Graphics Cards - Page 223

We're finally starting to see the performance of NVIDIA's purported GeForce GTX 1050 Ti, with the leaked information teasing that we're to expect GTX 960 like performance, without a PCIe power connector.

NVIDIA is expected to pack its GeForce GTX 1050 Ti with 768 CUDA cores, 48 TMUs, and 32 ROPs. We should see the GP107 GPU clocked at 1318/1380MHz for base/boost, respectively - a 250MHz increase over the GM107 GPU that powers the GeForce GTX 960. The texture fill rate on the GTX 1050 Ti is nearly doubled over the GTX 960, hitting 84 GTexel/s, while 4GB of GDDR5 at 7Gbps will be placed on a 128-bit memory interface providing around 112GB/sec memory bandwidth.

A post on Chiphell teases GTX 960 performance in 3DMark 11, but it'll be the price that determines if the GTX 1050 Ti is a winner. If it's priced at $149, it could be a great card for mid-range gamers, especially with a TDP of just 75W and no PCIe power connector required.

With all of the teases of NVIDIA's upcoming Volta architecture and Pascal 2.0 refresh aiming for 2017, we now have news of Team Red and their upcoming Vega GPU architecture. Our friends at Fudzilla have said that their "well-informed sources have confirmed that Vega 10, AMD's first HBM 2 card will by announced before the end of this year, at least for the professional market".

Now, the professional market launch arriving before the consumer Vega graphics cards makes sense, especially with HBM2 in tow. NVIDIA did the same thing with the Tesla P100 earlier this year, their first Pascal-based graphics card powered by HBM2, except it launched for the professional market first. Consumer graphics cards with HBM2 are still a while away, so expect them sometime in 2017.

AMD's upcoming high-end Vega 10 graphics card would feature 16GB of HBM2, the same type and amount of HBM2 used on NVIDIA's Tesla P100 professional graphics card.

It looks like the launch of AMD's new Polaris GPU architecture is doing wonders for the company, with Wells Fargo's David Wong taking a detailed look at Mercury Research's latest numbers on the discrete GPU market.

After reviewing the data on GPU market share from Mercury Research, Wong noted that AMD has already made significant gains. He said that the data proves AMD is stabilizing, where he said: "According to Mercury Research, AMD's unit share of the discrete GPU market increased from 26.2% in the December 2015 quarter to 29.4% in the March 2016 quarter. AMD's discrete GPU market unit share increased 4.8pp sequentially to 34.2% in the June 2016 quarter".

When it comes to the high-end graphics cards, AMD still isn't competing against NVIDIA yet, at least not with the Polaris architecture. AMD's new Vega architecture is the one that will compete with NVIDIA's enthusiast graphics cards, but won't hit consumers until 2017. Wong added: "Despite having apparently reached a plateau in unit graphics market share some quarters ago, NVIDIA's gaming GPU continued to demonstrate good revenue growth, in the 17-25% year/year range in each of the last 4 quarters (Oct 2015 through July 2016). AMD began to regain unit share in the graphics card market in 1H2016".

If the massive array of Volta GPU architecture leaks weren't enough yesterday, we had teases of the purported Pascal 2.0 refresh coming in 2017, as well as the massively fast graphics cards from NVIDIA sometime in 2018, and beyond - well, now we're hearing more concrete information about the GeForce GTX 1080 Ti... the card everyone has been waiting for.

The new report has NVIDIA releasing the GeForce GTX 1080 Ti in January 2017 (most likely at CES 2017 in early January), with very similar specs to the blazingly fast Pascal-based Titan X. NVIDIA's new GeForce GTX 1080 Ti will reportedly rock most of the same specifications as the Titan X, so we have 12GB of GDDR5X being used at 10Gbps on a 384-bit memory bus that will provide 480GB/sec of memory bandwidth.

NVIDIA's purported GeForce GTX 1080 Ti will have some CUDA cores shaved away, down to 3328 CUDA cores from the 3584 CUDA cores on Titan X. The GP102 GPU used will be clocked at 1503/1623 for Base/Boost clocks, respectively - providing 10.8 TFLOPs of compute performance, all with a 250W TDP. So we're looking at a very close competitor to Titan X, but cheaper.

We've just heard that NVIDIA will unveil a Pascal refresh in 2017, but the next-gen Volta GPU architecture is what everyone wants to know about on the high-end/enthusiast level of the scale.

Volta is expected to be unveiled at NVIDIA's own GPU Technology Conference, which takes place between May 8-11 next year. NVIDIA is expected to follow in the steps of its Tesla P100 unveiling at GTC 2016 earlier this year, unveiling Volta on a new HPC part first, with the Volta-based HPC product expected to be powered by the ridiculously fast HBM2 tech.

NVIDIA boss Jen-Hsun Huang is also rumored to unveil an updated GPU roadmap, which will include new codenames and technology details for future GPU technology ffrom NVIDIA. NVIDIA is expected to dive away from the 10nm node, and go right into 7nm once Volta is here and established on the 14nm node. Not only that, but the future GPUs will have support for both HBM3 and GDDR6.

NVIDIA is not only expected to unveil a Pascal 2.0 refresh in 2017 on Samsung's 14nm FinFET node, but the company is expected to make a big splash at its GTC 2017 event in May by unveiling their new Volta architecture, and more.

Well, today is the day of NVIDIA leaks, with the company reportedly aiming at a 2018 release for its consumer-focused GeForce graphics cards based on the next-gen Volta GPU architecture. These new cards will reportedly rock a huge 16GB of GDDR6, the new standard from Micron that's quite a large step on the already impressive GDDR5X standard that powers the GeForce GTX 1080 and new Titan X.

Micron's new GDDR6 has over 14Gbps of bandwidth, compared to the 10Gbps on GDDR5 and just 7-8Gbps on GDDR5. GDDR6 is much more efficient than GDDR5X, with lower power consumption allowing for more VRAM on higher-end graphics cards. The new 16GB GDDR6 cards will be based on the upcoming Volta-based graphics cards with a 256-bit memory bus, while the higher-end GV102 will use a faster 384-bit memory bus and possibly 24GB or even 48GB cards thanks to GDDR6.

A treasure trove of details on NVIDIA's GPU plans for 2017 and beyond have surfaced, courtesy of Baidu user USG Ishimura, who said that there is a Pascal refresh on the way, which will be followed by Volta, and so much more.

Starting with the Pascal 2.0 refresh, which will supposedly see more GPUs released on the GP102 core, meaning cheaper versions of the incredibly fast Titan X and GTX 1080 graphics cards. This is similar to what NVIDIA did with the GTX 700 series based on the Kepler architecture. NVIDIA recently inked a deal with Samsung to have their Pascal architecture made on the 14nm FinFET process, with the current Pascal cards being made on the 16nm FinFET node by TSMC.

Micron has been providing the super-fast GDDR5X clocked at 10GHz for the GTX 1080 and Titan X, but the yields didn't start well, and will only improve as we get closer to the purported Pascal refresh. GDDR5X would find its way onto the entire Pascal refresh, except for the entry-level GP107-based cards. NVIDIA has also had its 16nm-based Pascal cards hitting 2GHz GPU clocks easily, so we could see GPU Boost 3.0 keeping the clock speeds higher than we have now, allowing for proper differentiation between AIB partner cards.

AMD has been spreading their new Polaris architecture throughout the different markets that they serve. That started out with the RX 480, a gaming graphics card, and continued into the professional arena with their newly branded Radeon Pro brand of graphics cards. AMD also updated their data center GPUs with Polaris graphics cards with their MxGPU line of products, so the last place for AMD to update their lineup was in embedded.

AMD's embedded graphics products generally tend to sit within a few applications where the company has traditionally done well. These applications include the use of graphics in medical imaging, digital signage and casino gaming. These are all familiar applications for AMD embedded graphics chips, however those use cases are expanding beyond the standard expected applications with these new AMD embedded graphics chips.

Because the new Polaris architecture delivers up to 2.8x performance per watt compared to the previous generation, there are many new applications for AMD's newest embedded GPUs.

NVIDIA used to make graphics cards for Apple and some of their Mac products, but current-gen Mac systems are running AMD Radeon graphics cards, with this possibly changing in the future.

According to a job ad posted by NVIDIA, they are looking for a software engineer who would "help produce the next revolutionary Apple products", reports Bloomberg. The job post continues, adding that the role would require "working in partnership with Apple", as well as writing code that will "define and shape the future"of graphics-related software on Macs.

There also isn't just a single job posting, but three job listings on NVIDIA's database that reference Apple, with the latest posting appearing last week. One of the jobs specifies working with the NVIDIA Mac graphics driver team, which is interesting to see indeed. Apple has always been one to engineer their products to a very high standard, and with NVIDIA continuing to refine their GPU architecture, could we see Pascal or Volta in future Mac systems? I think so.

Following the release of NVIDIA's new GeForce 372.90 drivers tuned for Forza Horizon 3, AMD has released their new Radeon Software Crimson Edition 16.9.2 drivers that also have support for Forza Horizon 3.

There are a bunch of fixed issues as well, as well as known issues that AMD are aware of. The new RSCE 16.9.2 drivers fix a "small amount of corruption" in the lower right hand corner of the display on Radeon HD 7000 series cards in Deus Ex: Mankind Divided, and some stuttering issues with CrossFire mode in DX11 in Mankind Divided.

Here's the full list of fixed issues on RSCE 16.9.2, which you can grab here:

We've just heard some juicy details on AMD's next-gen Vega 10 and Vega 20 graphics cards - with 16GB of HBM2 and 32GB of HBM2, respectively. But it's the news on Navi that has me even more excited, and simultaneously disappointed.

VideoCardz.com's source says that Navi 10 and Navi 11 are "currently planned for 2019", meaning that AMD has delayed Navi by 12 months from its original 2018 release, and now into 2019. We haven't confirmed this yet, but I've sent some emails and will update this post when we hear back from someone at AMD.

We just reported on the first details of AMD's next-gen Vega 10 graphics card, but it seems as though that's the mid-range model, while the higher-end Vega 20 is shaping up to be a damn monster.

AMD's upcoming Vega 20 graphics card will feature 32GB of HBM2 memory with 1TB/sec of memory bandwidth, up from the 16GB of HBM2 with 512GB/sec bandwidth offered on Vega 10. Not only that, but the Vega 20 chip will reportedly support the upcoming PCI-Express 4.0 standard.

VideoCardz.com does note that the details were from AMD's internal server roadmap, so things might change for the consumer market. 32GB of HBM2 would be insanely expensive, so I would dare say these are higher-end cards not meant for the consumer market. Alternatively, maybe we will see these cards in the consumer market, as AMD will be needing to prepare for NVIDIA's next-gen Volta architecture in 2017.

We all know NVIDIA is preparing a GeForce GTX 1080 Ti, but we don't know when the company will announce, or release it. New rumors have emerged, from an enthusiast who spotted the specifications of the GTX 1080 Ti on NVIDIA's own website.

NVIDIA's new GeForce GTX 1080 Ti's leaked specifications teased the GP102 GPU, the same chip that powers the new Titan X, with 3328 CUDA cores and 12GB of GDDR5 RAM. The big thing to note here is that the purported GTX 1080 Ti wouldn't be using the faster GDDR5X that the new Titan X and GTX 1080 use, but it'll rock the 384-bit memory bus that the Titan X uses.

Equipped with the 384-bit memory bus, the purported GTX 1080 Ti would have 384GB/sec of memory bandwidth - sitting in between the Titan X and its 480GB/sec, and the GTX 1080 with its 320GB/sec - the GTX 1080 uses a 256-bit memory bus, while the Titan X uses a 384-bit memory bus. The GPU would be clocked at 1503/1623MHz for base and boost clocks, respectively - with a 250W TDP.

Can you believe it has been an entire year since AMD announced it was forming Radeon Technologies Group, a newly-focused team that had a hungry new thirst for the GPU market.

AMD formed the Radeon Technologies Group with a bang late last year in Sonoma, California, where it announced the new Polaris architecture and a tease of a new card that was capable of 1080p 60FPS using less than 100W of power. In that time, AMD has launched the exciting new Radeon RX 480 and landing new semi-custom designs with Microsoft and Sony for the Xbox One S that just launched, the new PS4 Pro coming soon, and the upcoming 4K-capable Xbox Scorpio console in 2017.

Radeon Technologies Group boss Raja Koduri (which on a personal note, I'm going to add is a super chill guy in real life who I've enjoyed a few ice creams with in the last 12 months after our random 3am icecream with the RTG guys and girls at CES earlier this year in the freezing night of Las Vegas) said: "We managed to bring the focus back to graphics and turned around the erosion that happened over the last few years".

MSI is celebrating its 30th anniversary with the announcement of an exclusive limited edition graphics card, featuring a unique custom-designed EK waterblock and the GeForce GTX 1080 at its core.

The MSI GeForce GTX 1080 30th Anniversary graphics card features an exclusive closed loop liquid cooler from EK Water Blocks, with the newly designed EK-XLC Predator 120 All-In-On, with EK Water Blocks' R&D boss Niko Tivar explaining: "EK Water Blocks is very proud to be part of MSI 30 th Anniversary celebration and exclusively for this project, we engineered and designed EK-XLC Predator 120 All-In-On liquid cooler, that is not available anywhere else".

As you can expect, MSI has tweaked the GTX 1080 Anniversary graphics card, with the GPU clocked at 1721/1860MHz for base/boost clocks, respectively. The same 8GB of GDDR5X is found on the card, and cooled by the full cover EK Predator liquid cooler.

NVIDIA is positioning themselves for a huge show at CES 2017, teasing that it will have major announcements on gaming and VR products, as well as the usual development updates in AI and autonomous car technology.

NVIDIA CEO Jen-Hsun Huang will be on stage for a keynote at the Consumer Electronics Show in January, with NVIDIA teasing: "When Huang takes the stage, you can be sure he'll break news in some of the areas we're focused on: artificial intelligence, self-driving cars, virtual reality, and gaming".

What could NVIDIA unveil? We could expect a big reveal of their next-gen Volta architecture, but I think this is something we'll see at NVIDIA's own GPU Technology Conference later a few months after CES 2017, but what about VR and gaming? NVIDIA could unveil the GeForce GTX 1080 Ti, a new Shield, or the Shield VR - which I've been saying for years that NVIDIA is working on a Shield-branded VR headset.

Samsung has been using various Qualcomm processors in its Galaxy-branded smartphones over the years, slowly shifting its reliance to the Exynos processor, a chip that Samsung created, and manufactures itself.

There are now reports surfacing from SamMobile that tease a possible future where Samsung could be using GPU technology from either AMD or NVIDIA, which could make for an interesting upcoming change in the smartphone industry. NVIDIA has its impressive Pascal architecture on the 16nm FinFET process made by TSMC, but it has also just signed a deal for Samsung to manufacturer its second wave of Pascal GPUs on their 14nm FinFET process.

AMD has launched its new Polaris architecture made on the 14nm FinFET process over at GlobalFoundries, and have just had major semi-contract wins with the new Xbox One S and upcoming Xbox Scorpio consoles from Microsoft, as well as the new PS4 Slim and PS4 Pro consoles for Sony.

AMD's Radeon Software Crimson Edition 16.9.1 driver is available now, and with it comes some much appreciated game optimizations and bug fixes, as usual.

Deus Ex: Mankind Divided supports DirectX 12 today from the game developer end, and with this driver, from AMD's end, too, so you should have no trouble running the game with buttery smoothness.

The other game to get some TLC is DOTA 2, which has been bestowed with a DirectX 11 CrossFire profile.

AMD has the lower- and mid-range GPU market tied with its Radeon RX 460 and RX 470 graphics cards, but NVIDIA is looking to enter the mid-range GPU game with its Pascal architecture in the purported GeForce GTX 1050.

Reports surfaced over the weekend of a GP107-powered GeForce GTX 1050, with its GPU featuring 768 CUDA cores, and 4GB of GDDR5 on a 128-bit memory bus with 112GB/sec memory bandwidth. It looks like the GPU will be clocked at 1316MHz with a boost clock of 1752MHz.

NVIDIA's upcoming GeForce GTX 1050 will reportedly have a TDP of just 75W, so we could expect GTX 1050s to arrive without an additional PCIe power connector. As for the price, considering the GeForce GTX 1060 is priced at $299 for the Founders Edition, the GTX 1050 will need to be priced at $150-$200 to better compete against the Polaris-based offerings from AMD.

GDDR5X debuted on NVIDIA's GeForce GTX 1080 earlier this year, and then we saw 12GB of GDDR5X powering the super-powerful Pascal-based Titan X. Now we have GDDR6 being prepped for a debut sometime in 2018.

GDDR6 will increase the bandwidth to over 14Gbps, up from the already generous 10Gbps offered by GDDR5X, and up greatly from the now-current bandwidth of new GDDR5-based cards at 8Gbps - before then, it was 7Gbps for GDDR5. GDDR6 is also more power efficient, with it being around 20% more efficient over GDDR5.

I'd expect to see GDDR6 in the cards for AMD and NVIDIA for 2018, with NVIDIA set to use GDDR5X on its flagship graphics cards into 2017 alongside HBM2 on the upcoming Volta architecture. AMD has Vega planned for the first half of 2017, which will utilize HBM2 memory - but only if HBM2 supply isn't ridiculously expensive at the time, and is available in high volume.