Add slots and watts instead of a new HPC workstation for increased GPU Compute performance!
1-800-829-0550
Email us
Email us
While at SIGGRAPH last month, I got the opportunity to attend a terrific demo at the Fusion-io booth, and speak with Fusion-io reps about SSD technology. SSD, or solid state disk, storage technology has made tremendous strides over the past few years, and Fusion-io is a leader in this market. They manufacture a line of PCI Express-based SSD products featruing non-volatile NAND flash technology, which outperforms conventional, SATA-based SSDs by as much as 2-3 times the speed on both reads and writes.
Fusion-io has been selling their products into many different verticals, but have recently committed to digital cinema and broadcast verticals. The ioExtreme and ioExtreme Pro solutions can dramatically speed up Windows or Linux-based digital workflows, much like NVIDIA CUDA-enabled products do for GPU-accelerated applications. Fusion-io's proprietary Xlink technology enables multi-card configurations for increased performance which scales almost linearly with each additional ioExtreme Pro card.
So, why am I mentioning this technology? As you scale performance up by adding more ioExtreme and ioExtreme Pro cards to your PC workstation or server, you run into the same problem which brought you to this website in the first place - slots & watts!
Cubix's GPU-Xpander provides artists and editors with the ability to add both NVIDIA CUDA-enabled GPUs and PCIe-based ioExtreme cards to your current PC workstation for dramatically enhanced Adobe CS5 or Autodesk 2011 application performance. Using only a single PCIe x8 or PCIe x16 slot, Cubix GPU-Xpander gives you an additional four PCIe x16 slots and an auto-ranging 750w power supply - all external to your current machine. Save money, save time, save slots.
Fusion-io provides some compelling data at their website, demonstrating how DCC and NLE customers who have adopted their technology have reduced lead times required for animation, compositing, and editing projects. For more details and pricing on ioExtreme and ioExtreme Pro, go to http://www.fusionio.com/products/ioxtreme/ - tell them Cubix sent you!
Eric Fiegehen, Director Visual & GPU Compute Solutions
ericc@cubix.com
Cubix and Refractive Software received a tremendous amount of interest at last week's SIGGRAPH show! The two buzz-word phrases you could hear in conversations were 3D and GPU rendering. Special thanks go out to Octane Render's Haven Sole for all of his help last week at the booth! We had approximately 300-400 visitors over the 3 day exhibition.
It's amazing how much the computer graphics markets have changed in just the past couple of years. We saw some incredible demos of bleeding-edge 3D visualization as close as the booth right across from where were at in the LA Convention Center.
NVIDIA announced their new Quadro lineup, all new Fermi-based cards ranging from the Quadro 4000 (replaces 3800), Quadro 5000 (replaces 4800), and the big boy Quadro 6000. Impressive products, and I found the single-wide Quadro 4000 to be a very interesting option for GPU-Xpander. We can load 4 of these $1100 Quadro 4000 cards per GPU-Xpander, each card with 2GB ECC VRAM and dual-link DVI plus Displayport connectivity.
$1100 may seem like alot for a professional-class graphics card, but you get alot for the money: 2X-3X the performance of Quadro FX 3800 or 4800, double the VRAM of 3800, and better support and warranty than GeForce.
More SIGGRAPH info to follow next week after I get caught up on things in the office.
A new client up in Canada (hi David!) contacted me last week with some interesting questions and points to make. His approach to deciding upon and deploying a physically-based rendering / 3D modeling workflow project struck me as being very methodical and pragmatic. He's been researching the technologies and vendors very thoroughly, and even though I've answered several of his questions in earlier posts (see GPU Xpander Technical Note from May 5th, https://www.cubixgpu.com/Blog/1/18), he made a couple of points of concern worth exploring in-depth:
(1) Design of a smooth workflow
(2) Hardware configuration options to consider
(3) What's bleeding edge and what's not
(4) Middle-ware considerations
Smooth workflow design: This point, in my opinion, is the hardest for me to answer. What are YOU trying to accomplish? What applications or activities are YOU planning to include in your digital content creation workflow, DI or NLE pipeline? Deciding upon this point is a prerequisite for any other planning or product research.
Hardware configuration options: As much as I hate to admit it, this is really something that, in most cases, needs to be looked at after the applications and software packages have been considered. However, thanks to NVIDIA and GPU Compute technology, hardware decisions are no less critical to designing and deploying a DCC, DI, or NLE workflow than the software running on the hardware platform you're trying to decide upon. What are the minimum hardware requirements specified by the software vendor(s) in their technical documentation? What's an optimum configuration? Do they support multi-GPU performance scaling (i.e., Refractive Software's Octane Render or Bunkspeed SHOT)? Is the application fully GPU-enabled or a CPU-GPU hybrid?
When talking about applications that are at least partially GPU-accelerated, bus bandwidth and drivers are the critical factors in determining what kind of hardware configuration you need to go with. Adobe's Creative Suite 5 and Autodesk's 2011 product line-ups feature lots of GPU-accelerated technologies. Even though neither vendor supports multi-GPU performance scaling at the moment, CS5 requires both a graphics card for GPU acceleration activities, and an OpenGL 2.0+ certified graphics card connected to your monitors. I haven't yet checked the requirements for Autodesk's 2011 products, but it seems logical that the same thing would apply since GPU-acceleration can take up most, if not all, of a GPU's performance capacity.
Fortunately, Cubix now offers the ability to scale PCI Express bus performance to whatever specs your particular applications require. Not hitting the PCI Express bus often? Go with GPU-Xpander with PCIe x1 adapter, ExpressCard, or PCI x8 adapter and data cable. Hybrid applications? Go with either GPU-Xpander x8 PCIe adapter or the new PCIe x16 option. If the application is very bus-intensive, you won't want to purchase a x1 or ExpressCard version of GPU-Xpander just to save a few hundred dollars or to justify using your laptop PC. Cubix can still sell you a GPU-Xpander x8 PCIe system with two NVIDIA GeForce GTX 480 cards for less money than you can purchase a new PC capable of running these high-performance beasts.
Bleeding Edge Technologies: Believe it or not, Cubix has been integrating and selling bus-expansion technologies into several markets for decades. Whatever you might think of GPU-Xpander, Cubix Visual & GPU Compute Solutions is very experienced and familiar with its underlying, 100% hardware-based technology.
Middle-ware considerations: Another tough one for me, or Cubix, to answer. The answer depends, again, on the software requirements. Cloud computing technologies and new products from software vendors such as Citrix, Microsoft, and VMware are in the process of re-writing the rules of thin-client and virtual PC deployments. Can the application benefit from GPU-acceleration? If so, adding Cubix GPU-Xpanders to legacy blade servers and pizza-box 1U servers will be the most cost-effective means of deploying these new technologies into cramped, highly-populated data centers.
Eric Fiegehen
Director, Visual & GPU Compute Solutions
Cubix Corporation
ericc@cubix.com