Tag: amd

AMD shows off Vega

P

AMD lifted the kimono on its next generation GPU architecture, codenamed Vega.

For those who came in late, Vega’s design is based on the concept that conventional GPU architectures have not been scaling well for diverse data types. Gaming and graphics workloads have shown steady progress and while their compute capability may have been increasing at a good pace, but memory capacity has not kept up.

AMD claims that Vega has the most scalable GPU memory architecture built to date with 512TB of address space. It also has a new geometry pipeline tuned for more performance and better efficiency with over 2X peak throughput per clock, a new Compute Unit design, and a revamped pixel engine.

The pixel engine features a new draw stream binning rasterizer (DSBR), which reportedly improves performance and saves power.

Vega should offer significant improvements in terms of performance and efficiency when products based on the architecture begin shipping in a few months.  AMD is pinning rather a lot on Vega and its forthcoming Zen CPU.  If the two pay off then it will give Intel a much needed kick to the bottom line and claw back AMD’s dismal market share. Both technologies look rather good on paper.

AMD’s Zen will be branded Ryzen

rizen-300x300AMD’s Zen will have a brand which sounds suspiciously like a drug for erectile dysfunction, a chemical weapon extracted from caster beans or a gospel group

The Ryzen brand will be applied to desktop and mobile variants. It had been formerly known as Summit Ridge. AMD has been telling the world that Ryzen is pronounced Rye-Zen (like horizon, without the “ho”). It has not said what the Rye means, but tired and emotional hacks suggested that it might be inspired by Jack Daniels – as indeed so are they.

Zen-based processors will target desktops, servers, and mobiles device, but the first wave of products will be targeted at the performance desktop market, where gamers and VR continue to spur growth. PC gaming hardware is predicted for 35 percent growth from 2015 to 2020, and VR-capable PCs are predicted to grow by a factor of 10 by 2020. Interest in eSports continues to skyrocket as well.

AMD has disclosed that one of the high-end options in the initial Ryzen line-up will feature eight cores (16 threads with SMT) and at minimum a 3.4 GHz base clock, with higher turbo frequencies. That processor will have 20MB of cache – 4MB of L2 and 16MB of L3. AMD is also hyping up what it is calling SenseMI technology. AMD’s SenseMI is a set of sensing and adapting technologies, including what AMD is calling “an artificial intelligence network” inside every Zen processor. There are five main features of SenseMI that include Pure Power, Precision Boost, and Extended Frequency Range (or XFR), along with the neural net prediction algos and smart prefetcher.

In fact, there is not that much exciting in SenseMI – it is just an updated branch predictor, prefetcher, and power and control logic in Zen. It might have some sexed up algorithms but it is not the ground breaking technology it is claimed.

AMD said that the AM4 platform for Ryzen will be have all the usual gubbins you expect to see on a modern PC enthusiast platform. AM4 motherboards will use DDR4 memory and feature PCIe Gen 3 connectivity, and support for USB 3.1 Gen 2, NVMe, and SATA Express.

Another thing that Ryzen will have is something called Pure Power which senses what’s going on with the processor’s current workload and provides a closed-loop control system to scale performance and power, over what AMD is calling its Infinity Fabric.

This monitors temperatures, frequencies, and voltage, and adaptively controls each element to optimize performance and power usage.

Another thing that Ryzen will have is something called Infinity Fabric which is interconnect technology used to link the individual bits of IP within a processor and the processor to other IO.

AMD wants this tech to scale its products, beyond moving to new process nodes, increase performance and efficiency, scale bandwidth, improve latency and the chip’s QoS.

Infinity Fabric consists of two key elements, a scalable control fabric and a scalable data fabric. The scalable control fabric has all the central control elements, with small remote elements that are dispersed in each different block of the SoC.

However the early indications suggest that Ryzen is rather good and should put the fear of Jehovah into Intel – particularly if AMD keeps the price down.

AMD releases AI based Radeons with basic instinct

BasicInstinct002AMD is announcing a new series of Radeon-branded products today, targeted at machine intelligence and deep learning enterprise applications.

Dubbed the Radeon Instinct, the chip is a GPU-based solution for deep learning, inference and training. AMD has also issued a new free, open-source library and framework for GPU accelerators, dubbed MIOpen.

MIOpen is made for high-performance machine intelligence applications and is optimized for deep learning frameworks in AMD’s ROCm software suite.

The first products are the Radeon Instinct MI6, the MI8, and the MI25. The 150W Radeon Instinct MI6 accelerator is powered by a Polaris-based GPU, packs 16GB of memory (224GB/s peak bandwidth), and can manage 5.7 TFLOPS of peak FP16 performance when the wind is behind it and it is going downhill.

It also includes the Fiji-based Radeon Instinct MI8. Like the Radeon R9 Nano, the Radeon Instinct MI8 features 4GB of High-Bandwidth Memory (HBM) with peak bandwidth of 512GB/s. AMD claims the MI8 will offer up to 8.2 TFLOPS of peak FP16 compute performance, with a board power that typical falls below 175W.

The Radeon Instinct MI25 accelerator uses AMD’s next-generation Vega GPU architecture and has a board power of approximately 300W. All the Radeon Instinct accelerators are passively cooled but when installed into a server chassis you can bet there will be plenty of air flow.

Like the recently released Radeon Pro WX series of professional graphics cards for workstations, Radeon Instinct accelerators will be built by AMD. All the Radeon Instinct cards will also support AMD MultiGPU (MxGPU) hardware virtualisation.

Why Kaby Lake and Zen is Windows 10 only

Windows 10Microsoft raised a few eyebrows when it announced that only Windows 10 will support Intel’s and AMD’s next-generation processor microarchitectures – codenamed Kaby Lake and Zen.

It appears that there are a few features on Kaby Lake and  Zen that require significant updates to Windows 10 to optimally function.

Kaby Lake uses Intel’s Speed Shift technology that make it possible to change power states more quickly than Skylake. Because Kaby Lake can make Speed Shift transitions faster, 7th Gen Core processors based on the architecture can increase and decrease clocks quickly. Speed Shift is hardware enabled but it uses the OS to function properly.

Intel’s Turbo Boost Technology 3.0  with Turbo Boost Max 3.0 technology improves single-thread performance by identifying the fastest core on a particular processor die and prioritising critical workloads for that core. This pushes up the processor’s frequency when needed and  workloads are also directed to the fastest possible core available. Support for that technology needs to be in the operating system.

AMD’s Zen-based processors have fine-grained clock gating with multi-level regions throughout the chip. Zen will bring in newer Simultaneous Multi-Threading technology for AMD chips. Microsoft will  have to make updates to the Windows kernel and system scheduler, which is more involved than a driver update. Vole did something similar to add proper support for Bulldozer-based processors with Windows 7.

So as far as AMD, Microsoft and Intel are concerned getting rid of support for older systems makes perfect sense. You can’t lock these chips into something which was released seven years ago. Windows 8 is similar to Windows 10 but about as popular as the Boston Strangler it is just not worth trying to update.

While corporate customers might like to remain on Windows 7 and incorporate next-gen hardware into their infrastructure, there will not be many of them. Older versions of Windows and alternative operating systems will still install and run on Kaby Lake and Zen, but they just won’t do the cool stuff.

AMD releases Radeon Pro WX 4100

Radeon-Pro-SSG-e1469514780361AMD has released its business end GPUs.

Dubbed the Radeon Pro, the  WX 4100 is the entry-level model with a half-height design for use in small form-factor workstations. The Radeon Pro WX 5100 is the mid-range model while the Radeon Pro WX 7100 is the top of the range. AMD says the 7100 can handle the most demanding design engineering and entertainment workflows and VR content creation.

The spec is still pretty much under wraps, but we know the following spec from copying down the slides:

  • Radeon Pro WX 7100: 32 compute units / > 5 TFLOPs / 8GB memory / 256-bit memory bus
  • Radeon Pro WX 5100: 28 compute units / > 4 TFLOPs / 8GB memory / 256-bit memory bus
  • Radeon Pro WX 4100: 16 compute units / > 2 TFLOPs / 4GB memory / 128-bit memory bus

The Radeon Pro WX 4100 has four mini DisplayPort connectors and the others have four full-size DisplayPort connectors.

The Radeon Pro WX 7100 will clost  just under $1,000 and the Radeon Pro WX 5100 and 4100 will slot in somewhat below that which fits in nicely to their mission critical role. Each has a ten year warranty.

AMD will consolidate its graphics product lines and kill off the FirePro range.  In its place will be the Radeon Pro WX Series cards to serve the professional market for AMD.

 

Nvidia releases answer to AMD’s Radeon RX 480

nvidiaNvidia just launched its Pascalish answer to AMD’s Radeon RX 480 mainstream card.

The GeForce GTX 1060 has about half of the resources of Nvidia’s super expensive GeForce GTX 1080 and the outfit claims it’s on par with a previous generation high-end GeForce GTX 980.

It runs on 120W and is a mix of low-power and high-performance. The new GeForce GTX 1060 features a new Pascal derivative GPU that’s somewhat smaller, called the GP106. It has 10 streaming multiprocessors (SM) with a total of 1280, single-precision CUDA cores and eight texture units.

The GeForce GTX 1060 also features six 32-bit memory controllers, for 192-bits in total. GeForce GTX 1060 cards with either 6GB or 3GB of GDDR5 memory will be available and offered performance that just misses the mark set by the pricier AMD Radeon R9 Nano.

The GeForce GTX 1060 has the largest leads over the Radeon RX 480 in the DirectX 11 tests, though the Radeon had a clear edge in OpenCL and managed to pull ahead in  some DirectX 12 tests.

The GeForce GTX 1060, however, consumes significantly less power than the Radeon RX 480 and is quieter too.

All up it means that Nvidia and AMD are squaring up with different offerings for a similar price

Polaris finally shines today

polarisAMD finally starts flogging its new RX 480 GPU today.

For those who came in late, the RX 480 uses the company’s latest Polaris architecture which is built lt on 14nm FinFET process technology.

The starting price is $199 for the 4GB model and $239 for the 8GB and has some interesting performance characteristics. Compared to the GeForce GTX 970 which sells for $280, the RX 480 performs is about five to ten percent better. But when it comes to DX12 games like Gears of War, Hitman and Rise of the Tomb Raider it is about 40 per cent faster.

Compared to previous AMD products, the RX 480 is as fast as a Radeon R9 390 but uses just 150 watts compared to 275 watts for the previous generation.

Rivals Nvidia are expected to have a competing product based on Pascal available sometime in July, so AMD’s advantage may be short-lived; but in the meantime, the Radeon RX 480 is clearly the best GPU for $200.

Of course the world is also waiting to see AMD’s entry into the CPU league tables with the much touted never seen Zen chip, which should be in the shops in December.

Apple gets Thor about Thunderbolt display and smites it

Thor.TDW.battle.cropApple is giving up on its Thunderbolt Display which it first introduced in the summer of 2011.

A spokesApple said that the the display will be available through Apple.com, Apple’s retail stores and Apple Authorised Resellers while supplies last, but it will not make any more of them.

The move has pundets puzzled and the Tame Apple Press thinks it means a new 4K or 5K display is on the horizon.

Stock shortages ahead of WWDC sparked rumors that Apple might be planning to introduce a new display at the event. Nothing happened and and Apple instead focused on dull software for iOS devices, Macs, Apple TVs, and Apple Watch devices.

But rumors that Apple was working on a 5K display have been around for a while. If they are true then it would have  resolution of 5120 x 2880 pixels, USB-C/Thunderbolt 3 peripherals.

It might have an AMD graphics chip inside so that anyone with the right connection could get a better looking screen.  Another suggestion is that it might have a DisplayPort 1.2 Multi-Stream Transport setup to stitch two halves of a display together to make one display.

However this is normal PR spinning after Apple kills off one product. The assumption is that it must have “something new” when it might just not be doing much at all.

 

AMD shows off Polaris-based Radeon RX 470 and RX 460

4528082378_4d5b9fb99e_zAMD has been showing off its latest Polaris based GPUs at E3 2016.

For those who came in late, Polaris is AMD’s bright new hope in the GPU world – a bit like Zen is for the CPU, only it appears to exist whereas Zen doesn’t.

THe Radeon RX 470 and RX 460  join the recently announced RX 480 as part of the company’s new Polaris family. Polaris is AMD’s newest GPU micro-architecture, which is based on the 14nm FinFet production process.

AMD is not telling us the prices of its new GPU, but it is possible to have stab at it. The  RX 480 is made for 1440p gaming, and the RX 470 will focus on delivering a “refined, power-efficient HD gaming” experience. The RX 460 will offer a “cool and efficient solution for the ultimate e-sports gaming experience.”

The 4GB version of the RX 480 will start out at $200, it’s safe to assume that these two other cards will launch at lower price points.

AMD says the chips are extremely thin, offering a very low Z-height, and will fit into thin and light gaming notebooks.

The entire new RX line will also support a wide variety of features that include DX12, Vulkan, HDR, HDMI 2.0b, DisplayPort 1.3/1.4, and H.265 encoding/decoding.

There is no release day  but since the RX 480 is scheduled to launch on June 29 the other two should be soon after. AMD is claiming that card outperforms $500 graphics cards in VR.

AMD finally gets its arm in

We_Can_Do_It!After more than a year of delays, AMD has finally released its ARM based enterprise server chip.

Dubbed the Opteron A1100 series these are not the X86 cores AMD has been producing for years and are designed for networking, storage, dense and power-efficient web serving, and 64-bit ARM software development.

The Opteron A1100 System-on-Chip (SoC), was formerly codenamed “Seattle” and was promised in the first half of last year and never showed up. Under the bonnet are off-the-shelf ARM Cortex-A57 processor cores, with integrated high-speed network and storage connectivity.

The SoCs have up to eight 64-bit ARM Cortex-A57 cores with up to 4MB of shared Level 2 and 8MB of shared Level 3 cache. They offer two 64-bit DDR3/DDR4 memory channels supporting speeds up to 1866 MHz with ECC and capacities up to 128GB, dual integrated 10Gb Ethernet network connections, 8-lanes of PCI-Express Gen 3 connectivity, and 14 SATA III ports.

The chip features an ARM TrustZone compliant crypto/compression co-processor, along with a Cortex A5-based system control processor. Each pair of Cortex A57s is linked to its own 1MB of L2 cache, hence the “up to” 4MB of shared L2 cache listed in the slide. Though the top-end A1100s feature eight Cortex A57 cores, quad-core models will also be offered that have a quartet of cores and their accompanying L2 cache disabled.

There will be three initial A1100-series Opterons. At the top end, the A1170 has 8 cores, with a max CPU frequency of 2GHz. The mid-range A1150 has a similar core configuration, but clocks in at a lower 1.7GHz peak. The A1120 has four cores and 2MB of cache, but also clocks in at 1.7GHz. All of the chips have the same memory limits and operating temperature range. The top two chips have higher 32W TDPs due to their higher core counts, versus the quad-core A1120’s 25W.

Pricing for the top-end Opteron A1170 will hover around the $150 mark it is not clear what the cheaper models will cost.

The Opteron A1100 series SoCs also work with both DDR3 or DDR4 memory types. DDR3 memory will be for lower-cost, and potentially lower-clocked solutions.