3D Artist
Apr
13

PNY Nvidia Quadro M6000 review

News & Features
by
Orestis Bastounis

Nvidia’s new top-end graphics card based on its Maxwell design is the first Quadro to offer really stunning OpenCL performance

PNY Nvidia Quadro M6000 review

It seems Nvidia updates its most high-end Quadro card according to a different timescale to the rest of the family.

Last year we were treated to a top-to-bottom refresh of every Kepler-based Quadro, except the K6000, which was left unchanged. The K5200 came along with double the memory and a large boost to its shader count, narrowing the gap between the best and second-best cards, and making the more expensive K6000 seem like an unwise purchase.

Until now. The M6000 is its replacement, being the first Quadro to use the new Maxwell design, which debuted in mobile and gaming cards. This one is a newer second-generation variant code-named GM200.

Looking at the card’s specs, it doesn’t immediately seem like an impressive upgrade. The K6000 had 2,880 shaders, the M6000 has 3,072. The 12GB of ECC-capable GDDR5 video memory hasn’t been increased at all, AMD’s FirePro W9100 and its 16GB of video memory retains that particular record. Memory bandwidth remains at 384-bit. It’s still based on a 28nm chip.

PNY Nvidia Quadro M6000 review

But the overall single-precision FP32 compute performance has ballooned from 5.2 TFLOPS to 7 TFLOPS, an impressive 34 per cent increase. This is due to the efficiency of the new architecture. There’s been a ten per cent rise in average power draw too, from 225W to 250W.

Double-precision FP64 performance has seen a very sharp decline though, but that’s a result that is probably more important to scientists and engineers that are running simulations using Quadros than 3D graphic designers.

The number of ROPs has doubled to 96 too, with the core clock increased from 902MHz to 988MHz and memory clock from 1,502MHz to 1,653MHz. There are also now 256 TMUs and more steam multiprocessors.

Physically, the M6000 has seen some slight changes. Once again a single eight-pin power connector is sufficient, with four full-sized DisplayPort outputs, capable of simultaneous 60Hz 4K output, and a single DVI port.

Nvidia is also touting a few new features. Most interesting is a new CUDA-accelerated iRay plugin for Maya, 3ds Max, Cinema 4D and other packages, which promises real-time ray tracing via a Fast Interactive rendering mode on the M6000.

As with previous top-end GPUs we’ve reviewed, the M6000 arrived in a workstation chosen to represent a typical system a customer might use it with. Such a powerful card is more likely to be purchased with a similarly pricey and high-end dual-Xeon workstation than an overclocked Core i7, so that’s what we received: two eight-core 3.2GHz Xeon E5-667 v3 processors, 64GB of DDR4 memory and a Samsung 850 Pro SSD, all provided by Workstation Specialists.

PNY Nvidia Quadro M6000 review

The first test was CINEBENCH. The OpenGL score clocked in at 185.38. It’s a great result but it doesn’t demonstrate what the M6000 can do.

The SPECviewperf 12 result was more interesting. The narrative of most Nvidia graphics card reviews has been one where the Maya section drops off considerably compared to results from AMD’s FirePro line.

Not this time.

The M6000 clocked a score of 110.8, the K5200 in Lenovo’s ThinkStation P900 scored 64.72, and AMD’s W9100 in Scan’s 3XS GW-HT20 managed 76.75 – meaning a huge boost for the M6000.

The other sections saw new records broken too with 138.65 in catia-04 and 139.21 in sw-03. These scores are off the charts.

But it’s OpenCL where the biggest performance increase comes. LuxMark’s Room render test has always chugged on Nvidia cards, even the K6000.

Historically, OpenCL is a competitor to Nvidia’s own CUDA API, and has never been given the same attention. The M6000 represents a change in Nvidia’s thinking. Again, the result breaks new records. The 2,231 points leaves every workstation we’ve tested with AMD’s FirePro cards in the dust. The previous best was 1,396 with a W9100, so there’s been some improvement!

But a few results are puzzling. ArionBench, a great CUDA-only benchmark, saw far less impressive gains. The K6000 managed 1,986 points in 2 minutes, 31 seconds. The M6000 was just eight seconds faster with 2,088 points. It was an improvement of around 5 per cent, which we confirmed is not an error.

Why this happened, we’re not sure. Along with the drop in FP64 performance, which is again, not a problem for most artists, it’s possible that not everyone will see such massive performance gains across the board by upgrading to an M6000. But smaller gains are generally overshadowed by the massive improvements.

The M6000 does put clear water between itself and the more affordable Nvidia cards, making the significant premium it carries well worth the upgrade.

It’s the most powerful professional GPU ever made and it’s also a new object of hardware lust for artists everywhere.

The M6000 can handle multiple 4K displays running at a full 60Hz, highly complex and detailed modelling, and even real-time ray tracing. What a shame this power carries such a high price though, but on the other hand, it’s at least at a comparable price level to its predecessor.

The W9100 is £1,000 cheaper, which may sway some, but really we can’t wait for AMD’s response and what we’ll get when Nvidia updates the rest of its more affordable GPU lineup with the new Maxwell chips. Hopefully it won’t be too far away.

Verdict

Features – 5/5
Performance – 5/5
Design – 5/5
Value for money – 4/5

Overall – 5/5