Benchmarking the Quadro K5200 and K2200

Nvidia kindly offered us two of the new Quadro cards to test performance, the high end Quadro K5200 and the mid-range K2200. Our benchmarking tool of choice was the same we've used extensively in the past: Viewperf 12. This latest version of SPEC's venerable Viewperf benchmark is designed to isolate the stress on the graphics card specifically, rather than the system as a whole. As a result, its scores reflect the GPU installed and do not (at least should not) reflect differences in other key system components like CPU, memory and storage.

It streams pre-defined viewsets (OpenGL and DirectX), representing typical, visual demands of popular workstation-caliber applications, including PTC Creo, Dassault Systemés CATIA and Solidworks, Siemens NX, and Autodesk Maya applications. New for version 12 are energy and medical viewsets, which exercise volume rendering and dynamic data generation.

In any benchmarking exercise, we would normally try to follow our standard practice of using only a production driver publicly available at the time of testing. However, since we were benchmarking pre-launch, we relied on Nvidia's commitment that the driver we used was the same one that would be posted for public download on the first day the product was available for sale.

The ideal comparisons for both the Quadro K5200 and K2200 are their predecessors, the K5000 and K2000, as their respective street prices are expected to be about the same. In both cases, the new SKUs bested their predecessors by substantial margins. The Quadro K5200's scores surpassed the K5000's by anywhere from around 40% (maya-04 viewset) to almost 120% (sw-03 viewset). Meanwhile, with one outlying exception, the Quadro K2200's scores topped those of the K2000 by a margin of from around 50% to 105%.

(That outlier was the K2200's score on the energy-01 viewset, which exceeded the K2000's score by almost 600%. It's possible the K2200's 4 GB frame buffer gave it a leg up on K2000's 2 GB, but we don't think there's that much sensitivity in memory size beyond 2 GB. Given the fact that the bump on energy-01 on the K5200 — with the same driver — was around 60%, we believe that either we had a problem with energy-01 running on the K2000 when we tested a while back, or there was some weird inefficiency going on with Nvidia's older driver when running on the K2000. Either way, that one result is not something we factor into our general assessment of the K2200's performance gains.)

Viewperf 12 benchmark scores of Quadro K5200 versus its predecessor, the K5000

Viewperf 12 benchmark scores of Quadro K2200 versus its predecessor, the K2000

The new SKUs' raw scores are interesting, particularly when compared to the scores of the products they are replacing. But of more importance to most buyers is price-performance, particularly those shopping in the more cost-sensitive mid-range where the K2200 plays. And increasingly, performance/Watt has become a higher-profile purchase decision criterion as well.

The Quadro K5200's performance and performance/Watt gains over its predecessor

The Quadro K2200's performance and performance/Watt gains over its predecessor

Now, we didn't bother posting performance/$ gains, since with the same expected ASPs, they should match the performance increases. What the charts do show is how the new products' higher power envelopes soften the gains made when looking at power efficiency. The Quadro K5200's performance/Watt gains were in the 20% to 80% range, as compared to the 50% to 105% range in raw performance gain. Similarly, the K2200's gains are not quite as dramatic when measuring performance/Watt, coming in between around 15% and 80% (not including that one energy-01 outlier).

The verdict?

Some are too quick to accuse vendors launching new products of trying to pull a fast one on customers, claiming what may be marketed as "new" are simply re-brands. Yes, that can and does happen (though in this case, we wouldn't call it a re-brand as even cards based on existing Kepler chips are built up quite differently). But this isn't the gaming space, where one vendor's fan base might get all up in arms and call out a competitor for perpetrating some kind of marketing fraud. This is the workstation space, where customers generally don't care, nor should they. Rather, the relevant question ultimately is — and always ought to be —how do these new products perform, compared both to the products they are replacing as well as what's currently on the market?

Regardless of whether the SKU at each price point now leverages Maxwell or Kepler, the fact is all come equipped with substantially more capable resources than their predecessor. And those substantially more capable resources have delivered a commensurate boost in GPU performance. The fact that some of these cards are not moving from generation to generation is moot.

Now, it's always worth emphasizing that just because a GPU can demonstrate a 50% or 100% gain in performance for a particular viewset doesn't mean an end-user should expect that type of gain, even if their application is the same one that created the viewset. By intent, Viewperf 12 stresses the GPU and the GPU alone (as much as it can, anyway), so its results reflect best-case performance gains possible.

Still, best-case gains are what we tend to measure when it comes to sizing up a new generation of GPU products. And given that context, this new set of Quadro SKUs is yielding speed-ups at least on the order of — and we think more — than what is typically seen from a generation-to-generation step.

We imagine some of that boost is coming from driver tuning, but probably not much. It's not like Quadro drivers are immature and hadn't already been dialed for these applications, and possibly Viewperf 12 specifically. Rather, the gains would appear to be more directly a function of more FLOPS, more memory, and more bandwidth to that memory.

The one tradeoff in this round of Quadros is the modestly higher power consumption. Increases in consumption are not enough to be problematic — for example, the don't surpass some key power limit (e.g. 225 W for max, high-end configurations) or push the cards out of an OEM's workstation chassis for which they are best suited. But still, it's a tradeoff, albeit not a dramatic one.

Nvidia's rival, AMD, has been delivering on some high-performance products recently, and Nvidia need a strong refresh cycle to maintain its dominant market position. With this set of new Quadro cards, it appears they have delivered on that goal ... regardless of whether their SKUs start with a "K" or an "M".