Performance Tests

We put the professional and consumer cards through the mill, with some surprising results…

To test the vendors’ theory that driver optimisation makes a considerable difference to the performance of professional graphics compared to consumer versions, we tried side- by-side tests of the Nvidia Quadro M4000 against a Geforce GTX 970, using Specviewperf 12 and the Opengl portion of Maxon Cinebench R15. The tests were performed with the cards installed in a YOYOTECH Renatus Pro M3 workstation, equipped with an Intel Core i7-5830k six- core processor clocked to 4.2GHZ and 32GB of 2,133MHZ DDR4 SDRAM.

As we stated elsewhere in this article, both of these cards sport 1,664 CUDA cores, but the clock speeds are slightly different. The Quadro M4000 runs its GPU at 773MHZ, whilst the Asus Geforce GTX 970 Directcu II OC Strix we were comparing operates at a considerably faster 1,114MHZ. However, the Quadro M4000 has 8GB of GDDR5 memory running at 6,000GHZ, whilst the Geforce GTX 970 has only 4GB, running at 7,010MHZ, although just 3.5GB operates at this speed, with the remaining 500MB is seven times slower. So, unless you need more than 3.5GB memory, the GTX 970 on paper has the upper hand, and looking at prices on Yoyotech.co.uk, it’s much cheaper, too. The Quadro M4000 will set you back £779.90 inc VAT, whilst the GTX 970 is only £ 279.99 in VAT – and this is a premium model, too.

Taking all this into account, the performance results are very surprising. In Maxon Cinebench R15’s Opengl test, the GTX 970 beat the Quadro M4000 by 167.78 to 152.56 – a ten per cent lead. But half of the Specviewperf 12 viewsets were significantly in favour of the M4000. In catia- 04, the M4000 managed 86.03, compared to 69.77 for the GTX 970 – 23 per cent faster. Similarly, in creo- 01, the M4000 achieved 73.72, where the GTX 970 could only manage 56.24 – 31 per cent faster. The gruelling energy- 01 showed the benefits of driver optimisation and the extra memory, with the M4000 almost twice as fast as the GTX 970, gaining 5.85 versus 3.01.

However, of particular interest to 3D animators will be the score in the maya- 04 viewset, where the M4000 achieved a healthy 68.77, but the GTX 970 was over 50 per cent faster still with 103.7. The GTX 970 was also 18 per cent faster than the M4000 in the clinical visualisation viewset, medical- 01, with 29.88 versus 25.26. The most understandable win for the GTX 970 was in showcase- 01, which is the first viewset to use Directx acceleration. Since this is the API used by most games, rather than Opengl, you would expect the consumeroriented GTX 970 to be optimised for it. The GTX 970 managed 61.08, where the M4000 could only muster 43.7, making the non- professional card 40 per cent faster.

However, the remaining two tests illustrate where driver optimisation for specific types of software is important. The snx- 02 viewset, which is based on the NX 8.0 CAD software from Siemens, completely flummoxed the GTX 970. The latter only managed 5.96, where the M4000 achieved 72.94, over 12 times faster, showing a complete lack of optimisation in the consumer card. The M4000 was also more than twice as fast in the Solidworks-based sw- 03, managing 101.3 compared to 50.1. So, for these applications, there are clear advantages to using professional hardware, with Siemens NX 8.0 being effectively unusable with consumer- grade graphics. Read the realtime rendering article here: www.bit.ly/206-real-time

It’s the performance results that really tell the story about why a pro card is the best choice for pro apps