Computer Music - Hardware,Test Labs Ryzen 7000 Series X3D Performance Testing.

Ryzen 7000 Series X3D Performance Testing.

Ryzen 7000 Series X3D in the studios

As we move through the early months of 2023, we see here today the first in a number of new CPU releases due to arrive with us this year. Over recent hardware generations, the competition in chip performance has been fierce and as it continues to escalate, we've also seen a divergence in design choices where both firms continue to aim at maximizing the performance of their respective platforms.

Intel like Apple before them have already made its move towards hybrid processing with its "big.little" core design, which features a mixture of high and low-performance cores, allowing them to pack in more threads and deal with varying workloads more efficiently. AMD so far has eschewed this design choice sticking with a more traditional arrangement of purely high-performance cores, but instead chosen to explore other avenues for getting the most with this way of working.

With the previous range of AMD 5000 series of CPUs, a late release into the market was the 5800X3D model, which proved to be an interesting chip that showed the potential of a new way of working. It had a slightly lower clock speed than the none "3D" edition of the 5800X, which meant that on raw processing performance alone it trailed behind. Instead, where the extra cache support of the design came into play was improved memory handling, which translated in our DAWBench VI testing as a boost to the low latency polyphony result when focusing on Kontakt, seeing leaps in performance that easily put it ahead of most other chips of its generation for sample-based workloads. This included the range's own 5950X flagship when working at the lowest ASIO buffer rates in our testing, something that always tended to be a bit of a weak spot for the Ryzen range previously. It had less effect at larger buffer sizes, with the other chips in the range once again overtaking it once we passed the 128 buffer, but it was certainly an interesting showcase of what could be achieved with the cache implementation on what was a fairly middle-of-the-range chip.

Of course, it left us wondering what could be achieved if this design change was rolled out across the rest of the Ryzen range, and when the 7000 series launch came and went without any 3D cache-based models amongst them, it almost felt like a missed opportunity.

Now only a few months later, we see this quickly rectified as both the 7950X3D and 7900X3D arrive with us. Announcements ahead of the launch proved interesting as they advised there were to be two different executions in design with this generation. The lower-end chips including the 7800X when it arrives with us next month will feature a single CCD core arrangement where all the cores share the same resources directly. Much like the 5800X3D before it, this layout ensures that the minimum of latency is easily achieved.

Alternatively, the upper models we're looking at today have an interlinked dual CCD design in order to support the total number of cores these chips have to offer. Normally we would expect the available resources to each block of cores to be replicated so that they can both access the required resources locally and make a minimal amount of calls to the other CCD block, a process which may add internal latency. Here we find instead that the implementation on these dual CCD chips has only 1 of the blocks that receive a direct connection to the 96MB of 3D cache, whereas the other CPU block runs with 32MB of cache but trades this off with cores that offer stronger performance under load.

With the older 5800X, we found a CPU that proved to be a super quick and great-value gaming chip although with mixed results in productivity applications. It almost seems strange that they didn't follow up such a popular and successful chip with its direct replacement the 7800X3D as the first model out of the gate this time. Even more so when in the CPU release info AMD's own configuration examples for the 7000 generation and in particular with the 7950X3D lean heavily on examples of game performance and the fact most games only support 8 or fewer cores. The idea is that the chips can be set up to favour either cache-accelerated loads like gaming or the faster cores block used for applications that favour raw core performance, although with examples that use a more limited number of cores. For audio application use where we want similar performance across all of the cores ideally, this does raise some questions about increased inter-CCD latency on these larger chips.

Intel's hybrid platform had some early teething issues with DAW support as the software engines didn't know quite what to do with the mixed core arrangements and a large part of Windows 11 release and subsequent updates were focused around improving support for this new style of hardware. AMD runs the risk of hitting the same obstacles here and a key point is going to be just how well the system and applications look to manage this alternate method of working.

Taking a look through the new BIOS options and you can choose to set the system up with a number of options, either choosing to force it to favour "Frequency" or "Cache" as the main cores in use. A third option switches it to full driver-based control where it will look to prioritize the system driver as the component making the choice on the best processing route at any given time, something supported in the latest AMD chipset package. AMD issued a guide to reviewers early on that outlined the best way to configure the chips and update the software, which whilst we observed it within our own testing, it was notable that the slant given to the optimization was very much in the direction of a gaming setup or frequency preferring applications that used a more limited number of cores to their fullest. Within our own testing here, I took multiple runs through the testing with both frequency and cache set as optimal, before another run through the "driver" software-based option.

For purely an audio-centric setup where the DAW software will be spreading the load, the testing showed up frequency over cache as the best default setting, with the DSP testing returning up to 20% more performance and the VI test a more modest but not insubstantial 10%. The driver-based load balancing came in a close third place here, which as it's trying to balance between both modes, it suggests a round of further optimization may be called for. Harking back to the Intel switch to hybrid handling before this there was a round of optimizations through Windows updates applied to bring it up to speed and undoubtedly over the coming weeks and months, I would expect this set of drivers to be fine-tuned in much the same fashion.

For an audio-dedicated setup currently and for our round of testing here, I would choose to run with the Frequency option as default, although for mixed workload systems, this may change. For anyone looking to build their own setup, it could prove worthwhile to do a little research on the latest developments when building and do a little testing of your own to establish the best setup for your type of workload.

The testing setup in use is the same AMD arrangement used for our prior round of testing (found here), with the latest BIOS revision update added in order to unlock the 3D cache configuration and features. The bench is based around an Asus X670E Gaming Tuf board, with 64GB of 5600MHz Corsair RAM and the be quiet! 280mm Silent Loop2 cooler.

DAWBench DSP Testing Results For AMD 7000X3D Range
DAWBench DSP Testing Results For AMD 7000X3D Range
DAWBench VI Testing Results For AMD 7000X3D Range
DAWBench VI Testing Results For AMD 7000X3D Range

Looking at the DSP results first and much like the 5800X3D last time around, we see the lower clock speed on offer places the 3D edition chips right behind their respective standard editions. On raw processor performance which this test is focused on, this was the expected outcome and the lowered maximum thermal ratings of these chips is also likely to play into this to some degree when being pushed under full load.

Pulling no surprises and much like the earlier 5800X3D testing, the VI results show where the extra cache support truly comes into its own. On the lowest buffer settings, the polyphony count on the new X3D chips takes the top two spots with convincing leads. However, much as before this benefit is lessened as we step up through the buffer sizes and whilst the 7950X3D remains comfortable at the top of the AMD options, the 13900K does take the lead again at 256 and above.

These chips have a lot to like about them, with high performance and reasonable power requirements. One of the shortcomings I felt previously with Ryzen as a platform was the low latency support and this cache change not only reverses that trend but truly blows it out of the water. If they had arrived with us at the launch of the 7000 series last year and been competitively priced then Intel's 13th generation launch may have turned out very differently.

As it is we have an already established 13900K chip holding onto the DSP crown across most of the results, only being pipped once we reach the 512 buffer setting. A situation that reverses in the VI testing that shows the AMD chips performing strongly at the lower buffer settings before being outpaced by the 13900K as we move above the 256 buffer setting.  

With the 13900K being priced in line with the existing 7950X standard edition, we see the 7950X3D launching with around a 30% premium, making the inclusion of 3D cache an expensive feature at launch and perhaps explaining the choice to concentrate on only expanding the cache on one CCD block. We did however see AMD revise pricing across their range shortly after the Intel launch last year and given the current strong state of market competition, there may be some changes in how the value lays as the market settles over the coming months and prices drift.

At the right price point, I can certainly see these making excellent recording system chips, where the performance at low latency and lower power use would both help to ensure a super quick and quiet setup. Intel's offerings still look to hold the crown on ultimate all-out performance in most scenarios right now, but at the cost of higher power usage. Whilst some users may be looking for simply the most powerful option available, there's always going to be an equal number of users who want perhaps not the most powerful setup but the best recording all-rounder to run within their studio environment.

Ultimately there is a wealth of performance here on both platforms and I dare say that making the choice over the best setup has never been more down to the situation and studio setup you find yourself in. 

If you would like to discuss your next studio system and spec out your perfect setup, you can contact our sales team with any questions and help configure your ideal solution.

Our online custom and fixed ranges can also be found through the following links.
Scan 3XS Custom Audio Range 
Scan 3XS Fixed Series Audio Range