Discussion in 'Videocards - AMD Radeon' started by OnnA, Jul 9, 2016.
Like everyone, who has Fury or Polaris and need some punch for Ultra 1440p Gaming ~70-120Hz
TSMC, is about to revolutionise graphics cards for both ATI/AMD (and Nvidia) GPU manufactures!
TSMC, the manufacturer of GPUs for both Nvidia and future AMD graphics cards, have announced a new wafer-stacking technology which could allow both companies to create massively more powerful graphics cards without increasing the size or density of their silicon.
The TSMC Technology Symposium is going on right now out in Santa Clara, and they’ve unveiled a new technology, called Wafer on Wafer (WoW) which allows them to connect chips together without mounting them laterally on the same interposer,
or using the likes of AMD’s Infinity Fabric or Intel’s EMIB designs to allow them to communicate.
The new wafer-stacking tech means chips can be piled on top of each other, much like the way NAND flash memory is being stacked in the latest solid state drives.
The individual wafers are directly connected using through-silicon vias (TSVs) and will not only allow many more cores to be contained in a single package, but also allow them to communicate incredibly quickly too.
There is also the potential that these WoW designs can be jammed onto an interposer with other silicon to create proper monster chips. Or even have more than two wafers stacked on top of each other. Skyscraper GPUs, please!
There are inevitable drawbacks to this sort of manufacturing, and most of that is down to the wafer yields TSMC are getting from different technologies.
With two wafers stacked on top of each other, if one of the chips comes out broken the connected chip is therefore useless, which would make wafer-stacking too costly on low-yield products.
That means it’s more likely to be used on established production nodes with high manufacturing yields. That would cut down the wastage and still produce chips with far more power than their forebears. Again, this is how it works in SSD land too.
Intel have also filed patents for stacked CPU technologies to allow them to combine older manufacturing processes with newer ones to create more advanced chips.
Given the high yield of something like TSMC’s mature 16nm lithography it might not be beyond the realm of possibility for an old Pascal era chip to be 3D stacked, instantly doubling the core-count of the newborn GPU. Tasty.
That would give multi-chip graphics cards a real boost, especially as they wouldn't actually need to appear as multi-GPU to the system.
Okay, right now that’s a little far-fetched, and TSMC aren’t going to divulge any further details about the new WoW tech until they’re good and ready.
But it does mean that TSMC, the biggest contract chip manufacturers in the world, will be able to offer their partners the capability to extend the lifetimes of their architectures on high-yield production nodes and deliver powerful silicon without struggling to stuff more cores into the same 2D space.
And that could be very exciting for future GPU refreshes. At the moment they’re little more than rebadges - think RX 480 to RX 580 - but if AMD had the option to stack two Polaris 10 chips on top of each other that would have been a genuinely exciting refresh.
Interesting enough. Idea is to separate those High Power Draw Cores and put them on top. And have un-Core (all low power IO/SoC/...) on lower layer.
Here come the bad side. On large GPU like Fiji/Vega64, Those power hungry (n)CUs take 80% of die area. So, your "Core" is 4 times as large than your "un-Core". Is GPU manufacturer supposed to add some additional features to lower die as it is wasting a lot of space? Maybe very large cache or massive die to die IO for dual-GPU cards? Hard to say how one should use this wasted space. But just fine for smaller GPUs like Polaris10.
But on Bright Side... CPUs. Ryzen's Core to un-Core area is almost perfect for this. That means Ryzen which is already quite cost efficient to make due to multi-die-CCX-IF design will cut area per chip on wafer to 1/2.
Similar thing applies to APUs, but since Vega iGPU is larger than CPU-CCX, un-Core area is now like 40% and CPU + GPU 60%. Not bad actually, there APU may have some L3 cache. And since iGPU is separated from CPU by un-Core, this can be separate CPU and iGPU die sitting on top of un-Core.
And apparently this require quite some redesign as chips are currently flat thing with I/O at edges and Cores in mid. It may be like game of paper folding.
This looks really interesting. This looks to be the future of the GPU. They had to come up with something like this or a new process altogether. You can only shrink so far and at 7nm, there isn't a Hell of a lot left. I'm quite excited by this news and am really intrigued.
In 5~10 years some of those other technologies for 'transistors' may already be ready for production. Some are meant to allow higher density / lower power consumption. Some may reach insane clocks in comparison to what we have.
Basically, cellphones in 10 years may have compute power equal to desktops today. All it takes is one success in one lab.
But still, most exciting technology for this is HBM as you can put HBM under Core die next to SoC die. Or some other tricks making I/O more better (good bye interposer).
New Tek? (THX to Tech PWR Up)
Machine learning version in 2018
Die-shrink to 7 nm. Tape-out in 2018
Release Date: 2019, probably H2
TSMC, 7 nm
GDDR6 Graphics Memory
Release Date: Early 2018 (chips), late 2018 (products using these chips)
Up to 16 Gbps speeds (2x GDDR5)
First products will launch with 12 Gbps and 14 Gbps
Voltage: 1.35 V (same as GDDR5X)
AMD is working on supporting GDDR6 for future graphics cards
HBM3 Graphics Memory
Release Date: Not before 2019
Double the memory bandwidth per stack (4000 Gbps expected)
Expected to be produced using 7 nm technologies
Specification released in late 2017
16 GT/s bandwidth per lane, per direction (2x the bandwidth of PCIe 3.0)
I/O virtualization capabilities
No chipsets/motherboards in sight
Release Date: Q1 2019
32 GT/s bandwidth per lane, per direction (4x the bandwidth of PCIe 3.0)
128/130 bit encoding (= 1.5% overhead)
Physical connector targeted to be backward compatible
AMD adds Netflix 4K acceleration in latest driver release!
AMD released its Radeon Software Adrenalin Edition 18.4.1 drivers a few days ago. In the release notes there was one stated significant addition to the drivers: “Initial support for Windows 10 April 2018 Update”.
Just that, a few fixed issues, and a list of known issues AMD is currently working on. However, it has since been discovered that AMD slipped in an extra feature that may be of great appeal to PC users,
especially those owning / making HTPC machines – support for Microsoft's PlayReady 3.0 DRM.
The importance of the above is that Microsoft's PlayReady 3.0 is one of the system requirements for Netflix 4K playback on a PC.
Hardware.info noticed (via HardOCP) the AMD Netflix 4K decode ability had been added when perusing a reviewer’s guide document for Raven Ridge APUs.
As with the Intel and Nvidia alternative routes, some other conditions must be met for Netflix 4K playback on the PC.
Users need the PlayReady 3.0 compatible hardware plus; the Microsoft Edge browser, a connection to the monitor via the hdcp 2.2 protocol, an existing h265 decoder, and a Netflix Premium subscription.
In the Dutch source’s own testing it was noted that Netflix 4K video streaming came with a considerably higher bit rate than lower resolutions, which is understandable.
Other than the demands on its internet connection, Hardware.info had no other issues watching Netflix 4k via a Radeon RX580 video card using the latest Adrenalin driver.
THX to Tech PWR Up
ATI/AMD is not standing still and is preparing its own sets of weapons for the new GPU war. In fact, it already has working 7nm Vega GPUs running its labs right now.
The red team’s next generation Vega 20 GPU built on TSMC’s cutting edge 7nm process technology is set to begin sampling later this year and we’ve already gotten a taste of what it can do thanks to a 3DMark benchmark leak just a couple of days ago. The new GPU has been leaked in the latest LLVM and clang compiler patches alongside a brand new GPU we’ve never heard of before, called Vega 12.
AI Instructions Confirmed for AMD’s 7nm Vega 20
The patch has revealed that unlike its predecessors, Vega 20 will feature support for intrinsic AI and deep learning instructions. We reported on Vega 20 in as early as January of 2017. We had viability into AMD’s plans to introduce a high performance Vega chip based on the 7nm process for quite a while. Confirmation of such a project, however, did not come until exactly a year later at this year’s CES.
Vega 20 (GFX906)
fdot2, sdot2, udot2.
The Sunnyvale California based chip maker announced earlier this year at CES that it’s building a new Vega chip using TSMC’s 7nm manufacturing technology to be deployed in deep neural networks for AI and machine learning workloads by the end of the year.
The patch confirms AMD’s previous announcements and claims about a 7nm Vega chip tailored for the highly lucrative machine learning and AI markets. The patch also confirms that the 7nm chip in question, while never mentioned by name by the company, is in fact Vega 20.
AMD’s Vega 20 GPU Specs & the Mysterious Vega 12
According to all the information available to date, Vega 20 builds on Vega 10 in several key aspects. Vega 20 is designed to deliver 8 times the double precision compute performance. It also features double the memory interface width and it features support for up to 4 HBM2 stacks.
This allows it to be equipped with up to 32GB of HBM2 vRAM, and have access to a whopping 1TB/s of bandwidth. In fact, the leaked Vega 20 GPU that we spotted last week on 3DMark’s benchmarking database confirms a 32GB memory capacity and more than a terabyte per second of memory bandwidth.
Unlike Vega 20, very little is known about Vega 12 (GFX904) other than the fact that it exists. Some speculate that it’s replacing Polaris at the entry level, but that’s pure conjecture at this point.
THX to WCCF
I really wish there was a gaming version.
The sooner the better when it comes to Freesync 2. I can't wait for this to become mainstream. Sick of monitors topping out at 75hz.
Yup, now good is 90-144Hz w/HDR, FS2 and min. for HDR is 600 (but i would like to have Dolby Vision one )
+ Sound Blaster Dolby Atmos 7.1.4 add-on Sound Card with Genius or Edifier Wooden Setup 5.1.2/7.1.2
CHUWI Announces HiGame NUC Featuring RX Vega Graphics
Chuwi HiGame is one of the rarely seen mini PC that delivers desktop-gaming-grade performance thanks to 8th Gen Intel® CoreTM I7-8709G with
RadeonTM RX Vega M GH graphics, which combines processing and discrete graphics on a single chip.
Alongside all the big players like Microsoft, EA, and Ubisoft with their own E3 presentations, the PC Gaming Show will once again return to the expo with plenty of new PC-focused announcements and reveals this year.
The show will be hosted live at the Los Angeles Wiltern Theatre on June 11 at 3pm PT with Sean 'Day' Plott also returning as the host.
As usual, there will be various showcases from sponsors, publishers, developers, and even hardware companies, such as Acer, Tripwire Interactive, Stardock Entertainment, Coffee Stain Studios, and Frontier Developments, with the show's theme this year being,
'See The Future of PC Gaming'.
Tickets to the show are available free of charge and can be reserved by going over here.
However, for those of us who can't make it there in person, the PC Gaming Show will also be livestreamed on Twitch, YouTube, Steam, as well as Facebook.
VESA's DisplayHDR Test software lets users double-check their monitors
When VESA announced its three-tiered DisplayHDR monitor certification program, the group promised it'd have a software utility for end users to check their displays' compliance with the standard by the end of Q2 2018. The outfit has delivered on that promise with this week's release of the DisplayHDR Test application for Windows 10.
VESA's announcement says the tool works with "off the shelf calibration colorimeters," but consumers will have to wait for the next update to get a list of colorimeters that meet the group's requirements for accuracy and tolerance.
The initial release's operation is a bit cumbersome, requiring that users run a command-line program that outputs a set of test patterns to the screen. The user is then tasked with operating their colorimeter hardware and software to measure the screen output while VESA's utility runs. To get the final results, the user must then pass the output from the colorimeter's software through a second utility provided by VESA.
The outfit says future versions of DisplayHDR Test may drive the sensor in the test tool and compute the results directly.
As a reminder, VESA's three DisplayHDR standards are based on the performance of difference luminance tests, two contrast tests, coverage of the BT.709 and DCI-P3 color spaces, color precision, and the speed of backlight's response to luminance changes.
So far, only about a dozen monitor models have earned DisplayHDR badging, and only a single display from Philips has reached the top-shelf DisplayHDR 1000 certification. Colorimeter owners can use VESA's DisplayHDR Test app to check and see where their displays measure up before making a decision on ponying up for a new screen.
Spoiler: Gaming w/AMD Store.MI HDD/SDD/RAM
Do you remember AMD's TrueAudio Next technology? If not, you'd be forgiven; it's not gained as much traction as it could (should?) have, considering its open nature. As a quick reminder, this is AMD's GPU-accelerated audio pipeline, which adds "audio raytracing" capabilities to audio by delivering true spatial positioning and object interactions in a given scene - at much higher performance than the usual CPU-based solutions.
The 1.2 version is being hailed as a "coming of age" for True Audio Next, which includes "a number of notable performance and feature improvements, and it reflects the enhancements built into the version of TAN supported in Steam Audio." Efficiency has also been improved, with minimized "memory, buffer transfer and synchronization overhead". The remainder of the blog post by AMD's Fellow Design Engineer Carl Wakeland follows."The TAN GPU utilities library now supports AMD Resource Reservation, in which a configurable part of the GPU may be reserved for audio processing apart from the normal GPU compute resources. As explained in earlier blogs, Resource Reservation protects audio and graphics queues and compute resources from blocking each other, allowing them to coexist on the GPU as never before possible. Developers can now call a function to query a system's TAN support and available resources, as well.
Finally, a number of new samples are added to exemplify and streamline the process of building audio applications using TAN:
Accelerated mixing. Mixing on the GPU with TAN can minimize buffer transfer overhead.
IIR (Infinite Impulse Response) filter.
Time domain convolution and doppler sample.
We continue to work on exciting new optimizations for future releases. Meanwhile, we welcome contributions from others - please feel free to make a pull request to submit your own examples and optimizations for TAN."
AMD’s RX Vega 56 finally gets miniaturised
Meet Nano 56
That’s where PowerColor’s Nano Edition card differs, as Hardwareluxx report. It’s the first design to offer the cut-down PCB and only a single fan design for mini-ITX builds, and while not the first time we've heard about it,
the card will supposedly make an appearance at Computex at the start of June. It seems only an RX Vega 56 version will be available for the time being.
That might be with good reason, too. AMD cards are pretty power-hungry, and with higher consumption comes higher temperatures, too.
The single axial fan of the PowerColor Nano Edition is going to have to be well-designed to cope with all that heat the RX Vega 56 is capable of pumping out.
It’s also a rather bland shroud design, and even the rather tame R9 Nano miniature graphics cards, from a few years back, stand out with their reference flair a little more than the PowerColor RX Vega variant.
The only feature that breaks up the shroud design are the 8+6 pin power connectors.
Hopefully we will get to see more in way of performance numbers from Computex to see how this ITX graphics card really copes. While other GPUs have been returning closer to MSRP in recent weeks,
AMD’s latest graphics architecture are still relatively pricey versus comparable performing cards. However, there is some hope that PowerColor’s miniature RX Vega 56 may be affordable,
as their Red Dragon RX Vega 56 is one of the only RX Vega 56 cards on the market right now for under £500.
Nice to see another Nano Vega, and this one is in form factor too although I wonder how well it cools as a result.
Going by the PCB teardown (As it's called.) the GPU is fairly similar to the standard PCB but with less voltage phases though still plenty powerful but as a result of everything being closer together heat build up is a bigger concern and some of the components don't handle heat very well which is probably why the other Nano variants opted for a extended cooler instead of keeping a smaller form factor.
Shame these much as any other Vega GPU fluctuate a lot in terms of pricing and overall availability can be pretty low, would be nice to see more reviews of these but I guess that might be a bit difficult if there's no cards in stock ha ha.