It’s easy to test graphics boards in desktop computers: you pop off the lid, disable the existing board, stick in the new one in a spare slot, reboot the computer, and install the new display drivers. But laptop computers are different, because the graphics circuitry cannot be disabled or replaced. There is no spare slot; you can’t change what’s inside of them.
Another issue is that high-end graphics boards consume a lot of power, often 100 Watts or more. The computer needs a power supply that can provide the extra 100-150W. If the desktop computer’s power supply is too weak, you can pop in a stronger one. In contrast, laptops are designed to sip power to maximize battery life.
So when it came time to test NVIDIA’s new top of the line mobile graphics, the no-spare-slot and insufficient-power problems would be solved if only NVIDIA could provide a loaner laptop already outfitted with their new graphics system. There was a bit of a delay as we waited for HP to start making and then begin shipping their brand new ZBook 17 engineering laptops.
Although this is a review of the K5100 graphics board, I have to also talk a bit about the ZBook, because the two are integrated as one computer.
As plain as its name may sound, the Quadro K5100M is a beautiful graphics subsystem from NVIDIA (see figure 1). The M is short for “mobile,” indicating it is meant for laptop computers. It uses NVIDIA’s new Kepler architecture, and so has more CUDA cores than earlier generations. And, it comes with up to 8GB of memory, which is useful for holding complex renderings entirely in its memory. NVIDIA reports that the mobile version is actually more powerful than its desktop equivalent.
|Figure 1: The K5100M’s GPU in the lower left corner, surround by eight 1GB RAM modules|
CUDA is short for “compute unified device architecture” and is NVIDIA’s system of performing parallel computations on the hundreds or thousands of GPU cores (graphics processing units) that power graphics boards. GPUs are designed naturally to run calculations in parallel, unlike the CPUs found in Pentium or iCore chips; these are better suited to sequential operations. While most CAD software does not lend itself to parallel computations, application such as finite element analysis and rendering are readily processed through parallel operations, and so suited to CUDA.
Here are is full specifications list of the graphics board:
The K5100M supports NVIDIA’s collection of utility software, including:
There are a whole lot of other features buried within the K5100M, but they pertain towards high-end rendering and gaming, two things that I don’t have enough knowledge about to speak to. However, I can talk about how well this card performed with SolidWorks. For instance, the 8GB of frame buffering means you won’t have lag with large assemblies and models.
After the HP ZBook 17 mobile workstation arrived on my doorstep (see figure 2), I was able to get started. Being able to test out a top-of-the-line graphics card on a top-of-the-line mobile workstation is awesome. Sending it back will not be easy.
|Figure 2: HP’s Zbook 17 mobile workstation|
First, let’s talk about what’s under the hood of the laptop, shall we? The ZBook I received came with Intel’s Core i7-4900MQ quad-core CPU running at 2.8GHz, with 16GB RAM and Windows 7 Pro 64-bit. It’s expandable to 32GB of memory and 2.8 TB of storage.
On the outside, it has a 17.3″ diagonal screen and a slew of I/O ports, such as three USB 3.0 ports and a Thunderbolt port. At 7.67 pounds, it’s just over a pound heavier than my 15″ Dell. Not too bad a tradeoff for the extra horsepower.
I’ll admit it, it’s a sexy machine. While my Dell is only a year old, the HP makes it seem completely out of date and loud. The HP is a quiet, yet powerful beast. It’s sleek, has a great display and performs as well as any desktop workstation I’ve had in recent memory.
My only other experience with an HP product was a tower which I was provided with at a former job, and I liked it well enough. But I loved how SolidWorks behaved on the new HP laptop. I messed around with it for hours, trying to bog it down – without success. HP has been a partner with SolidWorks for years, and so it stands to reason that they’d play well together; with the ZBook, they didn’t disappoint.
Honestly, I could sit here and heap on the accolades for the rest of this article, but I won’t. I will, however, point out that the ZBook I tested out was their top of the line model, upgraded to the K5100M card and Intel CPU, which would cost extra. For instance, the K5100M adds $2,150 to the price of the laptop. On HP’s Web site, I was able to custom spec the machine in prices ranging from a low of $2,280 to a high of $10,230. HP also has 15″ and 13″ versions of the ZBook.
The downside is that I’d have to compare the K5100M to the AMD FirePro M5950 graphics in my Dell laptop. So be it. For comparison’s sake, the specs for the AMD FirePro M5950 graphics on my year-old Dell are as follows:
So it’s not exactly a fair fight, but I think that the AMD’s specs are more common among graphics cards than those of the high-end NVIDIA. But by comparing the two, it’ll allow you to see how the other half lives, as it were.
The first 3D model I chose to test in SolidWorks was one that I had on my hard drive (see figure 3). While not huge, it has a large number of edges which tend to wreak havoc on the graphics when manipulating it.
|Figure 3: My 3D test model of stacked containers has lots of edges|
Here’s what I discovered: there wasn’t a whole lot of difference between the two cards. While the K5100M was slightly faster and smoother while rotating, I did not find it all that noticeable. To affect a difference, I applied different backgrounds and different materials, and changed system settings. But in the end it really made no difference.
This does not come as a big surprise to me. We’ve long known that unless we’re doing high-end renderings or animations, that a high-end graphics card isn’t going to help much. Time and again, it’s been proven that large assemblies will perform quite well with mid-range cards and that a high-end card is simply overkill.
Not content to accept such close results, I decided to run a test with a much larger assembly. Whereas the first assembly is about 3MB, the second assembly is a robust 61MB, chock-full of edges and patterns and configurations (see figure 4).
|Figure 4: : The 61MB test assembly of a lifting platform|
As one might expect, the K5100-enabled HP handled the assembly beautifully, while the Dell was sluggish. These results were present during Realview and shaded with edges. On the other hand, there wasn’t a noticeable difference between the two machines when the view was simply shaded.
While I’m no rendering guru, nor do I have any special rendering software outside of Photoview360, I did see a speed difference when I rendered the above assemblies into the pictures below (see figure 5). These containers took 8:36 on my Dell and 7:29 on the HP – 15% faster.
|Figure 5: Rendering of the stacked containers test assembly with SolidWorks|
The platform, however, was a much different story (see figure 6). On my Dell, the rendering 1:45:18 (hours:minutes:seconds); on the HP, 1:05:49 – 60% faster.
|Figure 5: Rendering the lifting platform|
Granted, the HP is a more powerful workstation, but that’s not really the point is it? Most of the rendering processing is handled by the graphics board, not the CPU.
NVIDIA and HP both have come out with a great offering for those of us who need to use mobile workstations. The M5100-line of graphics boards shows a significant time advantage for complex renderings, as well as overall smoother performance for everyday SolidWorks designing. Together, they make for a screaming system that I’d love to keep.