Recording Computer CPU Benchmarks – Sandy Bridge, Nehalem and Bulldozer Processors Compared

One of the toughest decisions to make when purchasing a computer for digital audio production is which processor will be best for your configuration. These days, the model numbers that designate the processors have very little to do with how fast the CPU actually is. Furthermore, the commonly used denominator for speed GHz means less than it ever has due to variables such as total number of cores, hyper-threading and turbo-boost support, hyper transports, and on and on…

As such, benchmarking digital audio workstation computers for audio production is not an easy task. There are many synthetic benchmarks available that help compare processors but these are generally geared towards gaming, 3-D animation and general productivity. What sets audio production apart from other computing tasks is that everything has to happen in virtual real time – low latencies mean that the CPU has less time to make a calculation, sometimes resulting in glitches and artifacts during recording and playback. Most synthetic benchmarks do not address audio glitches, instead they focus on the raw number of calculations a processor can accomplish in a given time. This is why we use a different, more hands-on approach to benchmark our systems.

Theoretically, a method to gauge the yield of a processor is to count how many plugins it can handle at different latencies. However, there are many variables that come into play. Different types of plugins are able to take advantage of different CPU technologies and many plugins will use up varying amounts of CPU depending on the content of the audio sent through it. Then there are VSTi’s (Virtual Instruments), which use giant hyper-sampled libraries and are very RAM dependent. To make things even more complicated different audio interfaces will give very different results.

Our goal was to simplify all of this. We chose not to benchmark a wide variety of plugins, that would be too complicated and time consuming. Instead, we focused on one plugin loaded many times. We also limited our benchmarks to one interface – the Tascam FW-1884.

Our test interface and benchmark method:

Operating system and system configuration:

All test results were measured on a systems running our specially optimized installations of Windows 7. We perform our benchmark after the computer is fully optimized for audio production. This includes BIOS and driver optimizations, removal of bloatware and unnecessary system services and other various tweaks to the system. All systems had a minimum of 4GB of RAM – however this is rather inconsequential as this is a CPU benchmark.

Interface: Tascam FW1884

We chose the FW-1884 for two reasons:

  1. Since we started Reyniers Audio in 2007 we have used the FW-1884 to test all our workstations. Through experience we have found that if it works with the Tascam FW-1884 it usually works with everything else, as this interface is extremely picky when it comes to which Firewire chipset you use. Obviously we have tested our system with many different interfaces but we use the FW-1884 as a control, to test all our computers against one standard.
  2. The FW-1884 provides very low round-trip latency. Many interfaces have a built-in safety buffer that will affect the true round-trip latency of your computer audio setup, from input to output via an ASIO driver. To test round-trip latency we use CEntrance ASIO Latency Test Utility – this measures the true round-trip latency and is an excellent tool to determine the true speed of an audio interface (don’t trust your host’s latency numbers – they are far from accurate). The number of plugins a project can utilize at low latency greatly depends on the buffer size you select in your interface ASIO settings. The ideal interface will allow a low buffer setting with a low round-trip latency and high plugin counts. The issue of round-trip latency is worthy of another blog post (coming soon) but below is a quick chart to show that different interfaces can have very different round-trip latency results. The lower the latency the more instantaneous audio will feel. As you can see, the Tascam FW-1884 scores very well, and because of this we think this an excellent interface to test with.

Round Trip Latency Measurements in Milliseconds

6.92
10.14
8.64
10.05
11.25
17.03
15.08
3.33
5.67
4.38
14.58
4.88
3.61
4.26
2.95
3.54
5.06
5.65
6.1
5.58
12
6.5
10.41
13.74
8.5
2.49
5.35
2.04
3.79
2.47
5.81
6.3
8.12
5.06
10.88
7.32
6.7
5.3
5.8
3.86
7.19
13.78
3.61
17.5
6.96
5.87
5.81
5.81
2.9
2.9
4.44
14.24
9.3
11
3.5
8.91
3.47
12.56
7.05
5.03
2.47
7.87

The bars in this graph represent the lowest round trip latency we were able to measure using the CEntrance Latency Utility. The number after the @ by the interface is the ASIO buffer size used. Lower=Better

Benchmark method

As mentioned before, audio benchmarks can be rather complicated given the number of variables involved. Luckily, the guys at www.dawbench.com have come up with several benchmarks to test a system with high track counts, extreme DSP loads and very CPU-intensive workloads. Although they offer benchmarks using many different DAW hosts and plugins we have isolated our test to one in particular: the RXC benchmark. This test uses the 32-bit version of Reaper, convenient for us as we already install it on all our workstations. We use this version of the benchmark for two reasons: 1) we love Reaper and use it daily; 2) it gives us a constant platform on which to run our tests and benchmarks.

The RXC benchmark is a Reaper project that consists of:

  • 40 tracks of sine waves
  • 4 stereo tracks of audio content
  • 320 ReaXcomp instances – 8 per sine wave

If you are curious how your DAW computer stacks up feel free to run the benchmark yourself. I have posted our slightly modified version of the DAWbench.com benchmark here.

Here’s how to run the benchmark:

  1. Install Reaper 32-bit (Although Cockos updates Reaper very frequently which could introduce slight variations in your results between versions, since we started testing using Reaper we’ve experienced less than a 3% variance between different Reaper versions)
  2. Extract the DAWBench-Reyniers_Audio-RXC.rar file to a location of your choice.
  3. Inside the extracted directory you will find a file named “reaxcomp-standalone.dll” – place this file in your VSTPlugins directory and make sure Reaper is able to find the plugin.
  4. Complete a fresh system boot
  5. Set the ASIO Driver setting to one of the following buffer settings: 32, 64, 128 and 256
  6. Open the project titled: “DAWbench-Reyniers Audio-RXC.rpp”.
  7. You’ll notice there are 44 tracks. Tracks 1-40 have since waves and are muted by the faders. Tracks 41-44 have the actual audio content – these tracks have no effects and will stay untouched throughout the process.
  8. Hit play and begin activating the ReaXcomp compressors one by one, track by track. You’ll notice the project is setup to loop continually. It is important the activated effects are spread evenly across the tracks so instead of turning on all the effects on one track, work across the project to ensure the workload is divided across all tracks. An easy way to toggle activation of plugins in Reaper is to hold the “Shift” key and clicking the ReaXcomp slot in the Track FX panels in the mixer.
  9. While playing back the project, continue turning on the ReaXcomp compressors until your audio stream gets interrupted by clicks and pops. Often times the clicks and pops will only begin after the project has looped a few times so once your CPU load is in the 80-90% range it is best to slow down the rate at which you turn on ReaXcomp instances to once per loop.
  10. Once a maximum has been reached, we take note of the number of plugins we were able to activate and repeat the process for each buffer size.

The Results:

CPU Benchmark results using ReaXcomp in Reaper

8389108110
94112122127
96110119142
110118143145
117134144153
126144153162
136158165170
156173181186
161176188198
168186194200
196204214222
224251265271
245266277283
253276288294
380415434445
32 Buffers
64 Buffers
128 Buffers
256 Buffers

Total number of ReaXcomp instances before CPU Overload. Higher=Better

By consistently running these benchmarks and collecting the resulting data for several years we were able to distill some very useful information that we now use in creating and optimizing our Digital Audio Workstation computers.

The results you see in our benchmark dictate the steps we take to optimize our workstations computers. In some systems we’ve seen an increase in performance by more than 50% after our optimizations. Through these benchmarks we have grown our understanding of how minor changes to BIOS settings, operating system and hardware drivers can make a huge difference in audio production capabilities. As a result, every system that leaves our workbench goes through a painstaking process of optimization to ensure it falls in line with results achieved with similar hardware. This not only helps us ensure our systems adhere to the highest standards, it also helps us guide our clients into a system that has both their budget and speed requirements in mind.

Some notes about the benchmark results:

  • Currently, the Intel Sandy Bridge processors provide the best “bang for the buck” performance – especially the 2600k processor.
  • The new AMD Buldozer (Zambezi) FX processors offer somewhat disappointing results compared to Intel’s Sandy Bridge line-up. It is however a less expensive platform which ultimately offers enough processing power for most mid-size projects.
  • Intel Nehalem 6-Core processors yield the highest plugin counts on a single CPU workstation.
  • Intel Dual Xeon systems featuring 12 CPU Cores (2×6) yield the highest plugins counts available to date.

It is hard to say which processor is best for everyone – again there are too many variables to make it that cut and dry. If you are a singer/songwriter and you mostly create projects by recording them one track at a time, a system based around the new AMD Zambezi processors or the Sandy Bridge i5 processor will usually be sufficient for projects requiring between 20 and 40 tracks. If you find yourself using lots of Virtual Instruments or track large bands with track counts over 40 you will undoubtedly benefit from extra CPU power, especially when monitoring directly from your DAW host software. For this we recommend the Sandy Bridge 2600k or the Nehalem i7 980 processor. For those composing symphonic movie scores using EastWest libraries or Vienna Orchestra libraries it is still the best to go with a dual Xeon setup – especially the 12 core configurations.

This entry was posted in Audio Interface, Benchmark, Recording Computer. Bookmark the permalink.

6 Responses to Recording Computer CPU Benchmarks – Sandy Bridge, Nehalem and Bulldozer Processors Compared

  1. Joseph Santoyo says:

    This article was VERY helpful – you should definitely continue to write more about these things. I was very impressed by it and I look forward to others!

  2. Keith says:

    Hey Guys, It’s been a year since this article was published. I’m not sure too much has changed, but it would be interesting to read about any impact on your recommendations the newest processors available might have. Are the 6, 8 and 12 core processors a big enough improvement to justify their cost or is the 2600K still the besyt bang for the buck running a bunch of tracks with plug-ins and/or virtual instruments at a low sample/latency setting? Thanks!

    • You are absolutely right in that this blog post is a bit dated. The landscape hasn’t changed a whole lot since this post however. Sure, Intel has come out with newer versions of their processors, but they are essentially slightly faster versions of the ones bench-marked in this review. We will likely be updating this post to reflect the newer sku’s but until then here’s some insight. The 2600k is still a very good bang for the buck processor and for +-$60 you can upgrade it to the i7 3770k which will yield an increase of about 10% in plugin counts. The 3770k is an Ivy Bridge processor, meaning it has a slightly more powerful graphics chip – this is relevant if you plan to use on-cpu graphics – it is completely irrelevant if you plan to purchase a designated video card. With regards to the the 6, 8 and 12 core processors (the 8 and 12 core options are only available on the dual processor xeon platform) – going to this high core counts does improve performance drastically. Note that the dual 6 core X5650 (no longer available) we tested is almost identical to the dual E5-2620. If you were to step up to 16 cores via dual E5-2650′s you’d increase your plugin counts by about 20% over dual X5650/E5-2620. Dual E5-2687W will give you about a 45% increase over dual X5650/E5-2620.

      • Keith says:

        Thank you for responding so fast!! The 3770K has its upsides for sure. I just worry about the way they designed the heat transfer vs the Sandy Bridge. I guess I could research a QUIET yet Powerful processor cooler. Do you have opinions on Overclocking at a stable level using DAWs? I am not sure about using the integrated graphics in a 3770K, if they use processor cycles. I’d rather save then for plug-ins and VIs, if that’s the case. If not, DAW graphics aren’t that intensive and I’d save the $$$ and use them for a Killer motherboard. Any recommendations for an LGA 1155 that just brings out the best in a 3770K? Memory speed that performs best? Thank you for being such a credible Recording source! Few and far between! Regards//Keith

        • We built many systems using the 3770k and various other Sandy Bridge and Ivy Bridge processors and paired with the upgraded from stock Noctua NH-U9B we’ve found heat dissipation to be very effective. Although, you can reach stable overclocks they are very hit or miss, especially if you need your system to be quiet. In fact, we offered overclocked configurations for some time, but found that we’d end up tossing out more CPU’s than we actually got to use because they required overly aggressive fan profiles. If you want to overclock, expect your system to be loud. If you want silence, it’s best to keep things at stock settings. Also, although it is possible to get stable performance out of an overclocked CPU I would still not want to depend on it for mission critical operations such as recording, because glitches will appear sooner or later.

          With regards to on CPU video processing, you don’t sacrifice any CPU cycles for video as the Sandy Bridge and Ivy Bridge CPU’s have a designated GPU portion on the processor that simply goes unused if you use a standalone graphics card. The only sacrifice you make is in the RAM department, you can designate how much of your RAM is used for Video, but no matter what you will at least give up 128MB of system RAM if you use on CPU video processing. Having said that for audio production the Sandy Bridge and Ivy Bridge processors offer enough GPU power to facilitate general use, some light 3D acceleration and video acceleration.

          I think the 3770K is an excellent bang for the buck processor and the motherboards we offer on our Virtuoso Ivy Bridge configuration take full advantage of all of its features. Furthermore we use high performance Corsair XMS3 RAM which utilizes all the bandwidth available on the platform. The only way to get more performance out of this type of hardware would be to overclock and like I said before, I generally don’t recommend it for mission critical operations such as audio production.

          • Keith says:

            Hi Wouter, Your hands on experience is priceless to the Recording world. Thank you for sharing that experience. I sure hope people realize “Perfection Is No Accident” and if they want a tried and tested system that will get great results for years to come, Reyniers Audio is certainly the place to start. You brought several perspectives to the forefront I probably would have overlooked. Nothing like going to a Pro! Thank you! Another great topic would be the Virtual Instrument and Plug-In world and how they are catching up to 64 bit, so we can use that now available memory! And then, how much memory is enough. Thanks again!

Leave a Reply

Your email address will not be published. Required fields are marked *

*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>