NVIDIA Quadro K2200 Graphics Card in Review

December 7, 2014 11:29 pm

Tags: 3D, Animation, Graphics Cards, K2200, NVIDIA



NVIDIA Quadro K2200 Graphics Card

A black Pelican case arrived on my desk. It was like the MacGuffin from every action movie. It looked like I had been given the bomb. I flipped the latches and inside was a sealed Quadro K5200 (see other review) and a Quadro K2200. Somebody wanted to know what I thought. So I tested. I played. I worked. I tinkered.

This review is the result of messing around with the NVIDIA Quadro K2200 workstation class card. It replaces its predecessor, the Quadro K2000 as part of an overall Quadro lineup upgrade. For those not familiar with the NVIDIA Quadro lineup, drumroll:

NVIDIA Quadro are graphics cards that cater to workstation requirements. This includes everything from graphics professionals, such as visual effects artists, to just about every engineering discipline you can imagine. It also includes the geospatial club, medical data sets, volumetric rendering, as is the case with MRIs and CT scans, atomic simulations - everything. They're literally everywhere.

One of the most noticable difference between Quadro and their gaming counterpart GeForce, is generally more memory and more CUDA cores. Programmers write kernels in CUDA or OpenCL and run those kernels in parallel on the massive number of CUDA cores Quadro cards provide. The Quadro K2200 provides 640 general purpose CUDA cores. Divide and conquer!


It occupies a single slot. It's about 8 inches long. It's a graphics card. Image courtesy me.


Basically, Quadro helps solve really big, really dense, computationally expensive problems. They're a supercomputer on a board, both for graphics and for general purpose computations. If your software makes use of them, things get much faster. Examples include the new OpenSubdiv library used as the default mesh smoothing algorithm found in Autodesk Maya 2015 and later, to select tools found in the NUKE compositing software by The Foundry, to many tools in the Adobe Creative Suite.

The new NVIDIA Quadro K2200 replaces its predecessor the Quadro K2000 as part of NVIDIA's ongoing commitment to keep the Quadro line up to date. In short, the lineup got a huge upgrade. Yay!

In their own words, NVIDIA says:

"Outstanding performance with a range of professional applications, substantial 4 GB of GPU memory to hold large models, and support for four displays for enhanced desktop productivity."

So, is it any good? Does it suck? Does it rock? Let's find out.

The Answer on high

It's great. I loved it. Despite its single-slot appearance it's more than sufficient for most graphics professionals. Perfect for freelancers. Yes, it's good enough for even you. It has double the memory of its predecessor and almost double the cores. Tens of millions of polygons in Autodesk Mudbox are a breeze. Sculpt like you mean it! Autodesk Maya loves it! More efficient architecture, bigger textures, same form-factor. Supports up to 4 displays. (Yes there's only 3 connectors: a DVI-I and 2x DisplayPorts. You can use Mutli-Stream of the DisplayPort connectors; i.e., drive more than one display from a single connector.) The card consumes only 68 watts! Go buy it. Be happy. The end.


There are 3 connectors, but you can run multiple displays through a single DisplayPort connector. Image courtesy me.


The Long Answer

The Quadro K2200 is a solid card and a substantial upgrade over the Quadro K2000. After using it for a couple of months I have zero complaints about it. Considering what you get, for a retail price of around $500-$600 (shop around), it's pretty amazing. I mean, 640 high performance CUDA cores on a single board that uses less than 68 watts of power and gives you 4GB of memory to play with - ! As a programmer, you have a supercomputer at your disposal for around $500. That said, the Quadro K2200 is pretty slick, any way you slice it, especially because...


Image courtesy me.


The Quadro K2200 is based on the new Maxwell architecture! It's the successor to Kepler architecture and the Quadro K2200 was the first NVIDIA GPU quadro card to use it. Why is Maxwell such a big deal? Power efficiency! More jollies, less joules.


The NVIDIA Quadro K2200 sporting its larger heat sink and 60mm ultra-quite fan. The heat sink features an aluminum base and fins, with a solid copper GPU contact on the underside (Not shown.) Image courtesy me.


Everything about this card and the new Maxwell architecture is about power efficiency. The CUDA cores are grouped into Streaming Multiprocessor units, each SM unit contains 32 cores. Maxwell based SM units actually have a lower performance than their Kepler based predecessors on a per-core basis, but a much higher power efficiency (around 2x) and are both physically smaller and simpler to implement. Who cares? What does this mean? Does this mean this card is slower?


Image courtesy me.


No, it's much faster and you care! Here's why: graphics and CUDA/OpenCL workloads are already embarrassingly parallel problems. By having a smaller, more efficient SM unit, we can fit more SM units on the GPU die; i.e., we can have more cores per unit area and per watt. We can run roughly 2x the number of cores on about the same amount of power. Thus, Maxwell based cards can be much faster than an equivalent Kepler card due to the increased number of cores - and that's exactly what happened; The Quadro K2200 saw an increase in the number of CUDA cores from 384 on its predecessor, to 640!


The NVIDIA Quadro K2200 uses the new Maxwell architecture as indicated by the GM107 stamped on the GPU. The M stands for Maxwell. Image courtesy me.


The other noteworthy improvement from its predecessor, the Quadro K2000, is that the new Quadro K2200 doubles the memory from 2GB to 4GB. You can work with bigger textures interactively in the viewport before quality or speed have to be sacrificed. The memory bandwidth also saw a small increase from 64GB/s to 80GB/s.

Finally, you won't need a power supply with a PCI-E power connector to run the K2200 like you would when dealing with it's bigger siblings. The Quadro K2200 draws all of its power through a standard PCI-E slot; No additional connectors required! Combined with it's smaller single-slot form factor, it will fit in even the most cramped of cases.


Check out how the Quadro K2200 stacks up next to some of its siblings as benchmarked in SPECviewperf® 12, the industry standard workstation benchmarking suite:


These scores will be lower than those released by NVIDIA. This is expected as their benchmark machine is much faster than mine. However the relative differences within this chart is what is important.


The subjective performance in Autodesk Mudbox was great. The card is more than sufficient for just about any artist.


Interactive sculpting performance was more than sufficient. Sculpt with over ten million polygons, no problem. Image courtesy me.


Performance in Autodesk Maya 2015 with the new OpenSubdiv library remained quite snappy as well. This image is taken in viewport 2.0 with Realtime ambient occlusion, motion blur, multi sample anti-aliasing lights, multiple 4096x4096 pixel textures and voxel based clouds. Things remained super great; i.e., above 60fps:




We can also see how the card stacks up against multiple cards from different generations as tested in the older SPECviewperf® 11 test suite:


Remember: These cannot be compared to their SPECviewperf 12 scores! The tests are entirely different. We also cannot compare these to SPEC results conducted on other test machines. Compare within this chart only.


Final Thoughts

Obviously, the NVIDIA Quadro K2200 isn't nearly as fast as its bigger brothers, especially for general purpose computing due to the comparatively smaller number of CUDA cores. However, compared to its predecessor, it's much improved - twice the cores, twice the memory.

Retailing for around $500-$600 also makes the Quadro K2200 significantly less expensive than its bigger brothers! For example, at the time of this writing the next step up, the Quadro K4200, will set you back roughly $1000 (shop around and you can find it for less). Despite the card's small size, its very efficient Maxwell architecture, its larger number of CUDA cores and its 4GB of memory makes it pack quite a punch. Performance-wise, it's almost in the same class as the older Quadro 5000 Fermi. (But overall much better, quieter, more efficient, etc. To be honest, I'd rather have this card.)

In my experience with the Quadro K2200 I found it more than adequate for the average users' everyday Autodesk Maya and Autodesk Mudbox demands. As a freelance artist I would be perfectly satisfied with it for all but the most demanding of tasks. It's perfectly adequate for high polygon counts (measured in tens of millions) and huge textures with little effort. If you're an artist trying to build a nice, capable workstation without totally breaking the bank - the Quadro K2200 is a viable option worth seriously considering. All things considered, two thumbs up!


Kurt Foster (Modulok) falls somewhere between programmer and visual effects artist. When not sifting through technical manuals, he takes on freelance roles in both programming and visual effects, attempting to create a marriage of technical knowledge with artistic talent. He can be seen helping out on the Renderosity Maya forum, when time permits.




December 7, 2014

Please note: If you find the color of the text hard to read, please click on "Printer-friendly" and black text will appear on a white background.

Create your masterpieces with products from the Renderosity MarketPlace
Check out What's on Sale and see the Hot New Products for more items!

Please take a moment to join Renderosity's Newsletter List to receive news and information from Renderosity!

Related Articles:

Article Comments

No article comments have been posted. Sign In or Register to share your thoughts.

Privacy Notice

This site uses cookies to deliver the best experience. Our own cookies make user accounts and other features possible. Third-party cookies are used to display relevant ads and to analyze how Renderosity is used. By using our site, you acknowledge that you have read and understood our Terms of Service, including our Cookie Policy and our Privacy Policy.