Intel Xeon E5 v3 Haswell processors review: we check out the fastest chips on the planet
The next generation of Intel Xeon chips are here. We benchmark them to discover if they can deliver more power to creative software such as After Effects, Cinema 4D, Premiere Pro, Blender and more.
The Intel Xeon E5-2600 v3 series of processors is the company’s latest line for dual-chip workstations for high-end animation, visual effects and video post-production. You’re also likely to see a choice of the new chips next time Apple updates its Mac Pro, which we expect to happen soon - tho with the Mac Pro you get only one chip.
We spent some time testing workstations based around two of the new chips before the launch at Boston, looking at a maxed-out workstation and more more modest machine - though Boston was unable to give us pricing pre-launch. But first off, how is the Xeon E5 v3 different from v2? (If you want to skip the science bit and get to the benchmark results, scroll down.)
These chips are based on the Haswell architecture, superseding Ivy Bridge and Sandy Bridge before that. In real terms, this means that the Xeon E5-2600 v3 offers more cores than the previous generations - maxing out at 18 cores per chip against v2’s 12 cores. As with v2, the v3 chips with the most cores have relatively slow clock-speeds - with the 18-core E5-2699 v3 running at 2.3GHz - and these are aimed primarily at server. Those with fewer cores run faster. Intel has picked out one chip as best for workstation performance – balancing core speed and number – and given it a W suffix: the 10-core, 3.1GHz E5-2687W v3. The previous generation had its own version of this - the E5-2687W v2 - which ran slightly faster (3.4GHz) but with eight cores.
This isn’t the only Xeon E5 v3 suitable for workstations: the E5-2667 v3 (eight-core, 3.2GHz) and E5-2643 v3 (six-core, 3.4GHz) offer fast clock speeds – while if it’s cores you want, the 10-core, E5-2660 v3 runs at a reasonable 2.6GHz and costs significantly less than Intel’s workstation picks.
Haswell vs Ivy Bridge
The Xeon E5 v2 'Ivy Bridge' chips were based on the same 'X9 Romley' platform as the 'Sandy Bridge' v1 chips, so could live on the same motherboards and sharing the same underlying features. The v3 'Haswell' chips are part of a new 'X10 Grantley' platform, which offers more than just the new chips to boost the performance of your application. There's 20 per cent faster connection between the chips and the rest of the platform (including the RAM), and support for faster-running DDR4 RAM. DDR4 RAM offers faster connection speeds and lower power needs, though the latter is more important for servers that workstations. One downside to DDR4 RAM currently is that it can be hard to get hold of - and therefore it's pricy too.
There's also built-in support for USB 3.0 and more SATA 3 connections for more hard drives - though again the latter is more for servers that workstations.
Haswell Xeon E5 v3 workstations
I spent a day at Boston's offices testing two Haswell-based dual-processor workstations. The 'supercar' Venom 2401-12T model (above) had two E5-2687W v3 Xeon chips (3.1GHz, 10 cores each) - plus 128GB DDR4 ECC RAM, a Nvidia Quadro K6000 graphics card with 12GB of its own RAM, a 240GB system SSD, and two 2TB media drives (mirrored). It was liquid-cooled to keep it quiet and powered by a 1200W PSU.
It's a monster.
The 'sportscar' Venom 2301-12T (above) had two E5-2650 V3 Xeon chips (2.6GHz, 10 cores each), 64GB DDR4 ECC RAM, a Nvidia Quadro K4000 graphics card with 3GB of its own RAM, a 240GB system SSD, and two 1TB media drives (mirrored). It's conventionally cooled with a 900W power supply.
This isn't as tricked out as the 2401-12T, but it's still more than powerful enough for the requirements most of us - and likely a lot more affordable too.
Haswell Xeon E5 v3 benchmarks
We tested both machines using our standard benchmarks that are partially or completely based on the processors - leaving out only those that test purely the graphics card or storage performance. The include the rendering part of Maxon's Cinema 4D-based Cinebench R15; video and VFX rendering in After Effects CC 2014 and Premiere Pro CC 2014; the SPECwpc workstation benchmarks based on Blender, Handbrake and LuxRender.
Haswell Xeon E5 v3 Cinebench benchmarks
Cinebench renders a relatively simple animated 3D scene, and is based entirely on CPU performance. It's freely downloadable for Mac and Windows, so you can test your own workstation.
Here the 'supercar's new E5-2687W v3 Xeon chips was over 41 per cent faster than two previous-generation E5-2687W v2 chips in a Dell Precision T7610. It was also 25 per cent faster than the E5-2650 V3 chips in the 'sportscar' model.
The difference over the current - though almost-certainly-soon-to-be-upgraded - single-chip Mac Pro (with a 3GHz 8-core chip) was massive. The 'supercar' model was 3.9x faster than the Mac Pro, and the 'sportscar' model was over twice as fast.
We also compared these dual-processor machines to a standard Dell single-chip T3610 workstation with a Xeon E5-1620 3.7GHz 4-core chip and 16GB of RAM - where the 'supercar' was 4.8x faster and the 'sportscar' 3.9x. These aren't really fair comparisons as the T3610 is much cheaper than either model - but it's worth being reminded of the power of dual-chip workstations for primarily CPU-based tasks when most of the buzz is around GPU-based rendering.
Benchmarking the Haswell E5 chips against the Ivy Bridge previous generation using our other tests was trickier, as SPECwpc and the CC 2014 versions of After Effects and Premiere Pro weren't yet released when we benchmarked the Dell Precision T7610.
Haswell Xeon E5 v3 After Effects benchmarks
Our After Effects rendering benchmark tests all-round performance of a PC: CPUs, RAM, graphics and storage. Here, the 'supercar' was 35 per cent faster than the 'sportscar' in our pure-AE scene and 21 per cent faster at working with a scene that included a 3D Cinema 4D scene through the CineWave plugin.
Haswell Xeon E5 v3 Premiere Pro benchmarks
This test outputs a three-minute HD timeline with effects and transitions and then encodes it as an H.264 file. We turned off CUDA and OpenCL processing so only the CPUs were used. Here the 'supercar' was 29 per cent faster than the 'sportscar', and they were 374 per cent and 267 per cent respectively faster than the single chip T3610.
The most notable thing here, though, is that graphics card acceleration is more important for fast rendering and encoding, with turning on CUDA and OpenCL making the biggest difference to processing speeds.
Haswell Xeon E5 v3 SPECwpc Blender benchmarks
This benchmark renders a more complex scene than the Cinebench test. SPECwpc is freely downloadable for Windows, so you can test your own workstation.
The 'supercar' was 12 per cent faster than the 'sportscar', and they were 87 per cent and 62 per cent respectively faster than the single chip T3610.
We were also able to test the 'supercar' with a single E5-2687W v3 chip. Here the dual-chip workstation was only 23 per cent faster.
Haswell Xeon E5 v3 SPECwpc Handbrake benchmarks
This benchmark encodes a series of videos using the popular open-source video encoding software.
The 'supercar' was 25 per cent faster than the 'sportscar', and they were 152 per cent and 101 per cent respectively faster than the single chip T3610.
The dual-chip 'supercar' was 17 per cent faster than with a single processor.
Haswell Xeon E5 v3 SPECwpc LuxRender benchmarks
This benchmark renders a scene using LuxRender, a physically accurate rendering engine that's very demanding - and where the power of these high-end chips really shines.
The 'supercar' was 25 per cent faster than the 'sportscar', and they were 421 per cent and 300 per cent respectively faster than the single chip T3610.
The dual-chip 'supercar' was 94 per cent faster than with a single processor.
Haswell Xeon E5 v3 benchmarks: What these results mean
From our tests, it's clear that these new Haswell Xeon E5 v3 chips are undeniably powerful - though we'll reserve judgement on their true worth until we see pricing for the full systems.
We saw great variance between the usefulness of having two chips - it made a massive difference to LuxRender but hardly any to Handbrake. Some applications just can't maximise themselves for 20 cores over 10. The Premiere Pro test also showed that graphics card performance can be as important as CPU power - if not more so - so if you having to choose between spending money on another chip or a more powerful graphics card, you need to consider your particular applications' needs carefully.
But if you need to maximum power - and money's no object we're guessing - then you want a workstation with two of these new Xeon E5 chips inside.