With the current DAWBench testing suite having been in service for a few years now, there is an ongoing discussion about how to keep expanding the benchmarking in order to support covering more hardware. With Vin being hard at work putting together new builds, we’ve been running an initial round of benchmarking here in Scan with a number of beta builds having passed over the testing bench in recent months. Whilst the original goal was to offer a new standardised test setup that could be applied natively across more platforms like the MacOS testing here, there has also been a move to using more readily available libraries as well.
So, this time around rather than look at new hardware specifically, we introduce these new tests and talk a bit about the changes involved, along with also taking a fresh look at the PC kit that’s already out there this generation. There’s been BIOS and driver updates in the interim between the original testing and this batch as well, so taking a fresh look helps to expand further on previous results and in some cases offers up a new perspective in how the chips line-up. I also have access to a 10 core M4 Mac Mini for the first time which is the first step to hopefully placing us in good stead for future testing across more platforms than we’ve been able to examine previously.
This round of testing is once more carried out with our regular Babyface Pro FS and Reaper setup, and consists of the 3 updated revisions to the standard DAWBench tests used previously.
DAWBench DSP
The new DAWBench DSP build adds the inclusion of Analogue obsession TUBA, replacing the stand-alone version of Reaper RXC compressor. The test spreads itself across all of the available cores and remains focused on establishing the overall overhead available when focusing on raw CPU performance as the key metric.
On the ASIO buffer 64 setting, the top two positions remain the same with prior 14900K third place not covered in this round of testing. The 9950X then switches place with the 265K, and we also see a stronger showing from the 9900X and 9900X3D after the 265K entry. There were a number of BIOS updates in Q1 which were reported to have skewed Intel performance to some degree and of course both parties have continued to optimize since then, so some movement in the results is to be expected as we see a positive upswing in these results for AMD overall in relation to the Intel hardware in this test.
The Apple M4 10 core with it’s 4 performance / 6 efficiency core layout landing just ahead of AMD’s 8 core / 16 thread SMT 9700X chip. With a mixture of physical cores clocking at different speeds, whilst mixed in performance they do impressively pull slightly ahead of AMD’s 8 physical core layout design. The Mac mini can scale in power draw up to around 65W, whereas the AMD chip is rated to 65w’s and scales up to around 90 – 95w’s when dealing with Turbo loads, so whilst the two chips trade blows in terms of performance in this test, the Apple hardware does so whilst maintaining its lower power profile advantage.
Stepping up from the baseline 64 buffer results and everything continues to scale smoothly, with no real surprises in terms of the performance spread as we move up through this chart.
DAWBench VI
This time around the test receives an update to the Kontakt 8 Player edition found in the Native Instruments “Komplete Start” package and then populated with a new library from Sonuscore, namely the excellent free “The Orchestra Elements” edition. The previous benchmark revision during the initial round of testing seemingly returned poor results on the Mac hardware with Kontakt 6, so it seemed worthwhile double checking this with an updated test and a much more modern library. Given Kontakt 6 itself also pre-dated the release of Intel’s 13th generation hardware, its older engine was written before the inclusion of hybrid style P/E chips from either Intel or even Apples M series. Ultimately, this didn’t look to change how the performance turned out on the Mac with results scaling in a similar fashion on both versions of the test. What turned out to be more interesting in the end was that when we take a look at the results, it appears that using the older build of Kontakt was leaving some performance on the table not only when comparing Intel and AMD, but even amongst AMD’s own chips.
The biggest surprise here is not so much Intel’s tighter spread up against AMD’s offerings but the elevation of the 9800X3D chip to the top of the table at the 64 buffer ahead of the other models within the range. The test tends to respond to a mixture of CPU and memory performance and with the 9800X3D having the fastest monolithic CCD block design currently from AMD, it’s coming out ahead of both the 9900X3D and 9950X3D which have a dual CCD design at the lowest buffer setting. This does appear to be some highlighting some internal inter-CCD latency between the CPU core blocks on two dual CCD arrangement chips, along with showing up the benefit the larger X3D cache addition brings over the regular variants on the 64 buffer.
The new VI test is more memory intensive than the prior revision and where we could max out the older test build in under 64GB for the full 12800 polyphony count, instead I found that this time around we’d pass the 64GB level much earlier. This new larger sound library is in no doubt more in-line with the generally larger sizable modern soundbanks found today than the older Kontakt 3 collection used previously, the inclusion of which shows that running a heavier testing load overall does appear to affect the chart layout to some degree.
Once we step up from the 64 buffer the performance on those chips switch places and everything starts to scale more as you would expect. The switch-up in plugin revisions and sound library here is noteworthy as it helps to highlight the differences in performance between modern plugin versions when compared with even builds from just a few years ago. It also goes some way to to outlining the heavy lifting being done by the X3D cache when working with larger memory intensive workloads and in particular running under the tighter buffer settings.
Above the 64 buffer setting we see AMD’s cache advantage fall away as Intel quickly pulls back into the lead, then continues to climb further ahead as we raise the ASIO buffer setting. There’s potentially a live use scenario that favours the AMD X3D chips here running as low a buffer as possible or one with live recording directly over a more limited number of library channels, but for users scoring with larger libraries and predominately inside the box Intel’s performance handling here comes out as the favourable option assuming the usage of larger buffer settings. Whilst the M4 Mini is not particularly showing a strong result in this test as was the case in the initial round of K6 testing, it should be remembered that it is limited to just 4 Performance cores and it is one of the lower models in the range.
DAWBench Bus VI
This brings us to the last of our charts and the reworked Bus test, now using the internal version of ReaXcomp so it’ll run under the native Reaper build on each platform. We looked at a few different plugins for the job here and this seemed to offer us the clearest example of scaling, although results were reflected across each test build we tried out.
The test here focuses on effectiveness when stacking effects within a single BUS chain, which should be an example of how far you can push a given single core and if the OS is managing the CPU load correctly it should highlight the strength of the “Performance” cores when dealing with “Hybrid” CPU’s as we find with the Intel and Apple silicon. In contrast the AMD chips are running a layout with cores of all similar performance, although the upper limit they can hit is likely to scale slightly differently depending on how many cores are being pushed by the turbo settings at any given time. The AMD monolithic none X3D chips came out strongly here whereas the X3D chips tend to turbo a little less due to heat from the cache causing them to limit a touch quicker, with Intel’s 285K winning out on the strength of its performance cores. However the overall winner in this test with a particularly strong showing coming from the M4 silicon with an outstanding performance, showing just what can be achieved with the silicon design on this platform when pushing those performance cores.
Overall, a welcome overhaul to the testing and hopefully this puts the DAWBench project into a good place in terms of future cross platform benchmarking. With the test charts updated and this initial batch test over, it should leave us with a solid platform to build upon as we head into the annual CPU release season over the coming months.
Scan 3XS Custom Audio Systems
Scan Fixed Series Audio Systems