VLSI technical publication Chipworks posted the first clear die-shot of AMD's "Fiji" silicon, revealing intricate details of the most technically advanced GPU. What makes Fiji the most advanced graphics chip is its silicon interposer and stacked HBM chips making up a multi-chip module. It's the die in the center of all that, which went under Chipworks' microscope.

The die-shot reveals a component layout that's more or less an upscale of "Tonga." Some of the components, such as the front-end appear to be entirely identical to "Tahiti" or "Tonga." The shot reveals the 64 GCN compute units arranged in four rows, on either side of the central portion with the dispatch and primitive setup pipelines. The pad-area of the on-die memory controllers appear to be less than the large memory I/O pads that made up the 384-bit interface of "Tahiti." The first picture below is the die-shot of "Fiji," followed by a color-coded die-shot of "Tahiti."

GIGABYTE finally got around to launching a Radeon R9 Fury graphics card (model: GV-R9FURYWF3OC-4GD). The company's air-cooled R9 Fury is cooled by a triple-fan WindForce 3X cooling solution, similar to the one featured on the company's GeForce GTX 980 Ti offerings. The card appears to be based on a full-length custom-design PCB by GIGABYTE, and doesn't reuse AMD's short reference-design board. This board has standard height, and the cooler features a thicker heatsink, making it a triple-slot solution. A back-plate is included.

GIGABYTE's Radeon R9 Fury WindForce comes with a slight factory-overclock of 1010 MHz core (compared to 1000 MHz reference), while the memory clock stays unchanged, at 500 MHz. The card draws power from a pair of 8-pin PCIe power connectors. Display outputs include one each of dual-link DVI, HDMI 1.4a, and three DisplayPort 1.2a connectors. There's no word on the pricing or availability, but expect GIGABYTE to charge a tiny premium over the SKU's recommended price of $550.

AMD's dual-GPU graphics card based on a pair of "Fiji" GPUs could be named the Radeon R9 Gemini, after the Greek-Roman mythological figures and twins Castor and Pollux, sometimes referred to as the "Gemini twins." This is consistent with AMD naming its single-GPU flagship product based on this chip after Furies. The term "Gemini" as associated by AMD to describe this product first came to light when an engineering sample making its way through India was parsed through the country's overly transparent customs, where its shipping manifest read "FIJIGEMINI." The same description also mentions that the sample had its cooling solution installed, so it can be assumed that the prototype AMD CEO Lisa Su held up at the company's "Fiji" silicon unveil event has matured into a mass-producible product.

NVIDIA's next-generation GPUs, based on the company's "Pascal" architecture, will be reportedly built on the 16 nanometer FinFET node at TSMC, and not the previously reported 14 nm FinFET node at Samsung. Talks of foundry partnership between NVIDIA and Samsung didn't succeed, and the GPU maker decided to revert to TSMC. The "Pascal" family of GPUs will see NVIDIA adopt HBM2 (high-bandwidth memory 2), with stacked DRAM chips sitting alongside the GPU die, on a multi-chip module, similar to AMD's pioneering "Fiji" GPU. Rival AMD, on the other hand, could build its next-generation GCNxt GPUs on 14 nm FinFET process being refined by GlobalFoundries.

AMD continues to push the boundaries of graphics card design, today announcing its category-creating AMD Radeon R9 Nano, the fastest Mini ITX graphics card ever to enable 4K gaming in the living room through ultra-quiet, ultra-compact PC designs. First previewed to gamers around the world during the Electronic Entertainment Expo (E3) in Los Angeles in June 2015, the AMD Radeon R9 Nano graphics card is based on the graphics chip codenamed "Fiji," and is the third "Fiji"-based product to launch this summer alongside the AMD Radeon R9 Fury and R9 Fury X graphics cards. The AMD Radeon R9 Fury graphics family, based on the "Fiji" chip, marks a turning point in PC gaming with the implementation of High-Bandwidth Memory (HBM) to deliver extreme energy efficiency and performance for ultra-high resolutions, unparalleled VR experiences, smoother gameplay, with the Radeon R9 Nano revolutionizing form-factors for enthusiasts everywhere.

With 30 percent more performance and 30 percent lower power than the previous generation AMD Radeon R9 290X card, the 175W AMD Radeon R9 Nano is the world's most power efficient Mini ITX enthusiast graphics card. The six-inch long, air-cooled board represents a new class of graphics card, enabling gamers, PC modders, and system integrators to build compact, unique, ultra-small form factors that have never before been possible, opening the door to new, sleek PC designs that are no bigger than a home DVR or videogame console, and look every bit in place beside them.

AMD's upcoming super-compact graphics card, the Radeon R9 Nano, will be faster than NVIDIA's GeForce GTX 980, and a whopping 30% faster than the GTX 970, according to the company. At its size, it will offer the fastest pixel-crunching solution for compact ITX/SFF gaming PC builders, and that is something AMD want to capitalize on. If what we're hearing is true, then not only will the R9 Nano have the same core-config as the R9 Fury X, but also its price - US $649.99. At this price, the R9 Nano definitely isn't going to affect sales of the GTX 970 or GTX 980, which are currently going for as low as $299 and $465, respectively; but serve as a "halo product," targeted at SFF gaming PC builders.

AMD's upcoming mini-ITX friendly graphics card, the Radeon R9 Nano, which boasts of a typical board power of just 175W, is not a heavily stripped-down R9 Fury X, as was expected. The card will feature the full complement of GCN compute units physically present on the "Fiji" silicon, and in terms of specifications, is better loaded than even the R9 Fury. Specifications sheet of the R9 Nano leaked to the web, revealing that the card will feature all 4,096 stream processors physically present on the chip, along with 256 TMUs, and 64 ROPs. It will feature 4 GB of memory across the chip's 4096-bit HBM interface.

In terms of clock speeds, the R9 Nano isn't too far behind the R9 Fury X on paper - its core is clocked up to 1000 MHz, with its memory ticking at 500 MHz (512 GB/s). So how does it get down to 175W typical board power, from the 275W of the R9 Fury X? It's theorized that AMD could be using an aggressive power/temperature based clock-speed throttle. The resulting performance is 5-10% higher than the Radeon R9 290X, while never breaching a power target. Korean tech blog DGLee posted pictures of an R9 Nano taken apart. Its PCB is smaller than even that of the R9 Fury X, and makes do with a slimmer 4+2 phase VRM, than the 6+2 phase VRM found on the R9 Fury X.

Top technologists from AMD are detailing the engineering accomplishments behind the performance and energy efficiency of the new high-performance Accelerated Processing Unit (APU), codenamed "Carrizo," and the new AMD Radeon R9 Fury family of GPUs, codenamed "Fiji," at the prestigious annual Hot Chips symposium starting today. The presentations will focus on new details of the high-definition video and graphics processing engines on the 6th Generation AMD A-Series APU ("Carrizo"), and the eight year journey leading to die-stacking technology and all-new memory architecture included on the latest top-of-the-line AMD Radeon Fury Series GPUs ("Fiji") for 4K gaming and VR. Using a true System-on-Chip (SoC) design, 6th Generation AMD A-Series processors are designed to reduce the power consumed by the x86 cores alone by 40 percent, while providing substantial gains in CPU, graphics, and multimedia performance versus the prior generation APU. The new AMD Radeon R9 Fury X GPU achieves up to 1.5x the performance-per-watt of the previous high-end GPU from AMD.

"With our new generation of APU and GPU technology, our engineering teams left no stone unturned for performance and energy efficiency," said Mark Papermaster, chief technology officer at AMD. "Using innovative design for our APUs, we've vastly increased the number of transistors on-chip to increase functionality and performance, implemented advanced power management, and completed the hardware implementation of Heterogeneous System Architecture. For our latest GPUs, AMD is the first to introduce breakthrough technology in the form of die-stacking and High-Bandwidth Memory. The results are great products with very large generational performance-per-watt gains."

AMD's ultra-compact graphics card based on its "Fiji" silicon, the Radeon R9 Nano (or R9 Fury-Nano), is nearing its late-August/early-September launch. At its most recent "Hot Chips" presentation, AMD put out more interesting numbers related to the card. To begin with, it lives up to the promise of being faster than the R9 290X, at nearly half its power draw. The R9 Nano has 90% higher performance/Watt over the R9 290X. More importantly, it has about 50% higher performance/Watt over the company's current flagship single-GPU product, the Radeon R9 Fury X. With these performance figures, the R9 Nano will be targeted at compact gaming-PC builds that are capable of 1440p gaming.

This is big. A new software is doing rounds on the forumscape, with lets you check exactly which GCN compute units (CUs) were disabled by AMD to carve out your "Pro" SKU graphics card (eg: R9 Fury non-X, R9 390, R9 285, etc.), and if you're lucky, re-enable some of those disabled CUs via a good old fashioned soft-mod. Called CUINFO, developed by OCN Forums member tx12, the tool has seen some success in unlocking disabled CUs in graphics cards based on "Fiji," "Hawaii," and "Tonga" chips.

The success is limited probably because AMD appears to be using two methods to disable CUs - laser-cutting them, and through firmware. Even in chips that let you unlock, you may not unlock all CUs, since some are genuinely damaged and disabled as part of the harvesting process, to maximize yield. There are no telltale signs of which chips let you unlock and which don't, and so you'll be playing a lottery. Cards with re-enabled CUs have shown increased performance, confirming that the soft-mod is real.

AMD's upcoming disruptive performance-segment graphics card, the Radeon R9 Nano, could be arriving sooner than its late-Summer expected launch. One of AMD's promotional heads Anthony "Elmy" Lackey posted two pictures of the card on his Flickr page, which reiterates just how compact the thing is. AMD earlier announced that the R9 Nano will be faster than the Radeon R9 290X, with typical board power well under 190W, making it an exciting product to look forward to. The R9 Nano will be based on the same "Fiji" silicon, which powers the R9 Fury X and R9 Fury. AMD could make a major announcement related to this product very soon, given how Elmy promised to release a few details next week.

AMD published a paper with the IEEE for a new high-density computing device concept, which it calls the Exascale Heterogenous Processor or (EHP). It may be a similar acronym to APU (accelerated processing unit), but is both similar and different to it in many ways, which make it suitable for high-density supercomputing nodes. The EHP is a chip that has quite a bit in common with the recently launched "Fiji" GPU, that drives the company's flagship Radeon R9 Fury X graphics card.

The EHP is a combination of a main die, housing a large number of CPU cores, a large GPGPU unit, and an interposer, which connects the main die to 32 GB of HBM2 memory that's on-package, and is used as both main-memory and memory for the integrated GPGPU unit, without memory partitioning, using hUMA (heterogeneous unified memory access). The CPU component consists of 32 cores likely based on the "Zen" micro-architecture, using eight "Zen" quad-core subunits. There's no word on the CU (compute unit) count of the GPGPU core. The EHP in itself will be highly scalable. AMD hopes to get a working sample of this chip out by 2016-17.

AMD announced the second graphics card based on its swanky new "Fiji" silicon, the Radeon R9 Fury. Positioned between the R9 390X and the R9 Fury X, this card offers higher pixel-crunching muscle than the R9 390X, while giving you cutting-edge 4 GB HBM memory. It can play any game at 2560 x 1440, and at Ultra HD (3840 x 2160), with reasonable eye-candy. The R9 Fury is designed solely for AMD's AIB partners to come up with their own air-cooled products.

AMD carved the R9 Fury out of the Fiji silicon, by enabling 56 of the 64 GCN compute units physically present, yielding 3,584 stream processors. Other specifications include 224 TMUs, 64 ROPs, and 4 GB of memory across a 4096-bit wide HBM interface. The core is clocked at 1000 MHz, and the memory at 500 MHz (512 GB/s). Custom-design boards will offer factory-overclocked speeds. AMD is pricing the R9 Fury at US $549.

Here are some of the first pictures of ASUS Radeon R9 Fury STRIX, detailed in no less than AMD's own [leaked] press-deck for the R9 Fury. It appears that only two AIB partners are going to launch the R9 Fury for whatever reason. These are the Sapphire, with its R9 Fury Tri-X card, and ASUS, with its R9 Fury STRIX. ASUS' card features the same new-generation triple-fan DirectCU III cooling solution that made its debut with the GTX 980 Ti STRIX, and is featured on the R9 390X STRIX. This cooler is mated to what appears to be the first custom-design PCB for AMD's "Fiji" silicon (Sapphire's card uses the reference AMD PCB carried over from the R9 Fury X). This card is firmly in the 30 cm-ish territory. Its display output configuration includes a DVI connector, apart from three DisplayPorts, and an HDMI connector. The cooler offers 0 dBA idle. AMD claims that the R9 Fury will offer higher performance than the GeForce GTX 980, and is hence expected to be priced in that range.

AMD's second graphics card based on its 28 nm "Fiji" silicon, the Radeon R9 Fury, will be an important SKU for the company. Ahead of its rumored mid-July launch, TweakTown got a whiff of its specifications from its sources. According to them, AMD will create the R9 Fury by enabling 56 of the 64 compute units on the silicon, yielding 3,584 stream processors. This sets the TMU count at 224. We doubt AMD will tinker with the render back-ends, and so the ROP count could remain at 64. The memory configuration could remain untouched, at 4 GB of 4096-bit HBM.

The clocks speeds on the R9 Fury will be the same as the R9 Fury X, at 1050 MHz core, and 500 MHz (512 GB/s) memory. One should expect temperatures of the R9 Fury to be higher, since it's being designed for air-cooled cards, although it's not expected to cross 75°C in typical gaming scenarios. Looking at the 12.5% drop in stream processors, one could expect the performance gap between the two Fury SKUs to be around 10-12%. This makes the R9 Fury a competitor to NVIDIA's GeForce GTX 980, if it's priced in its neighborhood (± $50).

GIGABYTE announced availability of its AMD Radeon R9 Fury X graphics card. Based entirely on AMD's reference design, the card features reference clocks of 1050 MHz core, and 500 MHz (512 GB/s) HBM clocks. Since the card lacks DVI connectors, GIGABYTE is including a DP-DVI adapter in its package. Based on the 28 nm "Fiji" silicon, the R9 Fury X offers 4,096 stream processors, 256 TMUs, 64 ROPs, and a 4096-bit wide HBM interface, holding 4 GB of memory. The compact card is cooled by a factory-fitted closed-loop liquid cooling solution. The card is generally available at US $649.99.

As the world's No.1 in GAMING graphics cards, MSI is pleased to announce the availability of the revolutionary AMD Radeon R9 Fury X. With the MSI R9 Fury X 4G, MSI adds a new high-end graphics card to its recently introduced and already heavily awarded AMD R9 and R7 300 series range. The MSI R9 Fury X 4G is equipped with 4GB of the groundbreaking High Bandwidth Memory, which is engineered directly onto AMD's Fiji XT GPU for unprecedented performance. The AMD Fury X is built for the future of ultra-high resolution gaming and Virtual Reality, which is available for enthusiasts to explore.

The card is housed in a much smaller form factor than we are used to from a top-level graphics card, which means that it fits into nearly any case. It is cooled by a closed loop liquid cooling solution which is included in the box which keeps the GPU around 50 degree Celsius during typical gaming sessions. The radiator can be mounted on the inside of the case to blow the heat directly outwards, it uses a single high quality 120mm fan for quiet performance.

AMD issued a statement on its upcoming Project Quantum SFF gaming desktop featuring Intel CPUs. The company clarified that it will offer both Intel and AMD processors, in the interest of consumer choice. AMD pointed out that the star-attraction in Project Quantum is the fact that it features two "Fiji" graphics processors in CrossFire configuration, to serve up graphics power unheard of, for this form-factor. Buyers will have a choice over the CPU, so if they feel that a CPU of a particular brand serves their needs better, they can make that choice.

Project Quantum will be an SFF gaming PC designed for super high resolution gaming. The company collaborated with liquid cooling majors to come up with a custom solution that cools the CPU and the two "Fiji" GPUs, over a common coolant channel, and a single large radiator. The system is split between two thermally-isolated chambers, with just coolant tubing and wiring running between the two. The bottom chamber features the mini-ITX form-factor motherboard, the dual-GPU video card, the SSD, and the power distribution PCB (PSU is external), while the top chamber is dominated by the radiator, and other liquid cooling components, such as the reservoir, pump, and a single large fan.

AMD's upcoming flagship GPU silicon, codenamed "Fiji," which is breaking ground on new technologies, such as HBM, memory-on-package, a specialized substrate layer that connects the GPU with it, called Interposer; features a hefty feature-set. More on the "Fiji" package and its memory implementation, in our older article. Its block diagram (manufacturer-drawn graphic showing the GPU's component hierarchy), reveals a scaling up, of the company's high-end GPU launches over the past few years.

"Fiji" retains the quad Shader Engine layout of "Hawaii," but packs 16 GCN Compute Units (CUs), per Shader Engine (compared to 11 CUs per engine on Hawaii). This works out to a stream processor count of 4,096. Fiji is expected to feature a newer version of the Graphics CoreNext architecture than "Hawaii." The TMU count is proportionately increased, to 256 (compared to 176 on "Hawaii"). AMD doesn't appear to have increased the ROP count, which is still at 64. The most significant change, however, is its 4096-bit HBM memory interface, compared to 512-bit GDDR5 on "Hawaii."

One of the three unexpected products based on the "Fiji" GPU, which AMD announced at its E3 event, Project Quantum, or the quest to design a 4K-worthy SFF gaming PC, which runs two "Fiji" GPUs in CrossFire, had the press assume that the rest of the system could be AMD-based, such as AMD-branded (albeit Patriot Memory manufactured) memory, AMD-branded (albeit OCZ manufactured) SSD; and importantly an AMD-made CPU or APU. Given its liquid-cooling, the prospect of a 95W "Godavari," or even upcoming "Carrizo" APU didn't seem far-fetched. Even a 95W FX CPU could have been deployed, since AM3+ on mini-ITX is not impossible.

When taken apart, Project Quantum was shown to be running an Intel Core i7-4790K "Devil's Canyon" CPU, on an ASRock-made mini-ITX motherboard, with its non-essential parts soldered out. The i7-4790K is neighbored by a pair of half-height Crucial Ballistix memory modules, which is excusable, since there are no half-height AMD Radeon memory modules, yet. The SSD is AMD-branded. The unit features a unified liquid cooling solution that's custom-made for AMD, by Asetek. A large (200 mm?) radiator, with a single fan, cools the CPU, the PCH, as well as the two "Fiji" GPUs.

It turns out that AMD's new "Fiji" silicon lacks HDMI 2.0 support, after all. Commenting on OCUK Forums, an AMD representative confirmed that the chip lacks support for the connector standard, implying that it's limited to HDMI 1.4a. HDMI 2.0 offers sufficient bandwidth for 4K Ultra HD resolution at 60 Hz. While the chip's other connectivity option, DisplayPort 1.2a supports 4K at 60 Hz - as do every 4K Ultra HD monitor ever launched - the lack of HDMI 2.0 support hurts the chip's living room ambitions, particularly with products such as the Radeon R9 Nano, which AMD CEO Lisa Su, stated that is being designed for the living room. You wouldn't need a GPU this powerful for 1080p TVs (a GTX 960 or R9 270X ITX card will do just fine), and if it's being designed for 4K UHD TVs, then its HDMI interface will cap visuals at a console-rivaling 30 Hz.

AMD's Radeon R9 Nano is shaping up to be a more important card for AMD, than even its flaghsip, the R9 Fury X. Some of the first pictures of the Fury X led us to believe that it could stay compact only because it's liquid cooled. AMD disproved that notion, unveiling the Radeon R9 Nano, an extremely compact air-cooled graphics cards, with some stunning chops.

The Radeon R9 Nano is a feat similar to the NUC by Intel - to engineer a product that's surprisingly powerful for its size. The card is 6-inches long, 2-slot thick, and doesn't lug along any external radiator. AMD CEO Lisa Su, speaking at the company's E3 conference, stated that the R9 Nano will be faster than the Radeon R9 290X. That shouldn't surprise us, since it's a bigger chip; but it's the electrical specs, that make this product exciting - a single 8-pin PCIe power input, with a typical board power rated at 175W (Radeon R9 290X was rated at 275W). The card itself is as compact as some of the "ITX-friendly" custom design boards launched in recent times. It uses a vapor-chamber based air-cooling solution, with a single fan. The Radeon R9 Nano will launch later this Summer. It could compete with the GeForce GTX 970 in both performance and price.

AMD's upcoming $650 Radeon R9 Fury X could have what it takes to beat NVIDIA's $999 GeForce GTX Titan X, while the $550 Radeon Fury (non-X) performs close to the $650 GeForce GTX 980 Ti, according to leaked 3DMark 11 and 3DMark (2013) benches by Korean tech publication ITCM.co.kr. The benches see the R9 Fury X score higher than the GTX Titan X in all three tests, while the R9 Fury is almost as fast as the GTX 980 Ti. The cards maintain their winning streak over NVIDIA even with memory-intensive tests such as 3DMark Fire Strike Ultra (4K), but buckle with 5K. These two cards, which are bound for the market within the next 30 days, were tested alongside the R9 390X, which is not too far behind the GTX 980, in the same graphs. The R9 Nano, however, isn't circulated among industry partners, yet. It could still launch in Summer 2015.

Here is the first reasonably detailed PCB shot of the dual-GPU graphics card based on "Fiji," which AMD announced at its E3 conference. The card is an inch taller than standard, but surprisingly short, for a dual-GPU board. This is thanks to the memory being relocated to the GPU package. All that's left on the PCB, besides the two GPUs, are the PLX PEX8747 PCI-Express gen 3.0 x48 bridge chip, and the 12-phase VRM, which draws power from a pair of 8-pin PCIe power connectors. Display outputs include one HDMI 2.0, and three DisplayPort 1.2a connectors.

AMD announced significant energy efficiency gains for "Fiji" over "Hawaii," and so this card could have a much lower than expected power-draw. The reference board could come with AIO liquid-cooling, much like the single-GPU Radeon R9 Fury X, some AIBs could even release cards with air-cooling solutions. The yet unnamed dual-GPU "Fiji" based graphics card could be available in Autumn 2015.Image Credit: Anshel Sag (Twitter)

AMD announced no less than five new products based on its swanky new 28 nm "Fiji" silicon, the company's most powerful GPU, packing over 8 TFLOP/s of raw compute power, and the first GPU to feature stacked HBM (high-bandwidth memory), moved to the GPU package, and communicating with the GPU die over a special silicon substrate called the interposer. The "Fiji" silicon will enable AMD to target NVIDIA's entire high-end GPU lineup.

The first product is Project Quantum. This is a console-sized SFF gaming desktop designed by AMD, which will be sold by the company's add-in board partners. Despite its diminutive size, the desktop packs two "Fiji" GPUs in AMD CrossFireX, and an AMD 64-bit x86 machine driving the rest. All main components (the CPU, the chipset, and the two GPUs), are liquid-cooled. This desktop will enable smooth 4K/5K gaming in the living room.