Did you ever wonder what all the fuss was regarding NVIDIA’s Fermi architecture, or what benefit is serves their already well established array of Quadro professional graphics cards? I did, and set off looking into what is going at the professional graphics side of NVIDIA.
When I was approached by NVIDIA to provide a review on some of the new releases in the new Quadro NVIDIA Fermi-based line of professional graphics cards, I was pleased to have the opportunity to see first hand what all the fuss was about. After all, I have been using graphics cards from the Quadro line-up for years, and have seen the good and bad side of professional graphics cards. Recently I killed the ATI graphics hardware in my Dell laptop (apparently that doesn’t take that much effort), and thought, “I wonder if I can kill a Quadro?”. Don’t get me wrong, I wouldn’t want to trash a Ferrari, but presented with one to beat on, why not? I’d get choked up, sure… right before I floored it.
So I started scheming, and decided that I would throw everything I had at the card. Autodesk Inventor, Civil 3D, Alias Design, Publisher, Fusion, and perhaps some other graphics-intensive applications , in order to see where not only the hardware limits are, but the graphics card drivers as well. So the kind people sent me a box with an NVIDIA Quadro 5000M, that came complete with its own handy carrying case, the Hewlett Packard EliteBook 8740w mobile workstation.
The Quadro 5000M is a complete High-performance mobile graphics processing unit (GPU) that is installed on various powerhouse mobile workstation solutions like the HP EliteBook and Dell Precision lines. The mobile specifications are a bit different than the main Quadro desktop workstation cards due to space limitations inherent in the laptop design. Here are some of the specifications:
- 320 CUDA Cores
- 2 GB GDDR5 Graphics Memory
- Memory Interface: 256 bit
- 76.8 GB/s Bandwidth
- Shader Model 5
- OpenGL 4.1
- DirectX 11
- NView Desktop Display Management
- Error Correcting Code memory (ECC)
- Fast 64 bit Floating Point Precision
- up to 2560×1600 resolution support
While the specifications are a bit less than that of its namesake, the Quadro 5000, it packs one hell of a lot of graphics capability, and includes ECC (parity checking) and 64 bit floating point precision, which some of the entry level Quadro’s do not.
A Chat with NVIDIA
Sean Kilbride, NVIDIA’s Technical Marketing Manager for Workstation Products, agreed to entertain my questions, and chatted about the history and evolution of professional workstation graphics cards, makeup of the new NVIDIA Fermi High-Performance Graphics architecture, and just about everything in between.
What about Old Technology?
My previous workstation graphics card technology research indicated that professional graphics cards in the industry at large were built on GPU technology that was as much as 1-2 years old. The drivers were updated for CAD design and the like, but the GPU is based on standard mainstream technology that is older. So, I asked Sean about the new Quadro pro graphics line, and if things were still the same.
Kilbride said “Our new cards (consumer and workstation) are built on our NVIDIA Fermi architecture, and Quadro GPUs are a professional variant of what we offer to the mainstream, or consumer market (i.e., GeForce). Our new Quadro professional graphics cards typically become available after our GeForce consumer products have launched in order to incorporate Quadro’s professional feature set.”
He went on to explain further, “Our workstation drivers are established for known design applications, and tuned for DirectX and OpenGL. The process takes about 6 months to validate the OpenGL drivers for the new Quadro GPU with new professional applications such as Autodesk Civil 3D 2011. Once that takes place, our Quadro procards are then released.”
What about OpenGL no longer being supported by Windows?
Last year, Microsoft stated that it would no longer support OpenGL. It would run it, but each developer would have to figure out how on their own. Currently, the open sourced language is not built on a strict standard, which offers flexibility for new technology to be adapted. The problem is that numerous design application programmers are writing instructions differently, and each application that communicates via OpenGL forces someone to adapt drivers to communicate properly with the graphics hardware like an interpreter. Additionally, each machine build tends to muddy up the waters, creating a tragedy of sorts for companies like Autodesk, who have to certify each video adapter for each software, so that their customers know what hardware they can trust.
My understanding was that many in the hardware department were happy to think that OpenGL would go away. The Inventor team successfully adopted Direct
”OpenGL is getting a shot in the arm from alternative operating systems like Mac OS X, Linux, and the evolving Google Android platform,” noted Kilbride.
These OS’s all utilize OpenGL for their graphics outputs and do not use DirectX. I think that Microsoft could see the train coming head on, and decided to continue support. Additionally, I suspect that the Mac OS will push Autodesk to continue its OpenGL use for AutoCAD on all platforms and OS’s, and subsequently the reliance on the NVIDIA Quadro pro graphics hardware and drivers.
“Fermi took 36 months to develop, substantially longer than previous platforms, and represents a radical shift in our graphics architecture.“
Graphics cards used to handle simple raster elements, and eventually utilized OpenGL to accelerate performance. This revolutionized the industry with the ability to handle actual triangles efficiently. This was a dream come true for AutoCAD, and its verticals like Civil 3D, which completely rely on triangular meshes. Eventually GPUs came on the scene and with them Texture and Lighting capabilities, where the applications could hand these off to the graphics hardware, instead of the computer’s CPU. Two years later, the GPUs were loaded with vertex and pixel shaders to autonomously deal with various tasks like opacity, etc. NVIDIA Fermi architecture goes beyond all of that, and permits the GPU to perform general purpose calculations. This enables the applications to divert some of the intense algorithmic calculations, again from the CPU to the GPU, to handle things like finite element analysis (FEA) and volumetric data analysis such as what’s used for medical imaging.
Quadro GPUs also contain a technology called Error Correcting Code (ECC) memory that acts like a parity check. A small amount of memory serves as a duplicate of instructions that the GPU then refers to, like a check list, to ensure that the outcome of each instruction is what was intended. Like a build-it-yourself home. They send the pictures of what it should look like when you are done, so you know if it is screwed up or not. If there is a problem, the information is re-sent and recalculated until the ECC compares properly. Sean told me how important this was in medical imaging, where exacting results are depended upon.
Take all that, and add fast 64-bit floating point calculations, and we get drastically increased triangle calculation performance. The newest Quadro solutions deliver up to 5x faster application performance, and 8x faster computational simulation.
Quadro GPUs use application profiles, setting up a separate one for each application that is running. Each profile exists in its own memory buffer, thus removing video conflicts and enabling different demanding applications like Civil 3D and Inventor to be run simultaneously in their own environments. This will be a welcomed feature for folks like me, who use a wide array of applications at the same time.
Did you know who makes your Graphics Card?
If it’s not a Quadro, you probably don’t. Sean explained that all the mainstream (i.e., consumer) graphics cards on the market today are actually made by someone else. Holy Crap. You know, I didn’t really think about that. For consumer cards, NVIDIA only makes the GPU. Here’s the interesting part:
How do you determine how each of the various NVIDIA consumer graphics cards will operate with AutoCAD, or Alias? YOU DON”T. That’s right, you have better payout odds in Las Vegas playing Keno than you have buying some vendor’s generic GeForce card with NVIDIA’s consumer GPU in it. It’s like buying a Yugo with a Toyota turbo 6 cylinder engine. How can I possibly trust this car? “Well partner, it’s got a Toyota race engine in it.”. Great, I can listen to the awesome sound after the wheels fall off. You know, that’s not enough when my livelihood is on the line.
The Quadro line is different. Each and every Quadro card is manufactured by NVIDIA in its entirety. Not only do their engineers know every detail of the NVIDIA Fermi architecture, but they also know your design applications very well. No other company can go that far.
“If you rely on your workstation to make you money, it makes sense to purchase professional graphics hardware, which is consistent, that will perform, and not fail you.”
It’s difficult for me to be able to validate the last comment regarding the increase in Quadro sales, as everyone around me has always used one Quadro or another. I can say that after losing my Dell laptop to a lousy ATI graphics card recently, the only graphics card that hasn’t failed me to date is the Quadro. In any event we’ll see just how well the NVIDIA Quadro 5000M will hold up to just about every other performance statement in that conversation. We’ll throw some benchmarks at it, and give it a run that most will never require. If she holds up to me, I’m betting she’ll hold up for you.
Let the beatings begin.