You are here

CPU Performance Tested

The Need For Speed
Published July 2015
By Pete Gardner

The Need For Speed

Our benchmark tests show you which CPUs offer the best audio performance, whether you’re on a budget or money’s no object.

Every few years, as resource-hungry software takes advantage of newer, more powerful computers, the same questions arise about the computer that lies at the heart of your studio: do I need to upgrade this machine, or should I replace it entirely? And what kind of improvement can I get for my money?

In this article, I’ll help you to answer those questions by taking you through the results of some CPU performance tests I carried out recently. I included a range of chips from the more cost-conscious to the fastest chips money can buy, and have looked at both mobile and desktop versions, as well as new chips and some that are now three or four years old. The chart makes clear which CPUs offer the best performance, but nothing about cost and certain other important considerations. So I’ll also explain in these pages a bit more about what makes a good CPU, and the implications of your CPU choice for the rest of the system (in terms of heat, cooling and noise, for example).

Download the full resolution chart here

CPU Basics

The CPU (Central Processing Unit) is just one factor that influences your computer’s audio performance. There are others, not covered here, including the type of storage used and the amount and speed of RAM, but the CPU plays a hugely important part.

The days of being able to identify the best CPU by looking at the headline clock speed are long gone. Despite the clock speeds not having increased dramatically in recent years, the processor manufacturers have continued to improve their chips in other ways. (See the Intel & AMD box.) This means that when comparing different generations of similarly named chips, the clock speed and core count may appear similar, but the real-world audio performance will vary significantly. When moving on from a product that’s three, four or perhaps even more years old to a new setup, it’s therefore very difficult to figure out what sort of performance improvement your money will buy — and it’s particularly difficult to establish at what point spending more money takes you past the point of diminishing returns.

Test Methodology

I spent quite some time comparing a wide range of CPUs commonly used in audio machines, using a standard test, the freely available DAWBench DSP Universal 2014 test. This reveals a lot about a CPU’s suitability for audio production: although it won’t give the full picture in regards to ASIO performance, which can vary according to other system factors, this test is designed specifically to ‘stress test’ a computer’s CPU. DAWBench has been discussed many times in SOS and there’s plenty of information at www.dawbench.com but, essentially, it loads instances of a plug-in in a DAW, measures the performance and generates a score. By restricting the tests to a standard plug-in (Cockos’s ReaXComp multiband compressor, in this case), DAWBench enables you to make meaningful comparisons in performance.

It’s important to keep other factors as consistent as possible, so I used the same USB audio interface (Native Instruments Komplete Audio 6) for all tests. Some expensive, better-performing interfaces may result in better performances overall, but the important thing is to establish a stable baseline. (It’s also worth mentioning that, while better interfaces are available, I have established via group testing that this particular model offers a great performance/price ratio for new users wanting to make music.) For the same reason, all tests were performed in an identical Windows 7 OS installation, and the same DAW software (Reaper) was used: we can therefore be sure that the differences in performance shown on the chart aren’t due to any differences in software.

I could have tested just the current generation of CPUs, but chose also to include legacy models because most users will keep a dedicated audio computer running for three to five years before upgrading — I hope the chart will give some indication of how the performance has progressed over the last several years, and how newer but lower-spec CPUs compare with older top-of-the-line ones.

As we’ll see, another important ‘take away’ from this test is that not all CPUs with the same nomenclature are created equal, even if they’re of the same generation and have the same naming system — this is particularly important to bear in mind when moving from a mobile to a desktop setup, or vice versa.

Finally, these tests are inevitably a snapshot: it won’t be too many months before a new bigger, badder, faster model comes along, so perhaps we’ll revisit this question when the next generation of chips becomes available. With all that in mind, let’s run through the results, from the bottom of the chart upwards...The Need For Speed

Intel ‘U’ & ‘M’ Series

First up, we have the Intel Core i3 4010U. It’s not surprising to see the U-series chips at the bottom of the chart: these chips were initially conceived to power the ‘ultrabook’ class of super-thin, lightweight laptops, were designed to offer longer battery life, and are closer in performance terms to the old Atom range than the desktop CPUs that share their ‘Core’ name. Nonetheless, they’re also found in many regular low- and mid-range full-size laptops as well as high-end tablets, including the Microsoft Surface Pro 3. For simple multitrack recording or editing of audio you‘ve recorded on location they offer all you need in a portable and inexpensive machine. If you wish to use virtual instruments, effects chains or other processing, though, their performance will soon prove frustrating, and you’d be better off considering one of the more fully featured Intel models discussed later.

The next (small) step up is another popular mobile chip, the basic M-series. This includes the dual-core models found in the current entry-level MacBooks. They outperform the U-class chips, but by a small margin: the performance levels are around 25 percent greater and so they’re not much better for audio work. I’d suggest that anyone wanting an Apple laptop for anything approaching serious studio work should find the funds for a MacBook Pro — a refurbished or secondhand model from a generation or two ago will probably outperform the current standard-issue MacBooks, and will also offer greater connectivity.

Early Core i7 & AMD

I’d hoped to test chips as far back as the Intel Core2 Quad Q6600. This offered the best bang-per-buck of its generation and was regarded as the CPU of choice for studio use. Unfortunately, the OS build I used refused point blank to work with this chipset, and to keep the tests consistent this model was excluded. Still, some useful data can be found on the CPUBenchmark web site (see box), which suggests that it offers roughly 40 percent lower performance than the Core i7 930, and only 20 percent higher than our bottom-of-chart CPU.

The Intel Core i7 930 and its family of processors ran on the X58-type motherboard, which saw a number of changes and significant improvements to the chipset. This is the generation of advancement that saw Intel move the memory controller onto the CPU die itself, as well as introduce the QPI (quick path interconnect) board design that emulated AMD’s HyperTransport bus design. By shortening the distance that the data had to travel before being processed, this yielded some significant performance gains — this was the first generation that saw Intel stride ahead of AMD in the performance race.

The next cluster of entries make for interesting comparison. The Core i7 4710MQ is possibly the most popular current option for production laptops, with good reason. It offers the best price-to-performance ratio in Intel’s mobile CPU range: there’s a handful of better-performing models but they roughly double in price for each five-to-eight percent increase in performance. Similar performance levels are offered by the AMD FX9370 and FX9590 (AMD’s current flagship models) and Intel’s Core i7 2600K, their consumer-level flagship in 2012.

As you can see, even today the i7 2600K is no slouch. Similarly, the Core i7 4710MQ currently powers many a laptop for on-the-road producers and performers. For those wanting to use an audio machine on-the-hoof, these processors are not bad by any means and should still hold appeal, but if you’re thinking of a long-term investment in a studio machine, note how startlingly different the benchmark score is between these and the highest-performing desktop chips!

The AMD chips cost less, as do related components, because the motherboards go through fewer socket changes over the generations, which means a computer can be upgraded over a longer period of time. They’re therefore a tempting proposition if you’re putting together your first dedicated setup on a budget. Their downside is that they’re rather power hungry, largely because AMD have found themselves squeezing hard to get every last drop of performance out of a platform that’s overdue a redesign. For example, the FX9590 requires almost twice as much power than the current crop of high-end Intel chips. You don’t have to be an eco-warrior for this to be relevant: more power means more heat, which means more effective cooling and, ultimately, a noisier system. If you can isolate your computer and ensure that any generated noise levels are taken care of, the AMD models are well worth consideration — but the limited budget that makes them attractive probably also means a smaller space in which to record and mix, in which case the extra noise will be a most unwelcome guest!

The Core i7 3770K is another consumer-level flagship chip from a couple of generations ago. Above this we have the Core i3 4370. You may be surprised to see an i3 chip included in the tests, as these are positioned as Intel’s budget-friendly low-end, low-cost CPUs and compete with AMD in the home-office market. Its place here is warranted, though, as the first of the current generation: it shows us what to expect of the i3 range when lined up against other currently available products in the range. Although perfect for the sort of role for which they’re intended, they tend to lack raw processing power when it comes to much more than basic multitracking and editing duties in the studio. Also, AMD offer better-performing options in the same price bracket.

The Core i5 4960 is a capable chip and a good entry point for those who are working on a budget and building a basic studio computer. Again, at this price the AMD products enjoy a better performance-to-price ratio but at the cost of more heat/noise; the i5 4960’s power draw is around a third of the high-end AMD chips’ in the same price bracket.

Newer Core i7s

Moving up the chart, we come to one of the most popular current options for the home studio, the Core i7 4790. I’ve included both its low-power ‘S’ and full-performance ‘K’ series revisions. The ‘S’ version of the chip has a total TDP (Thermal Design Power, the maximum heat generated by the CPU that the cooling system is typically required to dissipate) of 65W. By contrast, the TDP of AMD’s flagship chip is 220W! This shows just how efficient Intel have made their CPUs over the last few generations. Another model, below the ‘K’ and ‘S’ versions in performance terms, is the ‘T’ edition. This has an even lower TDP of 45W and is particularly well suited to smaller form–factor computers and in builds where passive cooling is desired (for instance, to create a truly silent PC.)

The ‘K’ edition is the highest-clocked version; it gains another 30 percent performance and still manages a respectable 88W TDP — it’s easy to build a low-noise machine around one of these. This, along with the competitive price, gives it amongst the best price-to-performance ratio and it’s probably the most popular current choice for those wishing to do everything in the box in a home-studio. It also means it’s a reasonable upgrade for users of the i7 930 (or previous) generation of CPU — although looking at the scores of the i7 2600K and above, it may not prove such good value as an upgrade path in the longer term.

Turbo Power!

Another key difference between the ‘K’ chip and the standard one lies in how they ‘Turbo’ the CPU. The Turbo feature on the Intel CPUs allows them to run above the standard clock speed automatically when the system determines that there’s the overhead to do so without overheating or otherwise causing the CPU to become unstable. The Core i5 4690, for instance, is advertised with a 3.5GHz base clock speed and a 3.9GHz Turbo clock speed, and these quoted figures are the same for both the standard and K-series chips. The reason for a £10$15 difference in price is a different implementation of the Turbo feature. The standard editions stagger the overclock across the four physical cores: in this instance, Turbo will overclock core 1 to 3.9GHz, core 2 to 3.8GHz, core 3 to 3.7GHz and Core 4 to only 3.6GHz. The ‘K’ version, on the other hand, will overclock all cores to the advertised limit, so all the cores can work at 3.9GHz resorting to traditional overclocking.

If we then compare the i7 4790K with its standard-edition counterpart, we see a 3.6GHz base clock with a 4GHz Turbo, versus a 4GHz base clock and 4.4GHz Turbo — the ‘K’ version represents much better value for audio users given the small increase in price. And that’s before we even begin to consider the benefits of a gentle overclock to 4.6GHz or above, which this chip is easily capable of.

More Than Four Cores

Our next group of chips includes Intel’s so-called ‘Enthusiast’ CPUs and a couple of entries from its Xeon server class. The Core i7 4960X was the flagship chip of the previous generation and is more powerful than the current consumer-class CPUs. Being a six-core design (the mainstream ones all have four cores) this didn’t come as a surprise. The same generation’s Core i7 5820K has a similar specification and offers a small increase in performance but is unlikely to prove worthwhile to owners of the previous generation’s six-core chips.

For sheer performance, the latest generation of Intel Core i7 chips cannot be beaten.For sheer performance, the latest generation of Intel Core i7 chips cannot be beaten.Intel’s current flagship desktop CPU is the eight-core Core i7 5960X. At its regular clock speed it’s more than a solid performer but, like previous flagship chips, it doesn’t offer great value for money compared with other processors in its class. It performs very well when overclocked, though, and it is, without a shadow of doubt, the best pick if CPU performance is the only criterion you’re interested in.

Along with the extra cores and overall performance, this chipset can support twice as many RAM DIMMs as previous ones, making it a perfect choice for those creating large, detailed arrangements using resource-intensive sample-based instruments such as Vienna Symphonic Library. In fact, it is users of such libraries (along with those who also do video work) who are usually pushing the limits of what systems can do.

This brings us nicely back to the Xeon server-class chips, which seem, at first glance, to hold so much promise. In principle, the differences between Xeon chips and their i7-series siblings are actually few and far between, the chief one being that Xeons benefit from an additional QPI link — something which enables data sharing between the two or more processors found in Xeon chips. Another is the sheer number of processors: there are already Xeon chips available with as many as 18 cores, although these are seriously expensive! The downside of the QPI link is that the chips remain ‘locked down’ — in other words, overclocking isn’t possible and there’s extremely little scope to tweak for performance enhancement. The reason for this is that these CPUs are designed for server applications, in which it is of critical importance that data sharing between the cores remains stable. The Xeon chips with a high core count are also optimised for parallelisation.

Clock Speed Versus Core Count

Unfortunately, in tests of Xeon CPUs, my colleagues and I have tended to find that the higher the core count, the slower each individual core becomes. With load balancing being handled by both the OS and your sequencer, each trying to squeeze the most out of the available performance, some headroom is inevitably lost. In fact, fewer cores with higher clock speeds will almost always yield better performance overall than a greater number of cores running more slowly. This goes some way to explaining the strong showing by the higher speed-per-core overclocked i7 5960X over the slower-clocked but higher core-count Xeons in the test results.

But it’s actually slightly more complex than that, because I’ve tested Xeon CPUs with a number of sequencers, and it’s clear that the Xeon architecture has yet to be fully exploited by any DAW manufacturer. Multi-core support varies from one DAW to another, and from one version of a DAW to another: some support a maximum of 16 cores, while a few can handle up to 40. This upper limit includes so-called ‘hyper-threaded’ cores, which mean in Intel’s case that there are twice as many ‘virtual cores’ as there are physical processors.

The Xeon server-class chips offer as many as 36 virtual cores per CPU, but DAW software has yet to be designed to take advantage. For now, they remain an expensive luxury, probably only of benefit to composers who have reached the RAM ceiling of the current Core i7 systems — but that itself is something that looks set to  change!The Xeon server-class chips offer as many as 36 virtual cores per CPU, but DAW software has yet to be designed to take advantage. For now, they remain an expensive luxury, probably only of benefit to composers who have reached the RAM ceiling of the current Core i7 systems — but that itself is something that looks set to change!By way of example, a hyper-threaded 18-core Xeon CPU makes 36 virtual cores available to your OS, and installing two such CPUs would give you a whopping 72 cores. Alas, as even the most multicore-friendly DAW can access only the first 40 cores, the other 32 would simply be released for other purposes, or remain idle. On the chart we can see this theory played out in the real world: the Xeon E5 2620 v3, a six-core 2.4GHz (3.2GHz Turbo) model, is comparable with the Xeon E5 2670 v3, a 12-core 2.3GHz (3.1GHz Turbo) CPU. When testing each of these, I used a pair of chips, so the E5 2670 v3s gave us a total of 48 threads (12 per chip makes 24, and hyperthreading doubles this to 48), and I witnessed the 40-thread wall in practice. While this pair of chips is capable of better performance than the chart shows, it requires a software developer to take advantage of it before you can reap the rewards, in practice.

Of course, having eight spare threads for Windows and other processes to use allows the DAW unimpeded access to the processing cores that are assigned to it. Yet this system sits pretty close to the top-of-consumer-range stock-clocked single-CPU eight-core offering... and the dual-Xeon solution costs roughly three times the 5960X! In fact, you could quite easily build a complete system around a Core i7 5960X for less than the cost of two of the Xeon CPUs alone, without any other system components. It’s safe to say, then, that while some hugely impressive Xeon CPUs are available, most users in a music-studio environment would get superior performance for less money with one of the high-end single-CPU setups. And if that’s really not enough for you, you’d probably be better off adding another machine and using software like Vienna Ensemble Pro to share the processing load across the two computers.

Until very recently, the Xeon platform offered one significant advantage over the X99/Core i7 systems: it could support a far larger amount of memory than any other chipset. Windows 8 Pro 64-bit can handle up to 512GB of RAM in theory, but the limit is lower in practice simply due to the motherboard’s ability to host that much RAM in the formats currently available. There are Xeon motherboards that can handle up to 192GB of RAM, whereas even the best ‘enthusiast’ motherboards could at the time of testing only accommodate a maximum of 64GB. Just as we were going to press, though, Corsair announced the availability of a (admittedly very expensive) 128GB RAM package. The only users really likely to require this much RAM are those who are scoring for film using the huge libraries made by firms like Vienna Symphonic Library, where some users’ templates can break the 64GB RAM limit of most consumer-level hardware. The Xeon solutions can, for those who require the very best performance and are willing to pay, offer a wonderful solution, but it looks very much like the Core i7 systems will soon be capable of similar performance.

Decisions, Decisions...

So, after all this, what processor should you have in your next studio PC? If your current machine was a mid-range model three or four years ago or a high-end one five years ago, or if you’re looking to move up from a mobile setup to a dedicated desktop machine, then most of the current desktop offerings from Intel will offer a very noticeable improvement. Looking at the upper mid-range Intel solutions will, without a doubt, give you a solid increase in performance. Anyone working below the six- or eight-core Intel ‘enthusiast’ models will currently also see a great performance increase by moving up to a system based around these chips. On the other hand, owners of the previous-generation six-core i7s or older Xeon systems may find it hard to see value in an upgrade below the current eight-core i7s — or the high-end Xeon products, if memory expansion is right up at the top of your list of requirements.

Intel & AMD

For many years Intel and AMD were locked in a CPU arms race: newer, more powerful designs were brought to market rapidly, and every five or six years a complete architecture redesign would propel one company ahead. Recent years have seen a less aggressive release schedule, but there’s still a constant stream of improvements and refinements that can benefit studio users.

The Need For SpeedThe Need For SpeedIntel’s last generational shift brought the first fully redesigned and optimised i-series CPUs (Core i5, i7 and so on). This family of chips has been with us since the turn of the decade, but while the headline figures show only modest improvements, the chips are becoming ever more efficient and thus offer better performance. Intel’s so-called ‘Tick-Tock’ release and update schedule benefits studio users. The ‘Tick’ stage refers to the CPU range receiving a ‘die shrink’, which allows for increases in performance and energy efficiency through design refinement. The ‘Tock’ brings a new micro-architecture, with the introduction of new features and functions. The full cycle completes once every 12-18 months and means that we see significant performance gains only every few years, but lower power-draw editions in between, which give us quieter cooling.

AMD’s base design predates Intel’s by a few years, and they don’t currently offer significant competition in the high-performance market. However, they’ve carved out niches with their well-established hardware and continue to hold market share, mostly through reasonably performing chips that compete strongly on price. Much of their recent sales growth is due to their provision of chips for games consoles, but they’ve sensibly been using these contracts to fund development of their next generation of hardware, and making a lot of noise about the planned release of new platforms and chipsets in 2016. Can they re-establish themselves as a provider of high-performance chips? It promises to be a very interesting year!

CPUBenchmark

CPUBenchmark is a web site that hosts a database of CPU test results. Results are updated daily, which means that even without specific DAW benchmarking it should still be possible for you to get some idea of the degree of performance improvement a proposed upgrade could deliver — even for CPUs not featured in my own tests.

www.cpubenchmark.net

Published July 2015