What benefits have the latest crop of CPUs brought for audio users — and what might the coming year have in store for us?
For those who keep a keen eye on the computer hardware market, it’s been an interesting few years when it comes to CPU development. Intel, the leaders for many years when it came to high‑performance applications such as audio and video production, were dethroned a few CPU generations ago. Indeed, for a while they appeared almost to be heading out into the wilderness due to delays in their platform updates, but the last time we took a look at how the leading Intel and AMD chips compared (a shade under a year ago, in SOS April 2022), Intel were showing strong signs of a recovery, and we found that they and AMD performed comparably at that time.
Intel’s step forward at that time was largely as a result of moving over to a so‑called ‘big.little’ core arrangement, with the CPU as a whole using both high‑ and lower‑performance cores in an arrangement similar to that in Apple’s current hardware. A timely switch to Windows 11 helped to support this new way of working and, now various DAW and plug‑in creators have had the chance to do a little tweaking and refinement, it seems largely to have proved successful. This, along with support for the newer DDR5 RAM, meant we saw large intergenerational gains in Intel’s testing results. Still, as I noted at the time, all of this was just enough to bring them back to level pegging with the well‑established high performance of the AMD platform.
Hot Gossip
Fast‑forward to today, and the situation has reversed: it’s Intel who are now looking to optimise an established platform, whereas AMD have brought out their new AM5 chipset which, along with any gain in raw performance, incorporates support for DDR5 RAM. While Intel’s last generation Z690 chipset continues to support their newly released 13th‑generation CPUs, the company have also brought out their new Z790 chipset, which includes PCIe 5.0 support. So too does the new AMD platform. All of which means that both companies’ latest ranges boast a similar range of features.
Although dedicated PCIe 5.0 devices remain thin on the ground — the first next‑generation storage options aren’t due to arrive until later on this year — the upgrade to PCIe 5.0 doubles the available bandwidth, which means there are other potential benefits. A number of mainboards have already appeared on both platforms which can take advantage of these extra resources, and offer up to five current‑generation PCIe 4.0 M.2 NVMe drives, along with the standard SATA drive connections. That will be a boon to anyone who’s looking to fit a large amount of fast storage for their ever‑growing collection of audio projects and sample libraries; those doing serious virtual orchestral work, for example.
Given the similar performance of their high‑end chips a year or so ago, it’s perhaps not surprising to discover that both companies have chosen to focus most strongly on headline improvements in the raw hardware performance. The performance gains are most notable with the companies’ flagship Intel 13900K and AMD 7950X models, but a similar trend can be seen across both ranges. Of course, these attempts to snatch this generation’s number one slot mean that power consumption has risen, and while that’s not a huge concern for many users, there’s a significant issue for audio users: while they might well be happy about the performance boost, higher power consumption invariably means more heat and thus potentially more system noise due to cooling.
Intel’s Z790 boards play host to CPUs which are already pushing the envelope, with a 253W TDP (Thermal Design Power — the maximum amount of heat generated) rating on the Core i9 13900K for its Turbo mode at stock ratings. This has been amplified by a number of new mainboards which, by default, already apply a small degree of performance boosting — which results in the Intel‑advised power limits being largely ignored! During our tests, we found that various boards apply chips’ PL2 voltage settings (Performance Level 2, intended for turbo performance) to regular PL1 workloads, whilst also unlocking the PL2 mode’s power draw completely, to allow the CPU to use as high a voltage as it wants when under heavier turbo‑boosted loads. This resulted in peak loads occasionally spiking in excess of 325W total draw, a level which would prove challenging for even the very best cooling options available today.
When the power consumption was unlocked in this way, the ‘real world’ performance gains that were achieved were surprisingly modest: typically it resulted in only an extra 100MHz being applied to each core. In terms of practical, everyday use, that seems a poor trade‑off. Making a few BIOS adjustments to set things in line with Intel’s specification vastly reduced the power spikes (to a much more manageable level) and brought the average temperatures down to a point that made quieter cooling far easier to achieve.
For the previous generation, AMD’s Ryzen platform was already the more power‑efficient solution, and although we’ve noted an increase in power usage this time around, the upper specified limit of 170W on the 7950X model ensures that it remains significantly less power‑demanding than the Intel equivalent. AMD have, to some degree, changed how performance is managed: typically, with Intel and previous AMD generations, you set your CPU target speed in the BIOS and the board would attempt to use whatever power it took to achieve it, unless you specified otherwise, even to the point of overheating. AMD have now switched to an approach whereby you set a maximum target temperature under full load, and it works to maintain a steady degree of performance depending on how well the system is cooled. Our tests suggest that this works very well.
Again, we’ve found that the AMD board settings push slightly beyond the official limits, with the CPU peaking above 200W when the overhead was available, but this is where AMD’s new Eco mode feature comes into its own. With a number of different low‑power settings, all the way down to 65W, making a change to the setting within the BIOS allows you to limit the chip’s overall power draw. With a workstation, you hardly want to be limiting your performance quite so dramatically, but the upper 170W Eco mode setting is of interest, as it quickly brings the CPU back into AMD’s official spec. In stress testing of the 7950X, we observed that, whilst unlocked, it would swing between roughly 5200 and 5300 MHz across all cores when under a 100% load. With the 170W Eco mode applied, this stabilised in the middle, with a more constant 5250MHz, allowing the system to run slightly cooler overall. The benefit extended to running a few extra plug‑ins in our DSP test (of which more later), because the removal of the larger dips in processor clock speed helped to achieve slightly greater performance overall.
With these observations in mind, we tested both companies’ current flagship chips, in setups that adhered strictly to their official specs, and with the same cooling system used for each: a Be Quiet! Silent Loop 2 280mm AIO (all‑in‑one) water cooler. Interestingly, the performance of the AMD Ryzen 9 7950X on the 170W TDP setting remained similar when substituting the Silent Loop system for our regular Dark Rock Pro 4 fan‑based cooler; switching to a triple‑fan water AIO model may help push the CPU performance slightly further, but at the cost of noticeably increased fan noise. The Intel 13900K, though, with its higher TDP figure of around 253W under maximum load, would still very briefly peak at a level a little beyond the rating of even the best of air coolers, as the turbo boosted and then settled again. So for this CPU at least, a water cooler AIO is definitely recommended.
Moving away from the flagship chips, the difference was striking. Even though AMD’s 7900X and Intel’s 13700K models have similar ‘on‑paper’ power usage ratings to those of their larger siblings, these less densely packed chips, with lower core counts and reduced maximum turbo ratings, typically proved much easier to cool — a good air cooler is an option for either chip (and those below them in either range).
Given the similar performance of their high‑end chips a year or so ago, it’s perhaps not surprising to discover that both companies have chosen to focus most strongly on headline improvements in the raw hardware performance.
On The RAM Page?
System memory is often of just as significant interest to music makers as the CPU itself — sample‑library players such as NI Kontakt, which is used in our DAWBench VI test, can benefit from more and faster RAM — and it’s another technology which has seen a significant jump in performance in recent years. If you read the previous article on this subject (the one linked to in the opening paragraph) you might recall that the 12th‑generation Intel chips were compatible with a mixture of mainboards, some supporting DDR4 RAM and some DDR5, and that in testing it became evident that switching to DDR5 could prove beneficial for anyone working with a lot of audio and library data. Yet this was also a source of frustration, since the initial supplies of DDR5 sold very fast, and the supply chain problems at the time meant many early adopters were left waiting a few months before they could implement this new technology. Thankfully, both the supply and cost slowly improved over the course of 2022, and now that AMD’s AM5‑chipset‑based boards support DDR5, the year ahead should see its more widespread availability and adoption.
The RAM for our latest tests was DDR5, and we ran comparable dual 32GB sticks for 64GB at a 5600MHz rated speed on each platform, with Intel XMP or AMD EXPO profiles as appropriate. Over the past few years, we’ve found that one way to give an easy bump to performance is to ‘overclock’ the RAM, simply by using kits that are faster than the chipset’s base recommendation. It yields a small but worthwhile boost to performance without any major increase to the voltage or heat levels associated with CPU overclocking. This time around, there are some additional things to note from the manufacturer’s recommendations which may affect anyone looking to maximise their available system memory.
For this latest generation, Intel advise RAM speeds of up to 5600MHz, while AMD lists support at 5200MHz. Note that, in both cases, these figures are for a pair of memory sticks running together in a dual‑channel arrangement. Since DDR5’s launch, ever‑faster RAM kits have become available, with some of those starting to appear on mainboard QVL (Qualified Vendors List) sheets. At the time of writing, the very best is capable of running all the way up to 7000MHz, but that’s only for dual‑RAM‑stick arrangements. For a full complement of four sticks, which is required if you wish to install the maximum allowed quantity of 128GB, the situation becomes a little murkier.
Once fully populated, the Intel chipset will tend to limit itself down to the platform’s actual base RAM clock speed of 4800MHz. A search through mainboard QVL charts for pre‑validated solutions suggests that this means there’s currently not yet a ‘real world’ 128GB configuration available (at the time of writing — this situation could change quickly!). AMD has a similar lack of pre‑validated options, specifying a supported limit of only 3600MHz when configuring a similar arrangement. That said, those who are willing to experiment and stray beyond the official specification will potentially find a little more wiggle room with the right hardware choices.
A quirk of the AMD architecture resulted in previous generations performing more optimally when RAM overclocking as you reached 1:1 parity with the speed of the memory controller. For this generation, 6000MHz is that target figure. Larger 32GB‑based kits that could run at 6000MHz were unavailable at the time of writing, but initial tests with 4x16GB‑based kits have gone well — this suggests that when 32GB sticks (these are required to install the full 128GB) become available, they could be well worth exploring.
Top Spot?
Returning to the present, let’s consider the latest set of test results. If you look over the charts for the ReaXcomp compressor‑based test, which puts the spotlight on digital signal processing (DSP) performance, you’ll notice straight away that competition for first place is fierce. As you might expect, the Core i9 13900K and Ryzen 9 7950X are firmly ahead of the pack, but there’s a tighter spread for the range of Intel models across the various buffer sizes, while the AMDs tend to come into their own at higher buffer settings. Indeed, the Ryzen 9 7950X is ahead overall on the uppermost 512 buffer setting.
When it comes to running the most effects and processing plug‑ins, Intel appear to have reclaimed the top spot — for now!
For the Kontakt‑based virtual instrument test, which asks more of RAM performance, Intel offered the strongest performer last time around and that’s repeated here. Not only does the Core i9 13900K put in a very strong showing, but the next‑model‑down Core i7 13700K holds its own against AMD’s top‑of‑the‑range R9 7950X. We saw in the comparison test last time that this type of workload would take advantage of DDR5 memory where it was available, but while this means the new AMDs certainly outperform their previous‑generation CPUs it seems they can’t yet quite keep pace with Intel here.
Intel also seem to be out in front again when it comes to sample libraries and virtual instruments — though signs point to some interesting developments by AMD in the coming months.
This pattern is repeated across both charts: Intel tend to appear stronger when placed up against AMD’s competing model. And while AMD’s CPUs used to be less expensive than the corresponding Intel models, that’s no longer the case — so evaluating the best bang for buck for your particular requirements should be pretty straightforward. Of course, the power draw could also be a consideration. Intel really have gone all out to secure the top result, and that comes at the cost of greater power consumption, although, having said that, both platforms do scale the power draw according to the required usage, and those headline upper limits are likely to be fairly uncommon in daily use. Still, with less heat being generated overall, there may still be benefit to AMD’s lower power consumption if configuring a particularly quiet system is a priority.
With the introduction of their new platform, AMD have lost perhaps the biggest competitive advantage they enjoyed last time around. Intel had suffered not just from stock availability but also the cost associated with the early adoption of new technologies like DDR5. Since both companies’ ranges performed comparably last time, it made sense to a lot of people to pick up a well‑established, lower‑priced AMD AM4‑based board along with the more freely available and cheaper DDR4 RAM. AMD’s move to DDR5 has levelled the cost and availability playing fields — the first range of AM5 boards cost broadly the same as their Intel equivalents. AMD launched their most recent chips a month before Intel’s went on sale, and the latter were able to price their range accordingly; priced similarly to the models they were replacing, Intel’s new CPUs are more keenly priced than most people had expected. It’s a move that has already caused AMD to revise their prices downward. (See? I told you an ‘arms race’ was on the cards!)
One result that sticks out like a sore thumb is the great low‑latency performance of the AMD Ryzen 5800X3D: this CPU isn’t even from AMD’s latest generation of chips.
The Need For Speed
One result from this latest round of tests that sticks out like a sore thumb is the great low‑latency performance of the AMD Ryzen 5800X3D: this CPU isn’t even from AMD’s latest generation of chips! it arrived towards the very end of the 5000 series’ reign, and it could well be a sign of what’s yet to come from AMD. Readers of past CPU round‑ups may remember me remarking that the low‑latency ASIO handling of AMD’s earlier ranges tended to be weaker than Intel’s (an issue for some music makers, though not all). AMD’s new 3D cache design, which is used in the 5800X3D, helps to reduce the CPUs’ internal latency, which is what lay at the heart of this issue. But while low‑latency performance has improved greatly in recent generations, the AMD chips do still lag slightly behind the equivalent Intels when using the lowest (64‑sample) buffer settings. The lower‑clocked 5800X3D ranks, as expected, slightly below the regular 5800X, but its cache comes into play where AMD has often proven at its weakest — and, as a result, we see an absolutely storming virtual instrument test result that, with the lowest buffer settings, effortlessly exceeds the performance of what even that range’s top‑end 5950X model can deliver.
At the start of January 2023, in a bid to follow up on this ‘experiment’, AMD confirmed that a number of 7000‑series 3D cache‑enabled models would become available later in Q1 2023. This includes the 7950X3D, a revision of the top‑end chip. This suggests that, even though it’s only a few months since the launch of the current range, AMD are already preparing their latest challenger.
Intel’s next development isn’t too far off, either. They’re looking to introduce their long‑awaited successor to the X299 high‑end chipset in the first half of 2023, in the form of the Fishhawk Falls chipset platform, as well as a workstation‑grade Sapphire Falls range. Indications are that these will remove the ‘efficiency’ cores found in the established 13th‑generation chips, and employ a more traditional array of full‑performance cores across the whole chip. That, along with support for up to 256GB of DDR5 memory, has the potential to make this the ideal platform for anyone working on the largest of projects, such as scoring with large libraries to film or other mixed media.
Takeaways
It’s clear that Intel have gone all out to deliver the highest performance. It’s been a bold move, and one that looks, for now, to have paid off for them. Their willingness to adopt an aggressive pricing strategy demonstrates their keenness to regain market share, after the past few years’ dip. AMD’s reluctance to bring us their 3D‑cache‑enabled 7000 series at the launch of their new range may have left them on the back foot for now, but with a mid‑cycle refresh already imminent, these could well shake up the charts once more. This is already shaping up to be another exciting year in tech!