AMD CEO Lisa Su unveiled the company's next flagship graphics processor at an event on the sidelines of Computex, in Taipei. Arguably the biggest GPU package, measuring in at 50 mm x 50 mm, Fiji will instead reduce graphics card PCB size, because memory has been moved to the GPU package, with four 1024-bit HBM1 stacks surrounding the GPU die on the package. We detailed the chip's memory implementation at great lengths in this article. The first graphics cards based on this chip could be unveiled at E3, two weeks from now; and market launch could follow a week later.

Since March, we've been hearing whispers that AMD could give the topmost tier SKU based on its swanky new HBM-equipped "Fiji" silicon a fancy name, just as NVIDIA names its top-dog the GTX TITAN. That name could be the AMD Radeon FURY. A similar name to the brand that launched the erstwhile ATI, with its Rage series, Radeon FURY will be AMD's (and probably the industry's) fastest GPU, and will compete with NVIDIA's GeForce GTX TITAN X.

The card itself is quite diminutive, but that's because of two reasons - with memory being moved to the GPU package, a large amount of PCB real-estate is saved, and so the card can make do with a smaller PCB; and because the rear-end of the card is where the fittings for its AIO liquid-cooling solution are located. These tubes lead to a 120 x 120 mm radiator, with a single 120 mm PWM fan. Given that such a contraption could cool the dual-GPU R9 295X2, it should be effective with the Radeon FURY, just as well. The card will draw power from a pair of 8-pin PCIe power connectors. Display outputs will include three DP 1.2a and one HDMI 2.0. The brand naming indicates that AMD wants to change the terms on which its top-end product competes with NVIDIA's. Low noise and high-performance will be the focus, not power draw. Nobody buys an Aventador for its MPG.

In the latest picture leaked of AMD's upcoming flagship graphics card, codenamed "Fiji-XT," we get a final confirmation of the reference-design card's length, particularly its short PCB. Since this card uses a factory-fitted AIO liquid cooling solution, and since the Fiji XT package is effectively smaller than that of Hawaii, with the surrounding memory chips gone (moved to the GPU package as HBM stacks), the PCB is extremely compact, with just the GPU package, and its VRM. Speaking of which, the card draws power from a pair of 8-pin PCIe power connectors. The coolant tubes stick out from the rear of the card, making their way to a 120 x 120 mm radiator, with a single included 120 mm PWM fan. With this card, AMD is doing away with DVI altogether. Connectors will be a mixture of DisplayPort 1.2a and HDMI 2.0.

Back in 2008, when it looked like NVIDIA owned the GPU market, and AMD seemed lagging behind on the performance and efficiency game, the company sprung a surprise. The company's RV770 silicon, the first GPU to implement GDDR5 memory, trounced NVIDIA's big and inefficient GeForce GTX 200 series, and threw AMD back in the game. GDDR5 helped the company double the memory bandwidth, with lower pin- and memory-chip counts, letting the company and its partners build graphics cards with fewer components, and earn great margins, which the company invested in development of its even better HD 5000 series, that pushed NVIDIA with its comical GeForce GTX 480, to hit its lowest ever in market-share. Could AMD be looking at a similar turnaround this summer?

Since the introduction of its Graphics CoreNext architecture in 2012, AMD has been rather laxed in its product development cycle. The company has come out with a new high-end silicon every 18-24 months, and adopted a strategy of cascading re-branding. The introduction of each new high-end silicon would relegate the existing high-end silicon to the performance segment re-branded, and the existing performance-segment silicon to mid-range, re-branded. While the company could lay out its upcoming Radeon R9 series much in the same way, with the introduction of essentially just one new silicon, "Fiji," it could just prove enough for the company. Much like RV770, "Fiji" is about to bring something that could prove to be a very big feature to the consumer graphics market, stacked high-bandwidth memory (HBM).

AMD's upcoming Radeon R9 390X graphics cards will ship in two SKUs - an air-cooled one, with a moderately long reference design board (though not as long as the R9 290X), and a new Water-Cooled Edition (WCE) SKU, which will feature a very compact PCB - one that could be no bigger than that of the GeForce GTX 970 reference. This is possible because of AMD's HBM implementation. The 8 GB of memory on this card is present on the GPU package, as bare 3D-stacked DRAM dies, surrounding the GPU die, with an IHS covering everything; rather than the GPU package being surrounded by memory chips. Below is a mock-up of the card by ChipHell. It's not a picture. The radiator is off-proportions, the Radeon logo is misaligned, and the PCIe I/O is misaligned, etc. It should still give you a good idea of what the card looks like, particularly its length. Other specs on hand so far, include 4,096 GCN 1.2 stream processors, 256 TMUs, 128 ROPs, and a 4096-bit wide HBM interface, which at 1.25 GHz memory clock, will offer memory bandwidth of 640 GB/s.

While Fiji package will be bigger than that of, say, "Hawaii," overall the setup is more space-efficient, and conserves PCB real-estate. The PCB hence only has the GPU package and the VRM. AMD is doing away with the DVI connector on its reference PCB. It will only feature three DisplayPort 1.2a and one HDMI 2.0a. The WCE variant will feature a pump+block covering the GPU package, which will come factory-fitted to a 120 x 120 mm radiator. The air-cooled R9 390X will be longer, but only to house a heatsink and lateral blower. The single-GPU card could offer performance comparable to the dual-GPU R9 295X2, which is faster than the GeForce GTX TITAN-X. AMD CEO Lisa Su, speaking at the Investor Day event, in New York, on 6th May, hinted that the product could launch on the sidelines of either Computex 2015 (early-June) or E3 (mid-June).Image Courtesy: ChipHell. Many Thanks to GhostRyder for the tip.

At its ongoing Investor Day presentation, AMD announced that will continue to make GPUs for every segment of the market. The company is planning to leverage improvements to its Graphics CoreNext architecture for the foreseeable future, but is betting on a huge performance/Watt increase with its 2016 GPUs. The secret sauce here will be the shift to 14 nm FinFET process. It's important to note here, that AMD refrained from mentioning "14 nm," but the mention of FinFET is a reliable giveaway. AMD is expecting a 2x (100%) gain in performance/Watt over its current generation of GPUs, with the shift.

AMD's future GPUs will focus on several market inflection points, such as the arrival of CPU-efficient graphics APIs such as DirectX 12 and Vulkan, Windows 10 pulling users from Windows 7, 4K Ultra HD displays getting more affordable (perhaps even mainstream), which it believes will help it sell enough GPUs to return to profitability. The company also announced an unnamed major design win, which will take shape in this quarter, and which will hit the markets in 2016.

Earlier this day, HP announced its newest line of desktop PCs, one of which comes with a curious-sounding Radeon R9 380 graphics card. HP's product pages for its new desktops aren't active, yet, leaving us to only speculate on what the R9 380 could be. One theory making rounds says that the R9 380 could either be a re-branded R9 285, or be based on its "Tonga" silicon, which physically features 2,048 stream processors based on Graphics CoreNext (GCN) 1.2 architecture, and a 384-bit wide GDDR5 memory interface. Another theory states that the R9 380 could be an OEM-only re-brand of the R9 280 or R9 280X, based on the 3+ year old "Tahiti" silicon.

The former theory sounds more plausible, because re-branding a "Tahiti" based product would be suicidal for AMD. Although based on GCN, "Tahiti" lacks a lot of architecture features introduced with "Hawaii" and "Tonga." AMD practically stopped optimizing games for "Tahiti," and some of its new features, such as FreeSync and XDMA CrossFire, can't be implemented on it. "Tonga," on the other hand, supports both these features, and one can create an SKU with all its 2,048 stream processors, and its full 384-bit GDDR5 memory interface unlocked. If the R9 380 is indeed an OEM-only product, then it's likely that the company's retail-channel products could be branded in the succeeding R9 400 series. GPU makers tend to re-brand and bump their SKUs by a series for OEMs to peddle in their "new" products at short notice.

AMD's next-generation GPU family, which it plans to launch some time in 2016, codenamed "Arctic Islands," will see the company skip the 20 nanometer silicon fab process from 28 nm, and jump straight to 14 nm FinFET. Whether the company will stick with TSMC, which is seeing crippling hurdles to implement its 20 nm node for GPU vendors; or hire a new fab, remains to be seen. Intel and Samsung are currently the only fabs with 14 nm nodes that have attained production capacity. Intel is manufacturing its Core "Broadwell" CPUs, while Samsung is manufacturing its Exynos 7 (refresh) SoCs. Intel's joint-venture with Micron Technology, IMFlash, is manufacturing NAND flash chips on 14 nm.

Named after islands in the Arctic circle, and a possible hint at the low TDP of the chips, benefiting from 14 nm, "Arctic Islands" will be led by "Greenland," a large GPU that will implement the company's most advanced stream processor design, and implement HBM2 memory, which offers 57% higher memory bandwidth at just 48% the power consumption of GDDR5. Korean memory manufacturer SK Hynix is ready with its HBM2 chip designs.

AMD's upcoming "Zen" architecture will see a major change in the way the company designs its CPU cores. It will be a departure from the "module" core design introduced with "Bulldozer," in which two cores with shared resources constitute the indivisible unit of a multi-core processor. A "Zen" core will have dedicated resources in a way things used to be before "Bulldozer," and only the last-level cache (L3 cache), will be shared between cores. "Zen" will also implement SMT, much in the same way as Intel processors do, with HyperThreading Technology.

The first implementation of "Zen" will be an insanely powerful APU (on paper anyway), featuring 16 physical "Zen" CPU cores, 32 logical CPUs enabled with SMT, 512 KB dedicated L2 cache per core, and 32 MB of shared L3 cache. The CPU's ISA instruction set will see a spring-cleaning, with the removal of underused instruction-sets, and the introduction of new ones. Other features on this APU are equally surprising - a quad-channel DDR4 integrated memory controller, a separate HBM (high-bandwidth memory) controller dedicated to the integrated graphics, with up to 512 GB/s bandwidth, and an integrated graphics core featuring "Greenland-class" stream processors. Given that AMD is able to build 7-billion transistor GPUs on existing 28 nm processes, building an APU with these chops doesn't sound far-fetched. The company could still have to rely on a newer fab.

NVIDIA reportedly adjusted launch of the GeForce GTX 980 Ti, to after Summer, 2015. The company could be using the time to let the market digest existing inventories of the GTX 980 (and avoid the repeat of slow sales on its GTX 770), particularly in the early-Summer season, when PC enthusiasts and gamers tend to upgrade or build afresh. The time could also be spent to watch what AMD comes up with, for its Radeon R9 390X.

The R9 390X, scheduled for a June-July launch, is based on a silicon that looks competitive with the GM200 on paper, and introduces a few new features, such as high-bandwidth memory (HBM). The GTX 980 Ti could feature an identical core-configuration to the GTX TITAN X, but feature half the memory amount at 6 GB, different clock speeds, and freedom for add-in card (AIC) partners to innovate custom-design cards.

AMD is preparing to time the launch of its next-generation Radeon R9 300 series with that of Computex 2015, in early June. The company had earlier planned to launch some products that are essentially price-adjusted rebrands of existing ones, such as the R9 380 series (being rebrands of R9 290 series on a slightly improved silicon), and the R9 370 series (being based on the "Tonga" silicon); but has decided to launch the two along with its flagship R9 390 series, based on a brand new silicon, around the same time. AMD's answer to the GTX TITAN-X from NVIDIA, the R9 390X will feature around 4,096 stream processors based on the Graphics CoreNext 1.3 architecture, and will implement an HBM (high-bandwidth memory) interface, with bandwidths in excess of 600 GB/s.

AMD's upcoming Radeon R9 380X and R9 380 graphics cards, with which it wants to immediately address the GTX 980 and GTX 970, will be based on a "new" silicon codenamed "Grenada." Built on the 28 nm silicon fab process, Grenada will be a refined variant of "Hawaii," much in the same way as "Curacao" was of "Pitcairn," in the previous generation.

The Grenada silicon will have the same specs as Hawaii - 2,816 GCN stream processors, 176 TMUs, 64 ROPs, and a 512-bit wide GDDR5 memory interface, holding 4 GB memory. Refinements in the silicon over Hawaii could allow AMD to increase clock speeds, to outperform the GTX 980 and GTX 970. We don't expect the chip to be any more energy efficient at its final clocks, than Hawaii. AMD's design focus appears to be performance. AMD could save itself the embarrassment of a loud reference design cooler, by throwing the chip up for quiet custom-design cooling solutions from AIB (add-in board) partners from day-one.

In its an investor conference-call following its Q4-2014 and FY-2014 results, AMD stated that it will release new GPU and APU products starting Q2-2015, or only after March. "Going into the second quarter and the second half of the year with our new product launches, I think we feel very good about where we are positioned there," said Lisa Su, chief executive officer.

Q2-2015 will start off with the company's "Carrizo" line of all-in-one and notebook APUs. These chips will integrate the company's new "Excavator" CPU cores, with an integrated graphics core based on Graphics CoreNext 1.2 architecture (the same one AMD built its "Tonga" GPU on). Around the same time, AMD will launch new Opteron "Seattle" enterprise CPUs, which integrate up to eight ARM Cortex A-57 64-bit cores, targeting the ultra-dense server market. In Q2-2015, AMD will launch its latest Radeon Rx 300 series graphics processors. Its performance-segment part, the R9 380, will feature 4,096 GCN 1.2 cores, double that of its predecessor, and 4 GB of stacked HBM (high-bandwidth memory). Its mid-range chip, codenamed "Trinidad" will succeed "Curacao," and offer performance competitive to the $200-ish price-point.

AMD's next flagship single-GPU graphics card, codenamed "Fiji," could feature High-Bandwidth Memory (HBM). The technology allows for increased memory bandwidth using stacked DRAM, while reducing the pin-count of the GPU, needed to achieve that bandwidth, possibly reducing die-size and TDP. Despite this, "Fiji" could feature TDP hovering the 300W mark, because AMD will cram in all the pixel-crunching muscle it can, at the expense of efficiency from other components, such as memory. AMD is expected to launch new GPUs in 2015, despite slow progress from foundry partner TSMC to introduce newer silicon fabs; as the company's lineup is fast losing competitiveness to NVIDIA's GeForce "Maxwell" family.

AMD is reportedly working on a new single-GPU graphics card SKU to compete with the likes of GeForce GTX 780 Ti, and perhaps even take a swing at the GTX TITAN Black, since it's not too far ahead of the GTX 780 Ti at single-display gaming. The new SKU will be more than just a clock-speed bump, it will leverage HBM (high-bandwidth memory), a cutting-edge new technology that relies on stacking multiple DRAM dies with dedicated memory paths into a single package, cutting down on power-draw, thermals, and PCB real-estate.

The first kind of HBMs to hit the market are 8 Gbit 4Hi, which will interface with the GPU over a 128-bit wide path, which means there will be just four memory packages on the card (since the "Hawaii" silicon features a 512-bit wide memory bus), with improvements in the area of power-draw and heat output. The memory could be clocked higher, too. Sadly, memory bandwidth is not the prime-mover in VGA performance, and AMD will have to offer higher GPU clocks for the card to stand a chance against NVIDIA's high-end single-GPU offerings. As for the name this card would bear, we get a sense of deja vu about how NVIDIA launched the single-GPU GeForce 7950 GTX alongside the dual-GPU 7950 GX2, replacing the 7900 GTX. And so, it wouldn't surprise us if AMD named it Radeon R9 295X.