All | Editorial | General Tech | Graphics Cards | Networking | Motherboards | Cases and Cooling | Processors | Chipsets | Memory | Displays | Systems | Storage | Mobile | Shows and Expos
Subject: Graphics Cards | February 20, 2017 - 02:54 PM | Jeremy Hellstrom
Tagged: nvidia, gtx 1080 Xtreme Edition, GTX 1080, gigabyte, aorus
Gigabyte created their Aorus line of products to attract enthusiasts away from some of the competitions sub-brands, such as ASUS ROG. It is somewhat similar to the Gigabyte Xtreme Edition released last year but their are some differences, such as the large copper heatsink attached to the bottom of the GPU. The stated clockspeeds are the same as last years model and it also sports the two HDMI connections on the front of the card to connect to Gigabyte's VR Extended Front panel. The Tech Report manually overclocked the card and saw the Aorus reach the highest frequencies they have seen from a GP104 chip, albeit by a small margin. Check out the full review right here.
"Aorus is expanding into graphics cards today with the GeForce GTX 1080 Xtreme Edition 8G, a card that builds on the strong bones of Gigabyte's Editor's Choice-winning GTX 1080 Xtreme Gaming. We dig in to see whether Aorus' take on a GTX 1080 is good enough for a repeat."
Here are some more Graphics Card articles from around the web:
- Gigabyte GTX 1080 Aorus Xtreme Edition 8 GB @ techPowerUp
- NVIDIA’s Fastest Graphics Card Ever: A Look At The Quadro P6000 @ Techgage
- Radeon Windows 10 vs. Linux RadeonSI/RADV Gaming Performance @ Phoronix
- Windows 10 vs. Ubuntu Linux Gaming Performance With NVIDIA GeForce GTX 1060/1080 @ Phoronix
Subject: Graphics Cards | February 17, 2017 - 07:42 AM | Scott Michaud
Tagged: nvidia, graphics drivers
Just a couple of days after publishing 378.66, NVIDIA released GeForce 378.72 Hotfix drivers. This fixes a bug encoding video in Steam’s In-Home Streaming, and it also fixes PhysX not being enabled on the GPU under certain conditions. Normally, hotfix drivers solve large-enough issues that were introduced with the previous release. This time, as far as I can tell, is a little different, though. Instead, these fixes seem to be intended for 378.66 but, for one reason or another, couldn’t be integrated and tested in time for the driver to be available for the game launches.
This is an interesting effect of the Game Ready program. There is value in having a graphics driver available on the same day (or early) as a major game releases, so that people can enjoy the title as soon as it is available. There is also value in having as many fixes as the vendor can provide. These conditions oppose each other to some extent.
From a user standpoint, driver updates are cumulative, so they are able to skip a driver or two if they are not affected by any given issue. AMD has taken up a similar structure, some times releasing three or four drivers in a month with only, like, one of them being WHQL certified. For these reasons, I tend to lean on the side of “release ‘em as you got them”. Still, I can see people feeling a little uneasy about a driver being released incomplete to hit a due-date.
But, again, that due-date has value.
It’s interesting. I’m personally glad that AMD and NVIDIA are on a rapid-release schedule, but I can see where complaints could arise. What’s your opinion?
Subject: Graphics Cards | February 16, 2017 - 03:35 PM | Jeremy Hellstrom
Tagged: msi, AERO ITX, gtx 1070, gtx 1060, gtx 1050, GTX 1050 Ti, SFF, itx
MSI have just release their new series of ITX compatible GPUs, covering NVIDIA's latest series of cards from the GTX 1050 through to the GTX 1070; the GTX 1080 is not available in this form factor. The GTX 1070 and 1060 are available in both factory overclocked and standard versions.
All models share a similar design, with a single TORX fan with 8mm Super Pipes and the Zero Frozr feature which stops the fan to give silent operation when temperatures are below 60C. They are all compatible with the Afterburner Overclocking Utility, including recordings via Predator and wireless control from your phone.
The overclocked cards run slightly over reference, from the GTX 1070 at 1721MHz boost, 1531MHz base with the GDDR5 at 8GHz to the GTX 1050 at 1518MHz boost, 1404MHz base and the GDDR5 at 7GHz. The models which do not bear the OC moniker run at NVIDIA's reference clocks even if they are not quite fully grown.
Subject: Graphics Cards | February 14, 2017 - 09:29 PM | Scott Michaud
Tagged: opencl 2.0, opencl, nvidia, graphics drivers
While the headline of the GeForce 378.66 graphics driver release is support for For Honor, Halo Wars 2, and Sniper Elite 4, NVIDIA has snuck something major into the 378 branch: OpenCL 2.0 is now available for evaluation. (I double-checked 378.49 release notes and confirmed that this is new to 378.66.)
OpenCL 2.0 support is not complete yet, but at least NVIDIA is now clearly intending to roll it out to end-users. Among other benefits, OpenCL 2.0 allows kernels (think shaders) to, without the host intervening, enqueue work onto the GPU. This saves one (or more) round-trips to the CPU, especially in workloads where you don’t know which kernel will be required until you see the results of the previous run, like recursive sorting algorithms.
So yeah, that’s good, albeit you usually see big changes at the start of version branches.
Another major addition is Video SDK 8.0. This version allows 10- and 12-bit decoding of VP9 and HEVC video. So... yeah. Applications that want to accelerate video encoding or decoding can now hook up to NVIDIA GPUs for more codecs and features.
NVIDIA’s GeForce 378.66 drivers are available now.
Subject: Graphics Cards | February 14, 2017 - 05:57 PM | Scott Michaud
Tagged: amd, graphics drivers
Just in time for For Honor and Sniper Elite 4, AMD has released a new set of graphics drivers, Radeon Software Crimson ReLive 17.2.1, that target these games. The performance improvements that they quote are in the 4-5% range, when compared to their previous driver on the RX 480, which would be equivalent to saving a whole millisecond per frame at 60 FPS. (This is just for mathematical reference; I don’t know what performance users should expect with an RX 480.)
Beyond driver overhead improvements, you will now be able to utilize multiple GPUs in CrossFire (for DirectX 11) on both titles.
Also, several issues have been fixed with this version. If you have a FreeSync monitor, and some games fail to activate variable refresh mode, then this driver might solve this problem for you. Scrubbing through some videos (DXVA H.264) should no longer cause visible corruption. A couple applications, like GRID and DayZ, should no longer crash under certain situations. You get the idea.
If you have an AMD GPU on Windows, pick up these drivers from their support page.
Subject: Graphics Cards | February 9, 2017 - 02:46 PM | Jeremy Hellstrom
Tagged: amd, nvidia
New graphics drivers are a boon to everyone who isn't a hardware reviewer, especially one who has just wrapped up benchmarking a new card the same day one is released. To address this issue see what changes have been implemented by AMD and NVIDIA in their last few releases, [H]ard|OCP tested a slew of recent drivers from both companies. The performance of AMD's past releases, up to and including the AMD Crimson ReLive Edition 17.1.1 Beta can be found here. For NVIDIA users, recent drivers covering up to the 378.57 Beta Hotfix are right here. The tests show both companies generally increasing the performance of their drivers, however the change is so small you are not going to notice a large difference.
"We take the AMD Radeon R9 Fury X and AMD Radeon RX 480 for a ride in 11 games using drivers from the time of each video card’s launch date, to the latest AMD Radeon Software Crimson ReLive Edition 17.1.1 Beta driver. We will see how performance in old and newer games has changed over the course of 2015-2017 with new drivers. "
Here are some more Graphics Card articles from around the web:
- Intel Celeron/Pentium/Core i3/i5/i7 - NVIDIA vs. AMD Linux Gaming Performance @ Phoronix
- PowerColor Radeon RX 470 Red Devil (4GB) @ Custom PC Review
- GeForce GTX 1080 @ Hardware Secrets
Subject: Graphics Cards | February 6, 2017 - 11:43 AM | Sebastian Peak
Tagged: video card, silent, Passive, palit, nvidia, KalmX, GTX 1050 Ti, graphics card, gpu, geforce
Palit is offering a passively-cooled GTX 1050 Ti option with their new KalmX card, which features a large heatsink and (of course) zero fan noise.
"With passive cooler and the advanced powerful Pascal architecture, Palit GeForce GTX 1050 Ti KalmX - pursue the silent 0dB gaming environment. Palit GeForce GTX 1050 Ti gives you the gaming horsepower to take on today’s most demanding titles in full 1080p HD @ 60 FPS."
The specs are identical to a reference GTX 1050 Ti (4GB GDDR5 @ 7 Gb/s, Base 1290/Boost 1392 MHz, etc.), so expect the full performance of this GPU - with some moderate case airflow, no doubt.
We don't have specifics on pricing or availablity just yet.
Subject: Graphics Cards | February 4, 2017 - 03:29 PM | Tim Verry
Tagged: micron, graphics memory, gddr6
This year is shaping up to be a good year for memory with the promise of 3D XPoint (Intel/Micron), HBM2 (SK Hynix and Samsung), and now GDDR6 graphics memory from Micron launching this year. While GDDR6 was originally planned to be launched next year, Micron recently announced its intentions to start producing the memory chips by the later half of 2017 which would put it much earlier than previously expected.
Computer World reports that Micron is citing the rise of e-sports and gaming driving the computer market that now sees three year upgrade cycles rather than five year cycles (I am not sure how accurate that is, however as it seems like PCs are actually lasting longer between upgrade as far as relevance but i digress) as the primary reason for shifting GDDR6 production into high gear and moving up the launch window. The company expects the e-sports market to grow to 500 million fans by 2020, and it is a growing market that Micron wants to stay relevant in.
If you missed our previous coverage, GDDR6 is the successor to GDDR5 and offers twice the bandwidth at 16 Gb/s (gigabits per second) per die. It is also faster than GDDR5X (12 Gb/s) and uses 20% less power which the gaming laptop market will appreciate. HBM2 still holds the bandwidth crown though as it offers 256 GB/s per stack and up to 1TB/s with four stacks connected to a GPU on package.
As such, High Bandwidth Memory (HBM2 and then HBM3) will power the high end gaming and professional graphics cards while GDDR6 will become the memory used for mid range cards and GDDR5X (which is actually capable of going faster but will likely not be pushed much past 12 Gbps after all if GDDR6 does come out this soon) will replace GDDR5 on most if not all of the lower end products.
I am not sure if Micron’s reasoning of e-sports, faster upgrade cycles, and VR being the motivating factor(s) to ramping up production early is sound or not, but I will certainly take the faster memory coming out sooner rather than later! Depending on exactly when in 2017 the chips start rolling off the fabs, we could see graphics cards using the new memory technology as soon as early 2018 (just in time for CES announcements? oh boy I can see the PR flooding in already! hehe).
Will Samsung change course as well and try for a 2017 release for its GDDR6 memory as well?
Are you ready for GDDR6?
Subject: Graphics Cards | February 3, 2017 - 05:58 PM | Scott Michaud
Tagged: nvidia, graphics drivers, vulkan
On February 1st, NVIDIA released a new developer beta driver, which fixes a couple of issues with their Vulkan API implementation. Unlike what some sites have been reporting, you should not download it to play games that use the Vulkan API, like DOOM. In short, it is designed for developers, not end-users. The goal is to provide correct results when software interacts with the driver, not the best gaming performance or anything like that.
In a little more detail, it looks like 376.80 implements the Vulkan 126.96.36.199 SDK. This update addresses two issues with accessing devices and extensions, under certain conditions, when using the 188.8.131.52 SDK. 184.108.40.206 was released on January 23rd, and thus it will not even be a part of current video games. Even worse, it, like most graphics drivers for software developers, is based on the old, GeForce 376 branch, so it won’t even have NVIDIA’s most recent fixes and optimizations. NVIDIA does this so they can add or change the features that Vulkan developers require without needing to roll-in patches every time they make a "Game Ready" optimization or something. There is no reason to use this driver unless you are developing Vulkan applications, and you want to try out the new extensions. It will eventually make it to end users... when it's time.
If you are wishing to develop software using Vulkan’s bleeding-edge features, then check out NVIDIA’s developer portal to pick up the latest drivers. Basically everyone else should use 378.49 or its 378.57 hotfix.
Subject: Graphics Cards | February 2, 2017 - 07:02 AM | Scott Michaud
Tagged: graphics drivers, amd
A few days ago, AMD released their second graphics drivers of January 2017: Radeon Software Crimson ReLive 17.1.2. The main goal of these drivers are to support the early access of Conan Exiles as well as tomorrow’s closed beta for Tom Clancy’s Ghost Recon Wildlands. Optimization that AMD has been working on prior to release, for either game, are targeted at this version.
Beyond game-specific optimizations, a handful of bugs are also fixed, ranging from crashes to rendering artifacts. There was also an issue with configuring WattMan on a system that has multiple monitors, where the memory clock would drop or bounce around. There is driver also has a bunch of known issues, including a couple of hangs and crashes under certain situations.
Subject: Graphics Cards | February 2, 2017 - 07:01 AM | Scott Michaud
Tagged: nvidia, graphics drivers
If you were having issues with Minecraft on NVIDIA’s recent 378.49 drivers, then you probably want to try out their latest hotfix. This version, numbered 378.57, will not be pushed down GeForce Experience, so you will need to grab them from NVIDIA’s customer support page.
Beyond Minecraft, this also fixes an issue with “debug mode”. For some Pascal-based graphics cards, the option in NVIDIA Control Panel > Help > Debug Mode might be on by default. This option will reduce factory-overclocked GPUs down to NVIDIA’s reference speeds, which is useful to eliminate stability issues in testing, but pointlessly slow if you’re already stable. I mean, you bought the factory overclock, right? I’m guessing someone at NVIDIA used it to test 378.49 during its development, fixed an issue, and accidentally commit the config file with the rest of the fix. Either way, someone caught it, and it’s now fixed, even though you should be able to just untick it if you have a factory-overclocked GPU.
Subject: Graphics Cards | January 31, 2017 - 11:18 AM | Ryan Shrout
Tagged: rx 460, radeon, giveaway, contest, buildapc, amd
As part of our partnership with AMD to take a look at the Radeon RX 460 as a budget gaming graphics solution, we are giving away the computer we built for our testing. If you missed our previous stories, shame on you. Check them out here:
- Building a Budget PC with the Radeon RX 460: Part 1
- Building a Budget PC with the Radeon RX 460: Part 2
Check out the embeded block below to see how you can win our system. It is a global giveaway, so feel free to enter no matter where you live! Thanks again to AMD for providing the hardware for this build!
Subject: Graphics Cards | January 27, 2017 - 09:19 PM | Scott Michaud
Tagged: microsoft, DirectX, llvm, dxil, spir-v, vulkan
Over the holidays, Microsoft has published the DirectX Shader Compiler onto GitHub. The interesting part about this is that it outputs HLSL into DirectX Intermediate Language (DXIL) bytecode, which can be ingested by GPU drivers and executed on graphics devices. The reason why this is interesting is that DXIL is based on LLVM, which might start to sound familiar if you have been following along with The Khronos Group and their announcements regarding Vulkan, OpenCL, and SPIR-V.
As it turns out, they were on to something, and Microsoft is working on a DirectX analogue of it.
The main advantage of LLVM-based bytecode is that you can eventually support multiple languages (and the libraries of code developed in them). When SPIR-V was announced with Vulkan, the first thing that came to my mind was compiling to it from HLSL, which would be useful for existing engines, as they are typically written in HLSL and transpiled to the target platform when used outside of DirectX (like GLSL for OpenGL). So, in Microsoft’s case, it would make sense that they start there (since they own the thing) but I doubt that is the end goal. The most seductive outcome for game engine developers would be single-source C++, but there is a lot of steps between there and here.
Another advantage, albeit to a lesser extent, is that you might be able to benefit from performance optimizations, both on the LLVM / language side as well as on the driver’s side.
According to their readme, the minimum support will be HLSL Shader Model 6. This is the most recent shading model, and it introduces some interesting instructions, typically for GPGPU applications, that allow multiple GPU threads to interact, like balloting. Ironically, while DirectCompute and C++AMP don’t seem to be too popular, this would nudge DirectX 12 into a somewhat competent GPU compute API.
DXIL support is limited to Windows 10 Build 15007 and later, so you will need to either switch one (or more) workstation(s) to Insider, or wait until it launches with the Creators Update (unless something surprising holds it back).
Subject: Graphics Cards | January 26, 2017 - 09:38 PM | Scott Michaud
Tagged: nvidia, graphics drivers
Update: There are multiple issues being raised in our comments, including a Steam post by Sam Lantinga (Valve) about this driver breaking In-Home Streaming. Other complaints include certain applications crashing and hardware acceleration issues.
Original Post Below
Now that the holidays are over, we’re ready for the late-Winter rush of “AAA” video games. Three of them, Resident Evil VII, the early access of Conan Exiles, and the closed beta of For Honor, are targeted by NVIDIA’s GeForce 378.49 Game Ready drivers. Unless we get a non-Game Ready driver in the interim, I am guessing that this will cover us until mid-February, before the full release of For Honor, alongside Sniper Elite 4 and followed by Halo Wars 2 on the next week.
Beyond game-specific updates, the 378-branch of drivers includes a bunch of SLI profiles, including Battlefield 1. It also paves the way for GTX 1050- and GTX 1050 Ti-based notebooks; this is their launch driver whenever OEMs begin to ship the laptops they announced at CES.
This release also contains a bunch of bug fixes (pdf), including a reboot bug with Wargames: Red Dragon and TDR (driver time-out) with Windows 10 Anniversary Update. I haven’t experienced any of these, but it’s good to be fixed regardless.
You can pick up the new drivers from their website if, you know, GeForce Experience hasn’t already notified you.
Subject: Graphics Cards | January 25, 2017 - 03:37 PM | Jeremy Hellstrom
Tagged: windforce, factory overclocked, GTX 1060 G1 GAMING 6G, GeForce GTX 1060, gigabyte
In their testing [H]ard|OCP proved that the Windforce cooler is not the limiting factor when overclocking Gigabyte's GTX 1060 G1 Gaming G6, even at their top overclock of 2.1GHz GPU, 9.4GHz memory the temperature never reached 60C. They did have some obstacles reaching those speeds, the cards onboard Gaming mode offered an anemic boost and in order to start manually overclocking this card you will need to install the XTREME ENGINE VGA Utility. Once you have that, you can increase the voltage and clocks to find the limits of the card you have, which should offer a noticeable improvement from its performance straight out of the box.
"We’ve got the brand new GIGABYTE GeForce GTX 1060 G1 GAMING 6G video card to put through the paces and find out how well it performs in games and overclocks. We will compare its highest overclock with one of the best overclocks we’ve achieved on AMD Radeon RX 480 to put it to the test. How will it stand up? Let’s find out."
Here are some more Graphics Card articles from around the web:
- GTX 1070 Overclocking Guide @ OCC
- Arctic Accelero Hybrid III-140 GPU Cooler @ Kitguru
- OCC's Top 3 Video Cards of 2016 @ OCC
- Benchmarking Radeon Open Compute ROCm 1.4 OpenCL @ Phoronix
Subject: Graphics Cards | January 18, 2017 - 08:43 PM | Sebastian Peak
Tagged: video, unlock, shaders, shader cores, sapphire, radeon, Polaris, graphics, gpu, gaming, card, bios, amd, 1024
As reported by WCCFtech, AMD partner Sapphire has a new 1024 stream processor version of the RX460 listed on their site (Chinese language), and this product reveal of course comes after it became known that RX460 graphics cards had the potential to have their stream processor count unlocked from 896 to 1024 via BIOS update.
Sapphire RX460 1024SP 4G D5 Ultra Platinum OC (image credit: Sapphire)
The Sapphire RX460 1024SP edition offers a full Polaris 11 core operating at 1250 MHz, and it otherwise matches the specifications of a stock RX460 graphics card. Whether this product will be available outside of China is unknown, as is the potential pricing model should it be available in the USA. A 4GB Radeon RX460 retails for $99, while the current step-up option is the RX470, which doubles up on this 1024SP RX460's shader count with 2048, with a price increase of about 70% ($169).
AMD Polaris GCN 4.0 GPU lineup (Credit WCCFtech)
As you may note from the chart above, there is also an RX470D option between these cards that features 1792 shaders, though this option is also China-only.
Subject: Graphics Cards | January 17, 2017 - 10:31 PM | Tim Verry
Tagged: SFF, pascal, low profile, GTX 1050 Ti, gtx 1050, gigabyte
Subject: Graphics Cards | January 10, 2017 - 10:11 PM | Tim Verry
Tagged: CES, CES 2017, aorus, gigabyte, xtreme gaming, GTX 1080, pascal
One interesting development from Gigabyte at this year’s CES was the expansion of its Aorus branding and the transition from Xtreme Gaming. Initially used on its RGB LED equipped motherboards, the company is rolling out the brand to its other higher end products including laptops and graphics cards. While it appears that Xtreme Gaming is not going away entirely, Aorus is taking the spotlight with the introduction of the first Aorus branded graphics card: the GTX 1080.
Paul's Hardware got hands on with the new card (video) at the Gigabyte CES booth.
Featuring a similar triple 100mm fan cooler as the GTX 1080 Xtreme Gaming 8G, the Aorus GTX 1080 comes with x patterned LED lighting as well as a backlit Aorus logo on the side and a backlit Eagle on the backplate. The cooler is comprised of three 100mm double stacked fans (the center fan is recessed and spins in the opposite direction of the side fans) over a shrouded angled aluminum fin stack that connects to the GPU over five large copper heatpipes.
The graphics card is powered by two 8-pin PCI-E power connectors.
In an interesting twist, the card has two HDMI ports on the back of the card that are intended to be used to hook up front panel HDMI outputs for things like VR headsets. Another differentiator between the upcoming card and the Xtreme Gaming 8G is the backplate which has a large copper plate secured over the underside of the GPU. Several sites are reporting that this area can be used for watercooling, but I am skeptical of this as if you are going to go out and buy a waterblock for your graphics card you might as well buy a block to put on top of the GPU and not on the area of the PCB opposite the GPU!). As is, the copper plate on the backplate certainly won’t hurt cooling, and it looks cool, but that’s all I suspect it is.
Think Computers also checked out the Aorus graphics card. (video above)
Naturally, Gigabyte is not talking clock speeds on this new card, but I expect it to hit at least the same clocks as its Xtreme Gaming 8G predecessor which was clocked at 1759 MHz base and 1848 MHz boost out of the box and 1784 MHz base and 1936 MHz boost in OC Mode respectively. Gigabyte also overlocked the memory on that card up to 10400 MHz on OC Mode.
Gigabyte also had new SLI HB bridges on display bearing the Aorus logo to match the Aorus GPU. The company also had Xtreme Gaming SLI HB bridges though which further suggests that they are not completely retiring that branding (at least not yet).
Pricing has not been announced, but the card will be available in February.
Gigabyte has yet to release official photos of the card or a product page, but it should show up on their website shortly. In the meantime, Paul's Hardware and Think Computers shot some video of the card on the show floor which I have linked above if you are interested in the card. Looking on Amazon, the Xtreme Gaming 1080 8GB is approximately $690 before rebate so I would guess that the Aorus card would come out at a slight premium over that if only for the fact that it is a newer release, has a more expensive backplate and additional RGB LED backlighting.
What are your thoughts on the move to everything-Aorus?
Follow all of our coverage of the show at https://pcper.com/ces!
Subject: Graphics Cards | January 9, 2017 - 02:32 PM | Jeremy Hellstrom
Tagged: Crimson Edition 16.12.2
Radeon Software Crimson Edition 16.12.2 is now live and WHQL certified, ready for you to grab here or through the version you already have installed, which supports the recommended clean installation option.
This particular update addresses Freesync issues with borderless Fullscreen applications as well as compatibility issues with Battlefield 1 and DOTA 2. There are also numerous optimizations and fixes for issues with Radeon ReLive, which you can read in more detail under the Release Notes tab.
Subject: Graphics Cards | January 8, 2017 - 03:53 AM | Tim Verry
Tagged: vega 11, vega 10, navi, gpu, amd
During CES, AMD showed off demo machines running Ryzen CPUs and Vega graphics cards as well as gave the world a bit of information on the underlying architecture of Vega in an architectural preview that you can read about (or watch) here. AMD's Vega GPU is coming and it is poised to compete with NVIDIA on the high end (an area that has been left to NVIDIA for awhile now) in a big way.
Thanks to Videocardz, we have a bit more info on the products that we might see this year and what we can expect to see in the future. Specifically, the slides suggest that Vega 10 – the first GPUs to be based on the company's new architecture – may be available by the (end of) first half of 2017. Following that a dual GPU Vega 10 product is slated for a release in Q3 or Q4 of 2017 and a refreshed GPU based on smaller process node with more HBM2 memory called Vega 20 in the second half of 2018. The leaked slides also suggest that Navi (Vega's successor) might launch as soon as 2019 and will come in two variants called Navi 10 and Navi 11 (with Navi 11 being the smaller / less powerful GPU).
The 14nm Vega 10 GPU allegedly offers up 64 NCUs and as much as 12 TFLOPS of single precision and 750 GFLOPS of double precision compute performance respectively. Half precision performance is twice that of FP32 at 24 TFLOPS (which would be good for things like machine learning). The NCUs allegedly run FP16 at 2x and DPFP at 1/16. If each NCU has 64 shaders like Polaris 10 and other GCN GPUs, then we are looking at a top-end Vega 10 chip having 4096 shaders which rivals that of Fiji. Further, Vega 10 supposedly has a TDP up to 225 watts.
For comparison, the 28nm 8.9 billion transistor Fiji-based R9 Fury X ran at 1050 MHz with a TDP of 275 watts and had a rated peak compute of 8.6 TFLOPS. While we do not know clock speeds of Vega 10, the numbers suggest that AMD has been able to clock the GPU much higher than Fiji while still using less power (and thus putting out less heat). This is possible with the move to the smaller process node, though I do wonder what yields will be like at first for the top end (and highest clocked) versions.
Vega 10 will be paired with two stacks of HBM2 memory on package which will offer 16GB of memory with memory bandwidth of 512 GB/s. The increase in memory bandwidth is thanks to the move to HBM2 from HBM (Fiji needed four HBM dies to hit 512 GB/s and had only 4GB).
The slide also hints at a "Vega 10 x2" in the second half of the year which is presumably a dual GPU product. The slide states that Vega 10 x2 will have four stacks of HBM2 (1TB/s) though it is not clear if they are simply adding the two stacks per GPU to claim the 1TB/s number or if both GPUs will have four stacks (this is unlikely though as there does not appear to be room on the package for two more stacks each and I am not sure if they could make the package bit enough to make room for them either). Even if we assume that they really mean 2x 512 GB/s per GPU (and maybe they can get more out of that in specific workloads across both) for memory bandwidth, the doubling of cores and at least potential compute performance will be big. This is going to be a big number crunching and machine learning card as well as for games of course. Clockspeeds will likely have to be much lower compared to the single GPU Vega 10 (especially with stated TDP of 300W) and workloads wont scale perfectly so potential compute performance will not be quite 2x but should still be a decent per-card boost.
Raja Koduri holds up a Vega GPU at CES 2017 via eTeknix
Moving into the second half of 2018, the leaked slides suggest that a Vega 20 GPU will be released based on a 7nm process node with 64 CUs and paired with four stacks of HBM2 for 16 GB or 32 GB of memory with 1TB/s of bandwidth. Interestingly, the shaders will be setup such that the GPU can still do half precision calculations at twice that of single precision, but will not take nearly the hit on double precision at Vega 10 at only 1/2 single precision rather than 1/16. The GPU(s) will use between 150W and 300W of power, and it seems these are set to be the real professional and workstation workhorses. A Vega 10 with 1/2 DPFP compute would hit 6 TFLOPS which is not bad (and it would hopefully be more than this due to faster clocks and architecture improvements).
Beyond that, the slides mention Navi's existence and that it will come in Navi 10 and Navi 11 but no other details were shared which makes sense as it is still far off.
You can see the leaked slides here. In all, it is an interesting look at potential Vega 10 and beyond GPUs but definitely keep in mind that this is leaked information and that the information allegedly came from an internal presentation that likely showed the graphics processors in their best possible/expect light. It does add a bit more hope to the fire of excitement for Vega though, and I hope that AMD pulls it off as my unlocked 6950 is no longer supported and it is only a matter of time before new games perform poorly or not at all!
- AMD Vega GPU Architecture Preview: Redesigned Memory Architecture
- CES 2017: AMD Vega Running DOOM at 4K
- AMD GPU Roadmap: Capsaicin Names Upcoming Architectures
- AMD's Raja Koduri talks moving past CrossFire, smaller GPU dies, HBM2 and more.