TweakTown NewsRefine News by Category:
CPU World is reporting the different configurations that Intel's upcoming Haswell processors will come in, with respect to graphics core, memory channels, and processing cores. Some of the configurations break from the current Ivy Bridge in that the highest performance CPUs don't have the highest performing GPUs.
This actually makes sense in that a user who buys a top CPU will likely be running a discrete GPU. Intel is offering at least two different graphics configurations which will be joined to dual- and quad-core parts. The dual-core parts will also have a variation in that they may only come with one memory channel.
The graphic above summarizes the different configurations we are reportedly going to see with Haswell. Each memory channel will support either one or two DIMMs, with the ULT (low power) chips seeing the one DIMM per channel memory controller. Each DIMM will have a maximum capacity of 8GB, so the lowest amount of memory (or highest, depending on point of view) will be 8GB on the ULTs.
We're still a good distance from the release of Haswell, so it's a good idea to take these configurations with a grain of salt. That said, they do make sense, so it's not unlikely that they are accurate.
Globalfoundries has just inked a multi-year deal with mobile chipmaker, ARM. The deal will see Globalfoundries produce ARM's next-generation mobile system-on-chip (SoC) processors.
Glofo's factories are equipped to handle 20nm and FinFET process technologies, with the deal also calling on ARM to develop its Artisan Physical IP platform, which includes cell libraries, memory compilers, and POP IP solutions.
The chips that will be made will be baked into future smart devices, and ultra-thin notebooks. ARM and Globalfoundries have been working together for a few years now, where we've seen them developing ARM Cortex A-series processors at Glofo's Malta, New York fab. The new deal should see the companies working on more energy-efficient ARM Cortex CPUs and ARM Mali GPUs for mobile devices.
Samsung have just announced their brand new Exynos 5250 SoC, which is now called the Exynos 5 Dual. Samsung's Exynos 5 Dual is the world's first Cortex A15-based chip, and should be baked into devices later this year. The first of which should be Samsung's 11.8-inch tablet that sports the 2560x1600 resolution, that we've talked about a few times now. The important specs are:
- Dual-core 1.7 Ghz Cortex A15 CPU
- Mali T604 GPU
- OpenGL ES 3.0
- OpenCL 1.1 full profile
- Support for WXQGA displays
- Wi-Fi display support
- 12.8 GB/s memory bandwidth with 2 port 800 Mhz LPDDR3 RAM support
- 1080p 60 FPS video performance and VP8 codec decoder
- USB 3.0 support
Intel is working on another high-end desktop (HEDT) chip codenamed Ivy Bridge-E. In many ways it is similar to Sandy Bridge-E in that they are increasing the core, cache, and memory channel count to produce the chip. According to a leaked Intel roadmap, this chip is sent to be released Q3 2013, sometime after Haswell makes its debut.
The roadmap above is the latest leaked version and I have some suspicions that it could be a fake. Look at the word Haswell. Do you see what I see? I find it highly unlikely that an Intel roadmap would have red, squiggly incorrect spelling marking underneath an actual roadmap. This makes me lean towards a fake.
That said, Ivy Bridge-E is said to run on existing X79 chipset motherboards and the LGA2011 socket. It is possible that Intel will produce a new chipset to go with Ivy Bridge-E, but they previously kept the X58 chipset over two generations of HEDT chips (45 nm Core i7 "Bloomfield" and 32 nm Core i7 "Westmere").
Athlon chips just won't die. While originally the top of the market, AMD changed the name usage to be used in conjunction with budget parts with the introduction of the Phenom parts. Even when Llano came out, the Athlon's got an upgrade into the FM1 socket and are basically the same, minus the on-board GPU.
And know we can say that the Athlon processor will continue to live on, even after AMD transitions to the FM2 socket. AMD is planning three new Athlon chips for the socket, and while we know only the most basic details, the main takeaway from this is that the Athlon brand will continue to live on for another day.
The three new chip models are the Athlon II X4 730, 740 and 750K. Athlon II X4 730 and 740 are 65 Watt processors and operate at 2.8 GHz and 3.2 GHz. The 750K is a 100 Watt product and should have an unlocked multiplier, due to it being a K processor. They will all be quad cores with 4MB L2 cache and will be based upon Piledriver.
There's not much more to report other than they will likely feature TurboCore, though frequencies are unknown, and the on-board GPU is likely to again be disbaled.
Websites are reporting that Intel is planning on releasing a new flagship processor in Q4 of this year. The new processor, called the i7-3970X, will utilize the LGA2011 socket and take the title of flagship. The report is claiming that the processor could ship with clock speeds of 3.5GHz and TurboBoost up to 4.0GHz.
Of course, as it is an Extreme Edition processor, it will feature an unlocked multiplier and should overclock like a dream. It will run off of the current SandyBridge-E silicon as the current LGA2011 processors and will be built on the 32nm production process. It will pack 6 cores and 12 threads through the use of HyperThreading.
It is rumored to sport 15MB of last-level cache and will pack a quad-channel DDR3 integrated memory controller, supporting up to 64 GB of DDR3-1600 MHz memory. Pricing will likely be in line with the current i7-3960X and this new chip will likely cause that chip to be gradually phased out in favor of this new, faster chip.
AMD has announced a new line of APUs for use in embedded applications. The new APUs are the first to combine a low-power CPU and a discrete-level GPU into a single integrated circuit for use in embedded applications. The high performance graphics capabilities allow the APU to power a wide array of devices such as slot machines and airline schedule screens.
AMD bills the part as being perfect for Digital Signage, x86 Set-Top-Box (xSTB), IP-TV, Thin Client, Information Kiosk, Point-of-Sale, and Casino Gaming, media servers and industrial control systems. The APU comes in a BGA package featuring a new, power-optimized CPU processing core to provide more bang for the user's buck.
The features are as follows:
- DirectX® 11 support lets you enjoy awesome graphics performance, stunning 3D visual effects and dynamic interactivity.
- Advanced discrete-level GPU with OpenGL 4.0 and OpenCL 1.1 support in an integrated device provides support to build the designs of tomorrow, today.
- Unprecedented graphics performance/watt thru advanced graphics and hardware acceleration delivering over 3X performance per watt of previous generation.
- Selective models, T56N and T40N, have additional boost capability enabled by AMD Turbo Core technology without additional power draw.
There's still a fair bit of time before Intel releases server chips based upon the Ivy Bridge architecture, but that isn't stopping avid readers and power users from questioning what comes next. Next, by the way, is Haswell, which is the successor to the current Ivy Bridge architecture and is a completely new design.
The leaked slides would appear to be real and detail some of the chip's features that we should expect when it releases. As far out as it still is, some of these details could be subject to change. The slide claims that there will be minimum of 10 cores per CPU and will sport up to 35MB of total LLC, about 2.5MB a core.
It is said to have 40 PCI-e lanes and will sport Hyperthreading and Turbo Boost. It appears that the new CPU will be getting a new socket (of course) called R3 or Socket R3. This is most likely so that it can handle the 4 channels of DDR4 operating from 1333MTS all the way up to 2133MTS. Additionally, it sports two QPI paths so there will be a max of two CPUs per system.
Intel DDR4 servers are supposed to debut in 2014, so don't be expecting Haswell-EP before then. That puts 2014 as the most probable time for when Haswell-EP will shed its NDA and more information will come available.
Toshiba have just made quite the leap, announcing they have developed a low-power, 64-core System-on-a-Chip (SoC) for embedded application in some areas which could be destined to automotive products and digital consumer products.
The insane chip sports 64 cores, claiming to feature eight times as many as its multi-core predecessor unveiled in 2008 and operates at an insane 14 times faster. The 64-core chip sits on a die area of 209.3mm2, two 32-core clusters are then integrated with dynamically reconfigurable processors, hardware accelerators, dual-channel DDR3 memory controller and other peripherals.
In one cluster, we'll find processor cores that share 2MB of L2 cache connected through a three-based network-on-chip (NoC). The high scalability and low power consumption is accomplished by the parallelized firmware for multimedia applications.
Take a quick stroll down memory lane and find Larrabee. Larrabee was Intel's failed discrete graphics creation designed to go head-to-head with the likes of NVIDIA and AMD. The Larrabee project was canned but Intel decided to continue the work with a change of focus. Instead of being a discrete GPU, Intel decided to target the HPC and supercomputing markets.
This new chip is an in-development chip known as Knight's Corner and is designed to be massively parallel. The product has clearly made significant progress as Intel has officially attached a brand name to the product. That brand name is Xeon Phi and will also be attached to future many-integrated-core (MIC) products.
Intel chose Xeon Phi as they wanted it to be part of the Xeon family. As for the Phi part, Intel says it's because it "evokes many concepts in science & nature including the 'golden ratio' in mathematics." Along with the brand name, Intel has released some technical details of the upcoming MIC chip.
The chip will be in production "in 2012" and be fabricated on Intel's 22nm process. The chip will feature more than 50 cores and "8GB+ GDDR5 memory" on a PCI Express card interface. Intel has asserted that the chip is capable of 1+ teraflops of double-precision compute throughput in Linpack. Intel has built a Knight's Corner cluster capable of producing 118 tflops as a proof-of-concept.