing further details on Nintendo's product. We are awaiting comment from Nintendo for more information.
We do know that the Tegra SOC is accelerating gameplay with hardware-acceleration for video playback, and that nVidia and Nintendo have deployed “custom software for audio effects and rendering.” We can confidently speculate that the Switch is not functioning as the previous Shield devices have (read: not streaming to handheld from a dock), mostly because the Switch is large enough to contain all necessary render hardware within its handheld state. The Switch is also shown in the advert to be playable on planes, which most certainly do not have fast enough internet to support up/down game streaming. This is processing and rendering locally.
Retail powerhouse Newegg is allegedly now majority-owned by Hangzhao Liaison Interactive Information Technology Co., Ltd, heretofore known as Over-compensator, Inc. UDN reports that the company with the big name now holds nearly 56% of Newegg through a $2.63B USD investment. This would allow China-based Liaison Interactive to claim hardware/software retail domination of US and Chinese markets. Granted, Google Translate does say that Newegg is the second-largest “US egg supplier in China,” so we can't interpret much beyond the alleged major investment made by Liaison.
We've already extensively looked at the GTX 1060 3GB vs. GTX 1060 6GB buying options, we covered the RX 480 4GB vs. 8GB options, but we haven't yet tested the 3GB & 4GB SKUs head-to-head. In this content, we're using the latest drivers to specifically benchmark the GTX 1060 3GB versus the RX 480 4GB cards to determine which has the best framerate for the price.
Each of the lower VRAM spec SKUs has a few other tweaks in addition to its memory capacity reduction. The GTX 1060 3GB, for instance, also eliminates one of its SMs. In turn, that kills 128 CUDA cores and 8 TMUs, dragging the 1060 down from 1280 cores / 80 TMUs to 1152 cores / 72 TMUs on the GTX 1060 3GB model. AMD's RX 480 4GB card, meanwhile, has a lower minimum specification for memory to assist in cost management. The RX 480 4GB has a minimum memory speed of ~1750MHz (or ~7Gbps effective), whereas the RX 480 8GB model runs 2000MHz (8Gbps effective).
NZXT's new Kraken X42, X52, and X62 liquid coolers were announced today, all using the new Asetek Gen5 pump with substantial custom modifications. The most direct Gen5 competition would be from Corsair, makers of the H115i and H100iV2, each priced to compete with the Kraken X42 ($130) and X52. The Corsair units, however, are using an unmodified Asetek platform from top-to-bottom, aside from a couple of Corsair fans. NZXT's newest endeavor had its components dictated by NZXT, including a custom (and fairly complex) PCB for fan speed, pump speed, and RGB control, planted under a custom pump plate with infinity mirror finish. The unit has gone so far as to demand a double-elbow barb for pose-able tubes, rather than the out-the-top setup of the Asetek stock platform – that's some fastidious design.
As for how we know all of this, it's because we've already disassembled a unit. We decided to dismantle one of our test-complete models to learn about its internals, since we're still waiting for the X52 and X62 models to be review-ready. We've got a few more tests to run.
Before getting to the tear-down, let's run through the specs, price, and availability of NZXT's new Kraken X42, X52, and X62 closed-loop liquid coolers.