Graphics Cards - Page 236

Micron has updated us on the status of GDDR5X, with the company receiving their first samples back from their fabrication plant, earlier than expected.

These samples are 100% operational at data rates passing 13Gbps, with the company still testing its GDDR5X production line, with GDDR5X samples being sent to partners in the spring (over the next few months). Right now, the 13Gbps-capable GDDR5X chips don't consume any more power than the current GDDR5 chips, so in certain scenarios, GDDR5X-based cards will actually use less power, when maximum VRAM bandwidth isn't required.

GDDR5 has an 8Gbps data rate, so the huge 13Gbps that GDDR5X has hit before it's even in volume production is a huge achievement by Micron. The company is hoping to reach 14Gbps, but hasn't said when this will happen. Micron explained on their blog: "Compared to GDDR5, these ultra-high data rates were achieved at an improved power consumption per transmitted bit due to VDD/VDDQ of 1.35V, while not gating maximum speed of the device. We also added features to improve system signal integrity: a new package with reduced ball pitch enables shorter PCB traces, which improves electrical performance".

While the Super Bowl 50 continues, we're seeing that AMD has been pushing its next-gen Polaris GPUs around - with a new Polaris-based flagship GPU, and the successor to the Fury X making an appearance.

The new C99 "FOC" (Full Operational Capacity) has been spotted on Zauba, with the C99 part being priced at $1700. This is an insane jump from the price of the previous Polaris-based C98 and C91/C92 chips. The C98 should be the successor to the R9 390 series, which will arrive as the new Polaris-based R9 490/490X. The C99 on the other hand could be the huge enthusiast Polaris, the successor to the Fury X, or some crazy dual-GPU beast.

The C98 board turned up on Zauba last month, and it holds a per unit value that is more than AMD's current Hawaii-based chips. The C91/C92 chips started shipping in January of last year, but the FOC designation was only attached in August 2015. The C91/C92 should be the lower-end Polaris-bases GPUs, which would most likely be the cards we were greeted to in Sonoma in December, and officially unveiled early January.

As promised, and a little quicker than expected perhaps, AMD's updated 16.1.1 driver has been released. This one includes all the fixes from yesterday, as well as a new one for flickering in Fallout 4, as seen post-version 1.3 and with Crossfire enabled.

If you missed the news yesterday, the driver includes Crossfire profiles for Rise of the Tomb Raider and Fallout 4, plus resolved issues for Radeon Settings and three high profile games (Battlefront, Assasin's Creed Syndicate, Battlefield Hardline). For all the details and the download, click here.

Now that Rise of the Tomb Raider is out, people are finding out just how useless their second GPUs are - except now, someone has found a fix for SLI users with multiple NVIDIA GeForce video cards.

3D Center's "Blaire" found some things to play around with in the NVIDIA Inspector Tool for SLI, which enables awesome SLI results. In order to do it, you'll need to search for Rise of the Tomb Raider's profile, and then change the SLI bits (DX11) to 0x080002F5. After that, click the magnifier icon that will reveal NVIDIA's Undefined options, and search for 0x00A0694B and change it to 0x00000001.

After you've done this, you've enabled full SLI support for Rise of the Tomb Raider, with DSO Gaming reporting that they've noticed 95% scaling on their GeForce GTX 690 - a damn good result for SLI scaling.

AMD was all systems go at VRLA last week, but during the VRLA Winter Expo keynote, the company teased its dual-GPU... the Radeon R9 Fury X2.

AMD's Roy Taylor said that the Radeon R9 Fury X2 has around 12 TFLOPS of SP, compared to the Radeon R9 295X2 which has 11.5 TFLOPS of SP compute performance. The big difference between the Fiji-based R9 Fury X2 and the Hawaii-based R9 295X2 is that the Fury X2 uses only 375W of power, compared to the R9 295X2 which would chew 500W. This means that the Fury X2 is around 40% more power efficient than the R9 295X2.

During his speech, Taylor said: "Last time I was here I also promised you that we would make the world's most powerful small computer for developers. We promised you we would take two of our highest end GPUs and put it inside that tiny box and if you go downstairs we actually have a demonstration of a dual GPU, 12 TeraFlops, fastest GPU solution in the world, inside of Tiki. It's a feat of engineering we are delighted with".

PC port specialist Nixxes Software has posted a tech support guide to Rise of the Tomb Raider on the Steam forums, just a few hours ahead of the game's launch. In the process, they've made it known AMD will be releasing a 16.1.1 driver "shortly", which they (Nixxes) recommends installing for the game.

It's possible 16.1.1 will include optimizations specifically for the game as is usually the case for big launches; NVIDIA did just this yesterday.

Asked for comment, AMD said, "We're working towards a hotfix that will have improvements for TR, but it's a work in progress. We can't comment on the timing or the details, but we'll keep you posted!"

NVIDIA has just launched its new GeForce GT 710 video card, a cheap $40 GPU that the company claims can deliver "up to 10x the performance of integrated graphics", and gaming up to "80% faster" than traditional iGPUs.

While the GeForce GT 710 gets beaten by Intel HD 530 integrated graphics found in Skylake chips, the card is an excellent option for budget gamers who have older rigs and want a cheap DirectX 12 card. As far as specs go, NVIDIA's inexpensive card uses DDR3 memory on a 64-bit bus with 14.4GB/s bandwidth, has a base clock of 954 MHz with 192 CUDA cores, and a memory clock speed of 1.8 Gbps.

The GeForce 710 supports a host of features including G-Sync adaptive sync, multi-monitor support for up to 3 displays, 3D Vision, NVIDIA's PhysX tech, and is OpenGL 4.5 and DirectX 12-ready out of the box. The card has a maximum resolution of 2560x1600 via HDMI and 2048x1536 on VGA, and features 1x Dual-Link DVI-D port, 1x HDMI and 1x VGA. It requires a minimum 300W power supply to function.

If you have been reading our GPU-related content, you should know that we are set for the biggest year in GPU history this year, from both sides: AMD and NVIDIA.

Well, at NVIDIA's GPU Technology Conference in April, we should see NVIDIA unveil the biggest GPU they've ever made - the successor to the GeForce GTX Titan X. The next-gen card could be called the GTX Titan X2, which would pull some of the wind out of AMD's sails with the dual-GPU Radeon R9 Fury X2, and we should see it featuring HBM2 - scaling up to 16-32GB with 1TB/sec of memory bandwidth. Insanity.

Back in September, we exclusively said that NVIDIA would release both a HBM2 and GDDR5X range of cards - something that will kick off with the HBM2-based Titan X successor. Towards June, we should see NVIDIA unveil a new GP104-based GeForce GTX 980 successor, based on GDDR5X - which offers 448GB/sec of memory bandwidth.

The last time we physically saw the dual-GPU version of the Fury X was at the launch event itself in Sydney, Australia - where we had our hands-on that beautiful PCB. But, the Radeon R9 Fury X2 has shown up again, this time at VRLA.

The VRLA expo was an event for all things virtual reality, held in LA last week. During the event, some of the HTC Vive demos were powered using the Radeon R9 Fury X2. Thanks to Facebook, we noticed Antal Tungler, PR Manager for AMD and all-round cool guy, posted on his Twitter account. He said: "Prototype Tiki from @FalconNW powering #htcvive with dual Fiji @AMDRadeon at the #vrla".

Someone asked Tungler "When you say "Dual Fiji" do you mean 2x Fiji cards, or 2x Fiji GPUs on 1 card? ;)", to which he replied with "One card". So we know that it wasn't 2 x R9 Fury X cards in the machine, but a single, dual-GPU beast. But with Polaris around the corner, I have to ask the question: where does the R9 Fury X2 fit in? It would only have 4GB of HBM1 per GPU, which really isn't enough VRAM considering it will be $1000+. VR headsets are pushing 90FPS, and a high-resolution to boot. I guess we'll see in the coming months, maybe AMD will launch the Fury X2 in between now and the release of Polaris in June/July.

The final specification for GDDR5X, the successor to GDDR5, has been decided, and though it doesn't allow for quite as much bandwidth as HBM or HBM2, though it's a technology that's a lot easier to implement than the latter, with fewer modifications needed to the GPU design to use.

GDDR5X allows for up to 14Gbps of total bandwidth and because it's based so heavily on its predecessor, it's pin compatible though highly internally revised in order to facilitate actual advancements in memory speed and bandwidth without making something entirely new. How JEDEC and Micron have done this is by increasing the prefetch by double, mandating the use of Phase Locked Loops and Delay Locked Loops as well as being able to transmit data at a rate that's quadruple the actual clock speed. In other words, it's fast. For comparison, GDDR5X running at the top-end 14Gpbs could potentially provide 448GBps of full bandwidth, which isn't too far off of the memory bandwidth of the R9 Fury X.

Micron, one of the leading manufacturers working on GDDR5X, estimates around a 10% power consumption decrease at the same VRAM size. VRAM sizes of 4Gb up to 16Gb can be used with the new specification. The reason for coming out with this new specification is to further address every segment of the market, especially those where HBM2 might not be economical, despite AMD's efforts to implement HBM in all segments of their GPUs. Now all GPUs can enjoy a healthy bandwidth increase for very little, if any, cost increase.

Now that HBM2 is beginning to flow into the market, thanks to Samsung making 4GB HBM2-based DRAM, NVIDIA is getting confident with Pascal - with the latest rumor stating that the company will unveil its next-gen GPUs in the first half of this year, with availability to follow in 2H 2016.

We know this will happen, where my sources tease that both AMD and NVIDIA will have next-gen GPUs prepared for June/July, but I've got a feeling NVIDIA will introduce a next-gen enthusiast GPU at their GPU Technology Conference in early April. NVIDIA is reportedly already playing around with the 16nm-based Pascal GPUs internally, but we should expect GDDR5X- and HBM2-powered offerings, with a GeForce GTX Titan X successor to be unveiled at GTC 2016. We might see the new Titan X with 16GB of HBM2, and possibly a professional-grade Tesla/Quadro GPU with 32GB of HBM2 teased, too.

As for the GP100, I don't think NVIDIA will unveil the GeForce GTX 980 Ti successor just yet as the GTX 980 Ti is still one of the best video cards you can buy. We should see a Titan X successor unveiled - powered by HBM2, followed by a successor to the GTX 980 - powered by GDDR5X. The HBM2-powered offerings will be able to pack 32GB of HBM2, and offer up to 1TB/sec of memory bandwidth, up from the 334GB/sec on the GeForce GTX 980 Ti and its GDDR5.

GPU's are fantastic tools for completing scientific computational work. They're effective at calculating math that's highly parallel, doing it far faster than any CPU could do alone. And now the GPU is going open-source with Binghamton University's new Nyami architecture that researchers have developed.

Timothy Miller and his colleagues have finally been able to test their open-source GPU design, called Nyami. It's essentially a GPGPU focused design that's borrowed a lot from Intel's Larrabee (Xeon Phi) while still being, at it's heart, a GPU.

This is the first open-source modifiable and synthesizable GPU made by anyone. The architecture has a measure of modularity so that any aspiring researcher or scientist can modify it to their hearts content, provided they have the expertise to do so, though. But really this is revolutionary because now software and hardware can reach a nexus and be developed with the help of the open-hardware community, which is a well supported community. The problem has always been getting an architecture started, which is a highly technical engineering problem. But now the first part is solved, and we might see some great scientific progress that could even spill over to consumer GPU's one day. Just keep in mind this isn't something you'll be playing Assassin's Creed Redundancy on.

There's a rumor floating around the Internet that seems to suggest that NVIDIA is very close to releasing their top-end mobile GPU, the GeForce GTX 980MX as well as the little brother, the GTX 970MX. Keep in mind that these are strictly mobile parts, and not related to the full-fat GTX 980 that's being stuffed into laptops.

There doesn't appear to be any actual source to confirm the imminent release, though they seem to be very adamant that NVIDIA is intent on releasing these high-end mobile parts soon. And these chips will be plenty fast and actually provide power efficiency that'll be necessary in thinner laptops.

The GTX 980MX is rumored to have 1664 CUDA cores, 104 texture units, 64 raster devices and a clock speed of up to 1048MHz on a 256-bit memory bus. This is slightly more CUDA cores than the slightly smaller part, the 980M. Oddly the TDP is only 25W less than the full-blow 980 laptop variant at 125W. That's still a lot of power, and you definitely wouldn't be gaming with a notebook powered by this monster without being tethered to the wall.

AMD's new Polaris chip might have just been caught on the Zauba import/export table, if we can truly believe what other sources have decoded while reading these manifests.

The price of per unit of this particualr "printed circuit board assembly for personal computer(video/ video card)" is such that it lines up with other AMD shipments in the past. It also indicate that the bigger Polaris die which we were able to see at CES, might be well into production. That's good news for us enthusiasts.

But then again, decoding the serial numbers and the entire manifest is very difficult, and even though some might claim that they know that these particular shipments are indeed for AMD and are a chip with a particular architecture, we don't actually know. But it is exciting that at least production appears to be marching on as you read this. Just don't forget the tablespoon of salt. And yeah, it's okay to be excited too. I know I am.

Whether you need to upgrade your outdated Radeon card or want to ensure your PC is ready for the incoming VR boom, we've found a duo of GPU deals that will fit the bill nicely. Today we have two different flavors of AMD's Radeon R9 390's discounted over at Newegg, both of which offer some impressive performances with 4K resolution support.

These sales are complimented by mail-in-rebates, which is par for the course for NewEgg sales. First up we have a PowerColor Radeon R9 390 for just $268 after a $20 mail-in-rebate (regular price $268), and the eggmen are tossing in a free $10 gift card to boot--but I don't think you can use the card for this purchase, only future purchases. The PowerColor R9 390 sports 1x HDMI, 1x Display Port and 2x DVI slots, and requires a 6-Pin / 8-Pin connectors with a 750W PSU.

Next up is an XFX Radeon R9 390 dropped down to $274 after a $30 MIR (original price $304). This card is similar to the PowerColor model, featuring a single HDMI and DisplayPort slot accompanied with two DVI ports. The power requirements are the same, with a minimum 750W PSU and the 6 and 8-pin PCIe power connectors.

During NVIDIA's GPU Technology Conference last year, NVIDIA unveiled its new NVLink interconnect that would find its BFF in their upcoming Pascal architecture.

At the time, we wrote that NVLink had 5x the bandwidth of PCIe 3.0, with NVLink opening up the possibilities for 8-way GPU setups - compared to the limit of 4-way SLI that we have now. Well, AMD is now talking about its upcoming next-gen coherent fabric, which will offer speeds of an insane 100GB/sec throughout multi-GPU setups. AMD has said that its new APUs will also be supported, with compute machines set to benefit greatly, too.

The big question for AMD is still in the air - but RTG boss Raja Koduri has said that he can't reveal if memory coherency and sharing between the GPUs and APUs will happen with the new interconnect. It would make sense to see it happen, but I'm sure AMD is rolling towards a big reveal in the near future with its Polaris architecture.

We are beginning to find out more and more about AMD's enthusiast-powered Polaris GPUs, thanks to a chat between RTG boss Raja Koduri, and VentureBeat.

During the chat, Koduri said: "We have two versions of these FinFET GPUs. Both are extremely power efficient. This is Polaris 10 and that's Polaris 11. In terms of what we've done at the high level, it's our most revolutionary jump in performance so far. We've redesigned many blocks in our cores. We've redesigned the main processor, a new geometry processor, a completely new fourth-generation Graphics Core Next with a very high increase in performance. We have new multimedia cores, a new display engine".

He added: "In summary, it's fourth generation Graphics Core Next. HDMI 2.0. It supports all the new 4K displays and TVs coming out with just plug and play. It supports display core 4.3, the latest specification. It's very exciting 4K support. We can do HAVC encode and decode at 4K on this chip. It'll be great for game streaming at high resolution, which gamers absolutely love. It takes no cycles away from games. You can record gameplay and still have an awesome frame rate. It'll be available in mid-2016".

Something I mused on about in September 2015 was that AMD and NVIDIA would use both HBM (and HBM2) as well as GDDR5, or GDDR5X on their next-gen cards throughout 2016. Well, AMD has confirmed this news.

Robert Hallock, the Technical Marketing Lead at AMD, explains: "We have the flexibility to use HBM or GDDR5 as costs require. Certain market segments are cost sensitive, GDDR5 can be used there. Higher-end market segments where more cost can be afforded, HBM is viable as well".

This will work out perfectly for AMD, as I think we'll see cards under $350 or so using GDDR5/GDDR5X while the HBM2-powered cards will arrive as enthusiast video cards, priced at $400 and above.

There were some fears when AMD split its GPU division into Radeon Technologies Group, but ATI 2.0 is really hitting the ground running with Polaris, and its continued commitment into VR. I think we'll see 'Big Polaris' powering VR, something we exclusively reported on last week.

During CES 2016 last week, RTG boss and cricket fan (yeah!) Raja Koduri spoke with VentureBeat about wanting to put the GPU division of AMD back into center focus. He said that 2016 is going to be a huge year for the company, as it hits its stride with the excited FinFET process, which will provide more performance per watt, HBM2 (which will provide 1TB/sec memory bandwidth) and the new Polaris architecture.

But when it came to the future of GPUs, Koduri said that VR will drive a large portion of this and that he won't be happy until we get 3D graphics that support 16K screens, and at 240Hz - yeah, 16K @ 240FPS. Just as Neo said in The Matrix: "whoa". Koduri says that this is when we'll reach the point of "true immersion that you won't be able to tell apart from the real-world". Koduri makes me excited about the future of not just RTG, but GPUs and VR, especially with quotes like that.

HBM2 is nearly here, with JEDEC releasing the official specification behind the fastest VRAM we've ever had on a video card, and this is incredibly exciting.

The official specification on HBM2 sees cards ramping up to 32GB, with up to 1TB/sec of memory bandwidth. This is a huge leap in the available VRAM over HBM1, which was limited to only 4GB on the Fiji-powered cards from AMD, including the super-small R9 Nano, Radeon R9 Fury and the R9 Fury X.

The power consumption of HBM2 will be lower than HBM1, with 8% saved over HBM1, on top of the 48% saved over GDDR5. The most important part of HBM is the amount of DRAM stacked, with HBM2 driving it up to 4/8 Hi Stacks (4/8GB) over the 4 Hi Stack of HBM1 (1GB total).