Maximum PC multi-core RSS Feed

Tilera Announces 100-Core CPU for Data Centers

Paul Lilly — Mon, 26 Oct 2009 08:17:01 -0500

Tilera today announced its new TILE-GX line of processors, including the TILE-Gx100, the world's first 100-core CPU. According to Tilera, the 100-core part offers the highest performance of any processor on the planet by at least a factor of four.

"The launch of the TILE-Gx family, including the world's first 100-core microprocessor, ushers in a new era of many-core processing. We believe this next generation of high-core count, ultra high-performance chips will open completely new computing possibilities," said Omid Tahernia, Tilera's CEO.

While the 100-core part is not meant to run Crysis (so please don't ask) or any other desktop application, it does offer 10 times the performance per watt as Intel's fastest Nehalem-based server chips. Assuming Tilera can convince customers to switch from Intel and Texas Instruments, The TILE-Gx100 will likely end up in data centers powering cell phone network equipment and cloud computing ventures.

Tilera says its 100-core chip will start shipping in Q4 of this year.

Image Credit: Tilera

IBM Gives Details on Octo-Core Power7 Processor

Paul Lilly — Wed, 26 Aug 2009 15:01:32 -0500

The CPU wars in the desktop market have grown pretty stale with Intel's Core i7 architecture kicking AMD's tail, but when it comes to the server sector, the battle is starting to heat up.

Enter IBM, who at this week's Hot Chips conference officially unveiled its muscular 8-core Power7 processor. The mighty chip is expected to pack 1.2 billion transistors onto a 45nm die. Each core will boast 12 execution units, as well as 32 threads per chip and advanced pre-fetching data and instruction sets.

"I am sure Power7 will be the fastest processor around, probably faster than Intel's Nehalem in some benchmarks," said Nathan Brookwood, principal of market watcher Insight64.

Other specs include scalability up to 32 sockets, 256KB L2 cache per core, 32MB of on chip eDRAM shared L3 cache, dual DDR3 memory controllers, 100GB/s memory bandwidth per chip, and 360GB/s SMP bandwidth per chip.

Image Credit: IBM via TGDaily

Windows 7 Gets Optimized For Intel Chips

Justin Kerr — Sun, 26 Jul 2009 16:05:45 -0500

Writing an operating system is no easy task, particularly for Microsoft. Having to design a piece of software that will maintain compatibility across a nearly infinite number of hardware configurations can be a daunting mission. It becomes even more difficult when hardware manufacturers implement new technologies, but the software lags behind.

Determined to not let that happen to Windows 7, Microsoft engineers have reportedly worked side by side with Intel to optimize support for its processors. In a posting made on the new Windows 7 partner blog, Microsoft’s Brandon LeBlanc highlights the results of the collaboration. “Working with Intel, Microsoft implemented a new feature called SMT parking, which provided additional support for the Windows 7 scheduler for Intel Hyper-threading Technology, enabling better performance on hyper-threaded, multi-core Intel processors.”Intel and Microsoft have also been working on optimizing boot/ shutdown/ sleep/ and resume times, which could end up giving the chipmaker a substantial edge in performance for the foreseeable future.

Intel’s rival Nvidia also posted a follow up on the partner blog outlining GPGPU functions in Windows 7, and encouraged upgraders to spring for dedicated graphics cards. Ultimately the Intel announcement is more interesting since the DX Compute found in DirectX 11 will favor both ATI and Nvidia equally and it certainly makes me worry about the growing performance gap between Intel and AMD.

Will this announcement influence your upgrade decision?

Edit: Brandon LeBlanc works for Microsoft, not Intel.

Study: More Cores Not Always Better

Paul Lilly — Thu, 15 Jan 2009 13:59:39 -0600

For a long time, both Intel and AMD relied on ever increasing clockspeeds for each new processor release. That still remains the case today, but to a much lesser degree. Case in point - Intel's long retired Northwood line topped out at 3.4GHz, or 200MHz faster than the zippiest Core i7 processor currently on the market.

The future of chip design has shifted to where multiple cores is now main factor, supported by larger cache, on die memory controllers, expanded instruction sets, and other secondary concerns. That's all well and good that AMD and Intel are on the same page, which puts the onus on software developers to catch up, but at least one group of researchers believes we're headed for an unpleasant surprise.

According to Sandia National Laboratories, performance is going to start tapering off significantly as chip makers keep piling on more cores. Sandia came to the conclusion by running a simulation consisting of "key algorithms for deriving knowledge from large data sets." During the simulation, few performance gains were had from moving from four to eight cores. But the real kicker is going beyond eight cores resulted in a performance drop. When that number gets as high as 16 cores, Sandia warns "a steep decline is registered as more cores are added."

So what the heck is going on that would cause multiple cores to stumble so unexpectedly? It comes down to a bottleneck in memory bandwidth. And what's scary is that the bottleneck isn't an unknown problem, but "it isn't an issue to which the industry has a known solution, and the problem is often ignored."

We're still quite a ways off from 16-core processors making it into the mainstream, and developers have yet to fully tap into even dual-core processors on a consistent basis. So while there's plenty of time to come up with a solution, chip makers haven't yet started doing so, according to Sandia.

Image Credit: TechReport.com

Report: Roadblock Ahead for Multi-Core Processors

Paul Lilly — Mon, 08 Dec 2008 13:30:06 -0600

In case you haven't noticed, multi-core processing has taken hold and the race is on to cram more cores onto a single die. But assuming developers can keep up, at some point, chip manufacturers are going to have address a potential major problem that could make adding more cores a useless endeavor. More specifically, a "memory wall" looms large in the not too distant future that, as Jon Stokes from ArsTechnica puts it, could make more than 16 cores pointless.

The problem stems from memory bandwidth not being able to keep pace with faster processors, whether those speed bumps come from a faster frequency or more cores. Put simply, memory is creating a bottleneck and can't feed the processor fast enough, a problem that has existed for some time. Intel and AMD have been able to mask the problem by adding more cache, but doing so doesn't overcome the memory wall, which looks poised to really rear its ugly head as more cores are piled on to new chip packages.

"Engineers at Sandia National Laboratories, in New Mexico, have simulated future high-performance computers containing the 8-core, 16‑core, and 32-core microprocessors that chip makers say are the future of the industry," writes Samuel K. Moore at IEEE Spectrum Online. "The results are distressing. Because of limited memory bandwidth and memory-management schemes that are poorly suited to supercomputers, the performance of these machines would level off or even decline with more cores."

According to the simulation, performance gains level out at 8-cores, with 16-core chips performing no better than a dual-core processor. As the core count increases, the performance starts to decline sharply.

One solution Sandia has proposed is to stack memory chips on top of the processor, something both Intel and IBM have been working on. Some might also remember reading about multi-core memory, a new memory architecture developed by cryptographer Joseph Ashwood. But no matter what the eventual solution, someone's going to have to come with something relatively fast. Sixteen core chips might not be as far off as you think, and the memory wall isn't going to crumble on its own.

Image Credit: Tpan.com

Hands-on with Deep Viewer, Intel's Potential Killer-App for Nehalem

Norman Chan — Thu, 21 Aug 2008 16:30:00 -0500

During a private briefing with Intel at IDF yesterday to talk about Nehalem, we were given a demo of some cool software in development that makes good use of the multi-threaded cores of the new CPU. Francois Piednoel, the Senior Performance Analyst (ie. benchmarking guru) at Intel describes Deep Viewer as a "science project" of sorts. It's an image sorting application that they acquired from an independent software developer that reminds us of Microsoft Live Labs' Seadragon technology (which is used in the recently released Photosynth online app). We're talking about near-infinite scaling of visual data (in this case photos and videos) being processed in real-time on your display.

In the first part of the demo, we were introduced to a Core i7-powered system running in tri-channel memory mode with a fancy-looking 30" display. The monitor was actually a touchscreen (enabled by that exposed silicon around the bezel) and one of the things Francois did first was move some objects around with his finger.

When it started up, the Deep Viewer app showed a small calendar grid next to a world map. Pinching the calendar -- like gesturing on an iPhone -- expanded the frame. And as the grid grew, we could see images populating the space within each calendar day. Francois kept zooming in, and more images revealed themselves on the fly.

This was actually 200GB worth of images, stored off of a 500GB hard drive. Over 224 terapixels of data was accessible in this demo, being streamed into view and processed in real-time. The scaling was incredibly fast, and we were told the app didn't store any cache or pre-process files. The images that popped up to full resolution as we zoomed in weren't just jpegs, either. RAW photos, bitmaps, and other image formats all work with Deep Viewer -- legacy file and codec support is one of the reasons a general-purpose CPU is optimal for this kind of app.

In addition to images, many of the thumbnails were actually videos, too! 640x480 (non-HD) clips shot with a power-and-shoot camera were split up into indexed scenes and played alongside the high-res jpegs. One really cool feature is the app's ability to run facial recognition algorithms on videos when you zoom in on them. We focused on a scene of some passengers walking off of a train, and a red circle highlighted each face to pick up details and find matches in other photos in the database.

Hit up the next page for more!

Looking at the CPU usage during the demo, we could see the Nehalem chip really being taxed as we zoomed in, but we never noticed any hiccups in the performance. Intel's argument is that this is the kind of application that users should be demanding from software developers -- something that'll really utilize the CPU's processing potential.

The magnetic hard drive was cited as the biggest bottleneck in the application, so we were shown another system running with 4 new Intel SSDs configured in RAID. Zooming into the Deep Viewer calendar on this machine looked just as smooth as on the drive-based HD equipped system, but this time, all the images being processed were large RAW image files.

Of course, we don't all have 30" touchscreen displays, so we were shown another technology that Francois is working on in his own off-time -- a motion sensor for Deep Viewer using a standard webcam. He demonstrated gesture controls by putting his hands two feet away from the webcam, pinching and waving in the air to navigate through the interface. His homebrew software looked responsive and accurate, and with a little more work, could bring that awesome motion-gesture tech from the Minority Report film to life.

Intel isn't making any promises that Deep Viewer will be released for free (or even at all). It's definitely an impressive visual computing application that innovates photo album sorting. Once they spruce up the GUI and optimize the software a little more, Deep Viewer could be a real killer-app for Nehalem.

Intel Release’s New Midrange Chips On Monday

Justin Kerr — Sat, 09 Aug 2008 22:20:15 -0500

In an attempt to tighten the screws on AMD, Intel is continuing to roll out new midrange processors early next week. Despite the fact that they aren’t officially released yet, online e-tailers are already taking orders. The CPU’s will be based on Intel’s 45-nanometer process and both Alienware & Falcon Northwest are preparing to announce systems featuring the new parts in tandem with its release. The 95 watt Q9650 currently retails for $559 on Newegg, and features a core clock speed of 3 GHz. This paired with a 12MB L2-cache, and a 1333MHz front side bus make it a solid performer for the price. In fact, this puts midrange consumers within striking distance of the Dream Machine's 150 watt QX9775 which retails for a much heftier $1550. The QX9775 runs only 200MHz faster with the same 12MB’s of L2-cache. The main difference between the two is the 1600 MHz FSB, Skulltrail support, and the subsequent overclocking potential that comes with the extreme series. Looking for something more modest? The Q9400 is rated for 2.66 GHz and will feature 6 MB of L2-cache on a 1333MHz FSB. For those who prefer the dual core design, the Core 2 Duo’s lineup will be receiving an update as well. The new E8600 clocks in at 3.33GHz,with 6 MB of L2-cache and a FSB of 1,333MHz, while the lower end E7300 will sport a 2.66GHz clock with 3 MB of L2-cache, and a 1066MHz FSB. The 65 Watt E8600 is e-tailing for $279 and the E7300 will go for $144.

Ready to upgrade? What are you going to use?

Nehalem Primer: What We Know So Far About Intel's Next-Gen Penryn Killer

Gordon Mah Ung — Thu, 31 Jul 2008 17:08:59 -0500

We’ve been calling Intel’s next-generation CPU family code-named Nehalem a Penryn-killer because, sadly, AMD’s best and brightest have hardly been that. For those who haven’t sifted the sands of the Internet, and picked the brains of OEM’s and hardware vendors for every detail of Intel’s next-gen microarchitecture, here’s your quick primer on Nehalem that’ll make you big man on campus at the next geekfest.

How Many Cores?

Most Nehalem’s will be native quad-core with all four compute cores on the same physical die. Intel says that the design of Nehalem will also let the company build an eight-core version, codenamed Beckton, for servers. Intel also hasn’t ruled out the possibility of a multi-chip version so could perhaps see a double-die Beckton with 16 cores as well.

HyperThreading Returns

An improved version of HyperThreading will find its way into the core of Nehalem. This improved simultaneous multi-threading, or SMT, will let the OS see a quad-core chip as eight cores. Although some still debate its merit, the implementation of HT in the Pentium 4 generally added 15 percent more performance in multi-threaded applications.

Integrated Memory Controller

Nehalem will integrate Intel’s second attempt at an integrated memory controller. The first IMC from Intel was with the ill-fated Timna CPU in the 1990s that cratered when the “PC as cell-phone idea” imploded with the dot-com bust. The IMC in Nehalem will be on-die and DDR3 for consumer CPUs. Beckton, the server-flavor of Nehalem, will continue to support fully buffered DIMMS and their ability to run metric ass-loads of RAM.

In addition to integrating the new IMC, Intel will finally ditch the front-side bus which forces all cores to share data along a shared bus. Instead, Intel will introduce its QuickPath Interconnect that connects all cores with dedicated paths. QPI was previously code-named CSI. Is Intel bashful about adopting two of the key ingredients used in Athlon64/Opteron? Nope. Intel says basically that yeah, they may look the same, but ours is way, way better so that makes it OK. In reality, does it matter who came up with the idea? To us, it’s who can make it faster.

Triple-Channel DDR3

Motherboards for Bloomfield will be a bit crowded thanks to the tri-channel DDR3 layout. As you can guess, you would need to populate three independent channels to get the tri-channel setup. Some boards may feature up to six DIMMs, while others may choose five slots – Three would be populated for tri, and you would have two free slots for additional RAM.

Socket Shuffle

All three Bloomfields will be based on the new LGA1366 socket that is completely incompatible with current LGA775 motherboards. Sometime next year, a value version of Nehalem using a different LGA1160 socket will hit the market. Codenamed Lynnfield, the chip will feature dual-channel DDR3 support instead of tri-channel. The big question is whether Intel will maintain a performance LGA1366 alongside the mainstream LGA1160 product or will LGA1366 get quickly put to rest the way AMD killed the original Socket 940 product in favor of the more mainstream Socket 939.

All indications are that Intel would do exactly that, since OEM’s don’t particularly like having to build two totally different machines. Many prefer to have a single SKU that can be sold as either a mainstream box or a performance machine based on a simple CPU swap out. Going from a mainstream LGA1160 machine to a “performance LGA1366” would mean swapping the motherboard out which means time taken to build the machine and less money in the pocket. However, since Bloomfield shares the same socket with the workstation version (codenamed Gainsetown) there is a possibility that Intel will keep the top-end Bloomfield in business to use as a foil against a possible resurgence by AMD.

How Will Nehalem Perform?

Early reports of Nehalem’s performance is quite good, with some sources telling us that Bloomfield is to Penryn as Conroe was to Prescott. That’s a pretty amazing speed bump, but we must point out that we’ve long thought that Penryn was being sandbagged. With many of the quad-core Penryn’s seemingly capable of running at or near 4GHz and most of the dual-core parts running beyond 4GHz, we’ve long suspected that Intel decided to keep the speeds of Penryn down to keep the yields up. With Bloomfield nearly here, true conspiracy nuts may also wonder if keeping Penryn slow wasn’t a plan to make Bloomfield look even better at launch. It’s not that Bloomfield won’t be fast, but keeping Penryn’s speeds down make the little bars for the next-gen chip look even better.

Availability and Configurations at Launch

The Nehalem CPUs intended for consumers will be offered in three different trims this fall under the Bloomfield codename: A performance chip in excess of 3GHz in the “Extreme” category, a performance version in the 3GHz range, and a mainstream version in the 2.5GHz range. Reported prices map directly onto Intel’s high, mid, low paradigm in use for Penryn.

The top-end chipset for Bloomfield will be called X58. Thanks to Intel and Nvidia burying the axe (although we’re not sure what limb the axe went into) some X58 boards will be the first desktops boards support both SLI and CrossFireX. This will be done by Nvidia selling motherboard makers nForce 200 chips that they can integrate into the boards to “enable” up to tri-SLI support on the boards. Because the deal was very last minute, not all X58 boards may have the chips. Be sure to plan carefully if you want SLI on your X58 machine.

Stay tuned to this page for updates as we find out more about Nehalem!