CPU, APU & Chipsets News - Page 139

It looks as though Intel is preparing to retire the Celeron name and replace it with Pentium. Intel's Pentium name has been with us for over ten years now, and debuted as the high-end of processors for notebooks and desktops when it first stepped into the limelight.

Intel has positioned the Pentium line between the high-end Core range and the low-end Celeron range, but are reportedly aiming a new processor for its Pentium line in the server market. The new processor is the Pentium 350 and it is aimed at low-end servers. The Pentium 350 is a dual-core part, featuring 3MB cache and a low power draw of just 15W.

The processor is bare bones, with no integrated graphics. The new chip will be targeting the microservers rather than high-end servers where performance is required. The market is served by other parts in the Xeon line.

Personally, I think we're at the edge of where we'll see more performance from CPUs as games become more reliant on GPUs, console ports and better coding. Why would we require 12-cores at 4GHz each? When will it end?

Well, Intel's Ivy Bridge is the next step for Intel, coming spring 2012. Chinese website Coolaler has gotten their mits on an engineering sample of Intel's 22nm-based Ivy Bridge platform. They've scored a quad-core chip, and have some screenshots of CPU-Z and Task Manager (without HyperThreading enabled).

AIDA64 Cache & Memory benchmarks have also been run. The chip is a 2GHz sample, with CPU-Z reporting it's running at 2.4GHz thanks to turbo boost (20-percent OC) on a Gigabyte GA-Z68X-UD7-B3 motherboard. What to expect? Roughly the same performance as current Sandy Bridge-based CPUs, but with ramped up northbridge and graphics core.

If you're just getting over this week's launch of Intel's Sandy Bridge-E CPUs, you might want to hold your breath just a little bit longer. Less than a week since its launch, Sandy Bridge-E is now sounding old with its successor taking shape over at Intel, with a release date for Q4 2012.

That's not the good part, the good part is that according to a leaked internal slide scored at XFastest, Ivy Bridge-E will be compatible with today's Intel X79 platform, and LGA2011 socket. This makes a SB-E upgrade that much better, as 12 months from now you can just buy an Ivy Bridge-E CPU, slot it in and away you go. You might need a BIOS flash to get there, but it makes a SB-E upgrade not so scary.

Ivy Bridge-E will be built on a 22nm silicon fabrication process, and shouldn't be much more than a shrink of SB-E. We could see the shrink allowing some headroom for enabling some of its components that have been locked away for SB-E. The fastest SB-E silicon physically holds 8 cores and 20MB of L3 cache, but the fastest SB-E-based Core processor has just six of those cores, and 15MB L3 cache enabled.

NVIDIA have just announced that the Barcelona Supercomputing Center (BSC) is developing a new hybrid supercomputer that, for the first time, uses energy-efficient, low-power NVIDIA Tegra ARM VCPUs, together with high-performance NVIDIA CUDA GPUs. BSC is in the planning stages and hopes to roll out the new ARM-based supercomputer, with a near-term goal of demonstrating two to five times improvement in energy efficiency compared with today's most efficient systems.

BSC's ultimate research goal is to establish exascale-level performance while using 15 to 30 times less power than current supercomputer architectures. This so-called EU Mont-Blanc Project will explore next-generation HPC architectures and develop a portfolio of exascale applications that run efficiently on these kinds of energy-efficient, embedded mobile technologies.

As ARM gains more support around the world with ARM-based initiatives, NVIDIA have also announced plans to develop a new hardware and software development kit. NVIDIA's new kit, with hardware developed by SECO, will sport a quad-core NVIDIA Tegra 3 ARM CPU accelerated by a discrete NVIDIA GPU.

While suited toward a different crowd for the most part, AMD has made efforts to steal a bit of the limelight strongly shining on Intel and its Sandy Bridge-E/X79 platform by kicking out a big announcement of their own at the same time.

Adter several weeks of being delayed, today finally sees the launch of AMD's first 16-core Opteron server processors, the Bulldozer based Opteron 6200 series (formally known as Interlagos). According to John Fruehe, the director of product parketing at AMD, these new processors are said to be 25 to 30% faster than their 12-cored Opteron 6100 predecessors.

The new Opteron 6200 family of processors includes 6262 HE, 6272, 6274, 6276 and 6282 SE models, running at clock rates of between 1.6 and 2.6GHz with pricing starting at $523 US and going up to $1,019 US. Power draw from the new processors starts at 85 watts and goes as high as 140 watts. The chips are compatible with existing server sockets to make upgrades from older 12-core chips a breeze.

Below, we have a picture of the AMD accelerated processing units in both FM1 and future FM2 packages, which the latter will be based on the next-generation Trinity APU. In the picture below, "Llano" FM1 APU is on the left, "Trinity" FM2 is on the right.

Both packages are pretty much identical, with the difference between them a set of blanked pins in the sub-central portion of the pin array. The FM2 package sports 904 pins whilst the FM1 package has 1 more at 905 pins. One pin is blanked, whilst a pair of blanked pins are arranged further away from the central cutout.

What does this mean? It means that FM1 and FM2 are completely incompatible, and won't allow you to run today's A-Series APUs in the FM1 package on future socket AM2 boards, nor will you be able to run future FM2-based CPUs on today's FM1 motherboards.

Intel's new Sandy Bridge-E flagship processor, the Core i7 3960X has been thrown through a handful of benchmarks by Chinese website Inpai.com. The benchmarks put the i7 3960X head-to-head against the Sandy Bridge-based i7 2600K.

Intel's Core i7 3960X will be the bees knees of the Sandy Bridge-E series, featuring 6-cores and 12 threads clocked at stock clocks of 3.6GHz and 3.9GHz when using Turbo Boost. The chip uses a maximum TDP of 130W, features 15MB L3 cache and launches on November 14.

Inpai.com tested the i7 3960X and i7 2600K against each other in various benchmarks, such as X264 Benchmark, which saw the i7 3960X score 47.4 versus the 2600K's 37.21, a decent improvement. Cinebench sees a bigger improvement, with the i7 3960X scoring 10.4 and the 2600K trailing far behind with 6.92.

ARM have just announced its next-generation mobile GPU, the Mali-T658. ARM have said that this design is set to offer ten times the performance of their current Mali-400 MP which is found in smartphones such as the amazing Samsung Galaxy S II handset.

T658 is ARM's second GPU using an architecture it calls Midgard. Midgard is designed to support both 3D workloads using modern APIs, with support for both OpenGL ES and Microsoft's Direct3D 11, and computation workloads using OpenCL, Microsoft DirectCompute, and Google RenderScript. Compute tasks get some love, with Midgard supporting full IEEE 754 floating point.

The first Midgard design, the T604, was announced last year, and was licensed by companies such as Samsung and LG. The new T658 is quite the powerhouse, with each core having four arithmetic pipelines, and one each of load/store for texture, doubling the number of arithmetic pipelines found in the T604. Up to eight cores can be integrated into a single GPU, again representing a doubling in performance relative to T604, which allows up to four cores to be ganged together.

Ivy Bridge isn't even here yet, and we're seeing details on Intel's next-generation processor architecture, Haswell. Haswell will be a brand new architecture based on a 22nm fan process which will be matured by Sandy Bridge's successor, Ivy Bridge. This is Intel's "tick-tock" product development model, where we see a "tock" being a new x86 architecture, and a "tick" miniaturizes it to a newer silicon fabrication process.

If Intel's 22nm process all goes to plan, Haswell has an ETA of Q2 2013, with Ivy Bridge taking the lead from Q2 2012 until then. Haswell will feature an all-new socket, LGA1150, which means it will not be compatible with LGA1155 boards. The reason behind this is that Intel have implemented drastic changes in the pin map of the package.

There has been a major change with the component arrangement in the platform that is affecting Haswell's pin map, more specifically, it's Haswell's higher bandwidth chipset bus, rearranged PCIe pins (with FDI pins), rearranged power pins, and miscellaneous pins. It also throws away the separate power domain for the integrated graphics controller.

AMD's Bulldozer just wasn't enough to not only topple Intel, but to impress customers enough to grab it while it was hot. AMD knows they need to turn it around, we know they need to turn it around and you know they need to turn it around. AMD has always focused on performance-per-watt, and while Fusion has pulled this off, Bulldozer when overclocked is pushing past an astonishing 400W.

The Bulldozer team needs to put their hard hats on and get dirty, with 2012 rolling into view, it needs to happen very soon. Piledriver is an update to current Bulldozer cores and the expected advantages are increased core capacity by up to 25-percent, making more cores available more of the time. Reduced power requirement, making Piledriver more efficient, real world performance boost by somewhere around 10-percent, making each core do more.

Will this be enough for AMD to at least catch up and possibly take a few swings at Intel's upcoming Ivy Bridge? Can they bring it before Ivy Bridge, or at the same time? Can AMD release it at a competitive price point, seductive enough to draw customers into their web of CPU goodness? Bulldozer was roughly the same price as an Intel 2600K motherboard/CPU combo, so they really failed in that way.