AMD's 7 nm "Navi 10" silicon may finally address two architectural shortcomings of its performance-segment GPUs, memory bandwidth, and render-backends (deficiency thereof). The GPU almost certainly features a 256-bit GDDR6 memory interface, bringing about a 50-75 percent increase in memory bandwidth over "Polaris 30." According to a sketch of the GPU's SIMD schematic put out by KOMACHI Ensaka, Navi's main number crunching machinery is spread across eight shader engines, each with five compute units (CUs).

Five CUs spread across eight shader engines, assuming each CU continues to pack 64 stream processors, works out to 2,560 stream processors on the silicon. This arrangement is in stark contrast to the "Hawaii" silicon from 2013, which crammed 10 CUs per shader engine across four shader engines to achieve the same 2,560 SP count on the Radeon R9 290. The "Fiji" silicon that followed "Hawaii" stuck to the 4-shader engine arrangement. Interestingly, both these chips featured four render-backends per shader engine, working out to 64 ROPs. AMD's decision to go with 8 shader engines raises hopes for the company doubling ROP counts over "Polaris," to 64, by packing two render backends per shader engine. AMD unveils Navi in its May 27 Computex keynote, followed by a possible early-July launch.

At its Computex presser, leading up to its 7 nm Radeon Vega series unveil, AMD touched upon the massive proliferation of the Vega graphics architecture, which is found not only in discrete GPUs, but also APUs, and semi-custom SoCs of the latest generation 4K-capable game consoles. One such slide that created quite some flutter reads that "Vega" shipments are over 10 times greater than those of the "previous generation."

Normally you'd assume the previous-generation of "Vega" to be "Polaris," since we're talking about the architecture, and not an implementation of it (eg: "Vega 10" or "Raven Ridge," etc.). AMD later, at its post event round-table, clarified that it was referring to "Fiji," or the chip that went into building the Radeon R9 Fury X, R9 Nano, etc., and comparing its sales with that of products based on the "Vega 10" silicon. Growth in shipments of "Vega" based graphics cards is triggered by the crypto-mining industry, and for all intents and purposes, AMD considers the "Vega 10" silicon to be a commercial success.

It has been a huge weekend of product announcements and launches from AMD, which expanded not just its client computing CPU lineup on both ends, but also expanded its Radeon graphics cards family with both client- and professional-segment graphics cards. This article provides a brief summary of everything AMD launched or announced today, with their possible market-availability dates.

AMD released the die-shot of its "Vega 10" ASIC. The multi-chip module (MCM) of the GPU die and two HBM2 memory stacks, sitting on a silicon interposer, is slightly smaller than the "Fiji" ASIC, as it features just two memory stacks. AMD didn't label the components of the GPU die, but we can make out 8 shader engines, holding 8 GCN compute units (CUs), each. This is unlike "Fiji," where the GPU holds four engines with 16 CUs, each. This would mean that each group of 8 CUs has its own dedicated geometry processor and rasterizer.

Since each CU holds 64 FP32 stream processors, we arrive at the total stream processor count of 4,096. Unlike "Hawaii," these stream-processors are FP16-capable, so simple compute tasks are handled at double the throughput. We also make out 32 render back-ends, double that of "Fiji" and "Hawaii," which could indicate 128 ROPs. The "Vega 10" ASIC features a 2048-bit HBM2 memory interface.Update: AMD stated that the die-shot appeared in one of its marketing slides, but may not be real.

AMD could be working on at least three SKUs based on its upcoming "Vega 10" silicon to make up its Radeon RX Vega series. Leaked 3DMark validations point to a device ID that's third in a series of possible device IDs of graphics cards based on the "Vega 10" silicon, the 687F:C1, 687F:C2, and 687F:C3. All three SKUs feature 8 GB of HBM2 memory, and according to leaked 3DMark TimeSpy scores, the "slowest" SKU is faster than NVIDIA GeForce GTX 1070. The fastest SKU is in the same league as the GTX 1080 Ti.

The three SKUs could differ with core-configuration and clock speeds. AMD carved four SKUs out of its "Fiji" silicon, the liquid-cooled R9 Fury X, the air-cooled R9 Fury (with 12.5% fewer shaders), the SFF-friendly R9 Nano (full core-config, but aggressive power-management), and the halo dual-GPU Radeon Pro Duo (1st gen). AMD could take a similar approach with "Vega 10." AMD is expected to launch its Radeon RX Vega series within Q2-2017.

A Linux patch for AMD's GPU drivers reveals that its upcoming "Vega 10" graphics processor bears numeric core-configuration similarities to the "Fiji" silicon which drives the enthusiast-segment Radeon R9 Fury series graphics cards. The patch bears configuration values which tell the software how to utilize the resources on the GPU, by spelling them out. The entry "gfx.config.max_shader_engines = 4," for example, indicates that "Vega 10" features four shader engines, like "Fiji." Another entry "Adev-> gfx.config.max_cu_per_sh = 16" signifies the number of GCN compute units (CUs) per shader engine. Assuming the number of stream processors per CU hasn't changed from 64 in the "Vega" architecture, we're looking at a total stream processor count of 4,096. This could also put the TMU count at 256.

At earlier reveals of the "Vega 10" package, you notice a large, somewhat square GPU die neighboring two smaller rectangular memory stack dies, which together sit on a shiny structure, which is the silicon interposer. The presence of just two memory stack dies sparked speculation that "Vega 10" features a narrower 2048-bit memory interface compared to the 4096-bit of "Fiji," but since the memory itself is newer-generation HBM2, which ticks at higher clocks, AMD could run them at double the memory clock as "Fiji" to arrive at the same 512 GB/s bandwidth. The 4,096 stream processors of "Vega 10" are two generations ahead of the ones on "Fiji," which together with 14 nm process-level improvements, could run at much higher GPU clocks, making AMD get back into the high-end graphics segment.

NVIDIA announced that its flagship GPGPU accelerator, the Tesla P100, will be available through Google Cloud Platform. The company's Tesla K80 accelerator will also be offered. The Google Cloud Platform allows customers to perform specific computing tasks at an infinitesimally lower cost than having to rent hardware in-situ or having to buy it; by offloading your computing tasks to offsite data-centers. IT professionals can build and deploy servers, HPC farms, or even supercomputers, of all shapes and sizes within hours of placing an order online with Google.

The Tesla P100 is a GPGPU with the most powerful GPU in existence - the NVIDIA GP100 "Pascal," featuring 3,584 CUDA cores, up to 16 GB of HBM2 memory, and NVLink high-bandwidth interconnect support. The other high-end GPU accelerators on offer by Google are the Tesla K80, based on a pair of GK210 "Kepler" GPUs, and the AMD FirePro S9300 X2, based on a pair of "Fiji" GPUs.

At its WX call, AMD focused on shifts in creativity from traditional design flows such as Solidworks, Adobe and Autodesk towards game engines as solutions for design visualization (Unreal Engine, Unity, CryEngine, or Autodesk's own Stingray platform), which signal changes in the creator ecosystem. Thanks to globalization, the Internet, and the available wealth of knowledge one can access through it, the line between amateurs and professionals is becoming more and more blurred. Now, those who would once be called amateurs are also using professional tools, and AMD plans to be at the forefront of technologies empowering creators to deliver their vision.

Radeon PRO serves to give creators more flexible and powerful solutions, leveraging open-source resources and centering the ecosystem back on creators and the tools they choose to use, with focused support on VR. As such, AMD is giving them the tools they need, by introducing three new products featuring the Polaris architecture, including 3 year standard + 7 year free extended warranty (including components such as the PCB itself, the PCI-Express slot, and the heatsinks), with AMD taking that extra 7 years as company commitment towards the quality of their products. Those three products are the WX4100, the WX5100, and the WX7100, and have planned, staggered availability throughout November.

PowerColor teased this image of its next "DEVIL" series graphics card, as part of its latest giveaway. The teaser states "a new devil is born," sparking off speculation that this could be a custom-design dual-GPU graphics card based on the "Fiji" silicon. It's unlikely that this is a product based on the upcoming RX 480 because the company reserves the "DEVIL" sub-brand for only its fastest solutions. Calling this card the Radeon Pro Duo Devil13 could dent the "gamer-professional" marketing of the Pro Duo, so if this is indeed a dual-"Fiji" graphics card, PowerColor could avoid using the Pro Duo SKU name.

At a presser in Taiwan for its Radeon Pro Duo launch, AMD talked extensively about its upcoming "Polaris" and "Vega" family of GPUs. The company appears to be betting heavily on two SKUs it plans to launch this June, Polaris 10 and Polaris 11. Polaris 10 is an internal designation to Radeon R9 490(X), based on the 14 nm "Ellesmere" silicon. It may be the biggest chip AMD builds on the "Polaris" architecture, but it won't exactly be a "big chip," in that it doesn't succeed "Fiji." That honor is reserved for "Vega," which debuts in early-2017.

The "Ellesmere" silicon is more of AMD's competitor to NVIDIA's GP104. It is rumored that the R9 490(X), based on this silicon, will offer consumers performance rivaling the GeForce GTX 980 Ti (ergo faster than the Radeon R9 Fury X), at a USD $300-ish price point. "Ellesmere" will be a lean-machine, physically featuring up to 2,560 4th generation GCN stream processors (2,304 enabled on Polaris 10), a possible 256-bit GDDR5X memory interface, and a deep sub-200W typical board power rating.

AMD finalized the GPU architecture roadmap running between 2016 and 2018. The company first detailed this at its Capsaicin Event in mid-March 2016. It sees the company's upcoming "Polaris" architecture, while making major architectural leaps over the current-generation, such as a 2.5-times performance/Watt uplift and driving the company's first 14 nanometer GPUs; being limited in its high-end graphics space presence. Polaris is rumored to drive graphics for Sony's upcoming 4K Ultra HD PlayStation, and as discrete GPUs, it will feature in only two chips - Polaris 10 "Ellesmere" and Polaris 11 "Baffin."

"Polaris" introduces several new features, such as HVEC (h.265) decode and encode hardware-acceleration, new display output standards such as DisplayPort 1.3 and HDMI 2.0; however, since neither Polaris 10 nor Polaris 11 are really "big" enthusiast chips that succeed the current "Fiji" silicon, will likely make do with current GDDR5/GDDR5X memory standards. That's not to say that Polaris 10 won't disrupt current performance-thru-enthusiast lineups, or even have the chops to take on NVIDIA's GP104. First-generation HBM limits the total memory amount to 4 GB over a 4096-bit path. Enthusiasts will have to wait until early-2017 for the introduction of the big-chip that succeeds "Fiji," which will not only leverage HBM2 to serve up vast amounts of super-fast memory; but also feature a slight architectural uplift. 2018 will see the introduction of its successor, codenamed "Navi," which features an even faster memory interface.

AMD's upcoming GPUs based on the "Polaris" and "Vega" architectures appear to be taking advantage of performance/Watt gains to keep stream processor counts low, and chips small, according to a VideoCardz analysis of curious-looking CompuBench entries. Assuming that a Graphics CoreNext (GCN) compute unit (CU) of the "Polaris" architecture, like the three versions of GCN before it, consists of 64 stream processors, AMD's Polaris 11 silicon, codenamed "Baffin," could feature over 1,024 stream processors, across 16 CUs; Polaris 10, codenamed "Ellesmere," could feature over 2,304 stream processors spread across over 36 CUs; and Vega 10 featuring 4,096 stream processors, spread across 64 CUs.

The "Baffin" silicon succeeds current generation "Curacao" silicon, driving mid-range graphics cards. It is expected to feature a 128-bit wide GDDR5 memory interface, holding 4 GB of memory. The "Ellesmere" silicon succeeds current-generation "Tonga" silicon, driving performance-segment SKUs. It could feature up to 8 GB of GDDR5(X) memory. These two chips could see the light of the day by mid-2016. The third chip out of AMD's stable, Vega 10, could succeed "Fiji," overcoming its biggest marketing shortcoming - 4 GB memory. Taking advantage of HBM2 interface, it could feature 16 GB of memory. It could launch some time in early-2017. AMD is claiming a massive 2.5X performance-per-Watt increase for "Polaris" over the current GCN 1.2 architecture, which drives the "Tonga" and "Fiji" chips, and so these stream processor counts could look deceptively insufficient.

AMD's flagship dual-GPU graphics card, the Radeon Pro Duo, is slated for retail availability from April 26, 2016, according to HardwareBattle. The card was announced at the company's Capsaicin, in mid-March. Reviews of this card could be posted much earlier. Based on a pair of 28 nm "Fiji" GPUs, with two sets of 4,096 stream processors, 256 TMUs, 64 ROPs, and 4 GB of 4096-bit HBM memory, each, the Radeon Pro Duo will be launched at the same price as the company's previous flagship product, the Radeon R9 295X2, at US $1,499. The card could feature on a variety of high-end gaming PCs, and VR content creation workstations. In the retail channel, AMD's traditional add-in board (AIB) partners could launch the reference-design card.

Although AMD's upcoming socket AM4 heralds new lines of processors and APUs based on the company's next-generation "Zen" CPU micro-architecture, some of the first APUs will continue to be based the current "Excavator" architecture. The "Bristol Ridge" is one such chip. It made its mobile debut as the 7th generation A-Series and FX-Series mobile APUs, and is en route to the desktop platform, in the AM4 package. What sets the AM4 package apart from the FM2+ package, and in turn "Bristol Ridge" from "Carrizo" is that the platform integrates even the southbridge (FCH) into the APU die. This could explain the 1,331-pin count of the AM4 socket.

The "Bristol Ridge" silicon is likely built on the existing 28 nm process. That's not the only thing "current-gen" about this chip. Its CPU component consists of two "Excavator" modules that make up four CPU cores, with 4 MB total cache; and its integrated GPU will likely be based on the Graphics CoreNext 1.2 "Volcanic Islands" architecture, the same one which drives the "Tonga" and "Fiji" discrete GPUs. The integrated memory controller supports dual-channel DDR4 memory. In its performance benchmarks, an AM4 APU based on the "Bristol Ridge" silicon was pitted against older 6th generation APUs, in which it was found to be as much as 23 percent faster.

The LinkedIn profile of an R&D manager at AMD discloses key details of the company's upcoming "Greenland" graphics processor, which is also codenamed Vega10. Slated for an early-2017 launch, according to AMD's GPU architecture roadmap, "Greenland" will be built on AMD's "Vega" GPU architecture, which succeeds even the "Polaris" architecture, which is slated for later this year.

The LinkedIn profile of Yu Zheng, an R&D manager at AMD (now redacted), screencaptured by 3DCenter.org, reveals the "shader processor" (stream processor) count of Vega10 to be 4,096. This may look identical to the SP count of "Fiji," but one must take into account "Greenland" being two generations of Graphics CoreNext tech ahead of "Fiji," and that the roadmap slide hints at HBM2 memory, which could be faster. One must take into account AMD's claims of a 2.5X leap in performance-per-Watt over the current architecture with Polaris, so Vega could only be even faster.

AMD unveiled its latest flagship graphics card, the Radeon Pro Duo. The card is designed for "creators who game, and gamers who create," as the tagline goes. It is a dual-GPU graphics card based on a pair of 28 nm "Fiji" chips, the same ones which drive the R9 Fury X and the R9 Nano. AMD is positioning this card in the gray-area between consumer graphics cards, and FirePro workstation products, as a new "workstation-class" product. Perhaps this allows the company to get away with things such as three 8-pin PCIe power connectors.

The Radeon Pro Duo features two "Fiji" GPUs in their maximum core configuration - 4,096 stream processors, 256 TMUs, and 64 ROPs, each; with 4 GB of HBM memory, each. The card hence packs a total of 8 GB HBM memory, and 16 TFLOP/s total single-precision floating-point performance. The card features a liquid-cooling solution designed by Cooler Master, with a thick 120 mm x 120 mm radiator that's similar to the one that ships with the R9 Fury X. The card's display output configuration is similar to the R9 Fury X, too, with three DisplayPort 1.2a and one HDMI 1.4a connectors. AMD is going ahead and claiming the title of "World's Fastest Graphics Card." The Radeon R9 Pro Duo is expected to be priced at US $1,499.

AMD, which has been timing its upcoming dual-GPU "Fiji" graphics card to launch sometime this year, may have demoed a production version of the card in one of its launch partners, Falcon Northwest's, Tiki high-end gaming desktop, as a "VR developer box." AMD's Roy Taylor, in a recent tweet, captions a picture of this dev box as being "the world's best DirectX 12 VR developer box," leading the press to speculate that it's running the company's dual-GPU "Fiji" card.

A close look at AMD's VR dev box, through its windowed graphics card compartment, reveals an air-cooled AMD reference graphics card, which VideoCardz' trigonometry pins as being shorter than a Radeon R9 390X reference board. It could be a reference R9 380X, but then a reference dual-GPU "Fiji" PCB is roughly of the same length, and a R9 380X wouldn't earn the title of being the "world's best" from a senior AMD exec while there are faster AMD cards, such as the R9 Fury. The ability of the full-spec "Fiji" silicon to cope well with a rather simple air-cooler in the R9 Nano fans even more speculation that a dual-GPU "Fiji" board could make do with a powerful air-channel cooler.

TechPowerUp released the latest version of GPU-Z, the popular graphics subsystem information, monitoring, and diagnostic utility. Version 0.8.7 adds support for new GPUs, fixes a variety of bugs, and improves the interface. To begin with, support is added for AMD Radeon R9 380X, R7 350, and the "Mullins" APU; NVIDIA GTX 980 8GB (notebooks), GTX 965M, GTX 750 (GM206), GT 710 (GK208), Quadro K1200, M5000, M2000M, M1000M, K2200M, GRID K160Q, and Tesla K80; and Intel "Skylake" Gen9 Graphics 510, P530, 540.

Among the bug fixes include improved Radeon Software version number detection, correct DirectX hardware-support readout on Intel "Skylake" IGPs, accurate 1.55V voltage reading for AMD "Fiji" GPUs in ULPS modes, a BSOD on Intel "Cloverview" Atom Z2760, SKU naming for AMD "Beema" chips; improved detection of CUDA devices running on Bus ID greater than 9, and a better explanation for OpenCL detection errors. The Armenian language pack has been improved.DOWNLOAD: TechPowerUp GPU-Z 0.8.7 | GPU-Z 0.8.7 ASUS ROG Themed

The change-log follows.

This week on TechPowerUp News: Jim Keller joins Tesla Motors as head of the team working on its self-driving tech; and prototype desktops with dual-GPU "Fiji" graphics cards.

At the Virtual Reality Los Angeles (VRLA) event, AMD along with a few gaming PC manufacturers demonstrated their desktop builds featuring the company's upcoming dual-GPU graphics card driven by a pair of "Fiji" GPUs. Among the desktops demoed include a prototype Falcon Northwest Tiki compact gaming desktop with a dual-Fiji graphics card, and HTC Vive HMD. Falcon Northwest commented that the dual-GPU "Fiji" graphics card is small enough to squeeze into its 4-inch thick Tiki, indicating that the card will be super-compact. AMD is expected to launch the dual-GPU "Fiji" graphics card some time in Q2 2016.

AMD gave its premium small-factor gaming graphics card, the Radeon R9 Nano, its first major price cut. The card now starts at US $499, down from its launch price of $649. At $499, the R9 Nano is priced on par with its similar-performing albeit bigger and noisier sibling based on the "Fiji" silicon, the Radeon R9 Fury. The company's flagship single-GPU card, the R9 Fury X, remains at $599, its price was gradually reduced from its launch price of $649.

The three SKUs appear to be positioned to compete with NVIDIA's GeForce GTX 980, and offer cost-effective alternatives to the $629 GTX 980 Ti. Elsewhere in the lineup, the Radeon R9 390X starts at $379, and has its guns trained on the GTX 980 and GTX 970. Its smaller sibling, the Radeon R9 390 starts at $299.

It has been confirmed that Samsung will be AMD's foundry partner for its next generation GPUs. It has been reported that AMD's upcoming "Arctic Islands" family of GPUs could be built on the 14 nanometer FinFET LPP (low-power Plus) process. AMD's rival NVIDIA, meanwhile, is building its next-gen "Pascal" GPU family on 16 nanometer FinFET node, likely at its traditional foundry partner TSMC.

It gets better - not only will Samsung manufacture AMD's next-gen GPUs, but also its upcoming "Zen" family of CPUs, at least a portion of it. AMD is looking to distribute manufacturing loads between two foundries, Samsung and GlobalFoundries, perhaps to ensure that foundry-level teething trouble doesn't throw its product launch cycle off the rails. One of the most talked about "Arctic Islands" GPUs is codenamed "Greenland," likely a successor to "Fiji." Sales of some of the first chips - GPUs or CPUs - made at Samsung, will begin some time in Q3 2016. Some of the other clients for Samsung's 14 nm FinFET node are Apple and Qualcomm. The company plans to speed up development of its more advanced 10 nm node to some time in 2017.

In this episode of TechPowerUp News, price drops across the meat of NVIDIA's GeForce GTX 900 lineup; the new upcoming AMD dual-GPU "Fiji" graphics card, and Michael Mantor's elevation in AMD.

AMD could launch its next-generation dual-GPU graphics card based on the "Fiji" silicon, some time in December. Codenamed "Fury-Gemini," the card features a pair of "Fiji" GPUs in an internal multi-GPU setup; and is cooled by an AIO liquid-cooling solution, much like the R9 Fury X. Prototype boards of this card were shown by AMD top-brass at some of the chip's earliest reveals. It's expected that the PCB (pictured below), will be mated with a liquid-cooling solution; and unless NVIDIA releases its dual-GPU GM200 graphics card any sooner, is on course to becoming the fastest graphics card you can buy. It remains to be seen if AMD can cash in on the Holiday shopping season.

Sapphire announced its latest version of the TriXX overclocking utility, which it bundles with its graphics cards. Version 5.2.1 is one of the first pieces of software to support voltage control on AMD "Fiji" GPU graphics cards (R9 Fury, R9 Fury X, and R9 Nano), and HBM overclocking. The utility comes with a new "dashboard" themed user interface that gives you an analog readout of your card's main sensors - clocks, voltage, and temperatures; and a simpler layout. In addition, it also supports voltage control on other Radeon R9 300 series GPUs.DOWNLOAD: Sapphire TriXX 5.2.1