By Neil Bennett | on July 14, 2014
Price: AMD FirePro W8100, £1,500 plus VAT. Nvidia Quadro K5000, £1,200 plus VAT
AMD FirePro W8100: 4.5
Nvidia Quadro K5000: 4.0
AMD’s FirePro W8100 has been developed with the Nvidia Quadro K5000 firmly in its sights. Both are aimed at 3D and VFX artists and video editors with serious but not outrageous performance needs – and similar budgets. They have roughly equivalent pricing – though you can pick up the K5000 for about £200 less than the W8100 from sites like Scan – but the W8100’s specs mark it out as being much more powerful. However, as usual, which is best is a more complex question than mere specs (as if it was you wouldn’t need me to tell you which one to buy).
AMD W8100 vs Nvidia K5000: Specs
Some of the specs differences are obviously in the FirePro W8100’s favour. The W8100 has 8GB of RAM to the K5000’s 4GB, so AMD’s board can work with larger numbers of more complex textures and video frames.
Other specs are less easy to draw conclusions from. The W8100 has 2,560 stream processors while the K5000 has 1,500 CUDA processors – but as the two are not directly comparable, these are essentially meaningless.
The same is largely true for the differences in 'peak single-precision floating-point performance’ – 4.2Tflops for the W8100 vs the K5000’s 2.1Tflops – or the size of memory interface – 512-bit for the W8100 vs the K5000’s 256-bit – or the throughput of memory bandwidth – 320GBps for the W8100 vs the K5000’s 173GBps. All of these exist only in the context of each card’s architecture, and once you throw in that applications work with each card slightly differently – or very differently in the case of non-3D tasks in video editing or VFX applications that tap the performance of each graphics chip using OpenCL (on AMD boards) or CUDA (on Nvidia) – it’s clear that the only real way to tell which board is better is by benchmarking them.
AMD W8100 vs Nvidia K5000: Benchmarks
We tested the FirePro W8100 and the Quadro K5000 in a Dell Precision T3610, which has an Intel Xeon E5-1620 four-core processor, 16GB RAM, a 256GB SSD system drive, a large 2TB media drive and Windows 7 – exactly the kind of system you’d put either of the boards in, and powerful enough not to throttle either of the cards’ performance.
There are three key benchmark tests that we ran on both cards: Maxon’s Cinebench R15 – which is based on Cinema 4D; Autodesk Maya 20XX – as part of the specWPC suite; and our own Premiere Pro CC 2014 test. The first two are both real-time 3D based – Cinebench is relatively simple but good for checking out day-to-day performance, while the Maya test is more strenuous. Our Premiere Pro test renders and encodes a sequence with multiple HD clips and effects, tapping the appropriate technology from each card to boost them (OpenCL for AMD’s W8100 and CUDA for Nvidia’s K5000). As you’d expect, we ran each test three times from a restart and took an average of the results.
In Cinebench, the results from the W8100 and K5000 were near identical, with the K5000 slightly ahead. In the more complex Maya test however, the W8100 pulled ahead – delivering 28% faster performance.
In Premiere Pro, the K5000 was moderately faster – taking 165 minutes to complete the test against the W8100’s 171 minutes (around two and 3/4 hours to you and me). Both were hugely faster than using the CPU alone though, which took 673 minutes (otherwise known as over 11 hours!). This surprised us – especially as AMD has promoted the W8100’s performance for video and other non-3D tasks – but could be as much to do with the interaction between Premiere Pro and the boards over two different technologies – AMD’s implementation of OpenCL and Nvidia’s own CUDA – as it’s based on the power of each board.
AMD W8100 vs Nvidia K5000: Installation
Both boards should fit easily in most workstations as both are relatively short, double height cards – and as they have reasonable power requirements. The W8100 doesn’t have ridiculous power consumption of its big brother, the W9100 – which required more power than our Dell T3610 could deliver (275W) – but it still requires 220W. This means you’ll need two free 6-pin outputs from your power supply.
The K5000 requires only a single 6-pin input, as it draws a modest 122W.
Where the W8100’s higher power need could cause issue is if you want to use two or even four in the same workstation – as you’ll need a heavy duty power supply and probably some additional cooling to keep things running smoothly. However, chances are that if you do want a workstation like this you’ll probably buy a complete new one and the likes of Armari or Boston – who sell four-card machines with the W8100 or the W9100 under the Ultra Workstation brand – will fit an appropriate (though expensive) power system for you.
You can’t buy an Ultra Workstation from big brand workstation vendors like Dell or HP, but you can buy Precision and Z-series workstations with two – but not four – K5000s. Nvidia also sells the Tesla series of boards to accelerate non-3D tasks such as rendering and encoding, though these are more expensive than just adding an additional graphics card.
Both boards feature four outputs: the W8100 has four DisplayPort outputs while the K5000 has two DisplayPort and two DVI (one DVI-D and one DVI-I). This might make a different to users in scientific or financial jobs who want four displays – but for animation, VFX and videos pros, it’s much of a muchness.
AMD W8100 vs Nvidia K5000: the final result
Both boards are capable and powerful, but right now we’d plump for the FirePro W8100. It’s more powerful at its core job – accelerating 3D graphics in reasonably complex scenes in the likes of Maya. While its Premiere Pro performance is lower than we’d expect from a board that’s brand new and apparently been designed for video post-production, past experience tells us that the W8100’s performance here will grow as AMD releases better drivers – we tested the W8100 with AMD’s first proper driver for it, while the K5000’s driver has been tuned since its launch almost two years ago.
The question that hangs over both these boards is how long traditional desktop workstations with hefty graphics hardware inside them will be a big part of creating 3D, animation, VFX and video content. Both hardware companies like Dell and Nvidia, and creative software developers like Autodesk and Adobe expect the future to see you hulking desktop to be replaced by a laptop or a modern equivalent of a dumb terminal – and all of the heavy-duty processing to be done on servers in your network or in the cloud and then streamed to your screen. We’re some way away from this becoming a reality – especially for smaller shops – but it’s something to keep an eye on for the future.