NVIDIA Quadro M4000 Review

1. Introduction2. NVIDIA Quadro M4000 Specifications3. Testing Methodology4. Cinebench R155. CATIA viewset (catia-04)6. Creo viewset (creo-01)7. Energy viewset (energy-01)8. Maya viewset (maya-04)9. Medical viewset (medical-01)10. Showcase viewset (showcase-01)11. Siemens NX viewset (snx-02)12. SolidWorks viewset (sw-03)13. Closing Thoughts14. View All Pages

The NVIDIA Quadro M4000 has been out since August, but as the successor to the K4200, it's the flagship mainstream professional graphics card. So we are reviewing it here to show how it compares to its predecessor (and the higher-end K5200). In terms of specification, it's another step forward, despite having a similar £777 inc VAT price.

The M4000 is based on the second spin of NVIDIA's Maxwell architecture, like the Quadro M6000 we saw in the immensely powerful Scan 3XS GW-HTX35, although a more recent revision. It has much more RAM than the K4200 – in fact twice as much – and considerably more CUDA cores too. This promises a significant leap forward in performance.

Here's where the M4000 fits into the current Quadro range and compares to its predecessors:

The Quadro M4000 sports 1,664 CUDA cores, compared to the previous K4200's 1,344 – so that's an increase of just under 25 per cent. The GPU uses the GM204GL revision of Maxwell, where the M6000 uses GM200GL. Whilst the K4200 has a GPU core speed of 780MHz, the M4000 uses a very similar 773MHz.

However, it's highly significant that FP64 (64-bit floating point) arithmetic logic units (ALUs) have been further reduced from 1/24 to 1/32 compared to the previous generation, which will put the M4000 at a disadvantage with applications using FP64 operations.

Most 3D content creation software doesn't, but some scientific visualisation software does, making the Kepler-based K4200 potentially still a better choice if you use software that definitely uses a lot of FP64 operations.

The other significant improvement over the K4200 is the quantity of memory. The M4000 sticks with the 256-bit memory path, but doubles the quantity of GDDR5 to 8GB, the same as the previous-generation ultra-high-end K5200.

This will give it substantially better abilities at handling huge texture sets compared to the K4200. The memory is also slightly faster, running at 1,502MHz compared to 1,350MHz.

In other words, the new M4000 has the same quantity and speed of memory as the former K5200, which cost more than twice as much. It offers identical bandwidth, too, at 192GB/sec. However, the K5200 has 2,304 CUDA cores – 38 per cent more than the M4000. Whilst these run at 650MHz, it should still be comfortably ahead of the new card. So if you purchased a system with one of these in not long ago, you don't have to feel too much like you wasted your money.

The M4000 sports four DisplayPort 1.2 connections, each one capable of driving monitors up to 4,096 x 2,160 4K resolution, with HDMI and DVD-D adapters also available. There's a separate 3D Stereo connector on an additional bracket.

There's support for all the latest 3D APIs, including DirectX 12, OpenGL 4.5, and Shader Model 5. Compute API supports includes CUDA (of course) plus OpenCL and DirectCompute, so this card can be harnessed for more general purpose number crunching than just 3D rendering. Power consumption has risen to 120W compared to the 108W of the K4200, but that's still not huge.

Only a single six-pin PCI Express power connector is required. Being a professional card, the M4000 comes with a standard three-year RTB warranty, although it's also possible to extend this to five years at time of purchase for a little extra.

We tested the Quadro M4000 in a IPW-SL workstation from InterPro. This is a modelling-focused system based around the Intel Skylake Core i7-6700K processor, backed with 32GB of 2,666MHz DDR4 SRAM.

For comparison, we pitted the M4000 against its K4200 direct predecessor and high-end predecessor the K5200. We used the industry-standard Maxon Cinebench R15's OpenGL test and SPECviewperf 12's suite of benchmarks.

Software:
Maxon Cinebench R15
SPECviewperf 12

SPECviewperf 12.0.2

The majority of this article will be focusing on results from the SPECviewperf test from the Standard Performance Evaluation Corporation. SPECviewperf 12 is the worldwide standard for measuring graphics performance, and is based on professional applications.

The latest version is SPECviewperf 12.0.2, which extends performance measurement from physical to virtualized workstations. SPECgpc members at the time of V12.0.2 release include AMD, Dell, Fujitsu, HP, Intel, Lenovo, Micron and NVIDIA. SPECviewperf 12 measures the 3D graphics performance of systems running under the OpenGL and Direct X application programming interfaces. The benchmark’s workloads, called viewsets, represent graphics content and behavior from actual applications.

InterPro IPW-SL Workstation Specifications:

• Intel Core i7-6700K @ 4.8GHz
• 32GB DDR4 SDRAM @ 2,666MHz
• ASUS Maximus VIII Motherboard
• 400GB Kingston Intel 750 Series NVMe PCI Express M.2 SSD
• 4TB Seagate Desktop HDD.15 SATA III 6Gb/s HDD
• Corsair H110i GTX Hyrdo Cooler
• Corsair RM750 750W Gold Modular PSU
• Fractal Design Define R5 chassis
• Windows 10 Pro 64bit

CINEBENCH 15 is a cross-platform testing suite that measures hardware performance and is the de facto standard benchmarking tool for leading companies and trade journals for conducting real-world hardware performance tests. With the new Release 15, systems with up to 256 threads can be tested.

CINEBENCH is available for both Windows and OS X and is used by almost all hardware manufacturers and trade journals for comparing CPUs and graphics cards. We only ran the Open GL portion of the test, as graphics power has no effect on rendering performance.

It is fair to say that things did not get off to a flying start with Maxon Cinebench R15. The M4000 placed significantly behind both the K5200 and K4200. We suspect this is related to the M4000's inferior FP64 performance, as Maxon Cinema 4D (upon which this benchmark is based) has supported double precision since 2010.

CATIA viewset (catia-04)

The catia-04 viewset was created from traces of the graphics workload generated by the CATIA V6 R2012 application from Dassault Systemes. Model sizes range from 5.1 to 21 million vertices. The viewset includes numerous rendering modes supported by the application, including wireframe, anti-aliasing, shaded, shaded with edges, depth of field, and ambient occlusion.

Viewset tests:

1. Race car shaded with ambient occlusion and depth of field effect
2. Race car shaded with pencil effect
3. Race car shaded with ambient occlusion
4. Airplane shaded with ambient occlusion and depth of field effect
5. Airplane wireframe
6. Airplane shaded with pencil effect
7. Airplane shaded
8. Airplane shaded with edges
9. Airplane shaded with ambient occlusion
10. SUV1 vehicle shaded with ground reflection and ambient occlusion
11. SUV2 vehicle shaded with ground shadow
12. SUV2 vehicle shaded with ground reflection and ambient occlusion
13. Jet plane shaded with ground reflection and ambient occlusion
14. Jet plane shaded with edges with ground reflection and ambient occlusion

Things get back on track with catia-04. The M4000 is 23 per cent faster than the K4200, although the K5200 is 17 per cent quicker still. So K5200 owners shouldn't feel depressed, but the M4000 is a significantly better option for this application than the K4200.

Creo viewset (creo-01)
The creo-01 viewset was created from traces of the graphics workload generated by the Creo 2™ application from PTC. Model sizes range from 20 to 48 million vertices. The viewset includes numerous rendering modes supported by the application, including wireframe, anti-aliasing, shaded, shaded with edges, and shaded reflection modes.

Viewset tests:

• Worldcar in shaded mode
• Worldcar in wireframe with anti-aliasing enabled
• Worldcar in shaded edges mode
• Worldcar in hidden mode
• Worldcar in shaded reflection mode
• Worldcar in shaded mode
• Worldcar in no-hidden mode with anti-aliasing enabled
• Worldcar in shaded mode with anti-aliasing enabled
• Plane in shaded mode
• Plane in shaded edges mode
• Plane in hidden mode
• Plane in shaded mode with anti-aliasing enabled
• Plane in shaded edges mode with high-quality edges enabled

The creo-01 viewset is clearly more sensitive to memory quantity and performance than brute GPU speed. The M4000 cruises past the K4200, beating its score by 31 per cent, and the K5200 is less than 4 per cent further ahead. The double helping of faster GDDR5 over the K4200 is clearly paying dividends here.

Energy viewset (energy-01)

The energy-01 viewset is representative of a typical volume rendering application in the seismic and oil and gas fields. Similar to medical imaging such as MRI or CT, geophysical surveys generate image slices through the subsurface that are built into a 3D grid. Volume rendering provides a 2D projection of this 3D volumetric grid for further analysis and interpretation.

At every frame, depending on the viewer position, a series of coplanar slices aligned with the viewing angle are computed on the CPU and then sent to the graphics hardware for texturing and further calculations such as transfer function lookup, lighting and clipping to reveal internal structures. Finally, the slices are blended together before the image is displayed.

The voxel in the 3D grid is a single scalar value. A transfer function — simply a 1D lookup table — maps the 3D density value to color and alpha values. For lighting calculations, the gradients are computed on the fly using the central differences at each voxel. These state changes exercise various parts of the graphics subsystem. This viewset makes use of hardware support for 3D textures and therefore trilinear interpolation.

The 3D datasets used in this viewset are all procedurally generated using a simple random function. There is a medium-res 1GB dataset and a large-res 3.2GB dataset. The large-res viewsets will exit on cards with less than 4GB graphics memory.

There is clearly something very beneficial for energy-01 from the Maxwell architecture. The M4000 is phenomenal in this test, and the M6000 was even more phenomenal when we tested that in a Scan workstation. With a 56 per cent faster score than the K5200, the M4000 is really in its element here.

Maya viewset (maya-04)

The maya-04 viewset was created from traces of the graphics workload generated by the Maya 2013 application from Autodesk. Model size is 727,500 vertices. The viewset includes numerous rendering modes supported by the application, including shaded mode, ambient occlusion, multi-sample antialiasing, and transparency.

Viewset tests:

1. Shaded mode
2. Shaded mode with screen-space ambient occlusion
3. Shaded mode with screen-space ambient occlusion and multi-sample anti-aliasing
4. Shaded mode with screen-space ambient occlusion, multi-sample anti-aliasing, and floating-point render target
5. Shaded mode with screen-space ambient occlusion, multi-sample anti-aliasing, floating-point render target, and transparency algorithm weighted average
6. Wireframe mode

Maya provides similar results to the creo-01 viewset, with the M4000 placing 23 per cent ahead of the K4200, and just 2 per cent behind the K5200. Memory is clearly the main factor here again. With Maya's popularity in 3D animation, this is extremely good news for content creators and NVIDIA alike.

Medical viewset (medical-01)

The medical-01 viewset is representative of a typical volume rendering application that renders a 2D projection of a 3D volumetric grid. A typical 3D grid in this viewset is a group of 3D slices acquired by a scanner (such as CT or MRI).

At every frame, depending on the viewer position, a series of coplanar slices aligned with the viewing angle are computed on the CPU and then sent to the graphics hardware for texturing and further calculations, such as transfer function lookup, lighting and clipping to reveal internal structures. Finally, the slices are blended together before the image is displayed.

The voxel in the 3D grid is a single scalar value. A transfer function – simply a 1D lookup table – maps the 3D density value to color and alpha values. For lighting calculations, the gradients are computed on the fly using the central differences at each voxel. These state changes exercise various parts of the graphics subsystem. This viewset makes use of hardware support for 3D textures and therefore trilinear interpolation.

There are two datasets in this viewset:

• A 4D heart dataset comprising multiple 3D volumes iterated over time. These were obtained from a phase-contrast MRI scanner. The 80MB dataset was contributed by the Department of Radiology, Stanford School of Medicine and Lucile Packard Children's Hospital.
• A stag beetle dataset provided by the Technical University of Vienna. The dataset size is 650MB and represents a workload with larger memory requirements.

For the medical-01 viewset, the M4000 sits squarely in between the two cards from the previous generation. It's 20 per cent ahead of the K4200 but 17 per cent behind the K5200. Again, K5200 owners shouldn't feel the need to replace their hardware, but the M4000 is a significant upgrade over the K4200.

Showcase viewset (showcase-01)

The showcase-01 viewset was created from traces of Autodesk’s Showcase 2013 application. The model used in the viewset consists of 8 million vertices. The viewset is the first viewset in SPECviewperf to feature DirectX rendering. Rendering modes included in the viewset include shading, projected shadows, and self-shadows.

The following tests are included in the viewset:

1. Shaded with self-shadows
2. Shaded with self-shadows and projected shadows
3. Shaded
4. Shaded with projected shadows

This test is unusual because it measures DirectX rather than OpenGL performance. But results are similar to medical-01, with the M4000 around 16 per cent ahead of the K4200 and 13 per cent behind the K5200. So our conclusion is the same – another win for the M4000, but K5200 owners can rest easy.

Siemens NX viewset (snx-02)

The snx-02 viewset was created from traces of the graphics workload generated by the NX 8.0 application from Siemens PLM. Model sizes range from 7.15 to 8.45 million vertices. The viewset includes numerous rendering modes supported by the application, including wireframe, anti-aliasing, shaded, shaded with edges, and studio mode.

Viewset tests:

• Powertrain in advanced studio mode
• Powertrain in shaded mode
• Powertrain in shaded-with-edges mode
• Powertrain in studio mode
• Powertrain in wireframe mode
• SUV in advanced studio mode
• SUV in shaded mode
• SUV in shaded-with-edges mode
• SUV in studio mode
• SUV in wireframe mode

Yet again, we see the M4000 sitting almost equally between its two predecessors. Again, it's 17 per cent faster than the K4200, but 10 per cent behind the K5200.

SolidWorks viewset (sw-03)

The sw-03 viewset was created from traces of Dassault Systemes’ SolidWorks 2013 SP1 application. Models used in the viewset range in size from 2.1 to 21 million vertices. The viewset includes numerous rendering modes supported by the application, including shaded mode, shaded with edges, ambient occlusion, shaders, and environment maps.

Viewset tests:

1. Vehicle in shaded mode — normal shader with environment cubemap
2. Vehicle in shaded mode — bump parallax mapping with environment cubemap
3. Vehicle in shaded mode — ambient occlusion enabled with normal shader with environment map
4. Vehicle in shaded-with-edges mode — normal shader with environment cubemap
5. Vehicle in wireframe mode
6. Rally car in shaded mode — ambient occlusion enabled with normal shader with environment map
7. Rally car in shaded mode — normal shader with environment cubemap
8. Rally car in shaded-with-edges mode — normal shader with environment cubemap
9. Tesla Tower in shaded mode — ambient occlusion enabled with normal shader with environment map
10. Tesla Tower in shaded mode — normal shader with environment cubemap
11. Tesla Tower in shaded-with-edges mode — normal shader with environment cubemap

SolidWorks' popularity with product designers and engineers make this a significant test, and the M4000 is comfortably 15 per cent ahead of the K4200, but 17 per cent behind the K5200. So this will be a significantly better card for creating product-focused engineering drawings than the K4200, but existing K5200 owners can remain happy that their more expensive purchase is still providing an even better workflow.

Apart from the disappointing result with Maxon Cinebench R15, the NVIDIA Quadro M4000 provides significantly better performance across the board than its K4200 predecessor. Despite the identical memory quantity and speed to the previous K5200, the M4000 doesn't quite offer enough to make you wish you hadn't bothered paying for the extra over the K4200 (the K5200 cost more than double the price of the K4200).

Nevertheless, the 8GB of GDDR5 memory is the real killer feature for the M4000. It does have a few more CUDA cores than the K4200, but its abilities really show themselves when applications are dealing with large texture sets, as shown by a few of the SPECviewperf 12 tests. These may only be in the minority at the moment, but the constant demand for increasingly realistic 3D will mean the huge amount of memory in the M4000 will make it a worthwhile investment over the years of its useful life.

Since the M4000 costs about the same as its predecessor has done for most of its lifetime, buying the new card is a no brainer, with the only caveat being if you're running FP64-heavy software. Otherwise, even at a greatly discounted “end of line price”, a K4200 won't be worth the economy over a M4000. We expect to see the NVIDIA Quadro M4000 gracing the majority of mainstream modelling workstations from now on – until its successor arrives.

Discuss on our Facebook page, over HERE.
Pros:

• Noticeably better performance than the previous K4200 in most applications.
• Twice as much memory as predecessor.
• Faster memory than predecessor.
• Same price as predecessor.

Cons:

• Worse FP64 performance than the previous generation.
• Slightly greater power consumption than K4200.

Kitguru says: Welcome to the new king of mainstream, high-end professional 3D content creation.

Become a Patron!