Using graphics processing units (GPUs) to perform general purpose computing (GPGPU) has been all the rage in the past few years. In fact, the second fastest high performance computer (HPC) in the world today uses NVidia GPUs. But for all their benefits, there is a problem with using GPUs to solve problems. The problem is that the bandwidth between the GPU and the rest of the system is limited to what the PCI Express bus can offer. Again, this can be seen in the second fastest HPC system in the world. That machine can only reach 46% of theoretical peak performance, while the fastest HPC machine can reach 80% of theoretical peak performance. The solution is to run as much of your code on the GPU as possible. If a researcher can do that, then they can potentially run their codes hundreds of times faster than on a CPU.
That is what makes the latest paper from Intel so interesting. In the paper, Intel showed that you could only achieve a 14x speedup. That is a nice, backhanded compliment, Intel.
The paper is called “Debunking the 100x GPU vs CPU Myth” and it is indeed true that not *all* applications can see this kind of speed up, some just have to make do with an order of magnitude performance increase. But, 100X speed ups, and beyond, have been seen by hundreds of developers.
via The NVIDIA Blog – “GPUs Are Only Up To 14 Times Faster than CPUs” says Intel.
Back on December 7th, we reported that Intel’s GPU chipset codenamed “Larrabee” was shut down. While later rumors claimed that Larrabee would rise from the dead like Lazarus, it appears that today the final nail has been pounded into the coffin. In a blog post, Bill Kircos, Intel’s Director of Product & Technology PR, has said that in the short term future, “Intel will not bring a discrete graphics product to market”. You can read the relevant portion of his blog post below.
In a nutshell, Intel has three visual computing efforts. The first is the aforementioned processor graphics. Since we began integrating graphics inside our chipsets back in 1999 (and now integrate graphics inside our processor products), the majority of PC users are now using integrated solutions. Second, for our smaller Intel® Atom™ processor and System on Chip efforts, and third, a many-core, programmable Intel architecture and first product both of which we referred to as Larrabee for graphics and other workloads. Here’s the latest:
1. Our top priority continues to be around delivering an outstanding processor that addresses every day, general purpose computer needs and provides leadership visual computing experiences via processor graphics. We are further boosting funding and employee expertise here, and continue to champion the rapid shift to mobile wireless computing and HD video – we are laser-focused on these areas.
2. We are also executing on a business opportunity derived from the Larrabee program and Intel research in many-core chips. This server product line expansion is optimized for a broader range of highly parallel workloads in segments such as high performance computing. Intel VP Kirk Skaugen will provide an update on this next week at ISC 2010 in Germany.
3. We will not bring a discrete graphics product to market, at least in the short-term. As we said in December, we missed some key product milestones. Upon further assessment, and as mentioned above, we are focused on processor graphics, and we believe media/HD video and mobile computing are the most important areas to focus on moving forward.
4. We will also continue with ongoing Intel architecture-based graphics and HPC-related R&D and proof of concepts.
via Technology@Intel · An Update On Our Graphics-related Programs.
The Huron River platform is the seventh-generation Centrino platform. The mobile processor associated with the Huron River platform is a Sandy Bridge CPU. The Sandy Bridge codename refers to a CPU that is the planned successor to Nehalem. In addition to the Sandy Bridge CPU, the Huron River platform will include a new version of Intel HD graphics. This graphics chip will be capable of supporting version 1.4 of HDMI.
Fudzilla is reporting that this Intel graphics chip will support 3D Blu-ray:
When it comes to Blu-ray support, Huron River plans to introduce a Blu-ray Stereoscopic 3D playback support but we can only hope that notebooks based on Huron River will come with a display with refresh rate that can support Blu-ray 3D on your notebook.
via : Huron River 2011 Platform supports 3D Blu-ray
NVidia’s Bill Dally has an article in Forbes where he discusses the impact of Moore’s law, the theoretical doubling of transistor counts every 18 months that would lead to the doubling of performance in CPU’s every 18 months. In recent years, this exponential growth has dropped as higher transistor counts have lead to higher power requirements and heat problems. While companies like Intel have attempted to compensate by introducting quad core, hex-core, and octal-core processors, they still find themselves constantly battling heat and power requirements. This leads Bill to say:
But in a development that’s been largely overlooked, this power scaling has ended. And as a result, the CPU scaling predicted by Moore’s Law is now dead.
This is being erroneously quoted out of context by some sites as NVidia saying “Moore’s Law is Dead”, when it’s actually the opposite.
The good news is that there is a way out of this crisis. Parallel computing can resurrect Moore’s Law and provide a platform for future economic growth and commercial innovation. The challenge is for the computing industry to drop practices that have been in use for decades and adapt to this new platform.
His suggested new platform, no surprises here, is streamline problem-specific processors like GPU’s. Massive core counts directed at specific problems rather than general purpose scenarios.
Life After Moore’s Law – Forbes.com.
Intel has offered Hyperthreading, a hardware level multi-thread optimization technique that turns a single physical core into two “virtual” cores, in their last several processors. In my experience, its use in CPU-heavy applications (like Rendering) hasn’t been worth the penalty, typically increasing render times as you watch CPU’s trade off performance. As one virtual core works, you see the other virtual core drop to 60-70% utilization. The new Nehalem processors claim to have greatly improved the Hyperthreading support, and luckily I just upgraded one of my computers to support these new processors.
My new machine has dual Intel Xeon 5560′s, that’s a total of 8 physical cores running at 2.8Ghz. My machine contains 24G of Ram and is running WindowsXP64 (I know, it’s old) and Autodesk’s 3D Studio Max 2010. A friend (someone much more experienced at 3dsMax than I) built me a scene that rendered using Mental Ray at a whopping 2600 pixels wide. The resulting scene rendered to completion in 23 minutes, 45 seconds.
After a reboot where I enabled Hyperthreading, we attempted the render again on the new 16-core configuration. Exact same setup, just clicked “Render” and it completed in 19 minutes, 14 seconds.
For the math novices out there, that’s a savings of 4 minutes, 30 seconds on the render with nothing but enabling hyperthreading. That’s a saving of almost 20%!
We’re still doing more tests and benchmarks, but so far the results are promising. If you have Nehalem processors, try it yourself and post your own results in the comments!
Cnet has posted an article with their take on the recent updates to the MacBook Pro update. The article comments on the recent graphics switching capability that Apple has included in the new lineup. They also note that on the 13 inch MacBook Pro, Apple updated the graphics chip but did not update the Intel CPU. This dovetails nicely with what Nvidia has been preaching for some time now: that many workloads can be offloaded to the graphics chip.
“Incremental [processing] workloads are being driven more by video and graphics and that’s where Nvidia comes into play,” said Ashok Kumar, an analyst at Rodman & Renshaw. Says Bajarin: “Apple is strategically writing software that is able to harness the GPU.”
Nvidia, Intel vie for lead role at Apple @ CNET
ComputerWorld has an interview with Sean Koehl, a technology evangelist with Intel Labs, who predicts that the Internet will look different in 5 to 10 years. He states that much of the Internet will be in 3-D. That is not much of a prediction, in my opinion. I predict that the weather will be different in 5 to 10 days. However, the title is a bit misleading. This leads me to believe that the fault lies more with the writer of the article than with Mr. Koehl himself. For example, at one point he says:
The Internet may never go fully 3-D, but making 3-D environments broadly accessible is probably capable within five years,” noted Koehl. “I think it remains to be seen but there are certain kinds of interactions people will want in two dimensions, like reading text. The things we’ll do in three dimensions may be things that we don’t do at all on the Internet today because it isn’t feasible.
via : Intel guru says 3-D Internet will arrive within five year @ Computerworld
TechEYE.net has an article up claiming to have an inside source on the Intel Larrabee engineering team that claims news of Larrabee’s death has been greatly exaggerated.
“We were literally hundreds of people, Intel picked some really big hitters and a lot of those people are still hovering around waiting for Larrabee to come online again,” he said.
And when it does resurface, don’t expect it to be poorly bunged together, either. According to our source, some disgruntled ex-Larrabee project employees had been actively spreading fud and lies, leaking bad information to press about Larrabee’s technical capabilities.
Frankly, the article doesn’t say anything we don’t already know. I don’t think anyone in their right minds believes it is truly dead, and even Intel claimed that it was first going to transition it to a parallelism SDK before returning to the original hardware strategy. The fact that ‘hundreds of people’ were involved doesn’t mean much, and that they’re ‘hovering around waiting for Larrabee to come online again’ means (to me) that it has been shelved temporarily and hasn’t changed since the SC09 demo.
The real question is will Larrabee come ‘out of the gate’ ready to stomp AMD & NVidia, or will it be playing a distant second (or third) fiddle to existing GPGPU technology?
via Intel’s Larrabee to do a Lazarus | Chips news | TechEye – All the technology news unfit for print.
Intel has a new “Visual Computing” demo online called ‘TickerTape’, which makes use of DirectX10 and the new Intel Core i7 processor capabilities.
Ticker Tape is a tech demo that showcases complex particle movement using aerodynamic calculations such as lift and drag. This is all done at high framerates by utilizing an n-way threaded framework and SIMD optimizations. Included in the demo is the capability to dynamically vary the number of threads and usage of SSE so the benefits of both can be easily seen.
Source code and binaries are available on their site, but you’ll have to be on Windows 7 or Vista, with a DX10 compliant video card.
via TickerTape – Intel® Software Network.
Not willing to live in the shadow of OnLive, Otoy is back with more details on their own “games-on-demand” service via a parnership with SuperMicro, AMD, and Intel. It’s not a complete replacement for OnLive, at least not at first:
The consumer service will be similar to what Steve Perlman envisions with OnLive, the well-financed games-on-demand service backed by big game publishers and AT&T. But Otoy’s approach is very different. The company is operating in a horizontal fashion, while OnLive is more vertical, doing each layer of the service itself. Otoy is licensing its technology to be used in the supercomputer, which is fueled by graphics and processors from Advanced Micro Devices. Hosting companies will offer the cloud-based service to publishers of games and other apps. And consumers will ultimately subscribe to the services.
The hardware is pretty impressive. Based on AMD’s Fusion Render Cloud technology (which frankly, I thought was dead.. Haven’t heard from this in a while), they estimate that 10 supercomputers could support 1-million users. What’s in their supercomputer?
A supercomputer will consist of 128 servers, with a total of 250 AMD “Mangy Cours” Opteron microprocessors and 500 graphics chips based on AMD’s Cypress designs. Each of those graphics chips can process 2.7 teraflops, or 2.7 trillion math operations per second. Each supercomputer could serve 3,000 high-definition users, or 12,000 standard-definition users. Otoy’s own software on a consumer’s own machine is tiny, taking up just four kilobytes of data.
So will 2010 be the year of Remote Gaming? So far we have OnLive, Otoy, and Gaikai all looking at summer releases.
via Otoy says supercomputer will enable revolutionary games-on-demand service | VentureBeat.