Posted on Monday, June 18 2018 @ 17:32:41 CEST by Thomas De Maesschalck
Another interesting deep learning research project from NVIDIA, the firm developed a system that can turn regular 30fps video into slow motion clips of 240fps or 480fps. To achieve this feat, the NVIDIA researchers used NVIDIA Tesla V100 GPUs and a cuDNN-accelerated PyTorch deep learning framework, to analyze over 11,000 videos of everyday and sport activities shot at 240 frames per second. Once trained, the convolutional neural network was able to predict the extra frames required to turn regular video into super slow motion.
“There are many memorable moments in your life that you might want to record with a camera in slow-motion because they are hard to see clearly with your eyes: the first time a baby walks, a difficult skateboard trick, a dog catching a ball,” the researchers wrote in the research paper. “While it is possible to take 240-frame-per-second videos with a cell phone, recording everything at high frame rates is impractical, as it requires large memories and is power-intensive for mobile devices,” the team explained.
“Our method can generate multiple intermediate frames that are spatially and temporally coherent,” the researchers said. “Our multi-frame approach consistently outperforms state-of-the-art single frame methods.”
You can check out a video of the work below. NVIDIA does not release the actual code.
Posted on Monday, June 18 2018 @ 11:46:20 CEST by Thomas De Maesschalck
With the arrival of the ASUS PG27UQ and Acer Predator X27, the market finally saw the launch of the first 4K HDR 144Hz G-SYNC screens. All good and well, except that these screens don't offer a true 4K 144Hz experience.
Various reviews of these new panels indicate there's a noticeable drop in image quality when using the panels at 144Hz, whereas 120Hz and below look pixel perfect. TechPowerUp explains this issue is caused by the fact that DisplayPort 1.4 has a maximum bandwidth of just 26Gbps. That's just enough for 4K at 120Hz but to achieve 144Hz the display makers have to resort to 4:2:2 chroma subsampling (YCbCr). Basically, the grayscale portion of the image is transmitted at full resolution (3840 x 2160), while the color information is transmitted at half the horizontal resolution (1920 x 2160).
This is a standard industry solution in the movie industry, but it doesn't work well for text and operating system interfaces. As TechPowerUp points out, the best solution for now is to either run these screens at 4K 120Hz, or to avoid being an early adopter, and wait for future 4K 144Hz screens with HDMI 2.1 support.
Alternatively, there's also DSC (Display Stream Compression) for DisplayPort 1.4, but that feature isn't completely lossless and it's not supported by a lot of display controllers at this moment. Either way, it's no fun paying top-notch dollars for a panel that pushes out sloppy image quality at the marketed use case.
Posted on Monday, June 18 2018 @ 11:24:20 CEST by Thomas De Maesschalck
So here's a weird snippet of news. AnandTech heard at Computex that due to BIOS limitations, various motherboard makers are considering to drop AMD Bristol Ridge APU support from upcoming Socket AM4 motherboards. Apparently, the motherboard makers see no other choice than dropping support for these pre-Zen processors on some models, or doubling the memory size of the BIOS chip from the standard 128Mb to 256Mb.
UEFI takes up a lot more space than the old text-based BIOS and it appears motherboard makers are running out of space on Socket AM4 motherboards. Due to the fact that Bristol Ridge APUs weren't widely distributed, and they're nearly up to two years old, an easy solution for motherboard makers seems to drop support. The reason why it's not possible to keep Bristol Ridge support isn't known, it could be because the base code of the platform or the microcode per CPU is too large, but there are no specifics.
The simple solution would be to use 256Mb BIOS chips, but AnandTech reports motherboard makers aren't keen to do this because of inventory, purchase agreements, and the higher cost.
The simple way to support all the processors is to increase the size of the BIOS chip, by moving from 128 Mb to 256 Mb. At least two vendors told us that the price of these larger chips is more than double the 128 Mb chips. Not only this, but they have so many 128 Mb chips in stock already (or purchase agreements), it would not be economically feasible to switch and dump. One company, Biostar, said that they were implementing motherboards with two BIOS chips in a striped configuration, giving an effective capacity of 256 Mb, in order to solve this issue. The only downside here is that the system is using two BIOS chips, effectively doubling the cost to implement a BIOS on the motherboard. For low end systems, this is an additional cost that might push the product out of a competitive price range.
Another infeasible solution is to bifurcate BIOS releases, but that results in the risk of a customer flashing the wrong BIOS version. Or perhaps motherboard makers could drop useless stuff from their BIOS...
Posted on Monday, June 18 2018 @ 09:23:02 CEST by Thomas De Maesschalck
Bleeping Computer reports various game developers have promised to remove RedShell, an analytics package created by Innervate. This software development kit (SDK) is embedded in a lot of new games, it's an analytics tool that sends data to publishers about the source of a new game install. However, Reddit and Steam users discovered the app also creates fingerprints for each gamer, with details about their gaming rigs and online habits.
To aggravate matters, most game publishers provided no warning about this, and do not offer an opt-out. A list of all games that use RedShell can be found at Reddit. Here's a look at the games that will soon remove RedShell:
Games who used Redshell which removed or pledged to remove it (as of 16.06.2018):
Elder Scrolls Online (Pledged to remove it)
Conan Exiles (Pledged to remove it)
Ylands (Pledged to remove it)
Holy Potatoes! We’re in Space?! (Pledged to remove it)
All Total War games, (Pledged to remove it)
Warhammer: Vermintide II, (Pledged to remove it)
Warhammer: Vermintide I, (might get removed also?)
My Time At Portia, (Pledged to remove it)
Dead by Daylight, (Pledged to remove it)
Battlerite, (Pledged to remove it)
AER Memories of Old, (Pledged to remove it)
Magic the Gathering Arena (closed beta & not on Steam), (Pledged to remove it for now)
Secret World Legends (Pledged to remove remains of it)
Posted on Monday, June 18 2018 @ 09:15:31 CEST by Thomas De Maesschalck
Over at the OCUK forums, an AMD employee confirmed that Raven Ridge will receive driver updates every 3 months.
Basically, AMD's APUs are updated every 3 months as WHQL releases only, whereas the firm's discrete graphics card often receive multiple driver updates per month. Obviously, this is not ideal for gamers as this leaves a lot of room between a game's release date and the time AMD provides performance optimizations and/or bug fixes.
This also leads to a somewhat funny situation. A couple of days ago, Intel rolled out driver version 18.6.1 for its Kaby Lake-G processor with the Radeon RX Vega M integrated graphics. Meanwhile, AMD's own APUs are still stuck with 18.5.1.
Posted on Sunday, June 17 2018 @ 20:25:37 CEST by Thomas De Maesschalck
News is spreading that after being for sale for roughly a year, venture capitalist Tallwood has finally found a buyer for MIPS. The company will be sold to Wave Computing, an AI startup that will use MIPS' technology to expand its AI offerings.
For their part, Wave Computing has stated that they intend to use MIPS technology to expand their machine learning product portfolio. Specifically, the company is looking to branch out from datacenter ML training to ML inference in edge devices, where MIPS’ existing products and their own work into inference products would be a natural counterpart to Wave Computing’s existing technology. Ultimately this would allow Wave Computing to offer a singular top-to-bottom machine learning product portfolio, with common technology ranging from datacenter training to the actual use of machine learning in edge and other end-user devices, giving them a leg up over vendors who can only supply one or the other.
Posted on Sunday, June 17 2018 @ 20:19:40 CEST by Thomas De Maesschalck
Thurrott picked up word that Microsoft is working on a next-gen Xbox console with an expected launch date in 2020. The new generation is codenamed "Scarlett" and it's described as a family of devices. So basically it looks like we can expect multiple Xbox products in 2020.
That date may seem relatively close considering that the Xbox One X launched last year but Microsoft is aggressively moving ahead with hardware in the console space to shake up the industry. But what we don’t know is if this is a brand new set of devices or simply a more ‘powerful’ Xbox One that is again, fully backward compatible with all the content available today.
Considering how much effort Microsoft has put into backward compatibility, I would expect the Scarlett devices to support current gen games but offer more advanced features and align possibly with Microsoft’s upcoming game-streaming services.
Posted on Sunday, June 17 2018 @ 20:17:17 CEST by Thomas De Maesschalck
Word is going around that AMD is planning the AMD Ryzen 5 2500X and Ryzen 3 2300X processors.
The Ryzen 5 2500X is a quad-core model with eight threads, it has a 3.6GHz base clock, 4.0GHz Turbo, 16MB L3 cache, and an expected price range of $150-$160. Leaked Geekbench scores reveal performance of 4782 points (single-core) and 17291 points (multi-core), putting it pretty close to Intel's Core i7-7700.
Next we have the Ryzen 3 2300X, this is a quad-core, quad-threaded chip with a 3.5Ghz base frequency, 4.0GHz Turbo, and an expected price level of $120 to $140. This one is close in performance to Intel's Core i5-7600K.
Availability is expected soon, but we don't know an exact date.
Posted on Friday, June 15 2018 @ 14:15:08 CEST by Thomas De Maesschalck
Not a lot of details here, but FUD Zilla managed to confirm that Xilinx helped AMD with HBM2. The site says both companies have quietly been cooperating for years on next-gen memory interfaces, as both firms realized they were after the same goal.
Increasing the memory bandwidth is the key for both computer graphics that AMD is perusing and for deep learning, high quality video processing and AI that both companies are now after. Stacking memory chips to increase the bandwidth was no walk in the park and both companies worked together to fix the obstacles.
Interestingly, the current CEO of Xilinx, Victor Peng, used to be a VP of silicon engineering at ATI and later AMD.
Posted on Friday, June 15 2018 @ 14:11:08 CEST by Thomas De Maesschalck
In a new blog post, Microsoft reveals it used artificial intelligence in an attempt to roll out Windows 10 April 2018 Update more smoothly than previous Windows 10 feature updates. Basically, the software giant trained an AI agent to collect update experience data, to learn which devices will have a good update experience, and to spot where problems may arise.
A pilot program was tried during the Windows 10 Fall Creators Update rollout but this was the first time Microsoft performed a large-scale test. The software giant says it's pleased with the results:
Our AI/Machine Learning approach started with a pilot program during the Windows 10 Fall Creators Update rollout. We studied characteristics of devices that data indicated had a great update experience and trained our model to spot and target those devices. In our limited trial during the Fall Creators Update rollout, we consistently saw a higher rate of positive update experiences for devices identified using the AI model, with fewer rollbacks, uninstalls, reliability issues, and negative user feedback. For the April 2018 Update rollout, we substantially expanded the scale of AI by developing a robust AI machine learning model to teach the system how to identify the best target devices based on our extensive listening systems.