Video Cards & GPUs - Page 398

Now here's something interesting that's going around the Internet just before the New Year hits in North America. Several sources are claiming that the next generation architecture from AMD, previously thought to be codenamed Arctic Islands, might actually be called "Polaris".

The sources of the supposed leak, HWBattle and WCCFtech are citing the information based off of a photo that doesn't appear to have any context whatsoever to corroborate the evidence. WCCFtech is also reporting that the Polaris name was tweeted about just before the RTG Summit was held by AMD.

The photo, however, has recently been taken down from HWBattle at the request from the leaker of the image and in its place is a cautionary statement from slides that are no longer under embargo from AMD that references the codename "Polaris" as well. The tweet above also specifically refers to Polaris being 2.5x brighter today than it was in the time it was first observed. That number might have significance. I can say that we'll just have to wait and see. Perhaps more information will be revealed at CES. I'd take the above information with a huge pile of salt, however. The information could very well be fabricated and not necessarily correlative to anything.

After promising that their dual GPU video card would be released in late 2015, AMD has delayed its Radeon R9 Fury X2. The reason? Because they expected the Oculus Rift and HTC Vive to be out in late 2015, but both VR headsets were delayed into early next year, forcing AMD to delay its dual-GPU efforts.

Hardware.fr talked with AMD, which said: "The product schedule for Fiji Gemini had initially been aligned with consumer HMD (Head Mounted Display) availability, which had been scheduled for Q415 back in June. Due to some delays in overall VR (Virtual Reality) ecosystem readiness, HMDs are now expected to be available to consumers by early Q216. To ensure the optimal VR experience, we're adjusting the Fiji Gemini launch schedule to better align with the market". AMD continued: "Working samples of Fiji Gemini have shipped to a variety of B2B customers in Q415, and initial customer reaction has been very positive".

AMD has reportedly sent out key people their new dual Fiji card, with these lucky people including DICE god Johan Andersson. We should expect AMD to show off the "Gemini" dual-GPU card at CES hopefully, but I think we're going to see an April-June launch - possibly at Computex in early June 2016.

An industry insider has indicated to the Korea Electronic Times that Samsung will be mass producing Greenland graphics chips for AMD starting in the second quarter of 2016. This follows the recent news GlobalFoundries will be producing for AMD, so it has two partners on its hands as of now.

Word is AMD will be "fluidly regulating" how many chips are produced by each company based upon yield and other factors.

It's said Greenland, which is based on the 14nm FinFET process technology, is twice as efficient as the 28nm Fiji. It's also said that shortly after Greenland begins production, Zen production will begin, and will also see production split amongst Samsung and GlobalFoundries facilities.

It looks like AMD's next-gen GPU will be arriving in Q2 2016, which is the rough timeframe that we expected from all the rumors over the last few months.

AMD's new part will be made on the 14nm process by Samsung and GlobalFoundries, with it reaching mass production at the end of Q2 2016 - or in June. This is perfect timing for Computex 2016, as we saw the rumors - and the TweakTown exclusive during Computex 2015 that the new card would be called the Radeon R9 Fury X, and not the Radeon R9 490X that the world thought.

The new GPU will feature HBM2, which will not only increase the amount of VRAM that the new GPU will have - up from the 4GB of HBM1 found on the current Fiji cards like the Fury X and R9 Nano - but it will also see an increase in memory bandwidth. HBM1 allows for 512GB/sec, while HBM2 drives the memory bandwidth numbers through the roof to over 1TB/sec. We should expect to see cards from AMD featuring up to 18 billion transistors, and up to 16GB and even 32GB on the professional cards in 2016.

Today Taiwanese PC hardware maker BioStar has launched its new NVIDIA GeForce GTX 750 Ti OC video card aimed at gamers on a budget.

BioStar's new GeForce GTX 750 Ti OC GPU is based on NVIDIA's Maxwell architecture and sports 2GB GDDR5 VRAM on a 128-bit bus. The card's main selling point is its new OC revision: the GPU's 640 CUDA cores are factory overclocked to 1127Mhz and can be further pushed to 1178Mhz speeds to boost frame rates and performance for DirectX 11 gaming. As far as a cooling solution, BioStar's GTX 750 Ti OC revision features the company's GAMING FPS dual-fan setup that's optimized to mitigate heat during overclocking or normal use.

The card can hit a max resolution of 4096 X 2160 via digital output, but doesn't come equipped with a Display Port outlet, instead featuring Dual-DVI ports and a single mini-HDMI. BioStar's GeForce GTX 750 Ti OC can perform alongside other NVIDIA GPU's in SLI, and is specifically designed to complement BioStar-branded motherboards.

SAPPHIRE Is preparing a new Radeon R9 Fury Nitro card, something that will be a slight step up on its R9 Fury Tri-X. SAPPHIRE's new R9 Fury Nitro will feature a custom PCB, and so much more.

The Fiji PRO GPU will reportedly be clocked at 1050MHz, with the card being a little longer than the current R9 Fury Tri-X model - the Nitro measuring in at 307 x 125 x 45mm, while the Tri-X model measured in at 300 x 110 x 400mm. SAPPHIRE will allowing the R9 Fury Nitro to expand a little wider too, moving to a 2.5-slot card, versus a 2.2-slot card from the Tri-X model. SAPPHIRE's new Radeon R9 Fury Nitro will still use 2 x 8-pin PCIe power connectors, but the custom PCB can be seen as significantly larger than the Tri-X model - so expect some higher power draw on the Nitro.

This means we can hope for some decent overclocking on the SAPPHIRE R9 Fury Nitro, and if SAPPHIRE has been provided approval by AMD for overclocking the Fiji GPU, that means other AIB partners can't be too far behind. The R9 Fury Nitro will still rock the same 3584 stream processors, 4GB of HBM1 on a 4096-bit memory bus providing 512GB/sec of memory bandwidth. The 2 x 8-pin PCIe power connectors will require 275W of power. The R9 Fury Nitro will include 3 x DisplayPort, 1 x HDMI and 1 X DVI-D (compared to the Tri-X which had the same display output configuration, minus DVI). As for pricing, that's TBD for now.

It looks like AMD's add-in board (AIB) partners are preparing new Radeon R9 390 video cards with 4GB of VRAM, compared to the 8GB of framebuffer on the current R9 390 models.

The new Radeon R9 390 4GB models will feature the Grenada Pro GPU, which is a revised version of the Hawaii Pro GPU on the 28nm process. This means we'll see 2560 stream processors, 160 TMUs, 64 ROPs, and a GPU clock speed of 1GHz. The 4GB of GDDR5 will be clocked at 1.5GHz (6GHz) with a 384-bit memory bus. The original R9 390 8GB models were aimed at 4K gaming, but the new 4GB models should hit the 1080p/2K gaming market, with more than enough horsepower to drive 60FPS at both resolutions.

The Radeon R9 390 8GB models still sell for around $329, while the new 4GB models will fall into the $299 price, or lower. This means that the R9 390 will be priced right against NVIDIA's GeForce GTX 970. As for the models, SAPPHIRE is preparing its Radeon R9 390 4GB Dual-X OC card, XFX has its Radeon R9 390 4GB Black Edition OC, PowerColor with its R9 390 4GB OC Edition and more. On average, the new R9 390 4GB model is around 10% faster than the Radeon R9 290, so if you're upgrading your old R9 290, now could be the time.

A couple of days ago we wrote a story about GDDR6 being teased, and that it could've been a rebranded version of GDDR5X... well, that was wrong. Micron reached out to us clarifying it all, saying that "the new memory advancements coming from Micron in 2016 are going to be called GDDR5X, not GDDR6".

They continued, adding that "GDDR5X and GDDR6 are not the same product and Micron has not announced any plans involving GDDR6". This fully clears things up for not only us, but the world. We will see GDDR5X released onto video cards next year, allowing for much higher memory bandwidth - and while it won't be high as HBM1 or HBM2, it will be significantly higher than the normal GDDR5 we have today.

AMD's big Crimson Edition driver launched a few weeks ago with many major changes and feature additions. As is often the case with significant releases such as this, there have been problems. First there was the fan speed issue, and now reports have surfaced about an underclocking issue caused by the driver.

Asked about the issue on Twitter, AMD yesterday stated it's aware of it and investigating. The fan issue was resolved very quickly with a hotfix driver, so the same is likely here. In the meantime, if you find yourself experiencing it and/or stutter in games, AMD says you can get around it by reverting to Catalyst 15.11.1 Beta or 15.7.1 WHQL drivers.

We heard about NVIDIA using GDDR5X on its next-gen Pascal-based offerings, and shortly after it was rumored that AMD would also use GDDR5X on their forthcoming 14nm-based parts. A new report has surfaced about GDDR6 arriving in 2016, which could be a rebranded name for GDDR5X - before it was even allowed to be released.

It makes sense though, as some users might think an 'X' isn't enough of an indication of increased performance between GDDR5 and GDDR5X, with a full number increase from GDDR5 to GDDR6 being a much better marketing strategy. The new GDDR6 is under development at Micron, but we don't know anything on the performance or specification side of things. We do know that the jump from GDDR5 to GDDR5X was actually quite significant, and while not quite as revolutionary as HBM, it was a great jump for cards not based on HBM.

Considering GDDR5 debuted on the 60nm process all the way back in 2007, it has matured greatly since then, where it is down to 20nm on current cards - including the GeForce GTX Titan X. I think GDDR5 and GDDR6 could be used starting next year in cards that are under $400, allowing the rest of the market above $400 to use the much more expensive HBM2 technology.