The final few months of the year tends to bring a refresh or two and this year we see new CPU’s arriving from both Intel and AMD. For Intel it’s the new Ultra range rebrand of their i-Series chips and for AMD it’s the initial offerings in their new 9000 series based X3D chips.
With Intel we find a company who having taken their previous platform as far as it could go, or it could even be argued further than it should have gone. Struck by what seems to have been silicon that was already being pushed hard, only to then see issues pile up further due to over-zealous board manufacturer profiles delivering overclocked voltages by default in order to extract every last drop of performance. The result has been a number of updates over the course of this year by Intel ensuring that the last few generations get reigned back in by applying stability fixes for developing issues.
Looking to reinvigorate their range the new 15th Gen or Ultra series as they are now known, Intel has not only gone through a die shrink but for the first time also moved over to a chiplet design (known as “tiles” in Intel parlance), not unlike AMD’s now very well established Zen series of CPUs. Whilst notably it took AMD a few generations to refine the sort of performance out of the range that we now take for granted today, it didn’t get there without a few bumps. The Intel Arrow Lake architecture found here has a mixture of tiles with the CPU cores themselves built around the latest 3nm process, whilst other components like the SoC tile (containing memory controllers and general PCIe I/O) on a older 6nm or the iGPU built around a 5nm process.
The Intel Ultra coverage itself has been delayed by a month whilst waiting for stock to arrive on the flagship 285K model, although it has allowed us to test with various pre and post launch BIOS builds, with notable performance gains appearing in the BIOS builds that arrived on or shortly after the the day of launch. With this being Intel’s first mixed tile design it has already hit a few notable pitfalls, with reviews commenting on the internal memory latency having risen quite dramatically between generations, something that also hampered AMD’s Zen platform for the first couple of generations after it’s transition. At best perhaps 50% higher when compared against the previous range, but at worst reaching 3 to 4 times higher than the prior generation in some reported scenarios. This has already been acknowledged by Intel, who are working on further optimizations which we may see re-tested in future updates over the coming months.
A big change of interest to us here with the new Intel CPU’s is the removal of Hyperthreading. A long standing feature designed to maximise a CPU cores performance by using unused clock cycles to run a second threaded load. With the move over to mixed hybrid cores and the optimizations done within the Windows OS in recent years, the efficiency of these split loads has began to become more questionable. With software looking to spread the load like the ASIO engine is designed to do, Hyperthreading has sometimes been controversial and even at times counterproductive in some DAW setups over the years and seeing how the performance is affected with it removed shall be interesting.
AMD in contrast recently released their own 9000 series chips and whilst they also experienced a number issues related to performance around launch, they do looked to have stabilized performance on the platform through a round of BIOS and OS updates in the weeks and months following the launch of the range. This time around we see their initial refresh of the X3D series, which historically haave traded off a bit of raw performance for a generous increase in available L3 cache.
AMD’s chiplet design sees the 9800X3D model and others below with a single CCD block layout whereas the models above feature a dual CCD block design. AMD’s focus has so far with the upper dual CCD chips been on mounting the cache on one slightly lower clocked CCD, then offering a regular faster clocked CCD block as the second half of the equation. Whilst a successful strategy for setups wanting to mix gaming and more clock sensitive workloads, for a purely workstation style that wants optimal uniform performance across all of the cores it can be less than ideal. AMD looks to have acknowledged the markets enthusiasm for the X3D gaming performance and chosen to run with the single block 9800X3D as the first available offering, with the gaming press once more having already been very impressed by the price & performance trade off.
Taking a break from the water AIO requirement of the pervious Intel generations, the hardware this time around is tested using a be quiet! Dark Rock Elite cooler. The most power hungry chip here in our round up is the 285K, which tops out around 250w’s and I tended to see it settle around 240W’s sustained under longer periods of stress testing. With the cooler itself being rated to around 280W’s TDP even after a few hours of maximum load it sat around 80 – 85 degrees, an acceptable figure under such an extreme load and with it being the sort of workload that far exceeds what you would expect to see occur within day to day usage.
First up we have the DAWBench DSP chart and with the cache failing to impact in this test focused on raw CPU performance, the 9800X3D lands squarely between the 9700X and Intel’s lowest CPU on test the Ultra i5 245K. The upper two Intel models both score strongly here as well, with the middle of the pack 265K sitting just ahead of last generations class topping 14900K, doing so whilst maintaining a maximum load of around 190W’s as opposed to the prior generations 260W+ load. This leaves the upper 285K chip with around 15% – 20% gains over the second placed 265K and Intel holding the top two places in this test overall.
In contrast the 9800X3D snatches the best performing chip on the tightest ASIO 64 buffer in testing. We’ve seen this previously with X3D chips when dealing with a test that focuses more on the CPU and memory handling working together, the L3 Cache helps accelerate the performance in the more stressful scenarios. This advantage soon drops away however as you step up through the buffer settings, with the final 512 buffer setting seeing the chip settling in similar position in relation to the AMD 9700X as we saw in the previous DSP test.
One notable trend here is that whilst the newer Intel chips lead their predecessors as you would expect on the 64 and 128 buffer ratings, the previous generation turns this on it’s head by overtaking the new models on buffer settings above 128. With changes to silicon and the earlier notes about advised memory latency, we could be seeing silicon efficiencies that give the new kit the advantage under tighter processing loads, but once that ceases to be the bottleneck the older lower latency memory design on the 14th gen shows it’s merit. If Intel get this under control over the coming month’s by tuning the CPU’s inter-tile interaction, there may be some scope for improved results in the VI test once any updates are rolled out.
The memory in use here is 6400MHz DDR5 which is the advised RAM speed on this Intel generation and a step up from the 5600MHz previously in use. With suggestions that the Ultra 200 series may prove optimal around the 8000MHz mark I did take a look at using faster kits as well. 8000MHz kit options were still thin on the ground at the time of testing, but I did try out a more readily available 7000MHz pack in it’s place. The result was a marginal 1.5% improvement on the VI results and no notable change to either the DSP or BUS result. This suggests the platform recommendation seems to be fairly well placed at the moment and that the memory itself isn’t really the bottleneck in this instance, although once again should they be able to make any improvement to the internal latency in future updates, this may also change.
The DAWBench BUS grouping test sees a small improvement the 14 series, but only draws the Ultra series level with AMD’s prior 7000 series. In comparison the 9800X3D fares slightly better and lands in-between the new Intel Ultras and AMD’s own recently released 9000X series sitting at the top of the chart.
Overall the results here in terms of the AMD 9800X3D are rather unsurprising, sitting just above the 9700X in the CPU centric benchmark and otherwise seeing a solid boost when dealing with memory hungry plugins at the tightest of latency settings. It is an impressive result at the lower ASIO buffer settings when dealing with sample libraries specifically and it continues to echo results seen on prior generations, but the market looks to have caught up and often far surpasses it once you start to raise the ASIO buffer setting.
The Intel Ultra 200 range on the other hand is perhaps a bit more of a surprising outcome. Whilst a lot of the public reviews focused on gaming have seen this very much as a sideways move, the testing focusing on packages like Cinebench or various other productivity applications have proven kinder in their results when the full chip is being leveraged. Here again we find that when the entire chip is allowed to flex the performance is there to be had, even with the healthy power savings obtained through the silicon update.
A slightly mixed bag but in general between the power savings and application performance, it seems that Intel is once more heading in the right direction again. With it’s latest range looking to hold a strong position in terms of application support as we close out the year, we contemplate just what may come in in 2025 and look forward to seeing any further optimizations that may originate from either side over the coming months.
Scan 3XS Custom Audio Systems
Scan Fixed Series Audio Systems