Welcome Visitor
Today is Sunday, December 10, 2017

 

NVIDIA GPU Quadro M5000 in Review

    Print

NVIDIA Quadro M5000 Workstation Graphics Card

The new NVIDIA Quadro M5000 workstation-class graphics card is the spiritual successor to the NVIDIA Quadro K5200. The 'M' in the name stands for Maxwell, named after James Clerk Maxwell - you learned his equations in physics. Maxwell succeeds Kepler with a focus on power efficiency. The NVIDIA Quadro M5000 features a second generation Maxwell GM204 core.

For those not familiar with NVIDIA Quadro, what is it good for:

  • Digital content creation (Autodesk Maya, Autodesk Mudbox, Nuke, AfterEffects, ...)
  • CAD/CAM (SolidWorks, Inventor, Catia, AutoCAD, ...)
  • Medical imaging (ImageVis3D, Vaa3D, volume rendering, ...)
  • Bioinformatics (GPU accelerated BLAST variants, microscopy stack alignment, ...)

It's more than just graphics today, through NVIDIA CUDA and OpenCL, it's general purpose compute kernels running in parallel on thousands of cores. Quadro cads excel at these types of workloads. This is especially true with the new Maxwell GM204 as we'll see on our tests. With continued demand for richer, more immersive media delivered on shorter deadlines, the graphics industry is rapidly moving toward NVIDIA CUDA and other GPU based solutions. This trend is expected to continue. Time for some quick stats:

Feature Quadro K5200 New Quadro M5000
GPU Memory 8 GB GDDR5 8 GB GDDR5
Memory Interface 256-bit 256-bit
Memory Bandwidth 192.0 GB/s 211.0 GB/s
NVIDIA CUDA® Cores 2304 2048
Core GK108 GM204 (2nd generation Maxwell)
System Interface PCI Express 3.0 x16 PCI Express 3.0 x16
Max Power Consumption 150 W 150 W
Display Connectors DVI-I, DVI-D-DL, 2x DP 1.2 4x DP 1.2 + DVI-I-DL
Max DP 1.2 Resolution 4096 × 2160 at 60 Hz 4096 × 2160 at 60 Hz
Max DVI-I DL Resolution 2560 × 1600 at 60 Hz 2560 × 1600 at 60 Hz
Max DVI-I SL Resolution 1920 × 1200 at 60 Hz 1920 × 1200 at 60 Hz
Max VGA Resolution 2048 × 1536 at 85 Hz 2048 × 1536 at 85 Hz
Graphics APIs Shader Model 5.0, OpenGL 4.54, DirectX 11.2 Shader Model 5.0, OpenGL 4.54, DirectX 12.05
Compute APIs CUDA, DirectCompute, OpenCL™ CUDA, DirectCompute, OpenCL™

Almost identical, right? Not exactly. The new Quadro M5000 boasts more memory bandwidth, more display port connectors, support for DirectX 12 and of course the new second generation Maxwell core - which is the big difference due to increased efficiency per core as we'll see.

Aesthetics & Build Quality

The form factor and physical dimensions of the new NVIDIA Quadro M5000 are almost identical to the older NVIDIA Quadro K5200. The new card is If the old card fit your case, the new one will too. The newNVIDIA Quadro M5000 has some minor aesthetic differences. There's arguably better airflow through the new triangular vents in the i/o plate and the fan shroud looks slightly less blocky. The stereo connector has been moved to be closer to the SLI connector. Lots of little changes.

You'll also notice there is now a single DVI connector instead of the 2x that were on the older NVIDIA Quadro K5200. The industry is moving toward DisplayPort due to the massive increase in bandwidth necessary to drive multiple high resolution displays.

The new NVIDIA Quadro M5000 sitting atop its spiritual predecessor the NVIDIA Quadro K5200.

The build quality of the NVIDIA Quadro M5000 I evaluated was excellent. The card itself, the heat sink and the connectors were rugged. For example, all i/o plate connectors were not only soldered to the PCB, but they're also screwed to the steel i/o plate - and those screws all had thread-locking compound on them. What's more, the plate itself isn't just screwed to the edge of the PCB. It is also attached to perpendicular cast aluminum struts as part of the mono piece cast aluminum block that runs the full length of the card.

NVIDIA Quadro M5000 build quality is excellent. All solid capacitors. Thread-lock fluid on all screws. Cast aluminum reinforced steel i/o plate.

Taking off the heat sink shroud took effort! The top half of the shroud itself is ABS plastic, it screws down to a full length cast aluminum block, mortise and tenon style. For 99.99% of people this is great; The stock fan and heat sink on the NVIDIA Quadro M5000 is excellent; There is no need to replace it.


The oversized stock heat sink. The squirrel fan runs silent most of the time. It's a very quiet card even under load and adjusts itself actively. Especially compared to the older Fermi core based cards. Everything in this picture that is black, other than the fan and heat sink clips, is a single hunk of cast aluminum.

For a reviewer photographing the bare core - it was challenging. One of the screws had to much thread-locking compound which bonded one of the plastic tenons in place. I managed to accidently break off said tenon - but I won and we have pictures!

The copper plate on the bottom of the stock heat sink after I cleaned off the thermal compound for the photo. It actually looks pretty good. The compound was quality stuff too. No dry crusties.

The second generation Maxwell GM204 core:


The amber color is a reflection in the GM204 perfect mirror finish. (After cleaning off the thermal compound. Remember, taking your shiny new Quadro card apart is probably something you shouldn't do.)

Why does it matter?

Solid build quality means there's noting on this card that will rattle, buzz or work its way loose. You can plug and unplug equipment all day long and the NVIDIA Quadro M5000 will be fine. You're highly unlikely to work loose a solder joint or break a connector. I don't recommend it, but you could probably use this card as an improvised hammer; The NVIDIA Quadro M5000 is built solid enough to withstand years of industry use.

Benchmarks and Performance

Remember, a benchmark can only be used to compare:

  • A system as a whole to another system as a whole.
  • Different hardware tested on the same system.

E.g., you can't compare an NVIDIA Quadro M5000 on one system to an NVIDIA Quadro K5200 on another system. From one system to another you'll get different results - even for the same graphics card. E.g., I can put an NVIDIA Quadro M5000 in a five year old machine and get a GPU score no better than it's predecessor. Or, I can put the NVIDIA Quadro M5000 in a new build and get a result that is significantly better. Remember: Compare these scores only to each other, not to anything else! Apples to apples.

All tests were run on the following test rig:

Component Description
CPU Intel Core i7-5820K Haswell-E 6-Core @ 3.3 GHz
Memory CORSAIR Vengeance LPX 32GB (4 x 8GB) 288-Pin DDR4 2133 (PC4 17000)
Motherboard MSI X99S SLI Krait Edition LGA 2011-v3 Intel X99
SSD Crucial BX100 2.5" 500GB SATA III MLC Internal Solid State Drive
OS Microsoft Windows 7 Professional SP1 64-bit

CINEBENCH R15 Benchmark

CINEBENCH, by MAXON, the makers of CINEMA 4D, is a standard benchmark. It provides both CPU and GPU testing. Here we'll use only the GPU scores. These scores provide a very basic OpenGL only test. It does not make use the card in its entirety.

A surprising result - maybe. The NVIDIA Quadro M5000 actually scores lower than the older NVIDIA Quadro K5200. Does this mean the Quadro M5000 is a slower card? No. It simply means its slower at running CINEBENCH on my test rig. Software written specifically with Quadro in mind will perform far better. Speaking of performing better, the new NVIDIA Quadro M5000 is about to clean up...

SPECViewperf 12 Benchmarks

SPECViewperf 12 is a set of very GPU intensive industry standard benchmarks. These benchmarks are far more resource hungry than most and are designed to provide a comparison of workstation class hardware designed to run popular software for medical, digital content creation and engineering applications.

Each test produces a single composite score that represents the overall performance for a given test. These scores do not represent frame rate. I ran each test on each Quadro card I could get my hands on, using the same hardware mentioned earlier to give a fair, apples-to-apples comparison. Below are the results comparing several NVIDIA Quadro cards, one graph for each test.

catia-04

creo-01

energy-01


When it comes to computation of extremely dense volume data sets the new NVIDIA Quadro M5000performs very well indeed. We'll see this echoed again in the medical-01 test. The GM204 excels at per-core efficiency compared to its predecessor. It performs far better, despite the lower core count.

maya-04


The general Maya performance is also a fairly large jump over it's nearest competitor. While this is a synthetic benchmark, it does a fairy good job of echoing the real world relative performance in Autodesk Maya.

medical-01


Again when dealing with volume heavy rendering, the NVIDIA Quadro M5000 cleans out the competition as is shown in this test from the volumetric heavy medical data sets.

showcase-01

snx-02

sw-03

3D Mark Professional Edition - Fire Strike

3D Mark is a ubiquitous GPU and system benchmark suite by Futuremark. It's typically used to benchmark gaming systems as it represents a typical gaming workload. Be careful reading these scores. These are for a single card systems in stock configuration, using driver defaults, using stock coolers and zero overclocking. As such, compare these results only to results in this article.

The following is the standard (normal) Fire Strike benchmark.

Graphics

Physics


Not a huge difference here, indicating a potential bottleneck elsewhere on the test system.

Combined Score

Autodesk Mudbox Sculpting

What's the subjective Autodesk Mudbox performance? With the NVIDIA Quadro M5000 it pretty much comes down to a system main memory limited endeavor - not a GPU limited one. I saturated 32GB of DDR4 on my test machine before I could drop below a usable frame rate in Autodesk Mudbox. That's over 110 million shaded, textured polygons, multiple 8k maps, etc. If and when I acquire more memory, I'll revisit the topic. However, the Intel Core i7 Haswell-E processor I was using can only address up to 64GB of memory and my test system already had 32GB. (Anyone want to send me a Xeon - for testing of course?)

The NVIDIA Quadro M5000 drawing 110 million textured shaded polygons at an average of over 30 fps with no signs of slowdown. (A Utonic goon from the Unlikely Heroes Studios comic Super!)

In short, the NVIDIA Quadro M5000 pretty much beat my new machine in this regard. How's that for subjective performance? Sculpt whatever you want, cake on the high frequency details - the Quadro M5000 will keep up. If you go crazy you'll have to move on to something in the Intel Xeon or AMD Opteron server family of chips just to be able to address enough memory to conceivably bog the Quadro M5000 down.

Autodesk Maya Usage

The subjective performance in Autodesk Maya was similar. Having 45 million polygons in the viewport, smooth shaded in viewport 2.0 with ambient occlusion, and multisample anti-aliasing, and multiple 8k textures and several thousand individual objects - performance was still pretty amazing, considering. The viewport remained above about 10 fps with all the viewport 2.0 pretty settings maxed out, still a workable for modeling.

For the record, this is a sick amount of detail and cranked up pretty settings for your average scene. I could have pushed things a lot further had I started combining objects into single shape nodes, point clouds, asset management, etc. Of course you can grind the NVIDIA Quadro M5000 to single digit frame rates and below if you stack on enough high resolution textures and mostly additional real-time lights with real time shadows and so on. I can always break things.

For doing high resolution sculpting, animation, modeling, etc - the NVIDIA Quadro M5000 is fantastic. I don't think anyone would be at all displeased with its performance. It's certainly a step above the olderNVIDIA Quadro K5200.


Despite just shy of 45 million triangles, x32 ambient occlusion on, multiple 8k textures, 16x multisampling anti-aliasing I still had a usable frame rate of about 10fps on my test rig and I could have further improved that. Under real-world conditions I see main system memory being the bottleneck for most people, not theQuadro M5000.

Final Thoughts

The NVIDIA Quadro M5000 is impressive. Its compute performance is most impressive. The fact that it draws no more than 150 watts and has fewer cores than its predecessor, yet often dramatically outperforms it, is pretty awesome. When it comes to sculpting in Autodesk Mudbox and general use in Autodesk Maya, it's well suited to dense, richly textured scenes and highly detailed models. The exact performance difference of course depends on the data set. In Autodesk Maya there was indeed a boost as echoed by the SPEC test. In general the new Quadro M5000 was anywhere from a 25%-40% faster.

Again, as we saw in the benchmarks, this is highly dependent on your specific workload. Overall it's a quality card with enough performance to please anyone. As the industry moves more toward CUDA and other GPU based solutions, having a good performing GPU at general purpose computing tasks is increasingly critical.

I'm currently beta testing the yet-to-be-released NVIDIA Iray renderer for Autodesk Maya and the Quadro M5000 has been an enormous number-crunching asset. Look for a future article about my experience with the Quadro M5000 coupled with NVIDIA Iray.

Learn More:

Read more from:
Reviews
Tags: 
GPU, Graphics Cards, M5000, NVIDIA
Share: 
Related Articles
     Print
Powered by Bondware
News Publishing Software

The browser you are using is outdated!

You may not be getting all you can out of your browsing experience
and may be open to security risks!

Consider upgrading to the latest version of your browser or choose on below: